8 ms·

Does not GPU compute

76 points by alexvoica 10y ago

mjevans 10y ago
A more optimal approach might be to look at what you are trying to accomplish and figuring out the scale that works best for what a local view appears to be. For example, if I imagine a game world where it's open, but there are natural limits to useful render distance, then it is possible to define absolute maximum scale sizes. My new to this problems space view is that even if the world is larger than those sizes, there is probably still some limited observer and scale that makes sense. Building in some spare room and padding in to the scale and it can then be transformed to center on different points. As movement towards one of those points happens the new centering for each object could be pre-computed in spare cycles (or at least spread out so it isn't a single noticeable hit).
- gue5t 10y ago
  > My new to this problems space view I'm going to be downvoted because this isn't particularly on-topic, but nonetheless I'd like to suggest you try hyphenating phrases like this to make them easier to read, so you don't construct garden-path sentences where the correct parse exists but isn't obvious. Thanks!
  - __s 10y ago
    Garden path sentences make pleasent experiences in parsing that accentuate the means to the end resulting communication. Smell the roses
    - dietrichepp 10y ago
      I like garden path sentences in my Joyce, but keep them off HN, please.
  - imron 10y ago
    Being off-topic isn't necessarily cause for downvotes, and correcting English usage is often welcomed by people with a different native language. Saying you're going to be downvoted for something on the other hand, is more likely to get people to downvote you. In fact, some people will always downvote posts that have words to that effect.
vessenes 10y ago
Even if you're doing fintech simulations, FP16 could well be plenty of precision for a first pass, and then you'd get all those extra cores and ops/watt. FP64 seems like a very small use case for most of the parallelized workflows I can imagine.
- Robi395 10y ago
  For scientific workloads, double precision is a must have. The 7 digits of FP32 is not enough. In my lab, we haven't updated our Kepler based GPU since 2013 for this reason.
  - bnegreve 10y ago
    I agree but the article has an answer to this. Since I don’t plan to run any DNA analysis or fintech simulations on my smartphone anytime soon, I am very satisfied having FP32/FP16 precision in mobile right now. And so should you.
    - Robi395 10y ago
      You are right, I thought vessenes meant: "FP64 seems like a very small use case for most of the parallelized workflows I can imagine" for any platform.
  - vessenes 10y ago
    That makes sense. I'm assuming that the majority of GPUs are not running scientific workloads, instead deep learning and matrix ops for finance.
manav 10y ago
For actual mobile devices, yes there is no need. FP64 only has a use in scientific research, maybe finance and a few other fields. Even there you would do a lot of mixed precision stuff. The reason that the support was there was probably that they wanted to design a single chip and either remove/disable cores for truly mobile or general purpose boards while having the logic available for customers that would actually want it.
dibanez 10y ago
It may be that mobile graphics gets by fine with FP32 or less, but I worry that if FP64 gets sidelined then none of the effort going into this hardware today will benefit applications that need real precision like science, weather prediction, GPS, etc.
- clevernickname 10y ago
  Not really a problem. NVIDIA sells cards based around 32-bit (and now increasingly, 16-bit) ALUs for desktop usage, while offering more expensive ones with more 64-bit-focused ALUs for workstations and compute. Compute is important enough to their bottom line to justify it. The real problem is that NVIDIA has compute locked down with CUDA. Mobile chipset vendors can't expand into compute if they're barred from entry at the API level.
  - techdragon 10y ago
    Given the quality of OpenCL and its cross platform nature, it's amazing that everything is still written directly for CUDA... There were no CUDA -> OpenCL toolchains last time I checked. Which is even more frustrating.
    - mattkrause 10y ago
      There is CU2CL: http://chrec.cs.vt.edu/cu2cl/ http://chrec.cs.vt.edu/cu2cl/ The cross-platform nature is actually part of the problem--the whole point of doing GPGPU work is that you're playing to the hardwares' strengths, which can be difficult when the hardware can be nearly anything from a CPU to a GPU to an FPGA. It doesn't help that until recently, AMD hasn't tried to push OpenCL nearly as hard as nVIDIA pushes CUDA.
      - dharma1 10y ago
        AMD management has completely missed the memo on deep learning. No mention of deep learning or FP16 perf yesterday when Polaris was announced - it was all around VR. They are just not turning up to the party and as a company are running out of time if Polaris and Zen dont sell.
sp332 10y ago
It's not just mobile either. The original Titan card released in early 2013 has much higher FP64 performance than any later Nvidia card.
Animats 10y ago
I once needed FP64 in a GPU for physics calculations. One reason that impulse/constraint won out over spring/damper is that spring/damper has a total loss of precision problem with 32-bit floats.
mon_insider 10y ago
I don't remember there ever was a time were FP64 was considered a big deal for mobIle GPUs. The article is trying to debunk a claim that was never popular to begin with.
- alexvoica 10y ago
  Allow me to refresh your memory then. How does the following sound to you? "We also support Full Profile and 64-bit natively, in hardware. After years of evangelising the benefits of such an approach it is nice to see other players in the industry join down this avenue." https://community.arm.com/groups/arm-mali-graphics/blog/2013/06/03/mali-t622--bringing-full-profile-gpu-compute-to-mid-range-devices https://community.arm.com/groups/arm-mali-graphics/blog/2013... "Mali-T622 was specifically tailored for this job. Mali-T622 also supports OpenCL Full Profile and includes double-precision FP64 and full IEEE-754-2008 floating-point support which are essential features in order to enhance the user experience" https://community.arm.com/groups/arm-mali-graphics/blog/2013/06/03/mali-t622--bringing-full-profile-gpu-compute-to-mid-range-devices https://community.arm.com/groups/arm-mali-graphics/blog/2013... I could go on with the examples but I think there's no need to spam the thread with tens of blog articles that say FP64 and "native 64-bit" (whatever that means) are essential to the mobile experience.
DiabloD3 10y ago
Is there any point in actually damning FP64 this hard anymore? There is no reason, imo, for a modern GPU to get worse than 1/3rd performance on FP64 over FP32. Side note: Non-Quadro/Tesla "professional" GeForces have FP64 performance specifically locked out, and AMD has started also doing this with GCN-era chips.
- alexvoica 10y ago
  My focus was on mobile, not desktop GPUs. However, I've noticed that energy and area efficiency are starting to become more relevant in desktop too. At the end of the day, I guess it's all about finding the right balance for your target application.

