Hacker News Reader

Comment by jenny91 | Hacker News Reader

NVIDIA knew a lot of their GPUs were bought for crypto mining, but attributed those sales instead to growth in gaming, potentially misleading some investors (maybe they wanted to invest in gaming, not the very volatile crypto industry).

This is the SEC doing their job, giving a public company a reminder to be honest about their disclosures. The fine is insignificant and nobody is in trouble.

It has nothing to do with whether people could've read between the lines, whether NVIDIA has a duty to know what their products are used for, or anything like that. Just that they knew what was going on and told a small lie to bolster their finances. You should be able to read company financial disclosures and trust them.

View on HN

My only problem here is once again the fine is meaningless. $5.5 million is couch change for Nvidia. The fines are supposed to deter these actions. This has no effect whatsoever.

The fine should be proportionate to the severity of the offense and the magnitude of the misbegotten gains.

In this case, the severity was fairly low: they didn't lie about what they were selling or how much they were selling, but they did purposefully omit some material information about their customers. If you were convinced that cryptocurrency is the future, then NVidia not mentioning cryptominers should have reduced your forecast of their earnings. If you were convinced that cryptocurrency will die any day now, then NVidia's omission should have increased your forecast for them.

But how much money did people lose? It's hard to say. This should be enough money that NVidia's investor relations people make sure not to do it again, but not so much money that NVidia can't recover.

In a hypothetical where NVidia told lies in order to manipulate its stock price, a much much larger fine should be levied.

It is not low impact. It makes a huge difference from an investor point of view if increased revenue comes from gaming (evergreen) or crypto mining (trend at best, hype).

SEC failed to disclose the actual numbers, thus righting the wrong Nvidia has done to investors. And 5.5M is a ridiculous fine for something that the insiders benefited from multifold. They missed at least one 0. Someone should have been fired over this.

I would be curious if those insiders that made the decision have sold shares that were inflated.

That’s not entirely true. A fine from a regulator, no matter how small, contains the threat that a much bigger punishment will come if you are a repeat offender. That includes trading suspensions:

https://www.sec.gov/files/tradingsuspensions.pdf

https://www.sec.gov/litigation/suspensions.htm

Fines against public companies are among the least effective deterrents I can possibly imagine. The officers make the mistake/misrepresentations, and the shareholders take the hit. If the government really wanted to deter these actions, they'd fine the officers making the 'wrong' statements.

This is a terrible idea. Mistakes of this magnitude are often the result of multiple people making mistakes - so blaming a specific person is wrong.

Not to mention that this would force a culture of attributing blame to individuals which is toxic, encourages people to hide mistakes and when mistakes happen the first thing everyone would scramble to is to find a scapegoat. This is objectively worse for both the company and the shareholders.

The right approach for the government is to appropriately penalize the companies. The right approach for companies is to put in processes that make sure no single person is even in a position to make mistakes like this.

>"The right approach for the government is to appropriately penalize the companies."

The government isn't penalizing the companies, it's penalizing the shareholders, who are the ones that suffered the inadequate disclosures!

>"Mistakes of this magnitude are often the result of multiple people making mistakes - so blaming a specific person is wrong."

Then either split the fine amongst the responsible parties, or just issue some sort of order that requires changes (such as some sort of audit) without a fine. If you can't find any specific person who did something clearly wrong, it's probably best to avoid blaming 'the company' (which is just a collection of individuals).

What makes you think the SEC can find the ‘guilty’ parties and that the company (i.e. possibly the very same people) would willingly cooperate with them and not just find a scapegoat to push under the bus? And just repeat this process every time something bad happens.

The point of the fine is to disincentive the company (and also other companies) from doing the same thing again, unless criminal behavior (i.e. fraud) was involved that’s purely an internal matter for the company (they may fire the people who supposedly did this or do whatever they think is necessary).

