Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AMD's Instinct MI455X: Aiming for the Sun

chipsandcheese.com

Post reply on HN
Print view View original

11–20 of 38 posts

Re: AMD's Instinct MI455X: Aiming for the Sun

#11

Post by dist-epoch »

stevefan1999 wrote: post #10
dist-epoch wrote: post #4

432 GB of RAM per chip. Tens of terabytes per rack. Clients queuing up to buy them. RAM prices are not coming down any time soon.

but you have to rewrite all your software to ROCm. And ROCm, to this day, still sucks. ZLUDA basically crash on high memory demand, and only accounted for ~70% of CUDA API coverage (and it is still buggy). But hey, at least it does run on Rust-CUDA. I'm one of the few who ported it and fixed a few bugs on ZLUDA. I used it to run a simple SHA256 kernel and it ran sure, but I gave it up because of those fundamental pro…

Co-founder and Chief Compute Officer Anthropic:

> I think the thing that we were thinking about originally was whenever we're bringing up a new hardware platform, it's a big effort. It's like a huge thing. And so as we were thinking about this, we started doing our own evaluation of MI 355. You guys generously got us a rack to start working. And we expected this to be kind of a big process.

> Our actual experience was we had one engineer who start doing it. They spun up Claude, asked it, hey, bring up this machine, left it going over the weekend. And we ended up with a graph of the actual performance of our leading model on it, just going up and up and up over the weekend.

https://x.com/austinsemis/status/2080336781782753635

Re: AMD's Instinct MI455X: Aiming for the Sun

#12

Post by paoliniluis »

dist-epoch wrote: post #11
stevefan1999 wrote: post #10

Earlier quoted context omitted.

but you have to rewrite all your software to ROCm. And ROCm, to this day, still sucks. ZLUDA basically crash on high memory demand, and only accounted for ~70% of CUDA API coverage (and it is still buggy). But hey, at least it does run on Rust-CUDA. I'm one of the few who ported it and fixed a few bugs on ZLUDA. I used it to run a simple SHA256 kernel and it ran sure, but I gave it up because of those fundamental pro…

Co-founder and Chief Compute Officer Anthropic: > I think the thing that we were thinking about originally was whenever we're bringing up a new hardware platform, it's a big effort. It's like a huge thing. And so as we were thinking about this, we started doing our own evaluation of MI 355. You guys generously got us a rack to start working. And we expected this to be kind of a big process. > Our actual experience wa…

Using AI slop in software that runs ML models is like buying a Ferrari with a motorbike engine. Might be good for demos or as a proof of concept, but that can’t be used for serving LLMs at scale

Re: AMD's Instinct MI455X: Aiming for the Sun

#13

Post by Farfignoggen »

thyristan wrote: post #2

> The basic GCN microarchitecture underpinned every single one of AMD’s compute accelerators for nearly 15 years [...]. With CDNA5 AMD has moved over to a microarchitecture that is based on the RDNA series putting a bookend to the long-lived line that was the GCN microarchitecture. So everything is new again. Will this be another "buy now, have ROCM work in 3 maybe years if you are lucky"?

At least going forward, with both AMD's AI accelerators and Consumer GPUs using the same RDNA based IP, there will be less reason for AMD to drop RDNA GPUs from the ROCm/HIP support matrix! And so with both MI series accelerators and consumer GPUs/Graphics all using the same basic RDNA Micro-Architecture the ROCm/HIP code base developed for "CDNA5"/later will work for consumer RDNA as well with minor changes required! Polaris and Vega GPUs/Graphics both have been dropped from the ROCm/HIP support matrix. But with AMD's "The Rock" software stack making use of SPIR-V in the same manner as Nvidia/PTX there made be little issues getting that to work with older Vega/Polaris GPUs. even if AMD's not validated/verified that software stack with Consumer Vega/Earlier Consumer GPUs and Integrated Graphics. AMD's biggest issue has been dropping its older GPU micro-architectures from the ROCm/HIP support matrix too soon, and just look at Blender 3D's HIP Back End that requires RDNA2/Later GPUs and Graphics for any Radeon iGPU/dGPU Accelerated Blender 3D Cycles rendering support!

