zachbee

7 ms·

1.
by zachbee 4mo ago
The interesting thing here isn't "how, logistically, is the Groq corporate entity able to raise more money?". That's straightforward. Rather, the interesting thing and the topic of most of the article is "how, after
2.
by zachbee 6mo ago
Their AES implementation uses old-school 2-share boolean masking [1], which has been shown to be insecure since 2005 [2][3]. A modern implementation would use domain-oriented masking [4], like OpenTitan does. Pretty bad look for Crossbar. [
3.
by zachbee 6mo ago
Their AES implementation uses old-school 2-share boolean masking [1], which has been shown to be insecure since 2005 [2][3]. A modern implementation would use domain-oriented masking [4], like OpenTitan does. Pretty bad look for Crossbar. [
4.
by zachbee 6mo ago
One thing that people rarely discuss about book publishing is a change to US tax law in the late 70s that meant that publishers couldn't write down the value of unsold inventory, but could write off that inventory by destroying it. Tha
5.
Why is it so hard for startups to compete with Cadence?(zach.be)
3 points by zachbee 1y ago|0 comments
6.
What Normal Computing is building(normalcomputing.com)
2 points by zachbee 1y ago|0 comments
7.
by zachbee 2y ago
It's a cute toy and a fun educational tool, but "computing for the future" seems like a bit of an overstatement.
8.
by zachbee 2y ago
hi, this is my article! thanks so much for the views, upvotes, and comments! :)
9.
by zachbee 2y ago
Author here -- I don't disagree! I actually noted this in the article: > Well, it turns out that LLMs are also pretty valuable when it comes to chips for lucrative markets -- but they won’t be doing most of the design work. LLM copi
10.
by zachbee 2y ago
I didn't get into this in the article, but one of the major challenges with achieving superhuman performance on Verilog is the lack of high-quality training data. Most professional-quality Verilog is closed source, so LLMs are generall
11.
by zachbee 2y ago
Yes, it would be a paid role. But I also don't necessarily want my job to be in jeopardy if my boss has a bad breakup.
12.
Ask HN: Would you be worried joining a startup where the founders are dating?
2 points by zachbee 2y ago|6 comments
13.
by zachbee 2y ago
If they're doing inference on edge devices, one challenge I see is protecting model weights. If you want to deploy a proprietary model on an edge AI chip, the weights can get stolen via side-channel attacks [1]. Obviously this isn&#x27
14.
by zachbee 2y ago
I'm curious what the motivation is here -- unfortunately, the dev blog is all in Chinese and I can't read it. If it's mostly to show a proof-of-concept of LLMs on a FPGA, that's awesome! But if this is targeting real-wor
15.
by zachbee 2y ago
I'm definitely not! I'm a hardware designer and I work with FPGAs all the time, for both work and for personal projects. Like with all things, there's a right tool for every job, and I think for modern DL algorithms like Tran
16.
by zachbee 2y ago
The big challenge when it comes to using FPGAs for deep learning is pretty simple: all of that reprogrammability comes at a performance cost. If you're doing something highly specific that conventional GPUs are bad at, like genomics re
17.
by zachbee 2y ago
If I was developing an AI app, I'd care about quality first before speed. And the open-source models just aren't as good as the closed ones.
18.
by zachbee 2y ago
Totally saw this one coming! [1] I think one major challenge they'll face is that their architecture is incredibly fast at running the ~10-100B parameter open-source models, but starts hitting scaling issues with state-of-the-art model
19.
by zachbee 2y ago
This research is awesome. And I think organic semiconductors are super promising, especially looking at the success of OLED. But the IEEE article is missing the biggest pitfall of this tech right now: it only works well at very low temperat
20.
by zachbee 2y ago
"Smart home" has really fallen flat across the board, from smart speakers to the Roomba. It would be great if Apple changed that, but I'm not optimistic. I just don't think people really want smart home devices all that
21.
by zachbee 2y ago
This doesn't appear to support any FPGAs other than their FPGA chiplet [1]. Also, like UncleOxidant said, the most complex parts of this toolchain are just existing open-source tools (Yosys and VPR). This seems like a useful toolchain
22.
by zachbee 2y ago
There are meaningful size and performance challenges to implementing post quantum crypto, but I don't think "oh no, HTTPS handshakes are 20% slower!" is something we should actually worry about. The real worry is all of the s
23.
What's the difference between Extropic, Normal Computing, and D-Wave?(zach.be)
1 points by zachbee 2y ago|0 comments
24.
by zachbee 2y ago
I'm really skeptical of Eteched. They make claims about how their chips are "optimized for transformers", as if Nvidia's Tensor Cores aren't optimized for those workloads too. Plus, if they really are only good at t
25.
by zachbee 2y ago
I'm a hardware security engineer, and I've actually used similar tools before! It's nifty, but overall, hardware security tooling is unfortunately seriously lacking. When you try to apply masking or hiding schemes to prevent
26.
by zachbee 2y ago
From a technical perspective, moving a mouse with your mind isn't entirely new: https://elifesciences.org/articles/18554 Personally, I think the most exciting part of Neuralink and other companies working on BCIs
27.
by zachbee 2y ago
There are so many different tools to help design and route PCBs now. You have DeepPCB, Flux, and Quilter doing auto-routing, and SKiDL, CuFlow, and JITX doing programmatic schematics (and also sometimes auto-routing?). To be honest, I don&#
28.
by zachbee 2y ago
When they originally announced tiny corp and the tinybox, the entire pitch was that AMD hardware were great, but their software was bad. [1] Now they're giving up on AMD hardware because the software is bad. Wasn't the whole point
29.
by zachbee 2y ago
100% right. GF is no longer gunning for leading-edge nodes. They've made their niche in mature nodes for low-power chips and emerging memories (FeFETS, MRAM, RRAM, etc). If you want to get leading edge nodes on American soil, Intel is
30.
by zachbee 2y ago
I dunno man, I think saying "this is a personal project of mine" and linking to an open source github repo implies that it's a casual side project. I never made any performance claims, it's just for fun. I'm not try

More ›