If the SEC or DoJ can't figure out who's done something wrong, they should probably refrain from any prosecution. I am not sure what the 'point of the fine' is, but it seems like most companies just treat it as a cost of doing business, that they pay before moving on with whatever they were doing.

If you think it's difficult to figure out who did what wrong, and that the companies are just scapegoating people, then the SEC implementing fines that cause the company to fire people is likely causing unjust firings.

Why not punish both? Humans get away with actions under the flag of a country, business, or organization. It makes no sense. But it also makes no sense to let the overlying structure get away with it. Hence my question: why not punish both? Yes, I do understand more than 1 person is responsible. But that's the flak you receive when you are high up in the hierarchy. You end up having responsibility. And with responsibility comes repercussions if you don't follow the rulebook (except in the corporate world, of course).

AMD and NVIDIA's numbers for crypto sales have always been dodgy. Like in 2017-2018 I think they said that crypto sales (including gaming cards used for crypto) made up like 5% or 10% of their revenue, which is total bullshit. I assume they had a range of potential numbers internally (there is always a range) and they picked the lowest possible number to avoid spooking investors.

There is very very little gaming demand for these cards at 2-3x MSRP. Yeah, there's a few, but not many. As mining demand has tapered off this quarter, you've seen prices absolutely plummet, and if there was truly organic gaming demand for cards at those prices, that wouldn't have happened.

When prices get super elevated (that 2-3x MSRP range), IMO the realistic number is at least 30% if not more like 50% of total sales are crypto-related.

Some caveats:

bear in mind that OEM sales are a huge thing. Something like half the cards on the market will never even hit a distribution channel. So even if nearly-100% of aftermarket sales go to miners, there's a significant number being moved in prebuilts. That said, at the height of crypto mania, miners will buy prebuilts and strip them of their GPUs and resell them on ebay/craigslist/facebook, so OEM sales are not 100% crypto-unrelated either.

I am taking that into account when I am guessing 30-50% - I think the number for retail sales are >75% if not >90% miner sales when that happens, and it's really only the OEM sales that hold things down to 30-50%. There are very few actual consumers willing to pay the $2400 that 3080s were going for at their peak, that's triple MSRP, and prices immediately plummeted when the LHR was created, because prices are driven by hash rate and expected crypto profit, not gaming sales.

AMD more or less didn't participate in crypto this time around though, because they simply didn't make any GPUs. 6800 and 6700XT are reasonably competent miners - about 3060 Ti/3070 performance, which is undesirable in a relative sense compared to Ampere and RDNA1 but that's still twice the hashrate of a RX 480, miners still use those cards very happily, and RDNA2 is more efficient because it's 7nm. But it took over a year after launch for the first RDNA2 card to show up on the Steam Hardware Survey - at 0.15% marketshare. Meanwhile Ampere was around 10% marketshare at that time. You can go back and look at the historical Steam results using the Wayback machine, or here's a contemporary article (you don't have to like Tom's, the numbers are from Steam):

https://www.tomshardware.com/news/rtx-3090-outsells-all-rx-6...

(If anything this probably understates it since miners did go after NVIDIA cards preferentially. It’s likely more like a factor of 20x in total shipments, vs the Steam numbers. I guess that’s where you go to the JPR numbers tho.)

Basically, AMD did not make hardly any GPUs because console sales were unexpectedly high and those contracts got first priority, at one point there were reports that console sales were eating 80% (!) of AMD's wafer supply. The remaining wafers went to CPUs since even a consumer CPU is ~10x more profit per wafer than a GPU - it makes complete sense to maximize their revenue on the remaining wafers after consoles got the lion's share.

https://www.hardwaretimes.com/in-q4-amd-supplied-a-total-of-...

