Supermicro is actually offer a system with dual GH in 1U but as a liquid cooled only option. As far as i'm aware they're offering 3 GH systems, 1GH air cooled, 1GH liquid cooled & 2GH liquid cooled and all of those in a 1U chassis
@@tonep3168to give us an option for ARM based CPU on Linux and possibly Windows. A slimmed down version of Grace Hopper might be the only thing competitive enough against Apple M chips in terms of power to performance ratio
Wow that Grace heatspreader is *way* bigger than I expected it to be. I know it is 72c but they're arm cores.. even with the interconnect I figured it would be about 30% smaller
Regarding internal layouts and rack sizes for dense servers, instead of a 1U 2Nodes system I would rather have 2U 4Nodes or even 4U 8Nodes due to more efficient fan options. You would get less noise for same fan performance the higher the unit is, I could see 1U being used with liquid cooling though.
im not a video editor or audio engineer so excuse my lack of knowledge but is it possible to "re-center" the audio? it feels a bit weird to get talked into the left ear all the time. Besides that little thing it's an awesome piece of content and I'm definitely looking forward to seeing more of grace hopper!
Nvidia Super Grace 144-core ARM CPU: - uses ARM Neoverse V2 CPU core: about 20% higher IPC than AMD Zen 4 - measured ARM Cortex X3 core in Geekbench 6 (V2 is server version of phone X3) - pure 64-bit CPU core, there is no 32-bit backward compatibility This ARM is a monster, no wonder it has 2x performance per watt than best x86 AMD Zen 4.
- uses ARM Neoverse V2 CPU core: about 20% higher IPC than AMD Zen 4 - measured ARM Cortex X3 core in Geekbench 6 (V2 is server version of phone X3) Source for the above statements? "This ARM is a monster, no wonder it has 2x performance per watt than best x86 AMD Zen 4." Are we really taking Nvidia's and ARM's claim at face value?
@@MemmoSJ Take Samsung Galaxy S23 GB6 ST result and calculate IPC as pts/GHz. Then do the same for Ryzen 7950x3d and you will get 20% higher IPC. Also do not forget ARM X3/V2 core has 6xALU + 2xBranchUnit = 8x INT scalar instructions/clock. While Zen4 has 4xALU + 1xBranchUnit = 5x INT scalar instruction/clock. So ARM X3/V2 core is 60% wider than Zen4. This means Zen4 is doing fine with such a narrow design however it's obvious the Zen4's design is inferior. And this year's Cortex X4 has 8xALU + 3xBranchUnit = 11x INT scalar instruction/clock. So next server ARM V3 core will be based on X5 in 2024 and will be even bigger beast. That's why Microsoft released their first CPU in history: 128-core ARM Cobalt CPU. MS knows that ARM is beating x86 really bad so they have to abandon x86 sinking titanic ASAP.
Why is it that CPUs (generally) use heatspreaders but GPUs don't? Even when they're mounted right next to each other and presumably share the same cold plate.
your voice over is coming only in the left channel, next time you might want to turn the audio track from your mic to a Mono track so it goes on both channels :)
Ah crap I see what might be the issue. Left channel is +6dB over right channel. That's the safety recording for peaking, and I didn't equalise in post. I personally barely notice a difference. But audio is always my achilles heel. As well as video.
@@TechTechPotato You're right, it is present on the right channel, but just quiet. I'm listening on Sennheiser HD600 headphones with an RME audio interface so it seems like quite a difference to me, probably less noticeable if you use speakers. But anyway, feed it back into the loop for improving :)
Taking the GPU out of this, what is the point of this ARM CPU? You cannot tell me that (or maybe you can?) this CPU is faster than an AMD Epyc CPU? What tangible benefits does this ARM CPU bring with it? 72 cores, 235w of power and 500GB/s memory bandwidth. I’m not seeing any kind of quantum leap here? I keep hearing fanatical cries of ARM is going to change the computing world, but nobody ever explains exactly how this is going to come about. For this to happen, the CPU would need to twice the performance per core, twice the memory bandwidth, lower latency, half the power and half the price of any x86 architecture, but not one of these things have come to pass. I get the feeling that the silence to this question kind of puts this subject in the “religious debate” area, as I think I’m correct, but want to be proven wrong. If I was stupidly wrong, I would be barely able to breathe under all the ARM aficionados facts proving me an idiot. And with this video being an advert, the silence speaks volumes.
@@christopherjackson2157 can you tell me what these advantages are? I simply cannot see it, and that might be my fault! Compare this with say an Epyc 9634 and tell me where it betters it? And you can’t tell me that having to make fully optimised ARM software is free or easier than making x86 software?
