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

S1: A $6 R1 competitor?

timkellogg.me

Post reply on HN
Print view View original

151–160 of 430 posts

Re: S1: A $6 R1 competitor?

#151

Post by ijidak »

sixothree wrote: post #82
jxdxbx wrote: post #69

Earlier quoted context omitted.

I hear this a lot, but what the hell. It's still computer chips. They depreciate. Short supply won't last forever. Hell, GPUs burn out. It seems like using ice sculptures as collateral, and then spring comes.

Year over year gains in computing continue to slow. I think we keep forgetting that when talking about these things as assets. The thing controlling their value is the supply which is tightly controlled like diamonds.

Honestly, I don't fully understand the reason for this shortage.

Isn't it because we insist on only using the latest nodes from a single company for manufacture?

I don't understand why we can't use older process nodes to boost overall GPU making capacity.

Can't we have tiers of GPU availability?

Why is Nvidia not diversifying aggressively to Samsung and Intel no matter the process node.

Can someone explain?

I've heard packaging is also a concern, but can't you get Intel to figure that out with a large enough commitment?

Re: S1: A $6 R1 competitor?

#152

Post by UncleEntity »

theturtletalks wrote: post #45

Deepseek R1 uses and wait and you can see it in the thinking tokens second guessing itself. How does the model know when to wait? These reasoning models are feeding more to OP's last point about NVidia and OpenAI data centers not being wasted since reason models require more tokens and faster tps.

From playing around they seem to 'wait' when there's a contradiction in their logic.

And I think the second point is due to The Market thinking there is no need to spend ever increasing amounts of compute to get to the next level of AI overlordship.

Of course Jevon's paradox is also all in the news these days..

Re: S1: A $6 R1 competitor?

#153

Post by ijidak »

syntaxing wrote: post #61
gessha wrote: post #56

Earlier quoted context omitted.

This is pure speculation on my part but I think at some point a company's valuation became tied to how big their compute is so everybody jumped on the bandwagon.

Matt Levine tangentially talked about this during his podcast this past Friday (or was it the one before?). It was a good way to value these companies according to their compute size since those chips are very valuable. At a minimum, the chips are an asset that acts as a collateral.

I asked this elsewhere, but, I don't fully understand the reason for the critical GPU shortage.

Isn't it because NVIDIA insists on only using the latest nodes from a single company (TSMC) for manufacture?

I don't understand why we can't use older process nodes to boost overall GPU making capacity.

Can't we have tiers of GPU availability some on cutting edge nodes, others built on older Intel and Samsung nodes?

Why is Nvidia not diversifying aggressively to Samsung and Intel no matter the process node.

Can someone explain?

I've heard packaging is also a concern, but can't you get Intel to figure that out with a large enough commitment?

(Also, I know NVIDIA has some capacity on Samsung. But why not go all out, even using Global Foundries?)

Re: S1: A $6 R1 competitor?

#154

Post by RobotToaster »

tyfon wrote: post #139
gessha wrote: post #56

Earlier quoted context omitted.

This is pure speculation on my part but I think at some point a company's valuation became tied to how big their compute is so everybody jumped on the bandwagon.

I don't think you need to speculate too hard. On CNBC they are not tracking revenue, profits or technical breakthroughs, but how much the big companies are spending (on gpus). That's the metric!

"But tulip sales keep increasing!"

Re: S1: A $6 R1 competitor?

#155

Post by aorloff »

syntaxing wrote: post #61
gessha wrote: post #56

Earlier quoted context omitted.

This is pure speculation on my part but I think at some point a company's valuation became tied to how big their compute is so everybody jumped on the bandwagon.

Matt Levine tangentially talked about this during his podcast this past Friday (or was it the one before?). It was a good way to value these companies according to their compute size since those chips are very valuable. At a minimum, the chips are an asset that acts as a collateral.

If you are a cloud provider renting them out

Otherwise you better keep them humming trying to find a business model because they certainly aren't getting any newer as chips

Re: S1: A $6 R1 competitor?

#156

Post by ALittleLight »

qup wrote: post #125
ALittleLight wrote: post #119

At 6 dollars per run, I'm tempted to try to figure out how to replicate this. I'd like to try some alternatives to "wait" - e.g. "double checking..." Or write my own chains of thought.

Like the ones they tested?

Yes, that is what "replicate" with my own ideas means.

Re: S1: A $6 R1 competitor?

#157

Post by janalsncm »

svantana wrote: post #49
swiftcoder wrote: post #22

> having 10,000 H100s just means that you can do 625 times more experiments than s1 did I think the ball is very much in their court to demonstrate they actually are using their massive compute in such a productive fashion. My BigTech experience would tend to suggest that frugality went out the window the day the valuation took off, and they are in fact just burning compute for little gain, because why not...

Besides that, AI training (aka gradient descent) is not really an "embarrassingly parallel" problem. At some point, there are diminishing returns on adding more GPUs, even though a lot of effort is going into making it as parallel as possible.

What? It definitely is.

Data parallelism, model parallelism, parameter server to workers, MoE itself can be split up, etc.

But even if it wasn’t, you can simply parallelize training runs with slight variations in hyperparameters. That is what the article is describing.

Re: S1: A $6 R1 competitor?

#158

Post by janalsncm »

> even the smartest people make hundreds of tiny experiments

This is the most important point, and why DeepSeek’s cheaper training matters.

And if you check the R1 paper, they have a section for “things that didn’t work”, each of which would normally be a paper of its own but because their training was so cheap and streamlined they could try a bunch of things.

Re: S1: A $6 R1 competitor?

#159

Post by pillefitz »

bloomingkales wrote: post #75

This thing that people are calling “reasoning” is more like rendering to me really, or multi pass rendering. We’re just refining the render, there’s no reasoning involved.

Which is related to multistage/ hierarchical/coarse-to-fine optimization, which is a pretty good way to find the global optimum in many problem domains.

Re: S1: A $6 R1 competitor?

#160

Post by eek2121 »

spamizbad wrote: post #149
sixothree wrote: post #82

Earlier quoted context omitted.

Year over year gains in computing continue to slow. I think we keep forgetting that when talking about these things as assets. The thing controlling their value is the supply which is tightly controlled like diamonds.

> Year over year gains in computing continue to slow. This isn't true in the AI chip space (yet). And so much of this isn't just about compute but about the memory.

From a per mm2 performance standpoint things absolutely have slowed considerably. Gains are primarily being eked out via process advantage (which has slowed down) and larger chips (which has an ever-shrinking limit depending on the tech used)

Chiplets have slowed the slowdown in AI, but you can see in the gaming space how much things have slowed to get an idea of what is coming for enterprise.

Post reply on HN