[yuan43]

> I [nvidia CEO] don’t think we could have seen it [massive effect of Ethereum merge on bottom line]. I don’t think I would’ve done anything different, but what I did learn from previous examples is that when it finally happens to you, just take the hard medicine and get it behind you…We’ve had two bad quarters and two bad quarters in the context of a company, it’s frustrating for all the investors, it’s difficult on all the employees.

This is not confidence inspiring. It was obvious that the Etherum merge would affect the bottom line in a big way. Why this professed ignorance? Does it have to do with the fact that to admit that it was visible a mile away would have been to admit the deep reliance the company had come to have on the short-term Ethereum mining boom?

[sophrocyne]

Couple of comments to this point suggest that it couldn't have been predicted since the official timing of the merge was only announced in 2022, and silicon supply chain requires planning in advance of that.

But that point is ignorant of this truth - Proof-of-stake has been on the roadmap since ~2017 if not earlier.~ Edit: 2016 - Thanks friend! :)

I think the reality is that the impact of Ethereum on Nvidia's business was not fully appreciated, and that 'veil of ignorance' may well have been intentional. They never truly served the crypto market directly (e.g., there wasn't really a "miner" line of cards), and as a result didn't do the due diligence to understand how those customers played into their business performance and strategy. Or they did, and just really underestimated the Ethereum devs on ever making the merge happen. But I lean towards the first.

Either way, I think that with crypto in the rearview, I'm actually more confident in their leadership team. They seem better suited to gaming and AI.

[paulmd]

> But that point is ignorant of this truth - Proof-of-stake has been on the roadmap since 2017 if not earlier.

it's been on the roadmap since 2016. That's actually still a problem though, a perpetually-rolling-deadline is effectively worse than not having a deadline at all.

Was NVIDIA just supposed to cut production for the last 6 years in anticipation of something that was continuously pushed back 6 months every 6 months? That's not a reasonable expectation.

[sophrocyne]

As an observer, I never got the sense that there were strong commitments being made on timelines until A) beacon chain was live (running in parallel), and B) testnets started getting merged successfully.

The moment of Genesis for the beacon-chain started a clock that Nvidia should have been paying attention to, and I think would have given them plenty of time to foresee the present situation.

[jonas21]

Regardless of whether they foresaw it or not, what should they have done differently?

[sophrocyne]

Great question, and perhaps at the point where decisions were being made, it'd be hard to argue a different path internally (hindsight being 20/20).

However, it seems clear that the business built both insane prices and the crypto lockup of devices (whether explicitly, or implicitly) into their forecasts for the business. They didn't have a good pulse on the actual demand/usage of their product, and when that usage pattern would shift.

The path they're taking right now, specifically regarding pricing towards & serving higher-end enthusiasts with newer products, makes sense while the used inventory gets cycled around the lower end of the market.

From a product perspective, I don't have any useful opinions to share because I'm not in hardware, and I don't have the information set they're operating from internally. But, they should have hoovered up as much cheap capital as they could while their stock price was high and the going was good to make the next period of heavy investments (to be fair, shares outstanding did grow, just not by a ton, %-wise, and they have a fair bit of cash on the balance sheet)

https://www.wsj.com/market-data/quotes/NVDA/financials/quart... is painful to see, and I don't foresee it getting better in the next year.

[_carbyau_]

> The path they're taking right now, specifically regarding pricing towards & serving higher-end enthusiasts with newer products, makes sense while the used inventory gets cycled around the lower end of the market.

This approach might take a beating if AMD is willing for a price war. And AMD can because of their chiplet approach being cheaper.

I will buy nVidia purely because of dabbling interests in ML images... so a price war might not be so effective thanks to a technical moat.

[paulmd]

people absolutely SCREAMED a year ago when there was a rumor going around that NVIDIA was pulling back on new chip starts, it was going all around that it was a plan to "spike prices during the holidays".

In the end 2021Q4 shipments were actually up according to JPR, of course. But people were mad, and I still see that MLID article brought up as proof that NVIDIA was deliberately trying to "worsen the shortage" and "spike prices during the holidays".

https://www.notebookcheck.net/Nvidia-allegedly-halting-RTX-3...

https://www.jonpeddie.com/press-releases/q421-sees-a-nominal...

Now, what MLID may not really know, is that wafer starts typically take about 6 months, so if he's hearing about reduced starts in October, it's probably more like NVIDIA is pulling back on expected Q1/Q2 production... which indeed did come down a bit.

But as to the public reaction... people were fucking mad about any sign of pulling back on production. People are just unreasonably mad about anything involving NVIDIA in general, every single little news item is instantly spun into its worst possible case and contextualized as a moustache-twirling plan to screw everyone over.

Like, would it have really been a bad thing to pull back on chip starts a year ago? That actually looks pretty sensible to me, and gamers will generally also suffer from the delay of next-gen products while the stockpile burns through anyway.