@@tonep3168 density for people building out racks. Bandwidth for feeding nvidia gpus. Software and driver efficiencies for people primarily interested in cuda. And the efficiencies and practicalities of building servers for companies like supermicro. Im not sure that someone who wasn't running nvidia hardware elsewhere in their system would see enough benefit to justify the price, however. But there are many many customers who are currently buying amd or Intel primarily as infrastructure to support their nvidia gpus. For those people, this can be very attractive as an alternative.
I think the Nvidia GH200 and Superchip are mostly for Nvidia's long-term strategy, but they also offer advantages such as a higher bandwidth. Nvidia ultimately wants to dominate all computing. For this, they need to introduce their CPUs at some point. Right now is a good time because they can ship it with their best AI GPU, and everyone will buy it. Furthermore, if they would not design their own CPU, they would practically give AMD free money, which they don't want.
As someone whose first computer was a Commodore 64, your t-shirt warms my nerdstalgic heart!
I'd be much more impressed by a ZX Spectrum t-shirt to be honest. Amiga 500, sure, but the Spectrum was clearly the better machine.
Supermicro is actually offer a system with dual GH in 1U but as a liquid cooled only option. As far as i'm aware they're offering 3 GH systems, 1GH air cooled, 1GH liquid cooled & 2GH liquid cooled and all of those in a 1U chassis
i hope they release a consumer grade cpu for us, and then we have a real Romance of Three Kingdom situation in the modern world.
Why?
@@tonep3168just my wishfull thinking. because consumer gpu realm has all three, why not in cpu realm too?
@@tonep3168to give us an option for ARM based CPU on Linux and possibly Windows. A slimmed down version of Grace Hopper might be the only thing competitive enough against Apple M chips in terms of power to performance ratio
@@genbu3834 ARM is a very different proposition for the general consumer. ARM on Windows is a disaster for a start.
Nvidia can fix windows with CUDA anytime@@tonep3168. Nothing stops them
Hopper rolls off the tongue nicely. Great job naming Nvidia
dont you mean, hops off the tongue 😂
@@lightninghell4 Off to pun jail with you!
Wow that Grace heatspreader is *way* bigger than I expected it to be. I know it is 72c but they're arm cores.. even with the interconnect I figured it would be about 30% smaller
BC Grace CPU swaps data to Hopper GPU often. Back & forth from DDR5 to HBM3.
Regarding internal layouts and rack sizes for dense servers, instead of a 1U 2Nodes system I would rather have 2U 4Nodes or even 4U 8Nodes due to more efficient fan options.
You would get less noise for same fan performance the higher the unit is, I could see 1U being used with liquid cooling though.
if your server is running fans you are doing it wrong
1:27 Epic pimpwalk in the background
did he say wood cooled?
I'm confused as well. I'm trying to think what rhymes with that.
im not a video editor or audio engineer so excuse my lack of knowledge but is it possible to "re-center" the audio? it feels a bit weird to get talked into the left ear all the time.
Besides that little thing it's an awesome piece of content and I'm definitely looking forward to seeing more of grace hopper!
Yeah check my other comment. I used a -6db safety channel in case of peaking. Didn't hear a difference myself, forgot to edit in post.
I clicked because I thought I read, "I touched the Nvidia Grasshopper."
My left ear enjoyed that video)
What NVIDIA uses in Geforce Now, in terms of CPU? If their own ARM-based chips, does that mean that we just one step away from PC gaming or ARM?
Yes, MSFT wants ARM PC laptops for windows 11 in 2024!
Grace Hopper? I'm all about Vera Rubin now! 😂
what link for Fugaku ranking page
Nvidia Super Grace 144-core ARM CPU:
- uses ARM Neoverse V2 CPU core: about 20% higher IPC than AMD Zen 4
- measured ARM Cortex X3 core in Geekbench 6 (V2 is server version of phone X3)
- pure 64-bit CPU core, there is no 32-bit backward compatibility
This ARM is a monster, no wonder it has 2x performance per watt than best x86 AMD Zen 4.
- uses ARM Neoverse V2 CPU core: about 20% higher IPC than AMD Zen 4
- measured ARM Cortex X3 core in Geekbench 6 (V2 is server version of phone X3)
Source for the above statements?
"This ARM is a monster, no wonder it has 2x performance per watt than best x86 AMD Zen 4."
Are we really taking Nvidia's and ARM's claim at face value?
