All That VRAM Needs a Bigger Brain

The day has finally come. I've been The day has finally come. I've been waiting for this for a while. Remember waiting for this for a while. Remember waiting for this for a while. Remember these these these >> flashback. >> flashback. >> flashback. >> Yeah, I'm glad I got the big case after >> Yeah, I'm glad I got the big case after >> Yeah, I'm glad I got the big case after all. Um, I'm ashamed to say they're all. Um, I'm ashamed to say they're all. Um, I'm ashamed to say they're still in a box, but they're coming out still in a box, but they're coming out still in a box, but they're coming out today. I've been feeling a bit of FOMO today. I've been feeling a bit of FOMO today. I've been feeling a bit of FOMO lately cuz I've been seeing a bunch of lately cuz I've been seeing a bunch of lately cuz I've been seeing a bunch of people with their Quad Pro 6000s posting people with their Quad Pro 6000s posting people with their Quad Pro 6000s posting all [music] kinds of X messages. all [music] kinds of X messages. all [music] kinds of X messages. Ah, finally I wield the power. to remove Ah, finally I wield the power. to remove Ah, finally I wield the power. to remove film before use. Okay, we'll do that. film before use. Okay, we'll do that. film before use. Okay, we'll do that. This is an RTX Pro 6000. This is the Max This is an RTX Pro 6000. This is the Max This is an RTX Pro 6000. This is the Max Q edition which runs at 300 watts Q edition which runs at 300 watts Q edition which runs at 300 watts maximum. And I also have two workstation maximum. And I also have two workstation maximum. And I also have two workstation editions. I'm going to have to pull them editions. I'm going to have to pull them editions. I'm going to have to pull them out of my previous build to do this one out of my previous build to do this one out of my previous build to do this one today. But it's going to be worth it today. But it's going to be worth it today. But it's going to be worth it because ASUS sent me this ET900 AX9 case because ASUS sent me this ET900 AX9 case because ASUS sent me this ET900 AX9 case with the ProWRX90E with the ProWRX90E with the ProWRX90E Sage motherboard already pre-installed. Sage motherboard already pre-installed. Sage motherboard already pre-installed. Now, this motherboard is no ordinary Now, this motherboard is no ordinary Now, this motherboard is no ordinary motherboard. it takes [music] a Thread motherboard. it takes [music] a Thread motherboard. it takes [music] a Thread Ripper Pro. So, yeah, now I kind of have Ripper Pro. So, yeah, now I kind of have Ripper Pro. So, yeah, now I kind of have to go and get one of those. There's a to go and get one of those. There's a to go and get one of those. There's a really good reason for it because the really good reason for it because the really good reason for it because the way we use LLM has completely changed in way we use LLM has completely changed in way we use LLM has completely changed in the last year since I built that the last year since I built that the last year since I built that machine. [music] So, that's why I need machine. [music] So, that's why I need machine. [music] So, that's why I need to go shopping. This is the home of my to go shopping. This is the home of my to go shopping. This is the home of my future processor and this is the Fairfax future processor and this is the Fairfax future processor and this is the Fairfax MicroEnter. I guess I'm kind of lucky MicroEnter. I guess I'm kind of lucky MicroEnter. I guess I'm kind of lucky because I have three micro centers near because I have three micro centers near because I have three micro centers near me within an hour's drive. Uh, but me within an hour's drive. Uh, but me within an hour's drive. Uh, but they're opening up new locations all the they're opening up new locations all the they're opening up new locations all the time, so you probably are going to get time, so you probably are going to get time, so you probably are going to get one soon, too. Maybe. Don't quote me on one soon, too. Maybe. Don't quote me on one soon, too. Maybe. Don't quote me on that.

[music] [music] Okay, the Fairfax mother wall is pretty Okay, the Fairfax mother wall is pretty Okay, the Fairfax mother wall is pretty impressive, too. impressive, too. impressive, too. This place is dangerous. The problem is This place is dangerous. The problem is This place is dangerous. The problem is there's six different Thread Ripper there's six different Thread Ripper there's six different Thread Ripper [music] Processors, ranging from [music] Processors, ranging from [music] Processors, ranging from expensive to financially irresponsible. This is Alan. Uh he works at MicroEnter This is Alan. Uh he works at MicroEnter in Fairfax. You can come see him. in Fairfax. You can come see him. in Fairfax. You can come see him. >> Yeah, come find me. >> Yeah, come find me. >> Yeah, come find me. >> We're in front of the Microsoft sign. >> We're in front of the Microsoft sign. >> We're in front of the Microsoft sign. But uh these will run Windows fine, But uh these will run Windows fine, But uh these will run Windows fine, right? right? right? >> Yeah. Maybe some Linux if you want. >> Yeah. Maybe some Linux if you want. >> Yeah. Maybe some Linux if you want. >> I want some Linux today because this is >> I want some Linux today because this is >> I want some Linux today because this is going to be in an AI [music] machine. So going to be in an AI [music] machine. So going to be in an AI [music] machine. So these are all the nonpro CPU. Orange these are all the nonpro CPU. Orange these are all the nonpro CPU. Orange box, nonpro, gray box, pro. Another way, box, nonpro, gray box, pro. Another way, box, nonpro, gray box, pro. Another way, if you're looking at just the models, if you're looking at just the models, if you're looking at just the models, the nonpros end in an X and the Pros end the nonpros end in an X and the Pros end the nonpros end in an X and the Pros end in a WX. You get that extra W for the in a WX. You get that extra W for the in a WX. You get that extra W for the win. Right now, we want to go for Pro win. Right now, we want to go for Pro win. Right now, we want to go for Pro because we want to get all those extra because we want to get all those extra because we want to get all those extra lanes for that Pro motherboard that I lanes for that Pro motherboard that I lanes for that Pro motherboard that I have have have >> and 2 TB of RAM. >> and 2 TB of RAM. >> and 2 TB of RAM. >> 2 TB of RAM. Now, I don't have the >> 2 TB of RAM. Now, I don't have the >> 2 TB of RAM. Now, I don't have the millions of dollars that it's going to millions of dollars that it's going to millions of dollars that it's going to cost me for that kind of RAM, but they cost me for that kind of RAM, but they cost me for that kind of RAM, but they have it here in the store. I just have it here in the store. I just have it here in the store. I just checked it out. And they have the ECC checked it out. And they have the ECC checked it out. And they have the ECC memory, too, right?

memory, too, right? memory, too, right? >> We have some. Yeah, >> We have some. Yeah, >> We have some. Yeah, >> that's the kind of memory you're going >> that's the kind of memory you're going >> that's the kind of memory you're going to need for this kind of system. I have to need for this kind of system. I have to need for this kind of system. I have some RAM already, so we're not going to some RAM already, so we're not going to some RAM already, so we're not going to do that. But how many threads can this do that. But how many threads can this do that. But how many threads can this one rip? one rip? one rip? >> This is the 16 core. After that, you go >> This is the 16 core. After that, you go >> This is the 16 core. After that, you go 24, 32, and 64. There is also a 96 core 24, 32, and 64. There is also a 96 core 24, 32, and 64. There is also a 96 core version, but we don't have that right version, but we don't have that right version, but we don't have that right now. now. now. >> That's the 9995, right? >> That's the 9995, right? >> That's the 9995, right? >> Correct. [music] >> Correct. [music] >> Correct. [music] >> That's the big boy. Now, here we got >> That's the big boy. Now, here we got >> That's the big boy. Now, here we got six. No, now I'm confused. 16 cores, 32 six. No, now I'm confused. 16 cores, 32 six. No, now I'm confused. 16 cores, 32 threads. This is a good entry point for threads. This is a good entry point for threads. This is a good entry point for a lot of people that are going to be AI a lot of people that are going to be AI a lot of people that are going to be AI enthusiasts. Two GPUs, pretty [music] enthusiasts. Two GPUs, pretty [music] enthusiasts. Two GPUs, pretty [music] good. No problem. Now, originally, I was good. No problem. Now, originally, I was good. No problem. Now, originally, I was thinking about the 24 core one. It's thinking about the 24 core one. It's thinking about the 24 core one. It's better for several GPUs doing data better for several GPUs doing data better for several GPUs doing data loading and pre-processing. But the 32 loading and pre-processing. But the 32 loading and pre-processing. But the 32 core, that's the sweet spot for multiple core, that's the sweet spot for multiple core, that's the sweet spot for multiple GPUs. I can put eight GPUs on there, no GPUs. I can put eight GPUs on there, no GPUs. I can put eight GPUs on there, no [music] problem. For 4 grand, do I get [music] problem. For 4 grand, do I get [music] problem. For 4 grand, do I get the threads and the ripper together? the threads and the ripper together? the threads and the ripper together? >> They do come in one package. It's a >> They do come in one package. It's a >> They do come in one package. It's a really small box, but it packs a lot. really small box, but it packs a lot. really small box, but it packs a lot. >> And that's the chip right there. That's >> And that's the chip right there. That's >> And that's the chip right there. That's what has to fit inside the motherboard. what has to fit inside the motherboard. what has to fit inside the motherboard. Then we [music] jump to the 64 cores and Then we [music] jump to the 64 cores and Then we [music] jump to the 64 cores and 128 threads. That won't make one model 128 threads. That won't make one model 128 threads. That won't make one model generate tokens twice as fast, but it generate tokens twice as fast, but it generate tokens twice as fast, but it can handle much heavier data processing, can handle much heavier data processing, can handle much heavier data processing, model conversion, [music] model conversion, [music] model conversion, [music] virtual machines, and multiple virtual machines, and multiple virtual machines, and multiple simultaneous users. At what point does simultaneous users. At what point does simultaneous users. At what point does it [laughter] at what at what point does it [laughter] at what at what point does it [laughter] at what at what point does it stop being a thread [music] ripper it stop being a thread [music] ripper it stop being a thread [music] ripper and become a thread annihilator?

