Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I think he is right about AMD but completely misses the mark when it comes to tenstorrent. He is ranting about exponential linear unit (elu), which hardly seems to be something that could possibly hold an AI company back. If the hardware is running and training models reliably, then it's just a matter of pricing to stay competitive. Poor optimization cuts your margins, but the incentives are aligned.

With AMD the experience is so poor that you have to save the company from itself if you want to make progress.



People refuse to believe it, but the AMD experience is getting better every day by leaps and bounds. Over the last few months, there is a brand new focus on improving the software. There is still a long way to go, but the company is absolutely trying to save itself.


AMD has legitimately been making great progress. They still have a long way to go, and I appreciate SemiAnalysis taking up the mantle of calling them out, but I ran:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.3

today on stream and it just worked. No external ROCm install, and just the amdgpu driver that's in Arch.

We also have our own complete driver/runtime now in tinygrad; it's so much nicer to build on a foundation when you can blame yourself for the bugs.


Regarding SA, I’m all for holding AMD accountable, but let’s at least get the facts right, and maybe don’t come at it with a history of cheerleading for Nvidia.

Maybe SA might set their sights on you next?


geohot got a lot of press and attention by coming out aggressively at AMD during a moment when their software really was weak.

The short-term payoff of that drama created a long-term problem where the only way they could look good was by outrunning the progress of the engineers who were inside the company, well funded, and already familiar with everything. It was an impossible goal from the start but he made it even more impossible by attacking AMD. AMD was smart to basically ignore them and wait for him to give up, as opposed to inviting that drama to crossover in-house or split their user base.


They didn't ignore him and he didn't give up. They gave him two boxes and he added MI300x support to tinygrad.


We got the MI300X box on MLPerf too, and every MLPerf from here on general tinygrad improvements should bring down the times. We're still quite focused on AMD.

Like it's strange people think I give up on things, I think they listen to the media too much. This is a 2+ year long project that I've worked on almost every day. https://geohot.github.io/blog/jekyll/update/2023/05/24/the-t...


He's not ranting about ELU. He's using ELU as an example of something that shouldn't be in the lower abstraction layers.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: