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

AMD YOLO

geohot.github.io

Post reply on HN
Print view View original

71–80 of 85 posts

Re: AMD YOLO

#71

Post by JohnBooty »

BitwiseFool wrote: post #61
georgeecollins wrote: post #40

I would really love it if people on Hacker News could weigh in on how much of a moat they think CUDA really is. As in: How hard is it to use something else? If you started a project today how much would you want to get paid to not use CUDA? A lot of readers on this site have a good insight into this and it is a key question financial people are asking without the knowledge many people here possess.

AMD's competitor to CUDA is ROCm. Historically, AMD has been hobbled by the quality of their drivers and because they sold less performant hardware. AMD has traditionally been the budget option for both CPUs and GPUs. Things have changed in the CPU space because of Ryzen, but sadly AMD has not been able to realize an equivalent competitive advantage in the GPU space. Intel has also entered the GPU market, but they ar…

    AMD has been hobbled by the quality of their drivers
I always hear this and I believe it, but I've never been able to find any insight about what exactly is holding them back.

Given the way nVidia is printing money, surely it absolutely cannot be a lack of motivation on AMD's part?

This is a very uninformed thought as I have no experience writing drivers, nor am I familiar with the various things supported by CUDA and ROCm. But how is AMD struggling with ROCm compute drivers, when their game drivers have been plenty stable as far as I have experienced? Surely the surface area of functionality needed for the graphics drivers is larger and therefore the compute drivers should be a relatively easier task? Or am I wrong and CUDA has a bunch of higher-level stuff baked into it and this is what AMD struggles to match?

     and because they sold less performant hardware.
Does anybody have and insight into specifically what part of compute performance AMD is struggling to match? Did AMD bet on the wrong architectural horse entirely? Are they unable to implement really basic compute primitives as efficiently as they want because nVidia holds key patents? Did nVidia lock down the entire pool of engineers who can implement this shit in a performant way?

I mean, aside from GPU compute stuff, it sure looks to me like AMD is executing well. It doesn't seem like they're a bunch of dunces over there. Quite the opposite?

Re: AMD YOLO

#72

Post by bayindirh »

xnx wrote: post #59
drexlspivey wrote: post #55

Earlier quoted context omitted.

AMD hired 1 guy to do it and then fired him

I wondered if John Carmack would take his strong knowledge of low level hardware and interest in AI and work on this, but he seems to be working on a non LLM flavor of AI.

From his talks, he never liked AMD/ATI hardware. Doom 3 code was infamously NVIDIA optimized, since he liked the hardware more.

Re: AMD YOLO

#74

Post by ein0p »

eightysixfour wrote: post #11
ein0p wrote: post #7

Earlier quoted context omitted.

> Nvidia is clearly way overvalued Tell me you haven't looked at Nvidia's financials (especially the margins) without telling me. It basically prints money, now and in the foreseeable future, and all of its products are permanently sold out, even at the insane prices Nvidia is charging.

> foreseeable future I think this is the arguable part. The more AI compute becomes valuable, the more reason there is to divest from their software moat. Their hardware is good, but not unassailable. I think their modest (by tech hype standards) P/E is recognition of this.

While I agree with you in theory, the practice of this has been somewhat less optimistic. So far AMD hardware _doesn't even work_. That's why Geohot had to write his own stack. Moreover, it also doesn't sell - there's maybe one or two obscure cloud providers for it. No major cloud provider has it, or is going to have it anytime soon, until software and driver problems are figured out. Why? Simple - they have no interest whatsoever in disappointing their customers and/or spending all of their support engineering time on supporting obscure GPUs that they'll have to charge less for in order for them to see uptake at all. Even Intel (!) has a more robust offering with Gaudi 2/3, and they're having a heck of a time getting large deployments anywhere outside Intel Developer Cloud.

Re: AMD YOLO

#75

Post by transpute »

empiricus wrote: post #38
hmm37 wrote: post #13

Earlier quoted context omitted.

AMD's PE ratio looks high because of their Xilinx acquisition and for tax reasons. You have to look at non-GAAP numbers instead, e.g. AMD's forward PE ratio which is only ~20 or so. Nvidia's forward PE is ~30.

just bought a little amd with zero dilligence, and now i see they aquired xilinx? why would anyone do that... remorse.

Isn't Xilinx the origin of "AI NPU" in AMD Ryzen?

Re: AMD YOLO

#76

Post by _DeadFred_ »

deadbabe wrote: post #21
Jalad wrote: post #4

Earlier quoted context omitted.

> AMD is probably undervalued, but Nvidia is clearly way overvalued. AMD trades at a price to earnings ratio of 99, NVDA trades at a PE of ~38. PE isn't everything when looking at companies, but I don't see other reasons to think AMD is undervalued

Is PE even anything? If you guided your investing based on sane PE ratios for the past decade you would have been consistently unhappy with missing out on huge gains in tech companies.

PE is for a stable intelligent market. 'The steady hand' type market. What you are talking about is called gambling. PE doesn't matter for gambling.

Re: AMD YOLO

#77

Post by ZeroTalent »

dlcarrier wrote: post #3

AMD is probably undervalued, but Nvidia is clearly way overvalued. I don't think the reckoning will come from AMD stealing Nvidia's market share, it'll come when the hype bubble collapses and businesses start treating neural networks like commodities, running whatever is cheapest instead of the absolute most powerful. AMD is in a great position, because they make both great GPU/NPU hardware and great CPU hardware.

NVIDIA is one of the most undervalued companies in the SPX looking at fundamentals, even disregarding what they are planning next (subscription services for CUDA, and humanoids, etc).

Take a look at their last quarter's income statement graph: https://i.imgur.com/mQwZ5o4.png - Once in a few years I see a Sankey graph looking like that. And it's only growing over the last 10 years.

Re: AMD YOLO

#79

Post by latchkey »

ChocolateGod wrote: post #52
georgeecollins wrote: post #40

I would really love it if people on Hacker News could weigh in on how much of a moat they think CUDA really is. As in: How hard is it to use something else? If you started a project today how much would you want to get paid to not use CUDA? A lot of readers on this site have a good insight into this and it is a key question financial people are asking without the knowledge many people here possess.

> could weigh in on how much of a moat they think CUDA really is. There's movement to implement CUDA libraries that work on non-Nvidia cards, but I guess adoption could be hindered by legal fears. https://github.com/vosen/ZLUDA

Wrong project. ZLUDA will never support enterprise.

What you're looking for is SCALE...

https://docs.scale-lang.com/

and they are making amazing progress.

Re: AMD YOLO

#80

Post by alecco »

I like the guy but he is wrong. Besides the CUDA ecosystem, Nvidia hardware has a lot of features and a lot of things to yet optimize. Like optimizations by DeepSeek (non-CUTLASS custom models and DualPipe). I think Nvidia's current hardware has plenty of legs and that's why they are not rushed on releasing next gen chips.

The actual challenger is Cerebras. No need to load (VRAM->SRAM) all the parameters for every batch . But they have yet to prove they can scale and support the custom stack. We'll see.

Post reply on HN