Re: AMD's Instinct MI455X: Aiming for the Sun

#14

Post by nolist_policy »

thyristan wrote: post #2

> The basic GCN microarchitecture underpinned every single one of AMD’s compute accelerators for nearly 15 years [...]. With CDNA5 AMD has moved over to a microarchitecture that is based on the RDNA series putting a bookend to the long-lived line that was the GCN microarchitecture. So everything is new again. Will this be another "buy now, have ROCM work in 3 maybe years if you are lucky"?

Not anymore: https://rocm.blogs.amd.com/software-tools-optimization/spir-...

Re: AMD's Instinct MI455X: Aiming for the Sun

#15

Post by nolist_policy »

stevefan1999 wrote: post #10
dist-epoch wrote: post #4

432 GB of RAM per chip. Tens of terabytes per rack. Clients queuing up to buy them. RAM prices are not coming down any time soon.

but you have to rewrite all your software to ROCm. And ROCm, to this day, still sucks. ZLUDA basically crash on high memory demand, and only accounted for ~70% of CUDA API coverage (and it is still buggy). But hey, at least it does run on Rust-CUDA. I'm one of the few who ported it and fixed a few bugs on ZLUDA. I used it to run a simple SHA256 kernel and it ran sure, but I gave it up because of those fundamental pro…

Spir-v is already happening: https://rocm.blogs.amd.com/software-tools-optimization/spir-...

Re: AMD's Instinct MI455X: Aiming for the Sun

#16

Post by ACCount37 »

fancyfredbot wrote: post #7

It's mind blowing seeing these multi exaflop single rack systems. The world's first exaflop supercomputer was Frontier. It was launched only 4 years ago in 2022. It's not a fair comparison of course. FP4 in Helios barely qualifies as floating point. Frontier was proper fp64, 16 times the bit width and probably 256x as many transistors. All the same just wow. Much compute.

Workloads did change over time. Back when we were first approaching practical exascale, the dominant workload for a supercomputer was thought to be physics simulations - and they often benefit from high numerical precision.

Now, the dominant compute-hungry workload is AI, where precision takes second place to the independent parameter count. To the point that the capacity of BF16, which were originally designed as a radical optimization for AI workloads, is sometimes considered wasteful now.

AI workloads have some truly peculiar and counterintuitive properties - the kind of things you might expect to see in biology instead of conventional computing. Intrinsic error tolerance, for one. It did necessitate some rethinking and reprioritization, and I'm not quite sure if we converged to the general shape of an "optimal" AI accelerator as of yet.

Re: AMD's Instinct MI455X: Aiming for the Sun

#17

Post by stevefan1999 »

dist-epoch wrote: post #11
stevefan1999 wrote: post #10

Earlier quoted context omitted.

but you have to rewrite all your software to ROCm. And ROCm, to this day, still sucks. ZLUDA basically crash on high memory demand, and only accounted for ~70% of CUDA API coverage (and it is still buggy). But hey, at least it does run on Rust-CUDA. I'm one of the few who ported it and fixed a few bugs on ZLUDA. I used it to run a simple SHA256 kernel and it ran sure, but I gave it up because of those fundamental pro…

Co-founder and Chief Compute Officer Anthropic: > I think the thing that we were thinking about originally was whenever we're bringing up a new hardware platform, it's a big effort. It's like a huge thing. And so as we were thinking about this, we started doing our own evaluation of MI 355. You guys generously got us a rack to start working. And we expected this to be kind of a big process. > Our actual experience wa…

That only works if you have all the unit tests and making sure the behaviors between different GPUs are expected, and can tolerate difference in implementation details such as numerical stability and complexity for some functions. AMD and Nvidia are doing vastly different computation for different implementation to the same algorithm for their APIs (can't tell due to NDA, but one algorithm Nvidia did it using broadcast, while AMD is using tile), and that's one of the reason I gave up. And perhaps only for light kernels doing basic BLAS operations and GEMMs. With a standard to follow, that's one of places where it indeed is giving it a good chance, for example on Tensorflow with ROCm or PyTorch. You basically have an insurance of the behavior and a translation layer for ROCm and CUDA by Python.