and become a thread annihilator? and become a thread annihilator? >> Definitely that 96, >> Definitely that 96, >> Definitely that 96, >> which I'm not going [music] to buy >> which I'm not going [music] to buy >> which I'm not going [music] to buy anyway because it's way too expensive. anyway because it's way too expensive. anyway because it's way too expensive. All right, Chad GPT. I need another All right, Chad GPT. I need another All right, Chad GPT. I need another silly joke uh to tell Allan about these silly joke uh to tell Allan about these silly joke uh to tell Allan about these thread rippers. I want him to like thread rippers. I want him to like thread rippers. I want him to like really react like he's being played. really react like he's being played. really react like he's being played. >> See what it comes up with. >> See what it comes up with. >> See what it comes up with. >> A few moments later. >> A few moments later. >> A few moments later. >> That's stupid. >> That's stupid. >> That's stupid. Now, all these have the same 128 PCIe 5 Now, all these have the same 128 PCIe 5 Now, all these have the same 128 PCIe 5 lanes, but the bigger chips don't unlock lanes, but the bigger chips don't unlock lanes, but the bigger chips don't unlock more GPUs. They just give GPUs more CPU more GPUs. They just give GPUs more CPU more GPUs. They just give GPUs more CPU workers, so they spend less time workers, so they spend less time workers, so they spend less time waiting. So, yeah, that big boy 9995WX waiting. So, yeah, that big boy 9995WX waiting. So, yeah, that big boy 9995WX is going to give you all the lanes you is going to give you all the lanes you is going to give you all the lanes you need and pretty much no bottlenecks with need and pretty much no bottlenecks with need and pretty much no bottlenecks with eight GPUs or even more. In fact, I know eight GPUs or even more. In fact, I know eight GPUs or even more. In fact, I know somebody's hooked up 16 Pro 6000s into somebody's hooked up 16 Pro 6000s into somebody's hooked up 16 Pro 6000s into one. Hear that, PewDiePie? That's more one. Hear that, PewDiePie? That's more one. Hear that, PewDiePie? That's more than you. Whichever CPU I choose, it than you. Whichever CPU I choose, it than you. Whichever CPU I choose, it does not include a cooler. And these does not include a cooler. And these does not include a cooler. And these chips can pull up to 350 watts, so I chips can pull up to 350 watts, so I chips can pull up to 350 watts, so I can't exactly strap on a normal desktop can't exactly strap on a normal desktop can't exactly strap on a normal desktop cooler onto it and hope for the best. I cooler onto it and hope for the best. I cooler onto it and hope for the best. I need something that's made for this need something that's made for this need something that's made for this Thread Ripper. Thread Ripper. Thread Ripper. >> This is the Silverstone one. You said >> This is the Silverstone one. You said >> This is the Silverstone one. You said you specifically care about noise. This you specifically care about noise. This you specifically care about noise. This is one of the louder ones, but it's cuz is one of the louder ones, but it's cuz is one of the louder ones, but it's cuz it has really high performance.

it has really high performance. it has really high performance. >> I like the high performance aspect >> I like the high performance aspect >> I like the high performance aspect >> cuz the fans go to like 2800 RPM. Okay. >> cuz the fans go to like 2800 RPM. Okay. >> cuz the fans go to like 2800 RPM. Okay. >> So, basically, it's never going to >> So, basically, it's never going to >> So, basically, it's never going to overheat, but it will get kind of loud overheat, but it will get kind of loud overheat, but it will get kind of loud under load. Where you putting the under load. Where you putting the under load. Where you putting the system? system? system? >> It's going to be on my desk. You could >> It's going to be on my desk. You could >> It's going to be on my desk. You could do the bits power then or the thermal do the bits power then or the thermal do the bits power then or the thermal take if you can fit a 420 radiator. Um take if you can fit a 420 radiator. Um take if you can fit a 420 radiator. Um they'll be a little quieter for like they'll be a little quieter for like they'll be a little quieter for like operating right next to you, but they're operating right next to you, but they're operating right next to you, but they're going to perform a little bit worse. going to perform a little bit worse. going to perform a little bit worse. They should still be fine, but They should still be fine, but They should still be fine, but >> Okay. All right. You know what? Let's go >> Okay. All right. You know what? Let's go >> Okay. All right. You know what? Let's go with Silverstone. It's a pretty with Silverstone. It's a pretty with Silverstone. It's a pretty well-known name. Even though the other well-known name. Even though the other well-known name. Even though the other brands I'm sure are going to do just as brands I'm sure are going to do just as brands I'm sure are going to do just as well, but I've already had a Silverstone well, but I've already had a Silverstone well, but I've already had a Silverstone one in my other build, so I trust it. one in my other build, so I trust it. one in my other build, so I trust it. >> Okay. >> Okay. >> Okay. >> Thank you very much. >> Thank you very much. >> Thank you very much. >> Of course, >> Of course, >> Of course, >> Dan. Welcome. We are building this thing >> Dan. Welcome. We are building this thing >> Dan. Welcome. We are building this thing today, which is the AMD Thread Ripper today, which is the AMD Thread Ripper today, which is the AMD Thread Ripper machine. Spoiler alert. I'm terrified of machine. Spoiler alert. I'm terrified of machine. Spoiler alert. I'm terrified of these things. these things. these things. >> I will be putting in that because Dan is >> I will be putting in that because Dan is >> I will be putting in that because Dan is afraid of bending some pins. I'm not afraid of bending some pins. I'm not afraid of bending some pins. I'm not afraid. I'm a little bit afraid. afraid. I'm a little bit afraid. afraid. I'm a little bit afraid. >> Wow, Alex, I didn't know you had an oil >> Wow, Alex, I didn't know you had an oil >> Wow, Alex, I didn't know you had an oil well or something. Probably move the ram well or something. Probably move the ram well or something. Probably move the ram out of the way. out of the way. out of the way. >> Hey, be careful with that. >> Hey, be careful with that. >> Hey, be careful with that. >> Yeah. Yeah. Yeah. My fault. [music] My >> Yeah. Yeah. Yeah. My fault. [music] My >> Yeah. Yeah. Yeah. My fault. [music] My fault. fault. fault. >> So, this is the W. >> So, this is the W. >> So, this is the W. It's like W uh R.