RDNA2 was the paperiest of paper launches. They delayed it as long as they could, and did a token paper launch to satisfy investors. I think the term "paper launch" is overused - it doesn't mean "I can't buy one on launch day" or even "I can't get one the first month" - Pascal took 6 months for inventory to normalize, 1080 was unobtanium until about November 2016 and even the 1070 was hard to get for a few months. Someone I know used the term "holodomor launch" at the time (cringe). But RDNA2 was a true paper launch, in the sense of only existing to satisfy commitments made to investors. AMD had no intention of producing serious quantities of it when they could be making 10-50x the money per wafer selling CPUs (and wafers are their bottleneck scaling profit). 6700XT and 6600XT/6600 were the only cards that shipped in any volume and those lower-end cards didn't launch until almost a year after the initial RNDA2 cards.

Not saying that wasn't the right decision in a financial sense, of course. Investors want them to make money, and AMD was very very wafer-bottlenecked. But the GPUs were launched just to say they launched.

But in a backhanded way, that was a good strategy this time around. No GPUs manufactured means very little crypto exposure. That's one solution to crypto boom-and-bust I guess - find a new market and don't sell GPUs anymore. NVIDIA is very, very subject to the boom/bust cycle of crypto and unlike AMD or Intel they don't have diversity of other products to smooth things out (not just other segments like enterprise/automotive, but other products like CPUs and mobile SOCs, etc).

AMD did participate last time though, and they had a huge inventory overhang that took almost a year to clear just like NVIDIA did. Not sure where people got the idea they didn't, but they discussed it openly in their late-2018 and early/mid-2019 financials calls... Here's a random one from that period I pulled up, ctrl-f "inventory" and look for yourself.

https://www.fool.com/earnings/call-transcripts/2019/01/30/ad...

Also, the bit about "NVIDIA selling cards directly to miners!" ended up being fake and based on tech media wildly misinterpreting a paywalled RBS article (that none of them wanted to pay for lol) that used an estimate of crypto network hash growth to estimate crypto gpu sales... after some massaging the line "miners bought $175m of NVIDIA GPUs based on hashrate estimates" became "NVIDIA sold $175M of their GPUs directly to miners outside retail channels" which simply isn't true.

Fake spin: https://www.kitguru.net/tech-news/featured-announcement/leo-...

Actual article: https://twitter.com/dylan522p/status/1332502890104188929

This is a great example of what I reference elsewhere in the thread, about how there's a clique that likes to manufacture fake or marginal rumors about NVIDIA, because a lot of people have viscerally negative emotional reactions to NVIDIA and it gets a lot of attention and clicks.

Mining card sales (not gaming cards used for mining, but actual dedicated mining cards) were a thing, but a very small thing overall (I'd guess less than 5% revenue as a segment, with gaming running around 50%). And many mining cards are still sold through partners or OEMs like HP... I used to see the EVGA cards coming through b-stock back in 2019.

Now in contrast partners were absolutely selling gaming cards in bulk direct to miners. Miners pay a premium for access to the cards, and partners require them to give up their warranties, so it's an advantageous deal for them, but as far as I know NVIDIA does not sell direct.

Just some baseline stuff that I often see come up in the nvidia/crypto discussions that is wrong, tried to source as much as I can.

Some gamers with high-end cards are likely doing a bit of mining. If the parents or dorms are picking up the electricity bill, that makes sense. And then there are some cards that are used for gaming and to study deep learning, I guess. It is likely a bit more of a mixed use of compute than just pure mining.

To give an example, I'm not a student, but I do have a box with 2 x RTX 2080 sitting in my garage, which I use for hobbyist deep learning research. Sometimes, I also need heating in my garage, it gets cold during the winter. It was pretty easy to flash a NiceHash usb stick. So I boot it from it and it becomes a 1.5kW heater which pays for itself. Not particularly efficient or climate friendly, but oh well.

Case in point: I do all of the above. These RTX cards are marvelous.

> Some gamers with high-end cards are likely doing a bit of mining. If the parents or dorms are picking up the electricity bill, that makes sense.

Yeah, I think a good number of gamers who bought cards are mining on them in their spare time as well. I'm counting that as a gaming sale since it's primarily for gaming rather than someone who set out to buy 10 cards and build a mining rig, but I do think a lot of people did the "well, I'll mine on it when I'm not using it and it'll soften the blow a bit..." approach.

