today at 6:19 PM
For people interested in this: I've been working on an open-source libtpu implementation for https://tenstorrent.com/ accelerators as a side project, it can already run most of torchtpu, see https://github.com/pcmoritz/libtt (based on the tenstorrent software stack). It shows all the things involved in lowering torch to tile based accelerators.
today at 7:03 PM
Very cool! I have been using Lean to drive PJRT like so: https://github.com/brettkoonce/lean4-mlir and in theory your code could unlock another platform!
today at 7:07 PM
Thanks for sharing, this is super cool! Right now my approach to correctness it to just run as many tests as possible and fix them (torchtpu greatly expands the amount of available tests), but formal verification is also super cool. Having a look at your book https://lean.brettkoonce.com/blueprint.pdf
The Tenstorrent ISA is documented btw https://github.com/tenstorrent/tt-isa-documentation, I wonder if one could do the verification end-to-end with the hardware.
today at 7:00 PM
This is super cool!
today at 7:04 PM
Thanks for the encouragement, if you have the hardware and end up trying it out I'd love to get your feedback (feel free to drop an issue in the repo) :)
today at 7:07 PM
I would love to do that, but unfortunately don’t have one on hand (yet?). In fact, I’m curious - did you buy a TT device to work on this, or are you using a cloud provider like Koyeb?
today at 7:11 PM
I first built my own to try it out (the exact hardware is described in https://relativity.cc/hardware/ under "one Blackhole card") and when I liked the hardware I got a quietbox 2. It is very high quality but a little expensive (but inexpensive compared to other LLM hardware and it has excellent interconnects which is very rare in the consumer/dev space).
today at 4:51 PM
Interesting, Amazon doing the same for AWS Trainium.
https://github.com/aws-neuron/torch-neuronx
today at 6:48 PM
Pytorch conference is happening today for explaining the coincidental timing
today at 6:51 PM
That explains the 5x speed increases huggingface mentioned for Mac using metal.
today at 8:03 PM
Link?
today at 8:18 PM
Google and Apple are going to move consumers and ordinary business users to edge inference. This is more disruptive to the upcoming IPOs than most investors understand.
today at 6:22 PM
Cool! XLA is pretty good + potentially a better compiler infrastructure than inductor which will lead to improved performance on all hardware.
Also I guess this is the final nail in the coffin of Tensorflow.
today at 7:19 PM
Amazing news, I remember the days when you had to fight XLA to make PyTorch run on TPUs, it seems we finally got first-party support!
Since HW specifics live mostly in the kernel layer (Pallas or CUDA), I imagine the Jax vs PyTorch question is more of a style preference now?
today at 5:27 PM
Google is selling TPU shovels via Cloud services and also physically to select partners. So this makes sense.
There will be an abundance of shovels. Nebius already crashed by almost half from its hype peak.
today at 5:33 PM
it doesn't seem like there's an abundance of shovels coming anytime soon
today at 8:30 PM
Shovels and gold aren’t the problem, humans are the problem
either AI achieves hyper-efficiency and triggers systematic economic demand destruction due to unemployment, or it underperforms and sparks a massive tech-sector asset bubble burst
If AI is as good as valuations - it crashes. If AI isn’t as good, it crashes. all eggs in one basket economy.
Every future leads to a massive financial shitstorm
AI is priced in as a technology to end all markets. The last bull market you will ever see.
It’s a miracle markets didn’t notice this “damned if you succeed damned if you don’t” type of inevitable checkmate
I have been building a bunker
today at 9:32 PM
It's possible. If goog's AI overview is accurate, it seems somewhat similar to speculative euphoria seen in the past (eg companies involved in electricity prior to the great depression). It's larger (currently 35% for AI versus 25% for electric/related cos), I'm guessing because of more credit. At the same time, the housing component of the market prior to the GFC was about 23%, and it wasn't as economically damaging as the GD, so hopefully the AI bubble unwinding won't be as bad.
today at 9:50 PM
> so hopefully the AI bubble unwinding won't be as bad
It's currently larger than the 1880s railroad boom.
"Overbuilding and speculative debt ultimately culminated in the Panic of 1893, plunging the nation into a severe economic depression."
today at 9:30 PM
[dead]
today at 4:48 PM
Are there TPUs outside Google Cloud?
today at 5:08 PM
Yes. Google is famously selling TPUs on prem to big customers (e.g., Anthropic) and Crux AI is now a venture to do just that.
today at 4:55 PM
Google coral USB and m.2 devices are TPUs.
today at 6:27 PM
"Have you seen any software run on Coral TPUs?"