It's like W uh R. It's like W uh R. >> God, these thread ripper names. Okay, >> God, these thread ripper names. Okay, >> God, these thread ripper names. Okay, this is the '90s Sage. This is the Pro this is the '90s Sage. This is the Pro this is the '90s Sage. This is the Pro version. This has seven PCIe 5 slots in version. This has seven PCIe 5 slots in version. This has seven PCIe 5 slots in it. You can put a lot of GPU in this it. You can put a lot of GPU in this it. You can put a lot of GPU in this thing. We're going to put in 128 gigs of thing. We're going to put in 128 gigs of thing. We're going to put in 128 gigs of memory. We're going to put in an AMD memory. We're going to put in an AMD memory. We're going to put in an AMD Thread Ripper 9975, which is a nice Thread Ripper 9975, which is a nice Thread Ripper 9975, which is a nice middle of the road chip. I mean, it's middle of the road chip. I mean, it's middle of the road chip. I mean, it's it's pricey. it's pricey. it's pricey. >> Yeah. $4,000 is middle of the road. >> Yeah. $4,000 is middle of the road. >> Yeah. $4,000 is middle of the road. >> Yeah, it is. It is middle of the road. >> Yeah, it is. It is middle of the road. >> Yeah, it is. It is middle of the road. Oh, yeah. This is the most expensive CPU Oh, yeah. This is the most expensive CPU Oh, yeah. This is the most expensive CPU per square inch that I've ever held. Not per square inch that I've ever held. Not per square inch that I've ever held. Not even per square inch. Just period. They even per square inch. Just period. They even per square inch. Just period. They give you give you give you >> the wrench and this. I don't know what >> the wrench and this. I don't know what >> the wrench and this. I don't know what that's for. I don't think we need that. that's for. I don't think we need that. that's for. I don't think we need that. >> We'll see. We'll read the manual when it >> We'll see. We'll read the manual when it >> We'll see. We'll read the manual when it comes to the cooler. comes to the cooler. comes to the cooler. >> I'm not reading that. You want to? >> I'm not reading that. You want to? >> I'm not reading that. You want to? >> Yeah, I'll read this. Not out loud, >> Yeah, I'll read this. Not out loud, >> Yeah, I'll read this. Not out loud, though. though. though. >> It's a liquid cooler bracket. We might >> It's a liquid cooler bracket. We might >> It's a liquid cooler bracket. We might need it. Did you find the instructions? need it. Did you find the instructions? need it. Did you find the instructions? >> Uh, yeah. I mean, it's right here. So, >> Uh, yeah. I mean, it's right here. So, >> Uh, yeah. I mean, it's right here. So, >> wait. Should I do without the >> wait. Should I do without the >> wait. Should I do without the instruction? instruction? instruction? >> I can do it if you want. >> I can do it if you want. >> I can do it if you want. >> No, no, no, no, no, no, no, no. >> No, no, no, no, no, no, no, no. >> No, no, no, no, no, no, no, no. >> You sure? >> You sure? >> You sure? >> Uh, >> Uh, >> Uh, no. Yeah. no. Yeah. no. Yeah. >> So, what it says here is you loosen up >> So, what it says here is you loosen up >> So, what it says here is you loosen up the bracket there with this.

the bracket there with this. the bracket there with this. >> Oh, first you loo >> Oh, first you loo >> Oh, first you loo >> Yes. >> Yes. >> Yes. >> Okay. It pops up. >> Okay. It pops up. >> Okay. It pops up. >> And then this is >> And then this is >> And then this is >> And then we slide in. It goes right >> And then we slide in. It goes right >> And then we slide in. It goes right there. there. there. >> Look at that. Look at that. >> Look at that. Look at that. >> Look at that. Look at that. >> Pro. And we got to take this thing out. >> Pro. And we got to take this thing out. >> Pro. And we got to take this thing out. >> Yes. >> Yes. >> Yes. >> Oh, look at all those pins. >> Oh, look at all those pins. >> Oh, look at all those pins. >> This is like if you're scared of heights >> This is like if you're scared of heights >> This is like if you're scared of heights and you're like on like a ladder or and you're like on like a ladder or and you're like on like a ladder or something and you have to look down. So, something and you have to look down. So, something and you have to look down. So, I'm just not even going to look at it. I'm just not even going to look at it. I'm just not even going to look at it. So, that's going to click into place. So, that's going to click into place. So, that's going to click into place. >> Click. [music] >> Click. [music] >> Click. [music] >> There it is. [laughter] >> There it is. [laughter] >> There it is. [laughter] >> That is the most terrifying like click >> That is the most terrifying like click >> That is the most terrifying like click I've I've I've >> now the top frame. >> now the top frame. >> now the top frame. >> Yes. And now you tighten it. >> Yes. And now you tighten it. >> Yes. And now you tighten it. >> And now we tighten it. closes 1, two, >> And now we tighten it. closes 1, two, >> And now we tighten it. closes 1, two, three, just the opposite way. three, just the opposite way. three, just the opposite way. >> And that's why you read the manual. >> And that's why you read the manual. >> And that's why you read the manual. >> I mean, I'm not arguing with that. This >> I mean, I'm not arguing with that. This >> I mean, I'm not arguing with that. This manual came in pretty handy this time. manual came in pretty handy this time. manual came in pretty handy this time. >> Well, to be fair, this is the first time >> Well, to be fair, this is the first time >> Well, to be fair, this is the first time we've had a manual since we've had to do we've had a manual since we've had to do we've had a manual since we've had to do anything in this space. anything in this space. anything in this space. >> What's next? RAM. >> What's next? RAM. >> What's next? RAM. >> Yeah, I mean, we can do the RAM. >> Yeah, I mean, we can do the RAM. >> Yeah, I mean, we can do the RAM. >> Now, the thing about RAM in this >> Now, the thing about RAM in this >> Now, the thing about RAM in this particular case or this particular particular case or this particular particular case or this particular motherboard is that since I don't have motherboard is that since I don't have motherboard is that since I don't have all eight channels, I I just couldn't all eight channels, I I just couldn't all eight channels, I I just couldn't get any. I don't know what's going on get any. I don't know what's going on get any. I don't know what's going on with that, Dan. Why couldn't I get RAM? with that, Dan. Why couldn't I get RAM? with that, Dan. Why couldn't I get RAM? I have no idea what could possibly cause I have no idea what could possibly cause I have no idea what could possibly cause be causing this sort of like issue with be causing this sort of like issue with be causing this sort of like issue with RAM.

RAM. RAM. >> We only have four to put in here and I >> We only have four to put in here and I >> We only have four to put in here and I had to rip this out from my other build had to rip this out from my other build had to rip this out from my other build unfortunately. So, we have to put it unfortunately. So, we have to put it unfortunately. So, we have to put it very carefully into certain slots. We very carefully into certain slots. We very carefully into certain slots. We have to put it in have to put it in have to put it in somebody installing a thread ripper over somebody installing a thread ripper over somebody installing a thread ripper over there. [laughter] there. [laughter] there. [laughter] I hear clicks. I hear clicks. I hear clicks. >> I think the coffee machine Wait, what? >> I think the coffee machine Wait, what? >> I think the coffee machine Wait, what? >> Yeah, the coffee machine is cleaning >> Yeah, the coffee machine is cleaning >> Yeah, the coffee machine is cleaning itself. itself. itself. >> Oh, okay. >> Oh, okay. >> Oh, okay. >> I think we got to put it into one. B1, >> I think we got to put it into one. B1, >> I think we got to put it into one. B1, A1, E1, and F1. We could try that. I A1, E1, and F1. We could try that. I A1, E1, and F1. We could try that. I mean, worst case scenario, like we can mean, worst case scenario, like we can mean, worst case scenario, like we can take the RAM out and then just keep take the RAM out and then just keep take the RAM out and then just keep putting it in. putting it in. putting it in. >> Um, >> Um, >> Um, >> no thanks. You want to do this part? >> no thanks. You want to do this part? >> no thanks. You want to do this part? >> Yeah. >> Yeah. >> Yeah. >> Can you tell which is A1, B1? >> Can you tell which is A1, B1? >> Can you tell which is A1, B1? >> It says up here. >> It says up here. >> It says up here. A B. So, this time I am going to consult A B. So, this time I am going to consult A B. So, this time I am going to consult the manual. the manual. the manual. >> Ah, smart. ASUS actually sent me this >> Ah, smart. ASUS actually sent me this >> Ah, smart. ASUS actually sent me this case and this board. So, thank you, case and this board. So, thank you, case and this board. So, thank you, ASUS. Eventually, I hope to run all ASUS. Eventually, I hope to run all ASUS. Eventually, I hope to run all eight channels because that's just going eight channels because that's just going eight channels because that's just going to make everything faster and it's going to make everything faster and it's going to make everything faster and it's going to take the full advantage of this CPU's to take the full advantage of this CPU's to take the full advantage of this CPU's capabilities. capabilities. capabilities. >> Oh my god, guys. If it if it weren't for >> Oh my god, guys. If it if it weren't for >> Oh my god, guys. If it if it weren't for whatever is going on in the world making whatever is going on in the world making whatever is going on in the world making RAM so expensive, I would just put this RAM so expensive, I would just put this RAM so expensive, I would just put this in no problem. But right now, there we in no problem. But right now, there we in no problem. But right now, there we go.