And you don't even need free electricity, the cards were and are still profitable, just not profitable enough (anymore) to gamble $10-20k setting up an industrial-style mining rig with a full loadout of cards. If you have the card anyway (because you game/train AI/etc) then it's still worth mining.

I question how many people cashed out regularly though, there may be a lot of people sitting with a bunch of paper profits from their gaming card...

> And then there are some cards that are used for gaming and to study deep learning, I guess. It is likely a bit more of a mixed use of compute than just pure mining.

I think that's a small market in general, there's not millions and millions of those people and they're not setting up hundreds of rigs with thousands of cards, but yeah definitely, that exists. I know people who bought a quad-1080 ti or quad-3090 rig and used it for gaming/DL/homelab and mined on it when they weren't using it.

It was perhaps a little less appealing this gen since there was never a proper Titan - the 3090 is still locked down in certain types of operations, not just FP64 but also some data types/operations that iirc might be relevant to DL users. bfloat16, and some types of FMA operations, or something like that? But lots of people can work around it or it isn't relevant to the thing they wanted to do, there are still a decent number of DL/homelab people who will buy it. It's again, not a gigantic market, but it's there.

imo with the way prices stacked up - it was $2400 for a 3080 or $3000 for a 3090 on ebay, and it was actually possible sometimes to get 3090s at decent prices ($2000) from partners (Zotac, EVGA, etc), where it wasn't really possible to ever score a 3080 from partners due to extreme levels of demand. So you could score a retail 3090 for cheaper than an ebay 3080, and actually even at ebay pricing, if you're going to spend $2.4k on a GPU anyway... just spend the last $600 and get the extra VRAM. 10GB is going to noticeably shorten the viable lifespan of the card anyway.

Given the pricing, it made a lot of sense imo to aim for the 3090 over the 3080, both for homelab/DL type stuff, and honestly also because the 3080 had a very stingy amount of VRAM out of the box even for gaming. I know the technical reasons (yield, memory bus width, etc) that make this complicated, but I'd really liked to have seen 12GB on the original 3080, if not 16GB. I think GDDR6X was a mis-step in a lot of ways and they would have been better off with a conventional 512b bus with GDDR6 instead of a super-fast 384b bus with GDDR6X... but that also complicates routing and board design.

Wait, but its AMD's RX 480/580 line that saw the most eye-popping price increases, up to 4x (FOUR times!) compared to 2020 prices !

https://web.archive.org/web/20220126135425if_/https://cdna.p...

It's starting to slowly come down now, but still 2x !

https://cdn.wccftech.com/wp-content/uploads/2018/04/Asus-RX-...

> There is very very little gaming demand for these cards at 2-3x MSRP

Anecdotally, I know 10 or so people that paid 2x+ for RTX3000 series cards and zero that are mining. A lot of people my age (about 40) got back into PC games for something to do during the pandemic.

Anecdotally I know a lot of people who are too ashamed to admit they're mining on it in their spare time but know that leaving a money printer that makes $5 a day sitting idle is stupid. ;)

(even today, a 3090 is still profitable even with absurd $0.30/kWh electricity or whatever - if you are around the national average of 10c they're still doing like $3 a day. Not "buy 10 of them and build a rig" profitable anymore, but at the peak they could do ~$11 a day after electricity and for a long time they were $5 a day after electricity. Pre-LHR 3080s were a beast too, almost the same hashrate as a 3090.)

Anyway the backlash against mining and miners in the 2021 timeframe was real and intense. I'm not saying there aren't legitimate reasons - it's a waste of electricity, it's screwing up the market for everyone else, etc - but I think a huge number of gamers who bought cards in the 2020-2021 timeframe mined on them when they weren't using them and just don't want to admit it.

