Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

How AlphaChip transformed computer chip design

deepmind.google

Post reply on HN
Print view View original

171–180 of 215 posts

Re: How AlphaChip transformed computer chip design

#171

Post by jebarker »

abc-1 wrote: post #92
jebarker wrote: post #67

Earlier quoted context omitted.

Because chip placement and the design of neural network architectures are entirely different problems, so this solution won't magically transfer from one to the other.

And AlphaGo is trained to play Go? The point is training a model through self play to build neural network architectures. If it can play Go and architect chip placements, I don’t see why it couldn’t be trained to build novel ML architectures.

Sure, they could choose to work on that problem. But why do you think that's a more important/worthwhile problem than chip design or any other problem they might choose to work on? My point was that it's not trivial to make self-play for some other problem work, so given all the problems in the world why did you single our neural network architecture design? Especially since it's not the transformer architecture that is really holding back AI progress.

Re: How AlphaChip transformed computer chip design

#172

Post by anna-gabriella »

smokel wrote: post #156
clickwiseorange wrote: post #81

Earlier quoted context omitted.

Oh, man... this is the same old stuff from the 2023 Anna Goldie statement (is this Anna Goldie's comment?). This was all addressed by Kahng in 2023 - no valid criticisms. Where do I start? Kahng's ISPD 2023 paper is not in dispute - no established experts objected to it. The Nature paper is in dispute. Dozens of experts objected to it: Kahng, Cheng, Markov, Madden, Lienig, Swartz objected publically. The fact that Ka…

Wow, you seem to be pretty invested in this topic. Care to clarify?

[flagged]

Re: How AlphaChip transformed computer chip design

#173

Post by therealcamino »

hinkley wrote: post #158
therealcamino wrote: post #113

Earlier quoted context omitted.

EDA software has long allowed trading off power, delay, and area during optimization . But TSMC doesn't produce those tools, as far as I'm aware.

https://www.tsmc.com/english/dedicatedFoundry/oip/eda_allian... They don’t produce but they are tailored for them just the same. “We have” doesn’t have to mean “we made”. They don’t say it as such here but elsewhere they refer to the IP they can make available, which can also be made in house or cross licensed and still count as “we have”.

Used in that sense, the same software could be called Samsung's and Intel's and any other foundry's, since it is qualified for use with those processes as well. But that's not really the main point I was making, which was that there have been 20+ years of cooperative effort in both process design and EDA software to optimize for power and trade it off against other optimization goals. While there are design and packaging approaches that are only coming into use because of "end of Moore's law, what do we do now" reactions, and some may have power implications, power optimization predates that by a good while.

Re: How AlphaChip transformed computer chip design

#174

Post by smokel »

anna-gabriella wrote: post #172
smokel wrote: post #156

Earlier quoted context omitted.

Wow, you seem to be pretty invested in this topic. Care to clarify?

[flagged]

As someone unfamiliar with the topic but trying to piece together information, I have to admit that they do a better job at convincing me of the potential of reinforcement in chip design.

As with most, if not all, applications of reinforcement learning, there are always traditional algorithms that outperform it. But that does not mean that the approach lacks promise, or is at least interesting.

Sure, the paper might have polished up some results, but if that is the case, it is better addressed through the appropriate channels. Engaging in public criticism does not build too much trust, at least not with this curious observer.

Re: How AlphaChip transformed computer chip design

#175

Post by hulitu »

pelorat wrote: post #101
hulitu wrote: post #84

Earlier quoted context omitted.

They don't. You cannot compare reality (Cadence, Synopsys) with hype (Google).

So you're basically saying that Google should have used existing tools to layout their chip designs, instead of their ML solution, and that these existing tools would have produced even better chips than the ones they are actually manufacturing?

> So you're basically saying that Google should have used existing tools to layout their chip designs, instead of their ML solution

Did they tested their ML solution ? With real world chips ? Are there any "benchmarks" that show that their chip performs better ?

Re: How AlphaChip transformed computer chip design

#176

Post by AshamedCaptain »

gabegobblegoldi wrote: post #170
AshamedCaptain wrote: post #166

Earlier quoted context omitted.

> Yet they chose an obscure problem in chip design It is not obscure (in chip design). If anything it is one of the most easily reachable problems. Almost every other PhD student in the field has implemented a macro placer, even if just for fun, and there are frequent academic competitions. A lot of design houses also roll their own macro placers since it's not a difficult problem and generally adding a bit of knowle…

Sorry. I meant obscure relative to the large space of combinatorial optimization problems not just chip design. Most design houses don’t write their own macro placers but customize commercial flows for their designs. The problem with macro placement as an RL technology demonstrator is that to evaluate quality you need to go through large parts of the design flow which involves using other commercial tools. This makes…

> Most design houses don’t write their own macro placers but customize commercial flows for their designs.

