Top all-time · #3503
Etched
Company blog · etched.com/progresshardwareai
Etched is the company blog at etched.com, writing about hardware and ai. Since 2024, 21 different Hacker News readers have linked it 21 times, most often “Sohu: The First Transformer ASIC”. No feed was found, so read it on the site itself.
Too few sampled comments (16) to say how often HN recommends it.?
Why HN links it
Company blog from Etched, an AI chip startup building transformer-specific ASICs (Sohu). Commenters mostly link its announcement, technical memo and progress posts as the source for its claims.
Summarised by Claude from a sample of 16 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Co-designed chips, racks, and software to push the pareto frontier to best-in-class throughput and interactivity.
Most-linked posts?
- Sohu: The First Transformer ASIC8 people · 8 mentions · first linked 2024-05-08
- Etched Is Making the Biggest Bet in AI7 people · 7 mentions · first linked 2024-06-25
- Oasis: An interactive, explorable world model2 people · 2 mentions · first linked 2024-10-31
- Frontier Inference Clusters2 people · 2 mentions · first linked 2024-12-04
What Hacker News says?
Oasis (the Minecraft world model) can serve about 5 players on 8 H100 in real-time at 20fps in 360p. This is a much more capable model with two orders of magnitude more context. They pretty much say it can't be played real-time, which I read as they generate less than 15fps@240p on 8 GPUs. Probably why they talk so much about using it for AI training and evaluation rather than human use. There is a distilled version…
Comment linking /
on “Genie 2: A large-scale foundation world model”Submitted Etched Is Making the Biggest Bet in AI as a story · 45 points
HN discussionIn case anyone's interested in these niche startups like taalas, here are a few more: 1. https://matx.com/ 2. https://www.d-matrix.ai/ 3. https://www.etched.com/ 4. https://www.positron.ai/ 5. https://hyperaccel.ai/ 6. https://axelera.ai/ 7. https://www.enchargeai.com/ 8. https://furiosa.ai/
Comment linking /
on “OpenAI Jalapeño: Better than Nvidia Blackwell”Submitted Sohu: The First Transformer ASIC as a story · 44 points
HN discussion>None of that is specialized to run only transformers at this point isn't this what [etched]( https://www.etched.com/ ) is doing?
Comment linking /
on “'Attention is all you need' coauthor says he's 'sick' of transformers”Submitted Oasis: An interactive, explorable world model as a story · 14 points
HN discussionThere is a fairly new ASIC named "Sohu" that is purpose-built for transformers. They have some bold claims that are impressive if true. I found a short discussion[2] you may find useful. [1]: https://www.etched.com/ [2]: https://www.lesswrong.com/posts/qhpB9NjcCHjdNDsMG/new-fast-t...
Comment linking /
on “AI's $600B Question”Submitted First playable AI model that generates open-world games as a story · 4 points
HN discussionYes, and startups do exactly that. Check out Etched https://www.etched.com/ where they made a Transformer specific GPU (basically a form of ASIC) where they bet that transformers would be the dominant GPU architecture for running AI / LLM workload.
Comment linking /
on “OpenTPU – An open-source AI accelerator, developed by AI”Submitted Frontier Inference Clusters as a story · 1 points
HN discussionEtched is a startup doing exactly this. https://www.etched.com/progress/frontier-inference-clusters
Comment linking /progress/frontier-inference-clusters
on “OpenTPU – An open-source AI accelerator, developed by AI”Do you mean something like this? https://www.etched.com/
Comment linking /
on “Why DeepSeek is cheap at scale but expensive to run locally”Congrats on the paper, very interesting. Can you opine on how the model will fare on hardware that is optimized for transformers? There is so much investment in accelerating the transformer arch[1][2], will xLSTM / sLSTM benefit as well, or will the hardware optimizations give transformers enough of an advantage that it’s hard to compete on general purpose hardware? 1. https://www.etched.com/ 2. https://www.embedded…
Comment linking /
on “xLSTM: Extended Long Short-Term Memory”We wouldn't build it ourselves, but there are several companies like Etched, Groq, and Cerebras working on purpose-built hardware for transformer models. Here's more: https://www.etched.com/announcing-etched
Comment linking /announcing-etched
on “Show HN: Lemon Slice Live – Have a video call with a transformer model”https://www.etched.com/announcing-etched I think there's another one but I can't remember the name of it. Also a bit further out is https://spectrum.ieee.org/superconducting-computer "Instead of the transistor, the basic element in superconducting logic is the Josephson-junction."
Comment linking /announcing-etched
on “Two new Gemini models, reduced 1.5 Pro pricing, increased rate limits, and more”Isn't Etched's Soho ASIC claimed to be much better than a GPU? https://www.etched.com/announcing-etched
Comment linking /announcing-etched
on “Microsoft CTO says he wants to swap most AMD and Nvidia GPUs for homemade chips”I’m not questioning Nvidia’s moat in gaming. But that’s pretty much irrelevant to its valuation. And sure, for small scale and research activities Nvidia makes sense, even for the long term. But that’s not where the money is at, either. The only way Nvidia can sustain its valuation is if it can win in the large scale production AI market long term, and that market turns out to be huge. I don’t really see how that ca…
Comment linking /
on “Nvidia Rides AI Wave to Pass Apple as Largest Company”Details from their technical memo at https://www.etched.com/announcing-etched ## How can we fit so much more compute on the silicon? The NVIDIA H200 has 989 TFLOPS of FP16/BF16 compute without sparsity. This is state-of-the-art (more than even Google’s new Trillium chip), and the GB200 launching in 2025 has only 25% more compute (1,250 TFLOPS per die). Since the vast majority of a GPU’s area is devoted to programmab…
Comment linking /announcing-etched
on “Sohu – first specialized chip (ASIC) for transformer models”Blog post: https://www.etched.com/announcing-etched I don't see a lot of detail on the actual architecture. Hard to evaluate if it makes sense. The idea of specializing in transformers has merit I think. It seems like this is only for inference, not training. Although inference accelerators are definitely important and potentially lucrative, I find it hard to get excited about them. To me the exciting thing is accel…
Comment linking /announcing-etched
on “Sohu – first specialized chip (ASIC) for transformer models”Official post: https://www.etched.com/announcing-etched
Comment linking /announcing-etched
on “Sohu: The First Transformer ASIC”They recently shipped their first rack to JS https://www.etched.com/progress/from-zero-to-one
Comment linking /progress/from-zero-to-one
on “AI chip startup Etched doubles valuation to $21B in under a month”
