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

Documentation for the AMD 7900XTX

github.com

Post reply on HN
Print view View original

31–40 of 154 posts

Re: Documentation for the AMD 7900XTX

#31

Post by ultra_nick »

roenxi wrote: post #8

So for reference, this geohot is George Hotz who has a company Tiny Corp [0] in the space. Among a long list of things, he had a moment in the spotlight recently for a long rant [1] where he "gave up on AMD" because their drivers sucked more than he expected. The situation is reasonably complex - AMD have some programmers working on ROCm who seemed to be operating at standard fare when that is really not what AMD nee…

NVidia is making boatloads of money because their driver works and they have a software library called CUDA that accelerates neural networks.

Nobody expects AMD to match them, but George thought they could at least write a GPU driver. If AMD can get that driver out, then George could provide a competitor to CUDA (for neutral networks only). They'd both make boatloads of money.

However, AMD was less capable than expected and their drivers were too buggy to run neural networks like those needed for the MLPerf benchmark. So now, it appears that AMD, Tinybox, and investors like me won't be making boatloads of money.

Re: Documentation for the AMD 7900XTX

#32

Post by georgehotz »

MuffinFlavored wrote: post #29
roenxi wrote: post #8

So for reference, this geohot is George Hotz who has a company Tiny Corp [0] in the space. Among a long list of things, he had a moment in the spotlight recently for a long rant [1] where he "gave up on AMD" because their drivers sucked more than he expected. The situation is reasonably complex - AMD have some programmers working on ROCm who seemed to be operating at standard fare when that is really not what AMD nee…

> The problem is if I spend an evening trying to do anything with OpenCL or ROCm the kernel hard-locks and I go to bed early. In a race against NVIDIA where every comment on here is about just how much better CUDA is than any alternative, why doesn't literally what you just said "also" have the attention of Lisa Su?

I have had multiple replies from Lisa Su on both Twitter and by e-mail. It doesn't help.

AMD is structurally incapable of fixing these issues. They don't even have a 7900XTX in CI. Crashing the firmware is so trivial there's no way anyone fuzzed anything.

They debug by the application, adding mitigations at many layers of the stack to make it work. While this strategy is fine for the 20 mainstream games that come out per year, unless you root cause issues and have a good CI, you'll never build a stable GPU for general compute.

Re: Documentation for the AMD 7900XTX

#33

Post by MuffinFlavored »

georgehotz wrote: post #32
MuffinFlavored wrote: post #29

Earlier quoted context omitted.

> The problem is if I spend an evening trying to do anything with OpenCL or ROCm the kernel hard-locks and I go to bed early. In a race against NVIDIA where every comment on here is about just how much better CUDA is than any alternative, why doesn't literally what you just said "also" have the attention of Lisa Su?

I have had multiple replies from Lisa Su on both Twitter and by e-mail. It doesn't help. AMD is structurally incapable of fixing these issues. They don't even have a 7900XTX in CI. Crashing the firmware is so trivial there's no way anyone fuzzed anything. They debug by the application, adding mitigations at many layers of the stack to make it work. While this strategy is fine for the 20 mainstream games that come out…

Genuine question for you then, why are you investing your time reversing their firmware if it sounds like you think they're failing/will fail? Why not just "adopt" the golden child/golden standard NVIDIA now for your usecase, reverse that (if needed) and move on? Genuinely curious what you gain from trying to make AMD work/investing in AMD on your end here. Is it that you see a value proposition through the "muck" of crashy-kernels that I'm missing?

Re: Documentation for the AMD 7900XTX

#34

Post by georgehotz »

MuffinFlavored wrote: post #33
georgehotz wrote: post #32

Earlier quoted context omitted.

I have had multiple replies from Lisa Su on both Twitter and by e-mail. It doesn't help. AMD is structurally incapable of fixing these issues. They don't even have a 7900XTX in CI. Crashing the firmware is so trivial there's no way anyone fuzzed anything. They debug by the application, adding mitigations at many layers of the stack to make it work. While this strategy is fine for the 20 mainstream games that come out…

Genuine question for you then, why are you investing your time reversing their firmware if it sounds like you think they're failing/will fail? Why not just "adopt" the golden child/golden standard NVIDIA now for your usecase, reverse that (if needed) and move on? Genuinely curious what you gain from trying to make AMD work/investing in AMD on your end here. Is it that you see a value proposition through the "muck" of…

This is my weekend hobby, I'm learning about GPUs. Besides, I'd like to see competition to NVIDIA.

From a business side, tiny corp is selling boxes with both AMD and NVIDIA, we'll let the user choose how much they value money vs pain.

Re: Documentation for the AMD 7900XTX

#36

Post by sashank_1509 »

atq2119 wrote: post #28
echelon wrote: post #12

Earlier quoted context omitted.