Most I don't know, but all the mid-to-large ones have automated macro placers. Obviously, the output is introduced into the commercial flow, generally by setting placement constraints. The larger houses go much further and may even override specific parts of the flow, but not basing it on an commercial flow is out of the question right now.

> The problem with macro placement as an RL technology demonstrator is that to evaluate quality you need to go through large parts of the design flow which involves using other commercial tools.

Not really, not any more than any other optimization such as e.g. frontend which I'm more familiar with. If you don't want to go through the full design flow (which I agree introduces noise more than anything else), then benchmark your floorplans in some easily calculable metric (e.g., HPWL). Likewise, if you want to test the quality of some logic simplification _in theory_ you'd have to also go through the entire flow (backend included), but no one does that and you just evaluate some easily calculable metric e.g. number of gates. These distinctions are traditional more than anything else.

Academic macro placers generally have limited access to commercial flows (either due to licensing issues or computing resource availability) so it is rather common to benchmark them in other metrics. Google paper tried to be too smart for its own good and therefore incomparable to anything academic.

Re: How AlphaChip transformed computer chip design

#177

Post by anna-gabriella »

smokel wrote: post #174
anna-gabriella wrote: post #172

Earlier quoted context omitted.

[flagged]

As someone unfamiliar with the topic but trying to piece together information, I have to admit that they do a better job at convincing me of the potential of reinforcement in chip design. As with most, if not all, applications of reinforcement learning, there are always traditional algorithms that outperform it. But that does not mean that the approach lacks promise, or is at least interesting. Sure, the paper might…

Hm... I also had to piece things together and agree that Google PR is pretty slick. The approach had promise 3-4 years ago, but the science seems clear now. Google is avoiding tests on shared chip designs but claims a breakthrough. There is no breakthrough as everyone is still using Cadence or Synopsys software tools.

Maybe you can make RL work for chip design at some point, but if the paper "polished up some results", why is it still getting any respect? You are right about "appropriate channels", that's what Chatterjee and Kahng tried, but Chatterjee was fired by Google as a whistleblower (red flag!) while Kahng is getting flak even these comments (another red flag!). Where would you look next as an independent observer?

Re: How AlphaChip transformed computer chip design

#178

Post by anna-gabriella »

smokel wrote: post #156
clickwiseorange wrote: post #81

Earlier quoted context omitted.

Oh, man... this is the same old stuff from the 2023 Anna Goldie statement (is this Anna Goldie's comment?). This was all addressed by Kahng in 2023 - no valid criticisms. Where do I start? Kahng's ISPD 2023 paper is not in dispute - no established experts objected to it. The Nature paper is in dispute. Dozens of experts objected to it: Kahng, Cheng, Markov, Madden, Lienig, Swartz objected publically. The fact that Ka…

Wow, you seem to be pretty invested in this topic. Care to clarify?

Reposting, as someone is flagging my comments. > People in the know are following this topic - big-wow surprise!

Re: How AlphaChip transformed computer chip design

#179

Post by joshuamorton »

anna-gabriella wrote: post #147
Workaccount2 wrote: post #51

Earlier quoted context omitted.

To be fair, some of these criticisms are a few years old. Which normally would be fair game, but the progress in AI has been breakneck. Criticism of other AI tech from 2021 or 2022 are pretty dated today.

Dated or not, if half of the criticisms are right, the original paper may need to be retracted. No progress on RL for chip design was published by Google since 2022, as far as I can tell. So, it looks like most if not all criticisms remain valid.

> No progress on RL for chip design was published by Google since 2022, as far as I can tell.

This makes sense given that both authors of the paper left Google in 2022. And one no longer seems to work in the chip design space, plausibly because of the bullying by entrenched folks.

Then again, since rejoining Google the other author has produced around one patent per month in chip design with RL in 2023 and 2024, so perhaps they feel there is a marketable tool here that they don't want to share.

Re: How AlphaChip transformed computer chip design

#180

Post by anna-gabriella »

FrustratedMonky wrote: post #163
anna-gabriella wrote: post #149

Earlier quoted context omitted.

I can tell you get excited by SciFi, that's where Google's work belongs - people have been unable to reproduce it outside Google by a long shot.

Alpha-GO was not sci-fi. And that was 2016 Protein Folding? That was against a defined data set and other organizations. Nobody can re-produce? Isn't that the definition of a competitive advantage? They are building something others can't, and that is bad? That is what companies do.

We are discussing AlphaChip in 2024, not AlphaGo from 2016. I don't know much about protein folding (there were some controversies there, but that's not relevant). Neither of these has been related to product claims.

As for "nobody can re-produce", no, that's not the definition. Imaginary things are not competitive advantage. They are exaggerating, and that's bad. But yeah, that's what companies do, you are right.

Post reply on HN