Hacker News Reader

Comment by jacquesm | Hacker News Reader

At this point the massive investment in software is what drives this, hardware differences may no longer be dominant but the only way to unseat NV at this point is drop in replacements and/or drop in replacement libraries. And that's getting there. Which is good because GPUs are too expensive and have too little memory, some competition might help move things along rather than to give NV more time to milk their precious digital cows.

View on HN

I really wish the ML researchers would have stayed with Vulkan or OpenCL instead of standardising on CUDA. Everyone must have known how it would end.

Those ml researchers wanted to get hired…

Those ML researchers (myself included) wanted to get the job done.

Sometimes people just have a job to do rather than to play politics and in this case doing it with OpenCL or Vulkan would have ceded the market for their application to the competition due the difference in speed and money spent for a given amount of compute. That's starting to change but the competition has reacted much too late. NVidia figured out that software is almost as important as the hardware early enough that betting on that with a sizeable dedicated team gave them a tremendous lead. I've used CUDA and have tried to use OpenCL but the number of hurdles I had to jump through with OpenCL made it a non-starter and that was before factoring in the cost. NV made it easy and made it work right out of the box. Quite often the most visible constraint is developer time and reducing developer friction is a great way to capture a larger slice of the market as well as to actually grow that market.

If AMD really wished ML researchers to use their hardware, they could have put in the effort to make OpenCL or Vulkan competitive with CUDA with respect to ease of developing ML systems. nVidia did put in that software effort, AMD didn't so now they both reap the consequences.

No argument there. AMD really shat the bed. But if you go with a single supplier, it’s pretty weird to make the pricing agreement “we’ll pay whatever you charge”.

Related, I’m putting some money into AMD. They can only improve, at this point.

huh? don't ML researchers use pytorch and tensorflow?

I’m guessing we’re talking about the pioneers here.

You have clearly not programmed in any of those. :)

OpenCL was almost unusable due to how difficult it was to set up, plus limited support. Vulcan is terribly ugly and verbose and provides no benefits for pure compute workloads. Writing a "hello word" is a few thousands lines of code, debugging impossible.

Working with CUDA is pure pleasure - immediate setup, super easy CPU/GPU interop and code sharing, all the modern (at the time!) C++ features, super pleasant debugging with stepping, profilers, everything. :)

I totally wish others invested as much into tools as NVIDIA did - but they didn't and made the whole experience miserable. :( If your velocity is 1/10 because of terrible tooling, you need to expect some insane benefits to pick it.

Reply on news.ycombinator.com