Unfortunately, crypto is a memeplex, it's like a corporation, it's a self-reinforcing living entity that supports itself (maintains homeostasis) through a series of rules and incentives, and now that it's been created it can't be easily killed. Every individual is financially incentivized to keep playing their part, so everyone would have to collectively agree to not make money by taking part, and as people fall out the incentives become steeper and steeper to participate. Would you mine on your card for $100 a day? $1000 a day?

Even if you pulled the plug on the exchanges, crypto is far enough along that it would be self-sustaining simply based on the black market. And the existence of PoS paradoxically makes it impossible to ever kill the PoW coins, because they can be traded on decentralized exchanges even if the real money exchanges ban them.

We will just have to live with destroying our planet, because now that it's been created it's virtually impossible to kill. That was the whole point. With apologies to Alien: it’s biologically engineered to be a survivor... the perfect predator (of certain other memeplexes).

>but know that leaving a money printer that makes $5 a day sitting idle is stupid. ;)

> Every individual is financially incentivized to keep playing their part, so everyone would have to collectively agree to not make money by taking part, and as people fall out the incentives become steeper and steeper to participate. Would you mine on your card for $100 a day? $1000 a day?

>We will just have to live with destroying our planet, because now that it's been created it's virtually impossible to kill.

You seem to have made up your mind on that front. If all you're worried about is making money at the cost of perpetuating behavior/consumption habits that are in the macroscale approaching pathological in terms of sustainability, that's on yo. Don't be surprised when the rest of society puts a punitive price tag on the activity as people are better able to quantify the tangible costs, and the layers of hype are peeled away.

Are you big in the crypto world? Cause one miner buying 250 cards matters a lot more then 10 gamers buying 1 each.

Also miners are typically buying the best card per a price point, which tends to be higher end or datacenter cards. Your average gamer is usually buying a medium tier.

The price point is also completely indifferent.

There are few miners so it's unlikely you would know one. But every miner buys a ton of cards, so even though there are few of them it matters a lot.

What about AMD?

AMD properly noted potential volatility due to cryptocurrency mining their SEC filings. [1]

>The demand for our products depends in part on the market conditions in the industries into which they are sold.... for example, our GPU revenue has been affected in part by the volatility of the cryptocurrency mining market...If we are unable to manage the risks related to the volatility of the cryptocurrency mining market, our GPU business could be materially adversely affected.

That's from their most recent one, but similar statements are in prior ones as well.

[1] https://ir.amd.com/sec-filings/filter/quarterly-filings/cont...

NVIDIA noted it too. This is about their earnings calls where they gave estimates of the magnitude of the crypto sales, and did what I consider (and really, most people who looked at it seriously) to be a massive massive undercount, by a factor of like 3-10x.

AMD was underselling it too though (in 2017-2018 it was a huge factor, not so much this time) and they had a huge inventory overhang lasting years as well... ctrl-f "inventory" in their late-2018/early-2019 earnings call transcripts.

https://www.fool.com/earnings/call-transcripts/2019/01/30/ad...

It is not about the earnings call. The SEC notice specifically cites 10-Q filings in 2018 as the reason for their action. Earning calls may have been deficient as well, but it's their filings that got them in trouble.

Earnings calls are not intended to be anywhere near as comprehensive as a Q-10 filing. They will typically last a shorter time that it would take to merely read the Q-10 itself. Further, earnings calls are not a legally mandated requirement, some companies may not have them at all. This makes them less susceptible to SEC scrutiny.

As for AMD, they were making similar comments to what I quoted above in their 2018 filings, NVidia was not. AMD downplayed it a bit in an informal non-regulated setting but gave better information in the 10-Q does. For NVidia's 10-Q, it was barely a mention that some earnings were impacted crypto, and they did not (per the SEC & a quick read through of their 2018 10-Q's) sufficiently note that it's gaming growth was still driven to a material extent by crypto, thereby exposing them to even more downside risk than what had already occured.

AMD may have disclosed it in their financial statements.

Did AMD attribute sales to growth in gaming, not crypto?

Reply on news.ycombinator.com