76 points by alexvoica 10y ago

A more optimal approach might be to look at what you are trying to accomplish and figuring out the scale that works best for what a local view appears to be. For example, if I imagine a game world where it's open, but there are natural limits to useful render distance, then it is possible to define absolute maximum scale sizes. My new to this problems space view is that even if the world is larger than those sizes, there is probably still some limited observer and scale that makes sense. Building in some spare room and padding in to the scale and it can then be transformed to center on different points. As movement towards one of those points happens the new centering for each object could be pre-computed in spare cycles (or at least spread out so it isn't a single noticeable hit).

gue5t 10y ago
> My new to this problems space view I'm going to be downvoted because this isn't particularly on-topic, but nonetheless I'd like to suggest you try hyphenating phrases like this to make them easier to read, so you don't construct garden-path sentences where the correct parse exists but isn't obvious. Thanks!
- __s 10y ago
  Garden path sentences make pleasent experiences in parsing that accentuate the means to the end resulting communication. Smell the roses
  - dietrichepp 10y ago
    I like garden path sentences in my Joyce, but keep them off HN, please.
- imron 10y ago
  Being off-topic isn't necessarily cause for downvotes, and correcting English usage is often welcomed by people with a different native language. Saying you're going to be downvoted for something on the other hand, is more likely to get people to downvote you. In fact, some people will always downvote posts that have words to that effect.

vessenes 10y ago

Even if you're doing fintech simulations, FP16 could well be plenty of precision for a first pass, and then you'd get all those extra cores and ops/watt. FP64 seems like a very small use case for most of the parallelized workflows I can imagine.