But anything other than that, nah.

Edit: Maybe I remember what algorithm that is. It is a trick to do coding theory stuff by using NTT (Number Theoretic Transform, the friend of DFT, the Discrete Fourier Transform, but on a ring and hence applicable to finite field) on GPU to do computation over a GF(2^8) field, for my former employment but I went back to using GFNI. Still I have a NDA to adhere to and can't really explain what happens.

Re: AMD's Instinct MI455X: Aiming for the Sun

#18

Post by ACCount37 »

paoliniluis wrote: post #12
dist-epoch wrote: post #11

Earlier quoted context omitted.

Co-founder and Chief Compute Officer Anthropic: > I think the thing that we were thinking about originally was whenever we're bringing up a new hardware platform, it's a big effort. It's like a huge thing. And so as we were thinking about this, we started doing our own evaluation of MI 355. You guys generously got us a rack to start working. And we expected this to be kind of a big process. > Our actual experience wa…

Using AI slop in software that runs ML models is like buying a Ferrari with a motorbike engine. Might be good for demos or as a proof of concept, but that can’t be used for serving LLMs at scale

That would be a very defensible take back in year 2023. Now though?

Anthropic has Mythos. That thing's low level code "AI slop" is better than the "meatbag slop" most software developers write, and it can keep cracking at a given problem with persistence.

OpenAI has GPT-5.6, and also that rabid dog of an AI model that was last seen out in the wild tearing HuggingFace open.

Modern LLMs are very, very capable - not just of writing raw code, but also of persistent, methodical problem solving. Which is what you want to tackle things like "port from an exotic system A to an exotic system B and smoke test the port". Persistently hunting for testable optimizations is a good fit too.

Re: AMD's Instinct MI455X: Aiming for the Sun

#19

Post by stevefan1999 »

nolist_policy wrote: post #15
stevefan1999 wrote: post #10

Earlier quoted context omitted.

but you have to rewrite all your software to ROCm. And ROCm, to this day, still sucks. ZLUDA basically crash on high memory demand, and only accounted for ~70% of CUDA API coverage (and it is still buggy). But hey, at least it does run on Rust-CUDA. I'm one of the few who ported it and fixed a few bugs on ZLUDA. I used it to run a simple SHA256 kernel and it ran sure, but I gave it up because of those fundamental pro…

Spir-v is already happening: https://rocm.blogs.amd.com/software-tools-optimization/spir-...

> PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx950;\ gfx1030;gfx1100;gfx1101;gfx1102;gfx1103;\ gfx1150;gfx1151;gfx1200;gfx1201"

That kicked off my PTSD to setup ROCm on my 6600XT and 9070XT, not even with different containers. I have to build an image for EACH architecture. That's duping 15GB for each arch on my already terribly small SSD.

Re: AMD's Instinct MI455X: Aiming for the Sun

#20

Post by InTheArena »

paoliniluis wrote: post #12
dist-epoch wrote: post #11

Earlier quoted context omitted.

Co-founder and Chief Compute Officer Anthropic: > I think the thing that we were thinking about originally was whenever we're bringing up a new hardware platform, it's a big effort. It's like a huge thing. And so as we were thinking about this, we started doing our own evaluation of MI 355. You guys generously got us a rack to start working. And we expected this to be kind of a big process. > Our actual experience wa…

Using AI slop in software that runs ML models is like buying a Ferrari with a motorbike engine. Might be good for demos or as a proof of concept, but that can’t be used for serving LLMs at scale

At the risk of noting that this is a PR statement from someone who needs something from AMD - you do realize that you just shat upon the CTO of the largest possible consumer of a technology talking about that technology? It's not 2022 anymore.

Just pointing out, there is a reason the large LLMs keep working to try to make it so their own tools can't attack their own moats, by preventing the exact behavior that you dismiss above.

Post reply on HN