5 ms·

Tensorflow is optimized for TPU's which isn't really consumer-grade hardware.

joseph_grobbles 3y ago
Google's first TPU was developed a year after Tensorflow. And for that matter, Tensorflow works fine with CUDA, was originally entirely built for CUDA, and it's super weird the way it's being referenced in here. Tensorflow lost out to Pytorch because the former is grossly complex for the same tasks, with a mountain of dependencies, as is the norm for Google projects. Using it was such a ridiculous pain compared to Pytorch. And anyone can use a mythical TPU right now on the Google Cloud. It isn't magical, and is kind of junky compared to an H100, for instance. I mean...Google's recent AI supercomputer offerings are built around nvidia hardware. CUDA keeps winning because everyone else has done a horrendous job competing. AMD, for instance, had the rather horrible ROCm, and then they decided that they would gate their APIs to only their "business" offerings while nvidia was happy letting it work on almost anything.
- HellDunkel 3y ago
  Best explanation so far. I am surprised OpenCL never gained much traction. Any idea why?
  - blihp 3y ago
    The same reason most of AMD's 'open' initiatives don't gain traction: they throw it out there and hope things will magically work out and that a/the community will embrace it as the standard. It takes more work than that. What AMD historically hasn't done is the real grunge work of addressing the limitations of their products/APIs and continuing to invest in them long term. See how the OpenCL (written by AMD) Cycles renderer for Blender worked out, for example. Something AMD doesn't seem to understand/accept is that since they are consistently lagging nVidia on both the hardware and software front, nVidia can get away with some things AMD can't. Everyone hates nVidia for it, but unless/until AMD wises up they're going to keep losing.
- onetimeusename 3y ago
  what did you do to get all your posts automatically dead?
caeril 3y ago
Unrelated question for the HN experts: My sibling commenter is shadowbanned, but if you look into their comment history, there are occasionally comments that are not dead. How does this happen?
- philipkglass 3y ago
  Somebody clicked on the timestamp of that post and used the "vouch" link to unhide it. I sometimes do that for comments from new accounts that been hidden by some overzealous anti-spam heuristic.
  - RobotToaster 3y ago
    Helpful to know, I've seen a few hidden posts that seem reasonable but didn't know I could do that.
RobotToaster 3y ago
Isn't the coral stick a TPU?
- giobox 3y ago
  Yes, although availability recently has been pretty bad following the chip shortage, and prices skyrocketed to ~300 dollars. Not sure if situation returning to normal yet. Similar woes to the Raspberry Pi etc. I needed two for a project and ended up paying a lot more than I wanted for used ones. For those not familiar, consumer/hobbyist grade TPUs: https://coral.ai/products/ https://coral.ai/products/

Google's first TPU was developed a year after Tensorflow. And for that matter, Tensorflow works fine with CUDA, was originally entirely built for CUDA, and it's super weird the way it's being referenced in here. Tensorflow lost out to Pytorch because the former is grossly complex for the same tasks, with a mountain of dependencies, as is the norm for Google projects. Using it was such a ridiculous pain compared to Pytorch. And anyone can use a mythical TPU right now on the Google Cloud. It isn't magical, and is kind of junky compared to an H100, for instance. I mean...Google's recent AI supercomputer offerings are built around nvidia hardware. CUDA keeps winning because everyone else has done a horrendous job competing. AMD, for instance, had the rather horrible ROCm, and then they decided that they would gate their APIs to only their "business" offerings while nvidia was happy letting it work on almost anything.

HellDunkel 3y ago
Best explanation so far. I am surprised OpenCL never gained much traction. Any idea why?
- blihp 3y ago
  The same reason most of AMD's 'open' initiatives don't gain traction: they throw it out there and hope things will magically work out and that a/the community will embrace it as the standard. It takes more work than that. What AMD historically hasn't done is the real grunge work of addressing the limitations of their products/APIs and continuing to invest in them long term. See how the OpenCL (written by AMD) Cycles renderer for Blender worked out, for example. Something AMD doesn't seem to understand/accept is that since they are consistently lagging nVidia on both the hardware and software front, nVidia can get away with some things AMD can't. Everyone hates nVidia for it, but unless/until AMD wises up they're going to keep losing.
onetimeusename 3y ago
what did you do to get all your posts automatically dead?

caeril 3y ago

Unrelated question for the HN experts: My sibling commenter is shadowbanned, but if you look into their comment history, there are occasionally comments that are not dead. How does this happen?

philipkglass 3y ago
Somebody clicked on the timestamp of that post and used the "vouch" link to unhide it. I sometimes do that for comments from new accounts that been hidden by some overzealous anti-spam heuristic.
- RobotToaster 3y ago
  Helpful to know, I've seen a few hidden posts that seem reasonable but didn't know I could do that.

RobotToaster 3y ago

Isn't the coral stick a TPU?

giobox 3y ago
Yes, although availability recently has been pretty bad following the chip shortage, and prices skyrocketed to ~300 dollars. Not sure if situation returning to normal yet. Similar woes to the Raspberry Pi etc. I needed two for a project and ended up paying a lot more than I wanted for used ones. For those not familiar, consumer/hobbyist grade TPUs: https://coral.ai/products/ https://coral.ai/products/