go. go. >> Just think about it this way. It's not >> Just think about it this way. It's not >> Just think about it this way. It's not yours. yours. yours. >> Well, that's not helping. [laughter] >> Well, that's not helping. [laughter] >> Well, that's not helping. [laughter] There we go. All right. Now, where does There we go. All right. Now, where does There we go. All right. Now, where does this thing go? this thing go? this thing go? >> Oh, you know what? This might be the >> Oh, you know what? This might be the >> Oh, you know what? This might be the heat sink. This has to be the first SSD slot. I This has to be the first SSD slot. I know there's more under here. So, know there's more under here. So, know there's more under here. So, there's nothing there. Whatever. Oh, I there's nothing there. Whatever. Oh, I there's nothing there. Whatever. Oh, I was right. was right. was right. >> And for that, your prize is installing >> And for that, your prize is installing >> And for that, your prize is installing it. it. it. >> Thank you. God, am I the only person who >> Thank you. God, am I the only person who >> Thank you. God, am I the only person who like struggles when it comes to like the like struggles when it comes to like the like struggles when it comes to like the plastic film on um thermal pads for our plastic film on um thermal pads for our plastic film on um thermal pads for our SSD slots? SSD slots? SSD slots? >> Yeah. >> Yeah. >> Yeah. >> Thank you for the honesty. >> Thank you for the honesty. >> Thank you for the honesty. >> Come on, Dan. >> Come on, Dan. >> Come on, Dan. >> I'm in. I'm in. >> I'm in. I'm in. >> I'm in. I'm in. >> Get with it. We don't know yet what the >> Get with it. We don't know yet what the >> Get with it. We don't know yet what the hardest part is going to be. Is it going hardest part is going to be. Is it going hardest part is going to be. Is it going to be changing out this um PSU or to be changing out this um PSU or to be changing out this um PSU or installing the cooler? installing the cooler? installing the cooler? >> I think the cooler is going to be harder >> I think the cooler is going to be harder >> I think the cooler is going to be harder to do. And that says a lot. It's going to do. And that says a lot. It's going to do. And that says a lot. It's going to be easy cuz the power supply, it's to be easy cuz the power supply, it's to be easy cuz the power supply, it's just like four screws and then a bunch just like four screws and then a bunch just like four screws and then a bunch of cables. I'm the worst person to talk of cables. I'm the worst person to talk of cables. I'm the worst person to talk about cable management. So, like this about cable management. So, like this about cable management. So, like this thing is going to look horrible from thing is going to look horrible from thing is going to look horrible from behind. So, we're not going to show you behind. So, we're not going to show you behind. So, we're not going to show you guys that, I think. Right. guys that, I think. Right. guys that, I think. Right. >> Well, I mean, I think we're going to >> Well, I mean, I think we're going to >> Well, I mean, I think we're going to have to. [laughter] have to. [laughter] have to. [laughter] Also, we need to make sure that the Also, we need to make sure that the Also, we need to make sure that the power supply feeds all the places on the power supply feeds all the places on the power supply feeds all the places on the motherboard, including the extra slots, motherboard, including the extra slots, motherboard, including the extra slots, cuz we're putting four GPUs in here. So, cuz we're putting four GPUs in here. So, cuz we're putting four GPUs in here. So, why are we doing this? Well, this is a why are we doing this? Well, this is a why are we doing this? Well, this is a 2,000 W power supply that comes with 2,000 W power supply that comes with 2,000 W power supply that comes with this case. But, ASUS sent us another one this case. But, ASUS sent us another one this case. But, ASUS sent us another one that's 30,000 W. [groaning] that's 30,000 W. [groaning] that's 30,000 W. [groaning] This monster right here is heavier than This monster right here is heavier than This monster right here is heavier than your 5090, your 5090, your 5090, >> and we're going to need to swap it out.

>> and we're going to need to swap it out. >> and we're going to need to swap it out. And Dan says we can't use the power And Dan says we can't use the power And Dan says we can't use the power cables interchangeably. So, we can't cables interchangeably. So, we can't cables interchangeably. So, we can't just leave that in there. We need to just leave that in there. We need to just leave that in there. We need to actually swap out all the power cables actually swap out all the power cables actually swap out all the power cables that came with this supply for the power that came with this supply for the power that came with this supply for the power cables that came with that supply. cables that came with that supply. cables that came with that supply. >> Depending on how much they zip tied this >> Depending on how much they zip tied this >> Depending on how much they zip tied this that the cables might be the hardest that the cables might be the hardest that the cables might be the hardest part to take out out of anything we've part to take out out of anything we've part to take out out of anything we've done so far. I mean, so far we really done so far. I mean, so far we really done so far. I mean, so far we really haven't done anything difficult. And the haven't done anything difficult. And the haven't done anything difficult. And the reason we're doing it is because there's reason we're doing it is because there's reason we're doing it is because there's going to be four Pro 6000s in here. Two going to be four Pro 6000s in here. Two going to be four Pro 6000s in here. Two of which can draw up to 600 watts. You of which can draw up to 600 watts. You of which can draw up to 600 watts. You want some headroom in your power supply want some headroom in your power supply want some headroom in your power supply so that it doesn't go crazy. Balls to so that it doesn't go crazy. Balls to so that it doesn't go crazy. Balls to the wall all the time. the wall all the time. the wall all the time. >> Am I allowed to say that? >> Am I allowed to say that? >> Am I allowed to say that? >> I I think that's fine. >> I I think that's fine. >> I I think that's fine. >> It's a It's a terminology that comes >> It's a It's a terminology that comes >> It's a It's a terminology that comes from airplanes and pilots, by the way. from airplanes and pilots, by the way. from airplanes and pilots, by the way. >> Really? Balls to the wall. >> Really? Balls to the wall. >> Really? Balls to the wall. >> Balls to the wall because pilots have >> Balls to the wall because pilots have >> Balls to the wall because pilots have the handle. the handle. the handle. >> They call that balls for some reason. >> They call that balls for some reason. >> They call that balls for some reason. >> And when they push it all the way up, >> And when they push it all the way up, >> And when they push it all the way up, that's balls to the wall. that's balls to the wall. that's balls to the wall. >> Okay, that's cool. See, in another >> Okay, that's cool. See, in another >> Okay, that's cool. See, in another timeline, this guy's a pilot. I hope you timeline, this guy's a pilot. I hope you timeline, this guy's a pilot. I hope you I [laughter] hope you guys know that. I [laughter] hope you guys know that. I [laughter] hope you guys know that. >> Yeah. Yeah, know that you're in the >> Yeah. Yeah, know that you're in the >> Yeah. Yeah, know that you're in the right timeline, in the good timeline. right timeline, in the good timeline. right timeline, in the good timeline. >> Now, this case, believe it or not, is >> Now, this case, believe it or not, is >> Now, this case, believe it or not, is actually sibling of the DJX Station actually sibling of the DJX Station actually sibling of the DJX Station case. This is pretty simple. I like it.