Anyone working on getting us out of the Nvidia monopoly on commerical AI should be praised. We need lots of competition in this space. I'm hoping Google spins off its TPU division and sells chips to everyone that wants them. I'm hoping AMD gets their act together and that they open up their drivers and stack. I'm hoping a lot of other hardware companies and compute stacks start making inroads on this. Nvidia is not j…

> I'm hoping AMD gets their act together and that they open up their drivers and stack. Their entire stack on Linux is open-source. That part isn't a problem.

Nope, they basically open sourced their api which is kind of a worthless proposition. You would think this is malicious, trying to pretend as open source, but the whole thing is done so ineptly, it is just incompetence. Likely management has no idea what open source means and thinks releasing api code means open source and is good enough for customers to use your GPU.

More concretely, what they open sourced calls closed source functions that run on the GPU which is where all the bugs are, and causes all the crashes.

Re: Documentation for the AMD 7900XTX

#37

Post by rudedogg »

georgehotz wrote: post #34
MuffinFlavored wrote: post #33

Earlier quoted context omitted.

Genuine question for you then, why are you investing your time reversing their firmware if it sounds like you think they're failing/will fail? Why not just "adopt" the golden child/golden standard NVIDIA now for your usecase, reverse that (if needed) and move on? Genuinely curious what you gain from trying to make AMD work/investing in AMD on your end here. Is it that you see a value proposition through the "muck" of…

This is my weekend hobby, I'm learning about GPUs. Besides, I'd like to see competition to NVIDIA. From a business side, tiny corp is selling boxes with both AMD and NVIDIA, we'll let the user choose how much they value money vs pain.

What are your thoughts on using Intel GPUs for deep learning?

Re: Documentation for the AMD 7900XTX

#38

Post by AnthonyMouse »

sashank_1509 wrote: post #36
atq2119 wrote: post #28

Earlier quoted context omitted.

> I'm hoping AMD gets their act together and that they open up their drivers and stack. Their entire stack on Linux is open-source. That part isn't a problem.

Nope, they basically open sourced their api which is kind of a worthless proposition. You would think this is malicious, trying to pretend as open source, but the whole thing is done so ineptly, it is just incompetence. Likely management has no idea what open source means and thinks releasing api code means open source and is good enough for customers to use your GPU. More concretely, what they open sourced calls clo…

> More concretely, what they open sourced calls closed source functions that run on the GPU which is where all the bugs are, and causes all the crashes.

In particular, what they did was open source the drivers but not the firmware, and then when the firmware has bugs nobody outside the company can fix it because it's not open source.

Re: Documentation for the AMD 7900XTX

#39

Post by doctorpangloss »

georgehotz wrote: post #34
MuffinFlavored wrote: post #33

Earlier quoted context omitted.

Genuine question for you then, why are you investing your time reversing their firmware if it sounds like you think they're failing/will fail? Why not just "adopt" the golden child/golden standard NVIDIA now for your usecase, reverse that (if needed) and move on? Genuinely curious what you gain from trying to make AMD work/investing in AMD on your end here. Is it that you see a value proposition through the "muck" of…

This is my weekend hobby, I'm learning about GPUs. Besides, I'd like to see competition to NVIDIA. From a business side, tiny corp is selling boxes with both AMD and NVIDIA, we'll let the user choose how much they value money vs pain.

Did you ever consider using NVIDIA? I thought the machine you were planning to sell originally had RTX 6000s.

Also, isn’t the real problem with the 7900 XTX is the lack of interconnect? I really appreciate the long list of stuff you are agitating AMD about. Infinity Fabric and blower fans would be great too. Another POV is none of that is ever going to happen, and everyone will have to wait for a Chinese GPU manufacturer to create real competition. I can’t recall a single time in my life a non-Chinese company - electronics or otherwise - delivered the “same” features for a lower price.

Re: Documentation for the AMD 7900XTX

#40

Post by DiabloD3 »

ultra_nick wrote: post #31
roenxi wrote: post #8

So for reference, this geohot is George Hotz who has a company Tiny Corp [0] in the space. Among a long list of things, he had a moment in the spotlight recently for a long rant [1] where he "gave up on AMD" because their drivers sucked more than he expected. The situation is reasonably complex - AMD have some programmers working on ROCm who seemed to be operating at standard fare when that is really not what AMD nee…

NVidia is making boatloads of money because their driver works and they have a software library called CUDA that accelerates neural networks. Nobody expects AMD to match them, but George thought they could at least write a GPU driver. If AMD can get that driver out, then George could provide a competitor to CUDA (for neutral networks only). They'd both make boatloads of money. However, AMD was less capable than expec…

Slight mistake in your description: CUDA is an out of date API that was replaced by Khronos's own official compute APIs. Khronos is a standards consortium that Nvidia is a founding member of.

Although the marketing department at Nvidia still pushes for greenfield CUDA codebases, no new code should be written in it, and they should opt for open source international standards only. Khronos APIs are implemented by over 120 vendors.

Post reply on HN