@@MemmoSJ Take Samsung Galaxy S23 GB6 ST result and calculate IPC as pts/GHz. Then do the same for Ryzen 7950x3d and you will get 20% higher IPC.
Also do not forget ARM X3/V2 core has 6xALU + 2xBranchUnit = 8x INT scalar instructions/clock. While Zen4 has 4xALU + 1xBranchUnit = 5x INT scalar instruction/clock. So ARM X3/V2 core is 60% wider than Zen4. This means Zen4 is doing fine with such a narrow design however it's obvious the Zen4's design is inferior.
And this year's Cortex X4 has 8xALU + 3xBranchUnit = 11x INT scalar instruction/clock. So next server ARM V3 core will be based on X5 in 2024 and will be even bigger beast. That's why Microsoft released their first CPU in history: 128-core ARM Cobalt CPU. MS knows that ARM is beating x86 really bad so they have to abandon x86 sinking titanic ASAP.
Hope they sell some slim down version which wont broke the bank
Somewhere near apple price tag is already fine for me
Why is it that CPUs (generally) use heatspreaders but GPUs don't? Even when they're mounted right next to each other and presumably share the same cold plate.
Super Micro product names just roll right off the tongue, don't they?
We thought Intel's naming schema was bad, Super Micro puts it in another orbit !!
Ask acer :)
your voice over is coming only in the left channel, next time you might want to turn the audio track from your mic to a Mono track so it goes on both channels :)
It's coming over both channels here? On multiple systems
Ah crap I see what might be the issue. Left channel is +6dB over right channel. That's the safety recording for peaking, and I didn't equalise in post. I personally barely notice a difference. But audio is always my achilles heel. As well as video.
@@TechTechPotato hahahah so thats why i was going crazy, like im becoming even more hard of hearing in the right side lol?
@@TechTechPotato You're right, it is present on the right channel, but just quiet. I'm listening on Sennheiser HD600 headphones with an RME audio interface so it seems like quite a difference to me, probably less noticeable if you use speakers. But anyway, feed it back into the loop for improving :)
@@TechTechPotato I guess you're not wearing headphones to edit? ....Or need to see an Audiologist haha.
Did Nvidia Grace Hopper consent? 🤔
Let's gooooo, CPU cores go brrrrrr
"space for activities" 😂😂😂
Doesn't play Crisis therefore not usable in a ""Real World"" scenario 🤣🤣🤣
Grace hoppah
Taking the GPU out of this, what is the point of this ARM CPU? You cannot tell me that (or maybe you can?) this CPU is faster than an AMD Epyc CPU? What tangible benefits does this ARM CPU bring with it? 72 cores, 235w of power and 500GB/s memory bandwidth. I’m not seeing any kind of quantum leap here? I keep hearing fanatical cries of ARM is going to change the computing world, but nobody ever explains exactly how this is going to come about. For this to happen, the CPU would need to twice the performance per core, twice the memory bandwidth, lower latency, half the power and half the price of any x86 architecture, but not one of these things have come to pass.
I get the feeling that the silence to this question kind of puts this subject in the “religious debate” area, as I think I’m correct, but want to be proven wrong. If I was stupidly wrong, I would be barely able to breathe under all the ARM aficionados facts proving me an idiot.
And with this video being an advert, the silence speaks volumes.
There are a lot of advantages at the system level.
I'm not sure that the CPU on its own is any great leap forward. As you say.
@@christopherjackson2157 can you tell me what these advantages are? I simply cannot see it, and that might be my fault! Compare this with say an Epyc 9634 and tell me where it betters it? And you can’t tell me that having to make fully optimised ARM software is free or easier than making x86 software?
@@tonep3168 density for people building out racks. Bandwidth for feeding nvidia gpus. Software and driver efficiencies for people primarily interested in cuda. And the efficiencies and practicalities of building servers for companies like supermicro.
Im not sure that someone who wasn't running nvidia hardware elsewhere in their system would see enough benefit to justify the price, however.
But there are many many customers who are currently buying amd or Intel primarily as infrastructure to support their nvidia gpus. For those people, this can be very attractive as an alternative.
I think the Nvidia GH200 and Superchip are mostly for Nvidia's long-term strategy, but they also offer advantages such as a higher bandwidth.
Nvidia ultimately wants to dominate all computing. For this, they need to introduce their CPUs at some point. Right now is a good time because they can ship it with their best AI GPU, and everyone will buy it. Furthermore, if they would not design their own CPU, they would practically give AMD free money, which they don't want.
@@xthesayuri5756 which bandwidth do you refer to?