case. This is pretty simple. I like it. case. This is pretty simple. I like it. It's clean. It's clean. It's clean. >> Oh, yes. Now, if we want to make it >> Oh, yes. Now, if we want to make it >> Oh, yes. Now, if we want to make it pretty, that's a different question. pretty, that's a different question. pretty, that's a different question. >> I thought you said this was going to be >> I thought you said this was going to be >> I thought you said this was going to be easy. easy. easy. >> Yeah. I mean, it is easy. Now, I just >> Yeah. I mean, it is easy. Now, I just >> Yeah. I mean, it is easy. Now, I just have to put in the new one. have to put in the new one. have to put in the new one. >> Oh my gosh. 3,000 watts. Insane. You >> Oh my gosh. 3,000 watts. Insane. You >> Oh my gosh. 3,000 watts. Insane. You hold the power in your hands. hold the power in your hands. hold the power in your hands. >> All right. >> All right. >> All right. There's our Silverstoneone XC360 TR5. There's our Silverstoneone XC360 TR5. There's our Silverstoneone XC360 TR5. >> Oh yes. >> Oh yes. >> Oh yes. >> Like this is a fan bracket. So I don't >> Like this is a fan bracket. So I don't >> Like this is a fan bracket. So I don't actually have to unscrew any of the actually have to unscrew any of the actually have to unscrew any of the stuff up here. stuff up here. stuff up here. >> So just like right here, you can just >> So just like right here, you can just >> So just like right here, you can just pop it off. pop it off. pop it off. >> Yeah. I mean I may still have to unscrew >> Yeah. I mean I may still have to unscrew >> Yeah. I mean I may still have to unscrew the screws that I'm screwing. That the screws that I'm screwing. That the screws that I'm screwing. That sounds interesting. And here is the sounds interesting. And here is the sounds interesting. And here is the cooler plate. cooler plate. cooler plate. So this thing comes with the uh thermal So this thing comes with the uh thermal So this thing comes with the uh thermal paste already in cute little squares. I paste already in cute little squares. I paste already in cute little squares. I don't know how you guys like to apply don't know how you guys like to apply don't know how you guys like to apply your thermal paste. Maybe I've said this your thermal paste. Maybe I've said this your thermal paste. Maybe I've said this in a video before, but I like to make a in a video before, but I like to make a in a video before, but I like to make a smiley face and then give it freckles. smiley face and then give it freckles. smiley face and then give it freckles. >> Yeah, we've seen how you do that. >> Yeah, we've seen how you do that. >> Yeah, we've seen how you do that. >> You guys liked it, right? I was pretty >> You guys liked it, right? I was pretty >> You guys liked it, right? I was pretty happy with it. Where's the manual? happy with it. Where's the manual? happy with it. Where's the manual? >> Here. >> Here. >> Here. >> All right. >> All right. >> All right. >> We don't need no stinky manual. >> We don't need no stinky manual. >> We don't need no stinky manual. >> Oh, go right on top. This goes right on >> Oh, go right on top. This goes right on >> Oh, go right on top. This goes right on top. I'm telling you, I got just the fan top. I'm telling you, I got just the fan top. I'm telling you, I got just the fan to put on the bottom there.

to put on the bottom there. to put on the bottom there. >> Oh, perfect. A cooler master fan. >> Oh, perfect. A cooler master fan. >> Oh, perfect. A cooler master fan. >> This is a brand new fan. And this >> This is a brand new fan. And this >> This is a brand new fan. And this [music] thing is supposed to be pretty [music] thing is supposed to be pretty [music] thing is supposed to be pretty quiet. 27 dB. All right. And I was quiet. 27 dB. All right. And I was quiet. 27 dB. All right. And I was saving this for another build, but maybe saving this for another build, but maybe saving this for another build, but maybe we can do this now. One is an ARGB fan, we can do this now. One is an ARGB fan, we can do this now. One is an ARGB fan, the M120 ARGB, and the other one is the M120 ARGB, and the other one is the M120 ARGB, and the other one is A120, but it has a cool metal grill on A120, but it has a cool metal grill on A120, but it has a cool metal grill on it. Which one should we do? it. Which one should we do? it. Which one should we do? >> I think we do know RGB. >> I think we do know RGB. >> I think we do know RGB. >> Come on. This is a Thread Rub build. It >> Come on. This is a Thread Rub build. It >> Come on. This is a Thread Rub build. It needs an RGB. needs an RGB. needs an RGB. >> But they don't have a connector for it. >> But they don't have a connector for it. >> But they don't have a connector for it. >> All right, you got to feel this thing. >> All right, you got to feel this thing. >> All right, you got to feel this thing. >> Oh my god. [laughter] >> Oh my god. [laughter] >> Oh my god. [laughter] >> It's not what you expected, right? It's >> It's not what you expected, right? It's >> It's not what you expected, right? It's all metal. all metal. all metal. >> That is insanely cool to feel. Hey, >> That is insanely cool to feel. Hey, >> That is insanely cool to feel. Hey, Cooler Master, can you sell me these, Cooler Master, can you sell me these, Cooler Master, can you sell me these, please? Now, this is a kind of fan that please? Now, this is a kind of fan that please? Now, this is a kind of fan that I would not want to stick my finger I would not want to stick my finger I would not want to stick my finger into. into. into. >> No. Yeah, it's made of metal. >> No. Yeah, it's made of metal. >> No. Yeah, it's made of metal. >> It's metal. >> Nice. >> Nice. >> It's nice coffee, Dan. I'm glad. I'm >> It's nice coffee, Dan. I'm glad. I'm >> It's nice coffee, Dan. I'm glad. I'm glad. If this thing doesn't turn on, glad. If this thing doesn't turn on, glad. If this thing doesn't turn on, though, I think you will have to do some though, I think you will have to do some though, I think you will have to do some push-ups. push-ups. push-ups. >> Good thing I don't have to do push-ups >> Good thing I don't have to do push-ups >> Good thing I don't have to do push-ups because this will turn on and it will because this will turn on and it will because this will turn on and it will work flawlessly, guaranteed. Or your work flawlessly, guaranteed. Or your work flawlessly, guaranteed. Or your money back. There we go. One card in.

money back. There we go. One card in. money back. There we go. One card in. >> Oo, nice. You sure you don't want me to >> Oo, nice. You sure you don't want me to >> Oo, nice. You sure you don't want me to just plug in the other ones? It's not just plug in the other ones? It's not just plug in the other ones? It's not going to be going to be going to be >> sure. Let's just do the other one. Yeah. >> sure. Let's just do the other one. Yeah. >> sure. Let's just do the other one. Yeah. I want to spread out my risk factor I want to spread out my risk factor I want to spread out my risk factor here. here. here. >> Yes, it lives. >> Yes, it lives. >> Yes, it lives. >> Dan, what's up with all this? >> Dan, what's up with all this? >> Dan, what's up with all this? >> I uh Look, man. I warned you from the >> I uh Look, man. I warned you from the >> I uh Look, man. I warned you from the very beginning about my cable very beginning about my cable very beginning about my cable management. Oh my god. Don't show them management. Oh my god. Don't show them management. Oh my god. Don't show them the front, please. the front, please. the front, please. >> All right, we're looking at the front. >> All right, we're looking at the front. >> All right, we're looking at the front. Uh Uh Uh >> what's going on here? >> what's going on here? >> what's going on here? >> Uh I'm fixing it. I'm fixing it. >> Uh I'm fixing it. I'm fixing it. >> Uh I'm fixing it. I'm fixing it. Okay, I can explain. Okay, I can explain. Okay, I can explain. This thing is not long enough to [music] This thing is not long enough to [music] This thing is not long enough to [music] just comfortably go in the way we want just comfortably go in the way we want just comfortably go in the way we want it to, you know. it to, you know. it to, you know. >> So, >> So, >> So, >> don't make the joke. [laughter] >> don't make the joke. [laughter] >> don't make the joke. [laughter] >> Nobody's making any jokes >> Nobody's making any jokes >> Nobody's making any jokes >> now. It looks good. >> now. It looks good. >> now. It looks good. >> This is a joke-free zone. >> This is a joke-free zone. >> This is a joke-free zone. >> This Yeah, we don't we don't joke around >> This Yeah, we don't we don't joke around >> This Yeah, we don't we don't joke around here. Everything is super serious. here. Everything is super serious. here. Everything is super serious. [music] There we go. That's it. Looks [music] There we go. That's it. Looks [music] There we go. That's it. Looks good. good. good. >> There we go. >> There we go. >> There we go. >> This thing got so much heavier now. Oh >> This thing got so much heavier now. Oh >> This thing got so much heavier now. Oh my god. my god. my god. >> For this one, we need special power. >> For this one, we need special power. >> For this one, we need special power. This is um 240 volts. Yeah, because this This is um 240 volts. Yeah, because this This is um 240 volts. Yeah, because this power supply requires it. So, if you power supply requires it. So, if you power supply requires it. So, if you live anywhere else in the world, you're live anywhere else in the world, you're live anywhere else in the world, you're okay. But if you live in United States, okay. But if you live in United States, okay. But if you live in United States, well, you need a little mini. You need well, you need a little mini. You need well, you need a little mini. You need to call your electrician to get you one to call your electrician to get you one to call your electrician to get you one of these. That's because regular US of these. That's because regular US of these. That's because regular US plugs are 115 volts. I'm going to kick plugs are 115 volts. I'm going to kick plugs are 115 volts. I'm going to kick things off with just two Pro 6000s.