It's nowhere near the "sure miners may be annoying, but deal with it for 6 months and then we all get cheap GPUs and everyone holds hands and sings" that LTT and some other techtubers presented it as. Like, yeah, if you want a cheap 30-series card at the end of its generation/lifecycle great, but, you'll be waiting for 4050/4060/4070 for a while. Even AMD pushed back their midrange chips and is launching high-end-only to allow the miner inventory to sell through.

And people hate that now that they've realized the consequence, but they were cheering a year ago and demanding the removal of the miner lock / etc. More cards for the miners! Wait, no, not like that!

It's just so tiresome on any article involving NVIDIA, even here you've got the "haha linus said FUCK NVIDIA, that makes me Laugh Out Loud right guys!?" and the same tired "turn everything into a conspiracy" bullshit, constantly.

[cma]

Probably not much, but different investor guidance.

[doikor]

Nvidia did (does?) have a miner line of cards (CMP HX). Though they were mainly their server cards that failed QA but could still work as a miner.

https://www.nvidia.com/en-us/cmp/

These also had models that you can’t find anywhere on Nvidia website like HX170 that is basically a A100 with less memory

A lot of miners preferred consumer cards though as those can be sold to gamers once the bust comes again (and with crypto it always will every few years)

[sophrocyne]

Completely unaware of this line. Thanks for sharing.

Frankly, doesn’t paint a better picture for an Nvidia…

[doikor]

NVidia (and AMD for their own line) did not talk about these in the public much as they were really bad PR during the worst moments of the gpu shortages.

To the point that they never even updated the site for the later/better models (like the already mentioned 170HX). Only mentions of it you can find are the sellers you have to go to to get one (NVidia did not sell these directly to anyone as far as I know)

https://www.viperatech.com/product/nvida-cmp-170hx-professio...

[paulmd]

AMD developed a line of mining cards as well as well, and also some PS5-based mining rigs from asrock (it's a PS5 apu reject on a motherboard with VRAM...) that utilized AMD BC-250 mining processors.

https://wccftech.com/amd-rdna-crypto-mining-graphics-cards-r...

https://www.tomshardware.com/news/amd-navi-12-bc-160-mining-...

https://www.techspot.com/news/93980-14800-asrock-mining-rig-...

We regret to inform you, the milkshake duck was selling mining cards too.

Welcome to capitalism, nobody cares about consumers all that much.

[throw101010]

> They never truly served the crypto market directly (e.g., there wasn't really a "miner" line of cards)

There were deliberate, and miserably failed, attempts to make lines of cards that could not be used for mining, while in parallel keeping their non-limited lines in production, making them the defacto miners' lines.

So yes they did know about it and tried to address it by catering to both markets, but were unable to do it correctly.

[izacus]

What would "correct" approach look like?

[sophrocyne]

ASICs probably.

The point you responded to is also fair; the fact they tried to lock miners out was a de facto acknowledgment that crypto had an impact on their sales, and is more critical evidence they handled it poorly.

[asciimike]

This is the route Intel is going down: https://bitcoinmagazine.com/business/intel-launches-new-bitc...

Time will tell if it works out (different chain, etc., etc.)

[opportune]

I think Huang does not want to draw investors’ attention to crypto because he doesn’t want people to equate Nvidia’s performance as a company with crypto performance. He doesn’t want Nvidia to just be a crypto company.

At the same time, he also definitely wants to cash in on any future crypto booms, because they are lucrative.

It is best for him to take a position that mostly ignores crypto. I think he legitimately doesn’t want crypto to be the future of Nvidia and doesn’t want to build for that use case, nor does he want to be financially reliant on it, but there is also no point in him talking shit or spreading doom about crypto when he can just shut up and still sell gpus.

[swalsh]

Nvidia doesn't have a role in any future crypto boom, unless it's being used for analytics or AI. All modern chains use PoS.

[Jorge1o1]

Except for every single VR “metaverse” project (of which there are thousands) and all of the graphics hardware and software that goes into them.

[kranke155]

They do now but they didn’t just a few days ago.

[vineyardmike]

I think we forget that Silicon manufacturing is planned a lot farther out than Silicon shopping.

They were likely trying to make TSMC purchase orders in the start of the pandemic, before a crypto boom. They also tried to handicap their GPUs wrt crypto. They likely didn’t expect the absolute shit show of a chip shortage (because who predicted or understood the pandemic early).

The rest of the market was desperate, and they probably expected it to be more robust than it ended up being. The merge would have been so far away at the time that they wouldn’t predict if it would happen at all nevermind when.

[SkyMarshal]

Many folks in crypto expected those Ethereum-mining GPU farms to just switch to some other GPU-minable cryptocurrencies. It wasn't a certainty all those farms would just close up and dump their GPUs on the market en masse. But Fed interest rate policy hitting at the same time, driving down the crypto market across the board (and all other risk assets), may have unexpectedly changed the ROI calculation there and resulted in the dump.

[Tenoke]

Many poorly educated folk maybe. All other chains were already less profitable and had a small fraction of Ethereum's hashrate. Even 5% of ETH's hashrate moving to them is plenty to make them unprofitable for most. This was never a likely outcome.

[paulmd]

> Many folks in crypto expected those Ethereum-mining GPU farms to just switch to some other GPU-minable cryptocurrencies.