Robi395 10y ago
For scientific workloads, double precision is a must have. The 7 digits of FP32 is not enough. In my lab, we haven't updated our Kepler based GPU since 2013 for this reason.
- bnegreve 10y ago
  I agree but the article has an answer to this. Since I don’t plan to run any DNA analysis or fintech simulations on my smartphone anytime soon, I am very satisfied having FP32/FP16 precision in mobile right now. And so should you.
  - Robi395 10y ago
    You are right, I thought vessenes meant: "FP64 seems like a very small use case for most of the parallelized workflows I can imagine" for any platform.
- vessenes 10y ago
  That makes sense. I'm assuming that the majority of GPUs are not running scientific workloads, instead deep learning and matrix ops for finance.

manav 10y ago

For actual mobile devices, yes there is no need. FP64 only has a use in scientific research, maybe finance and a few other fields. Even there you would do a lot of mixed precision stuff. The reason that the support was there was probably that they wanted to design a single chip and either remove/disable cores for truly mobile or general purpose boards while having the logic available for customers that would actually want it.

dibanez 10y ago

It may be that mobile graphics gets by fine with FP32 or less, but I worry that if FP64 gets sidelined then none of the effort going into this hardware today will benefit applications that need real precision like science, weather prediction, GPS, etc.

clevernickname 10y ago
Not really a problem. NVIDIA sells cards based around 32-bit (and now increasingly, 16-bit) ALUs for desktop usage, while offering more expensive ones with more 64-bit-focused ALUs for workstations and compute. Compute is important enough to their bottom line to justify it. The real problem is that NVIDIA has compute locked down with CUDA. Mobile chipset vendors can't expand into compute if they're barred from entry at the API level.
- techdragon 10y ago
  Given the quality of OpenCL and its cross platform nature, it's amazing that everything is still written directly for CUDA... There were no CUDA -> OpenCL toolchains last time I checked. Which is even more frustrating.
  - mattkrause 10y ago
    There is CU2CL: http://chrec.cs.vt.edu/cu2cl/ http://chrec.cs.vt.edu/cu2cl/ The cross-platform nature is actually part of the problem--the whole point of doing GPGPU work is that you're playing to the hardwares' strengths, which can be difficult when the hardware can be nearly anything from a CPU to a GPU to an FPGA. It doesn't help that until recently, AMD hasn't tried to push OpenCL nearly as hard as nVIDIA pushes CUDA.
    - dharma1 10y ago
      AMD management has completely missed the memo on deep learning. No mention of deep learning or FP16 perf yesterday when Polaris was announced - it was all around VR. They are just not turning up to the party and as a company are running out of time if Polaris and Zen dont sell.

sp332 10y ago

It's not just mobile either. The original Titan card released in early 2013 has much higher FP64 performance than any later Nvidia card.

Animats 10y ago

I once needed FP64 in a GPU for physics calculations. One reason that impulse/constraint won out over spring/damper is that spring/damper has a total loss of precision problem with 32-bit floats.

mon_insider 10y ago

I don't remember there ever was a time were FP64 was considered a big deal for mobIle GPUs. The article is trying to debunk a claim that was never popular to begin with.

alexvoica 10y ago
Allow me to refresh your memory then. How does the following sound to you? "We also support Full Profile and 64-bit natively, in hardware. After years of evangelising the benefits of such an approach it is nice to see other players in the industry join down this avenue." https://community.arm.com/groups/arm-mali-graphics/blog/2013/06/03/mali-t622--bringing-full-profile-gpu-compute-to-mid-range-devices https://community.arm.com/groups/arm-mali-graphics/blog/2013... "Mali-T622 was specifically tailored for this job. Mali-T622 also supports OpenCL Full Profile and includes double-precision FP64 and full IEEE-754-2008 floating-point support which are essential features in order to enhance the user experience" https://community.arm.com/groups/arm-mali-graphics/blog/2013/06/03/mali-t622--bringing-full-profile-gpu-compute-to-mid-range-devices https://community.arm.com/groups/arm-mali-graphics/blog/2013... I could go on with the examples but I think there's no need to spam the thread with tens of blog articles that say FP64 and "native 64-bit" (whatever that means) are essential to the mobile experience.

DiabloD3 10y ago

Is there any point in actually damning FP64 this hard anymore? There is no reason, imo, for a modern GPU to get worse than 1/3rd performance on FP64 over FP32. Side note: Non-Quadro/Tesla "professional" GeForces have FP64 performance specifically locked out, and AMD has started also doing this with GCN-era chips.

alexvoica 10y ago
My focus was on mobile, not desktop GPUs. However, I've noticed that energy and area efficiency are starting to become more relevant in desktop too. At the end of the day, I guess it's all about finding the right balance for your target application.