things off with just two Pro 6000s. things off with just two Pro 6000s. Even that makes me nervous already. Uh, Even that makes me nervous already. Uh, Even that makes me nervous already. Uh, and then I'll add two more later, and then I'll add two more later, and then I'll add two more later, hopefully, if everything goes well. All hopefully, if everything goes well. All hopefully, if everything goes well. All [music] right. Whoa, Whoa, that is quite the fan that is quite the fan that is quite the fan on the back of this thing. All right, on the back of this thing. All right, on the back of this thing. All right, the AIO is working up on top. This the AIO is working up on top. This the AIO is working up on top. This insane fan is very loud back here, but insane fan is very loud back here, but insane fan is very loud back here, but that Cooler Master fan, it's pretty that Cooler Master fan, it's pretty that Cooler Master fan, it's pretty quiet right now. It's training the quiet right now. It's training the quiet right now. It's training the memory hopefully. memory hopefully. memory hopefully. So, it's going to take a few minutes. So, it's going to take a few minutes. So, it's going to take a few minutes. Ah, that feels nice. You know, on a hot Ah, that feels nice. You know, on a hot Ah, that feels nice. You know, on a hot day, this is going to be great. A little day, this is going to be great. A little day, this is going to be great. A little bit of an aside, I had to update the bit of an aside, I had to update the bit of an aside, I had to update the BIOS by flashing it. There's a little BIOS by flashing it. There's a little BIOS by flashing it. There's a little flash utility back here. Just download flash utility back here. Just download flash utility back here. Just download the file, put it on the flash drive, the file, put it on the flash drive, the file, put it on the flash drive, FAT32 format, push that button. After FAT32 format, push that button. After FAT32 format, push that button. After that, it rebooted. And listen, it's that, it rebooted. And listen, it's that, it rebooted. And listen, it's quiet. Press yes to reset FTPM. Since quiet. Press yes to reset FTPM. Since quiet. Press yes to reset FTPM. Since this is brand new, I'm going to go ahead this is brand new, I'm going to go ahead this is brand new, I'm going to go ahead and do that. Press yes. Boom. And there and do that. Press yes. Boom. And there and do that. Press yes. Boom. And there it goes. By the way, this kind of it goes. By the way, this kind of it goes. By the way, this kind of keyboard, the old school type wired keyboard, the old school type wired keyboard, the old school type wired function keys, they're a must. Whenever function keys, they're a must. Whenever function keys, they're a must. Whenever I'm installing a new system, this baby I'm installing a new system, this baby I'm installing a new system, this baby comes along with me. That's the right comes along with me. That's the right comes along with me. That's the right BIOS version that I just installed. We BIOS version that I just installed. We BIOS version that I just installed. We want to confirm we have the total memory want to confirm we have the total memory want to confirm we have the total memory correct. This is DDR5 128 GB. Pretty correct. This is DDR5 128 GB. Pretty correct. This is DDR5 128 GB. Pretty decent. I think we can do better. not decent. I think we can do better. not decent. I think we can do better. not speedwise, but uh more memory under speedwise, but uh more memory under speedwise, but uh more memory under advanced NVME configuration and it's

advanced NVME configuration and it's advanced NVME configuration and it's detecting the correct size for that detecting the correct size for that detecting the correct size for that drive we installed. So, that's good. drive we installed. So, that's good. drive we installed. So, that's good. Everything looks good there. And then Everything looks good there. And then Everything looks good there. And then PCI subsystem settings. We want to make PCI subsystem settings. We want to make PCI subsystem settings. We want to make sure that resize bar support is enabled sure that resize bar support is enabled sure that resize bar support is enabled because we're going to have multiple because we're going to have multiple because we're going to have multiple GPUs in there. Good to go. Let's install GPUs in there. Good to go. Let's install GPUs in there. Good to go. Let's install the OS. Nice and simple for that. Uu the OS. Nice and simple for that. Uu the OS. Nice and simple for that. Uu 2404. It's a very well supported by 2404. It's a very well supported by 2404. It's a very well supported by Nvidia, by the way. I love these things. Nvidia, by the way. I love these things. Nvidia, by the way. I love these things. I'll link to this down below if you're I'll link to this down below if you're I'll link to this down below if you're curious. Oh, uh, almost forgot. Secure curious. Oh, uh, almost forgot. Secure curious. Oh, uh, almost forgot. Secure boot. I want to disable that for now. boot. I want to disable that for now. boot. I want to disable that for now. Under OS type, instead of Windows, I'm Under OS type, instead of Windows, I'm Under OS type, instead of Windows, I'm going to go to other and then under boot going to go to other and then under boot going to go to other and then under boot configuration, fast boot. Disable fast configuration, fast boot. Disable fast configuration, fast boot. Disable fast boot. F10. Save and exit. Done. Now, boot. F10. Save and exit. Done. Now, boot. F10. Save and exit. Done. Now, it's getting a little toasty. I'm going it's getting a little toasty. I'm going it's getting a little toasty. I'm going to share my trick with you. I was a to share my trick with you. I was a to share my trick with you. I was a casual Linux user, so I don't know all casual Linux user, so I don't know all casual Linux user, so I don't know all the ins and outs. But within the last the ins and outs. But within the last the ins and outs. But within the last year or so, I suddenly became a Linux year or so, I suddenly became a Linux year or so, I suddenly became a Linux guru, an expert. And that's because the guru, an expert. And that's because the guru, an expert. And that's because the first thing I do is install [music] first thing I do is install [music] first thing I do is install [music] cloud code. Not sponsored by them, but cloud code. Not sponsored by them, but cloud code. Not sponsored by them, but you can install any agent that'll know you can install any agent that'll know you can install any agent that'll know all the commands. So, you [music] just all the commands. So, you [music] just all the commands. So, you [music] just speak English. So, when I need to speak English. So, when I need to speak English. So, when I need to upgrade my drivers, all I do is say upgrade my drivers, all I do is say upgrade my drivers, all I do is say upgrade the drivers. Pro tip. Now, let's upgrade the drivers. Pro tip. Now, let's upgrade the drivers. Pro tip. Now, let's get the rest of the GPUs up in this get the rest of the GPUs up in this get the rest of the GPUs up in this thing. Huh, that's satisfying. Now, thing. Huh, that's satisfying. Now, thing. Huh, that's satisfying. Now, [music] I know a lot of you are going to [music] I know a lot of you are going to [music] I know a lot of you are going to yell at me for one reason or another, yell at me for one reason or another, yell at me for one reason or another, but I'm going to install [music] but I'm going to install [music] but I'm going to install [music] GPUs in slightly weird way. noise.