"No, but I have seen the house that the project built, and it is terrific." ~ some Google director
today at 8:39 PM
Is a pretty successful open source NVR that used corals when it launched (and many people still use them with it)
today at 5:07 PM
Only in marketing name. They're not related otherwise.
today at 5:55 PM
Maybe just marketing, but they call the NPUs in their Pixel phones "TPUs"
today at 4:55 PM
TPU / neural compute accelerator / Math coprocessor;
"OpenTPU: Open-Source Reimplementation of Google Tensor Processing Unit (TPU)" https://news.ycombinator.com/item?id=44111452
From https://news.ycombinator.com/item?id=45848676 :
> How does Cerebras WSE-3 with 44GB of 'L2' on-chip SRAM compare to Google's TPUs, Tesla's TPUs, NorthPole, Groq LPU, Tenstorrent's, and AMD's NPU designs?
There are TPUs made out of silicon carbide with standard fab.
There is in-flight tensor processing with photons, but that's not quite a TPU.
today at 5:59 PM
I think the commenter was asking if Google sells TPU chip hardware to use outside of Google's own infrastructure and cloud.
today at 7:33 PM
Pardon my ignorance, but who else besides Google has TPUs? Are they commercially available somewhere? What’s their main application right now outside of Google?
today at 7:44 PM
You can rent them on cloud + they sell directly to Anthropic. Its pretty well known that Anthropic is a heavy user of TPU's (along with Trainium & GPU's)
today at 7:43 PM
today at 5:07 PM
Dumb question from non-AI engineer.
Is this somehow Google's answer to today's best DeepSeek open-sourcing inferencing engine?
today at 5:15 PM
I dont think big companies like Google react to news like that. Very likely this was meant to go out on Tuesday for weeks now. And it just happened to have coincided with another news.
today at 5:58 PM
Yeah they read the news this morning and whipped up a quick Torch->XLA translation layer and coordinated with four different inference engines (who also all wrote their own blog posts) and shoved this out the door quick to stay on top of the news cycle
today at 6:11 PM
Why be rude on purpose? OP clearly prefaced by saying it was a dumb question
today at 9:59 PM
because the question was incredibly stupid and deserved some sarcasm
today at 6:41 PM
Welcome to the internet, great to have you here.
today at 6:01 PM
they're releasing this today because the pytorch conf is this week...
today at 6:24 PM
No, these are two very different objectives.
today at 5:20 PM
They do different things and sit in different parts of the stack (training vs inference). TorchTPU seems more like a play to get more people familiar with Googles TPU chips as they want to sell them to people outside of google.
today at 6:22 PM
Yeah I'd assume this is to allow us to train on their TPUs with torch instead of needing to rent a GPU
today at 6:57 PM
ty. that helps a lot and know where to research next.
today at 5:50 PM
so if the moat isn't the model and the moat isn't the inference layer, i guess the moat is the hardware now? what does this mean for frontier labs?
today at 6:10 PM
hardware has always been the moat in the AI space
today at 7:15 PM
Yeah a lot of people bought into the idea of CUDA being a moat but we’re well past the point where it’s pretty trivial to jump to another equivalent stack on the software side.
today at 7:29 PM
Those are the two main interfaces: the inference hardware and the endpoints. Endpoints are pretty much standardized to the point that you can switch them out in seconds. CUDA still has an edge but every release like this chips away at that. Even so, TPUs are unobtanium so while I appreciate Google's effort here I also recognize that it is not something that will be useful to me.
In that sense AMD is still better positioned to give NVidia a run for their money but given the family connection and AMD's underdog status I am not very hopeful that we'll see meaningful competition in inference hardware any time soon.
NVidia is a much harder nut to crack than OpenAI or Anthropic. They are easily a decade ahead of viable competition. Check out the GB300 if you don't believe that.
today at 7:58 PM
Even if it's allegedly possible, the devil has always been in the details. It always sucks to find out that there's some quirk of your model architecture that breaks an implicit assumption that underpins the export process.
today at 7:02 PM
Will always be true
today at 5:57 PM
Products can also be a moat.
today at 7:28 PM
Is the Linux driver available?
today at 5:05 PM
[flagged]
today at 5:52 PM
[dead]