This was a pretty common take but if you did the math Ethereum had about 90% of the GPU-mining market (by hashrate) so it was obvious the profitability was going to tank on those other currencies as soon as Ethereum switched.

In the long run yes, there will probably be another big spike in another cryptocurrency that starts another GPU boom. But it's not magic where one instantly springs up to absorb all the ethereum hardware at equivalent profitability.

A GPU crash was inevitable regardless of the interest rate drop hitting at the same time.

[swalsh]

I hoped there would be a rise in proof of work like chains, where in the work was something useful like training an AI or brute forcing a hard but useful problem. Like a SETI@Home, but paying crypto for successful solutions as opposed to relying on altruism.

[omegalulw]

It's hard to pull this off, if not impossible. A key attribute of proof of work systems is that the difficulty should be dynamically adjustable and that everyone has perfect consensus on what "work" is. Doing meaningful work, while admirable, puts the owners of those projects in control of defining "work" and adjusting difficulty, i.e., people in the loop. That's not trustworthy from a currency POV, no matter who the people are.

[asciimike]

There are a few of these type of things, e.g. RNDR token (https://rendertoken.com) and rent a flop (https://rentaflop.com) in rendering, and golem (https://www.golem.network) and sonm (https://sonm.com/) in the "general purpose computing on the blockchain"

[kmeisthax]

The problem is that in order for work to be useful to secure a blockchain, the work needs to have special properties. Specifically...

- It must be difficult to compute. It should be possible to get exponential increases in difficulty out of problem size increases.

- It must be easy to verify. It needs to be verifiable solely through data that can be copied on-chain, with no reference to external sources[0], and an easily-understood computer program needs to exist to verify that data.

- There must be no ambiguity in interpretation or valuing of the work. Every node should be able to look at the data and agree that work has been done just by having a connection to other nodes and valid consensus algorithms.

Most PoW is hashing-based, there are a few weirdos like Primecoin that use prime searches as a PoW; but these are both forms of numerical work. Number problems are easy, they have neat exponential-blowup properties, we can validate that the work has been done far quicker than they take to produce, and we only have one axis of value: how hard is it to produce that work? ML training and SETI@Home don't work that way. Verifying that someone did training requires testing the entire training set, and verifying that a Fourier transform worked involves just running the inverse, which doesn't have that same performance gulf between verification and work.

Furthermore, useful work has a value ambiguity problem. What ML training sets are considered valid proof of work, and how do we put them in a nice total ordering so that we can resolve chain splits? Who gets to inject signal data from SETI@Home into SETICoin? The nature of the work is not value neutral; if we just allow people to say "here, I trained an AI on a training set and it got better", then that allows deliberate construction of "easier" problems that look hard. Imagine like being able to unilaterally change Bitcoin from an ASIC-friendly hash function into one that is more akin to, say, Burstcoin and Chia. That would be a major trust problem for the chain.

[0] This is why most smart contract proposals fail. If we want to have the network, say, track container ships, we can't have every node operator actually board every ship to punch in data and make sure it was done correctly. There has to be trust, which disqualifies us from using an unpermissioned blockchain.

[AceJohnny2]

As the article says, the timing on the Ethereum dropping proof-of-work was shorter than Nvidia's production pipeline.

[colechristensen]

I think it's more like there are so many games at play as CEO in that position that anything but vague denial would be far more trouble than it's worth. Anything you say is going to attract a lot of criticism so the only thing you can say is the least damaging one.

In other words, most public statements are mostly nonsense engineered for response and have only a casual association with the truth.

[modeless]

They have consistently underestimated the effects of crypto, it's been screwing up their demand forecasts for a long time. I think what happened was they had all these efforts to prevent miners from buying cards so gamers could buy them instead, and they thought they were successful. So they attributed strong demand to gaming, but they were actually failing and miners were still buying all the cards. I don't know why they thought they were successful...

[ashafer]

I'm guessing it is less about proof of stake and more about the fact that all of crypto is in the gutter and it's not as economical to mine anymore. I would guess that is the big core issue that is enhanced by things like proof of stake. Kind of a perfect storm when paired with inventory issues and crazy inflation. I think that storm is what he is saying they couldn't have predicted.

[KaoruAoiShiho]

It was sudden and couldn't have been predicted. Ethereum ended PoW just this month but the GPU crash was 7 months ago. In reality the PoW transition had nothing to do with the GPU crash, it was the end of WFH and the crypto decline caused by the russian invasion that resulted in the GPU crash.

[mccorrinall]

> crypto decline caused by the russian invasion

What? One of the biggest recent decline in crypto happened when LUNA foundation dumped multiple billions as BTC in order to keep terra stable (didn’t work out). The other dump is caused because borrowing money for leverage won’t be as cheap as it was for at least the next 4 years (taking bloomberg projections of Fed rates here).

How is the invasion related to the dump at all?

[KaoruAoiShiho]

These are all caused by general asset declines due to the russian invasion, inflation spiking, and fears of fed reaction. General asset declines put pressure on crypto whose more dramatic moments is really traced back to the russian invasion.