GPUs in slightly weird way. noise. GPUs in slightly weird way. noise. Let's see what happens here. Don't blow Let's see what happens here. Don't blow Let's see what happens here. Don't blow up. Don't blow up. You ready? It's booting. You ready? It's booting. Nvidia SMI. Huh? Max Q Max Q Nvidia SMI. Huh? Max Q Max Q Nvidia SMI. Huh? Max Q Max Q workstation. But where is the other one? workstation. But where is the other one? workstation. But where is the other one? >> A few moments later. >> A few moments later. >> A few moments later. >> All right. Don't judge me. This is >> All right. Don't judge me. This is >> All right. Don't judge me. This is weird. I know. It's on a riser cable and weird. I know. It's on a riser cable and weird. I know. It's on a riser cable and I thought there might be an issue with I thought there might be an issue with I thought there might be an issue with the connection to the PCI slot and I was the connection to the PCI slot and I was the connection to the PCI slot and I was right because this is a two slot card right because this is a two slot card right because this is a two slot card and the bottom slot is right up against and the bottom slot is right up against and the bottom slot is right up against all those connections for the front all those connections for the front all those connections for the front panel USB and audio. So, it's not panel USB and audio. So, it's not panel USB and audio. So, it's not sitting in there quite nicely. And sure sitting in there quite nicely. And sure sitting in there quite nicely. And sure enough, Nvidia SMI now reports four enough, Nvidia SMI now reports four enough, Nvidia SMI now reports four GPUs. We did it. Except uh now I need to GPUs. We did it. Except uh now I need to GPUs. We did it. Except uh now I need to somehow find a way to get that in there. somehow find a way to get that in there. somehow find a way to get that in there. But progress. You know what I'm itching But progress. You know what I'm itching But progress. You know what I'm itching to do right now? to do right now? to do right now? Running a model. Yeah, just like this. Running a model. Yeah, just like this. Running a model. Yeah, just like this. Deepseek V4 Flash, one of the models Deepseek V4 Flash, one of the models Deepseek V4 Flash, one of the models that people are talking about right that people are talking about right that people are talking about right [music] now. It's always changing. Okay, [music] now. It's always changing. Okay, [music] now. It's always changing. Okay, but this is FP4 experts with FP8 but this is FP4 experts with FP8 but this is FP4 experts with FP8 attention. On all four cards, we're attention. On all four cards, we're attention. On all four cards, we're getting 33 tokens a second. That's with getting 33 tokens a second. That's with getting 33 tokens a second. That's with one agent. And then it goes up from one agent. And then it goes up from one agent. And then it goes up from there. 62 at concurrency of two. That there. 62 at concurrency of two. That there. 62 at concurrency of two. That means two agents at the same time.

means two agents at the same time. means two agents at the same time. Concurrency of 4, 116, all the way up to Concurrency of 4, 116, all the way up to Concurrency of 4, 116, all the way up to 16 agents at the same time with 364 16 agents at the same time with 364 16 agents at the same time with 364 tokens a second. And then it kind of tokens a second. And then it kind of tokens a second. And then it kind of drops off from there. But that's not at drops off from there. But that's not at drops off from there. But that's not at all interesting to you. Well, it might all interesting to you. Well, it might all interesting to you. Well, it might be interesting to you. We are doing this be interesting to you. We are doing this be interesting to you. We are doing this on four GPUs. We can do cool things on four GPUs. We can do cool things on four GPUs. We can do cool things fast, but that's not the most fast, but that's not the most fast, but that's not the most interesting thing. So, there's four RTX interesting thing. So, there's four RTX interesting thing. So, there's four RTX Pro 6000 Blackwell GPUs in here, and Pro 6000 Blackwell GPUs in here, and Pro 6000 Blackwell GPUs in here, and each one [music] has its own PCIe root each one [music] has its own PCIe root each one [music] has its own PCIe root complex. A root complex is basically a complex. A root complex is basically a complex. A root complex is basically a private on-ramp to the CPU. So, no card private on-ramp to the CPU. So, no card private on-ramp to the CPU. So, no card has to share a lane with another one. has to share a lane with another one. has to share a lane with another one. And that's great for running LM. But And that's great for running LM. But And that's great for running LM. But when you're using this with an agent, when you're using this with an agent, when you're using this with an agent, let me show you where all that actually let me show you where all that actually let me show you where all that actually goes. First, just a quick note about the goes. First, just a quick note about the goes. First, just a quick note about the power budget. Since two of these are Max power budget. Since two of these are Max power budget. Since two of these are Max Q cards, those two max out at 300 watts Q cards, those two max out at 300 watts Q cards, those two max out at 300 watts each. The workstation edition, these each. The workstation edition, these each. The workstation edition, these black ones on the outside, those can go black ones on the outside, those can go black ones on the outside, those can go up to 600 watts. So, all that together up to 600 watts. So, all that together up to 600 watts. So, all that together could add up to 1,800 watts and I could could add up to 1,800 watts and I could could add up to 1,800 watts and I could run it at that, but I capped the total run it at that, but I capped the total run it at that, but I capped the total at 1,200 watts. And why am I fine doing at 1,200 watts. And why am I fine doing at 1,200 watts. And why am I fine doing that? Well, here's the thing. My actual that? Well, here's the thing. My actual that? Well, here's the thing. My actual workload is not GPU processing. It's workload is not GPU processing. It's workload is not GPU processing. It's agents, coding agents. And each agent agents, coding agents. And each agent agents, coding agents. And each agent turn goes like this. This little turn goes like this. This little turn goes like this. This little timeline right here represents what each timeline right here represents what each timeline right here represents what each turn looks like. And first we have the turn looks like. And first we have the turn looks like. And first we have the LLM call. Then we have scaffolding, LLM call. Then we have scaffolding, LLM call. Then we have scaffolding, compile, run tests, analyze a that's the compile, run tests, analyze a that's the compile, run tests, analyze a that's the syntax tree, hash, and then package it syntax tree, hash, and then package it syntax tree, hash, and then package it all up. All those pretty little colors all up. All those pretty little colors all up. All those pretty little colors in there. Some of them are so small you in there. Some of them are so small you in there. Some of them are so small you can't even see it. See, everybody thinks can't even see it. See, everybody thinks can't even see it. See, everybody thinks about the model call, right? What's about the model call, right? What's about the model call, right? What's running on the GPU? But nobody really running on the GPU? But nobody really running on the GPU? But nobody really talks about the rest of that list of

talks about the rest of that list of talks about the rest of that list of things that happen when you're running things that happen when you're running things that happen when you're running an agent. Let's switch over to this one an agent. Let's switch over to this one an agent. Let's switch over to this one right here. This is more clear cuz we right here. This is more clear cuz we right here. This is more clear cuz we have a lot of that LLM call space. have a lot of that LLM call space. have a lot of that LLM call space. That's the cyan color right here. That's That's the cyan color right here. That's That's the cyan color right here. That's when it calls the GPU and the GPU does when it calls the GPU and the GPU does when it calls the GPU and the GPU does some processing. Everything after that, some processing. Everything after that, some processing. Everything after that, the uh orange and the pink and the the uh orange and the pink and the the uh orange and the pink and the purple stuff, that's the CPU, the whole purple stuff, that's the CPU, the whole purple stuff, that's the CPU, the whole CPU doing actual work. And I put this CPU doing actual work. And I put this CPU doing actual work. And I put this graphic in because this is more of an graphic in because this is more of an graphic in because this is more of an illustrative approach. What really illustrative approach. What really illustrative approach. What really happens is more something like this. happens is more something like this. happens is more something like this. Look at the little slivers of GPU work Look at the little slivers of GPU work Look at the little slivers of GPU work compared to the rest of it. 71% of each compared to the rest of it. 71% of each compared to the rest of it. 71% of each turn is actually not happening on the turn is actually not happening on the turn is actually not happening on the GPU at all. But what about that crazy GPU at all. But what about that crazy GPU at all. But what about that crazy CPU that I have in there? Do you really CPU that I have in there? Do you really CPU that I have in there? Do you really need that? Well, I decided to do a need that? Well, I decided to do a need that? Well, I decided to do a little race. little race. little race. Wa. Uh, that's going to be okay. It's Wa. Uh, that's going to be okay. It's Wa. Uh, that's going to be okay. It's just a switch. I built this little just a switch. I built this little just a switch. I built this little machine. It's one of my open benches. machine. It's one of my open benches. machine. It's one of my open benches. This has an AMD 9700X in it. Same This has an AMD 9700X in it. Same This has an AMD 9700X in it. Same benchmark, same Zen 5 architecture as a benchmark, same Zen 5 architecture as a benchmark, same Zen 5 architecture as a Thread Ripper, but this time it's only Thread Ripper, but this time it's only Thread Ripper, but this time it's only eight cores, 16 threads. One worker is eight cores, 16 threads. One worker is eight cores, 16 threads. One worker is one agent running at a time. 32 workers one agent running at a time. 32 workers one agent running at a time. 32 workers means all fighting over the same cores.

means all fighting over the same cores. means all fighting over the same cores. Calm down, boys. And this chip, it's uh Calm down, boys. And this chip, it's uh Calm down, boys. And this chip, it's uh still pretty cool, actually. This one still pretty cool, actually. This one still pretty cool, actually. This one has 64 of them to hand out. Take a look has 64 of them to hand out. Take a look has 64 of them to hand out. Take a look at the consumer CPU and one worker. at the consumer CPU and one worker. at the consumer CPU and one worker. [music] We're at 89.7 [music] We're at 89.7 [music] We're at 89.7 iterations or turns per minute versus uh iterations or turns per minute versus uh iterations or turns per minute versus uh the Thread Rippers. 85.6. the Thread Rippers. 85.6. the Thread Rippers. 85.6. That's about 5% difference. Not bad. That's about 5% difference. Not bad. That's about 5% difference. Not bad. Pretty close even for the little guy. Pretty close even for the little guy. Pretty close even for the little guy. I'd even call that a dead heat. And if I'd even call that a dead heat. And if I'd even call that a dead heat. And if that's your workload, one worker, then that's your workload, one worker, then that's your workload, one worker, then you're golden. Save your money. you're golden. Save your money. you're golden. Save your money. Seriously. But watch what happens when I Seriously. But watch what happens when I Seriously. But watch what happens when I turn the workers up. Here's two workers. turn the workers up. Here's two workers. turn the workers up. Here's two workers. Thread ripper versus consumer. We're Thread ripper versus consumer. We're Thread ripper versus consumer. We're still about the same. Let's keep going. still about the same. Let's keep going. still about the same. Let's keep going. I'm going to go to eight. Oh, now we're I'm going to go to eight. Oh, now we're I'm going to go to eight. Oh, now we're starting to see a little bit of a starting to see a little bit of a starting to see a little bit of a difference there. 11.57 seconds total difference there. 11.57 seconds total difference there. 11.57 seconds total versus the consumers 14.96. And it versus the consumers 14.96. And it versus the consumers 14.96. And it becomes even more obvious at 16 workers. becomes even more obvious at 16 workers. becomes even more obvious at 16 workers. 16.11 seconds wall clock for the Thread 16.11 seconds wall clock for the Thread 16.11 seconds wall clock for the Thread Ripper. And then double that for the Ripper. And then double that for the Ripper. And then double that for the consumer CPU. In fact, the consumer CPU consumer CPU. In fact, the consumer CPU consumer CPU. In fact, the consumer CPU pretty much stalls out at 16. If we go pretty much stalls out at 16. If we go pretty much stalls out at 16. If we go to 32 here, it stays the same. 568 is to 32 here, it stays the same. 568 is to 32 here, it stays the same. 568 is the peak. But the 9975 thread ripper, it the peak. But the 9975 thread ripper, it the peak. But the 9975 thread ripper, it keeps ripping those threads. It goes up keeps ripping those threads. It goes up keeps ripping those threads. It goes up to 2113 up here and then peaks out at 64 to 2113 up here and then peaks out at 64 to 2113 up here and then peaks out at 64 workers at 2239 workers at 2239 workers at 2239 turns per minute. That is that's almost turns per minute. That is that's almost turns per minute. That is that's almost four times 3 1/2 seconds versus 13 1/2 four times 3 1/2 seconds versus 13 1/2 four times 3 1/2 seconds versus 13 1/2 seconds. Now, let's take a look at the seconds. Now, let's take a look at the seconds. Now, let's take a look at the efficiency here at 32 workers. The 9700X efficiency here at 32 workers. The 9700X efficiency here at 32 workers. The 9700X is not even running agents at that

is not even running agents at that is not even running agents at that point. It's just thrashing 0.21 21 per point. It's just thrashing 0.21 21 per point. It's just thrashing 0.21 21 per worker against 0.74 worker against 0.74 worker against 0.74 for the Thread Ripper. Not even close. for the Thread Ripper. Not even close. for the Thread Ripper. Not even close. So that was the CPU. Now, I also ran So that was the CPU. Now, I also ran So that was the CPU. Now, I also ran this with the model call turned back on, this with the model call turned back on, this with the model call turned back on, so it's going to go to the GPU. Pass 16 so it's going to go to the GPU. Pass 16 so it's going to go to the GPU. Pass 16 workers. It all kind of falls apart workers. It all kind of falls apart workers. It all kind of falls apart here. Average model latency went from here. Average model latency went from here. Average model latency went from half a second to 12 seconds. Here's why. half a second to 12 seconds. Here's why. half a second to 12 seconds. Here's why. All 32 agents talk to one server sharing All 32 agents talk to one server sharing All 32 agents talk to one server sharing all four GPUs. The CPU work spreads all four GPUs. The CPU work spreads all four GPUs. The CPU work spreads across 64 threads. The model calls don't across 64 threads. The model calls don't across 64 threads. The model calls don't spread anywhere. They just line up. spread anywhere. They just line up. spread anywhere. They just line up. Every request waits behind 31 other Every request waits behind 31 other Every request waits behind 31 other requests. That's how half a second requests. That's how half a second requests. That's how half a second becomes 12. I don't know, maybe I'm becomes 12. I don't know, maybe I'm becomes 12. I don't know, maybe I'm doing something wrong here. Most likely, doing something wrong here. Most likely, doing something wrong here. Most likely, of course, somebody will tell me in the of course, somebody will tell me in the of course, somebody will tell me in the comments, but that's what I measured. comments, but that's what I measured. comments, but that's what I measured. Look, we chase GPUs and we have been Look, we chase GPUs and we have been Look, we chase GPUs and we have been because that's what run models runs because that's what run models runs because that's what run models runs models and that's how it's been up until models and that's how it's been up until models and that's how it's been up until this year. The big thing that happened this year. The big thing that happened this year. The big thing that happened this year is agents. GPUs are still very this year is agents. GPUs are still very this year is agents. GPUs are still very important. If you're running things important. If you're running things important. If you're running things locally, if you're running your models locally, if you're running your models locally, if you're running your models locally, that's where your money goes locally, that's where your money goes locally, that's where your money goes right now. It goes there first. But now right now. It goes there first. But now right now. It goes there first. But now there's different agentic workflows there's different agentic workflows there's different agentic workflows where you can do things locally. You can where you can do things locally. You can where you can do things locally. You can also call cloud models with your local also call cloud models with your local also call cloud models with your local agents. And in that case, the GPU does agents. And in that case, the GPU does agents. And in that case, the GPU does nothing for you. It's your CPU that nothing for you. It's your CPU that nothing for you. It's your CPU that determines how many agents and how many determines how many agents and how many determines how many agents and how many workers you could run at once. And if workers you could run at once. And if workers you could run at once. And if you're only running one at a time, an 8 you're only running one at a time, an 8 you're only running one at a time, an 8 core CPU matches a 32 core CPU. Just core CPU matches a 32 core CPU. Just core CPU matches a 32 core CPU. Just save your money. What are you running save your money. What are you running save your money. What are you running your agents on? Let me know down in the

your agents on? Let me know down in the your agents on? Let me know down in the comments below. And if you want to see comments below. And if you want to see comments below. And if you want to see these two GPUs, how I built it on a TRX these two GPUs, how I built it on a TRX these two GPUs, how I built it on a TRX board instead, that's [music] the nonpro board instead, that's [music] the nonpro board instead, that's [music] the nonpro version, watch this video right over version, watch this video right over version, watch this video right over here. Otherwise, thanks for watching and here. Otherwise, thanks for watching and here. Otherwise, thanks for watching and [music] I'll see you in the next one.