2022 in GPUs: The shortage ends, but higher prices seem here to stay | Hacker News Reader

2022 in GPUs: The shortage ends, but higher prices seem here to stay

arstechnica.com

27 pointsmetadat23 comments

I have a family friend that owns gas stations. He always says that gas prices have a tendency to "take the elevator up and the stairs down".

There doesn't have to be explicit collusion. People that have large amounts of capital know whats best for them and are very judicious when it comes to lowering prices.

There's a famous economics paper arguing this based on microeconomic factors, and that this is largely a reason why recessions happen.

As you implied, "sticky prices can be both privately efficient and socially inefficient."

and " Private incentives produce too much price adjustment following an expansion in aggregate demand and too little price adjustment following a contraction in aggregate demand."

https://scholar.harvard.edu/files/mankiw/files/small_menu_co...

I’m trying to think of a reason why the players in this market wouldn’t utilize a shortage like this to implicitly collude and fix the prices at the higher ranges until some additional competitor forces them down.

Is there really a player other than Nvidia? AMD has <20% market share. Nvidia doesn’t really have a competitor for standalone cards, especially in GPGPU.

My AMD GPU performs quite well, but not for GPGPU I agree. NVIDIA did a great job locking everybody in

Because it's illegal? Because it wouldn't maximize profits?

Price fixing increases profits as long as nobody undercuts you, that's the whole point

Not always. I for one am just not upgrading at all due to costs.

It would be a lot better to make $500 profit for each of the 100k people buying your product than $1000 profit for each of the 20k buying.

Not upgrading due to costs? Luckily for you, we have invented planned obsolescence to solve this problem!

I mean, you just push customers out of the market. I've still got the same 960 I've had for years, and that's when I bother to use my desktop, which isn't often anymore.

Is _implicit_ collision illegal? I’m thinking in the game theory sense when there are two companies who look at the decision space of increasing, leaving the same, of decreasing prices. It seems pretty easy to reason about when there are two players and simultaneously likely that leaving the prices high would increase profit.

> Is _implicit_ collision illegal?

Probably not.

https://www.winston.com/en/competition-corner/antitrust-101-...

Thanks for the link! I should’ve used “tacit collusion” instead of “implicit” in my previous comments

This brings up the topic of algorithmic tacit collusion which is something I hadn’t considered before but seems likely to occur

It has occurred. See Real Page, which is currently being sued by the government.

I have no idea why the sibling comment would be dead.

With the exponentially accelerating relevance of ML/AI and compute resources even for casual users (see DLSS and RTX), NVIDIA really doesn't have any meaningful competition on multiple axes in the GPU space.

So they don't really need to collude with anyone to fix the prices, they can just raise them.

>I have no idea why the sibling comment would be dead.

Taking this statement in good faith:

1. 20% is a huge fraction that shows a commercially viable alternative. Pricing shifts can shift that quickly to AMD being the preferred option for use cases, just like AMD was as the inferior but cheap CPU ca. 2015, and as Intel is in that space now.

2. Computing demand is incredibly elastic based on market conditions. For every game-changing application of parallel computing, there's a bitcoin mining rig in a place with subsidized electricity. Price hikes kill many applications at the margin, reducing the demand that needs to be met at higher prices.

3. There are well-funded new entrants to various segments of parallel computing at all times, and juggernauts in adjacent spaces will move in if encumbants leave huge spaces. Intel were the only serious CPUs, then the only serious laptop CPUs, and now they're competitive bjt bargain parts. Remember when Apple didn't make their own chips? And in another space, Qualcomm are going to lose their flagship reputation to Mediatek soon if they don't stop turning out trash Android SoCs in the next generation or two.

> Pricing shifts can shift that quickly to AMD being the preferred option for use cases, just like AMD was as the inferior but cheap CPU ca. 2015, and as Intel is in that space now.

No. They can't. It's hard for me to take your comment seriously when you're:

- Painting the Intel to AMD shift as quick

- Painting it as a pricing thing when it took AMD moving mountains from Bulldozer to Zen

- Painting 20% as a huge fraction when the discussion is about "colluding to drive prices up", as if the 80% needs the 20% to define prices...

> Computing demand is incredibly elastic based on market conditions. For every game-changing application of parallel computing, there's a bitcoin mining rig in a place with subsidized electricity. Price hikes kill many applications at the margin, reducing the demand that needs to be met at higher prices.

This is a long form way of saying: You can't raise prices to infinity because people don't have infinite checkbooks. Ok?

That has literally no bearing on their ability to hike prices substantially unless you're under the impression that AMD being so far behind doesn't increase their margin. If you're not aware the biggest killer of margin in modern times has consistently been... competition: https://quoteinvestigator.com/2019/01/13/margin/

NVIDIA isn't a mom and pop shop throwing darts at a price board, they'll ruthlessly optimize their prices taking in mind that their main competition holds a "huge" 20% at what's shaping up to be a pivotal moment for gpgpu...

> 3. There are well-funded new entrants to various segments of parallel computing at all times, and juggernauts in adjacent spaces will move in if encumbants leave huge spaces. Intel were the only serious CPUs, then the only serious laptop CPUs, and now they're competitive bjt bargain parts. Remember when Apple didn't make their own chips? And in another space, Qualcomm are going to lose their flagship reputation to Mediatek soon if they don't stop turning out trash Android SoCs in the next generation or two.

This again, is a very long statement... with no bearing on anything anyone actually said.

Did you forget the sibling comment?

> Is there really a player other than Nvidia? AMD has <20% market share. Nvidia doesn’t really have a competitor for standalone cards, especially in GPGPU.

This isn't about "AMD is dead", this is about "NVIDIA does not need to collude with AMD to massively hike prices right now". That's not even something anyone should have to tell you, it's so painfully self-evident that I'd question how familiar with the space anyone trying to say otherwise actually is...

According to the article, evga is out of the gpu business for a season at least. Evga made most of my cards over the last decade; they were the best

Time will tell if the market can longterm normalise $500+ as the entry level for beyond-basic PC graphics.

It's a tough sell from a mainstream gaming perspective, given the relative prices of popular consoles, which have stayed fairly static by comparison.

Therefore I believe the sustainability of high GPU prices is reliant upon proliferation of non-gaming uses.

What are they? ETH 2.0 is here to stay, there is little profitability left. Deep learning? Sure maybe for some small scale companies or labs. Rendering farm? there can't be that many people who needs that, right?

Copium is a powerful drug

I don't disagree.

To paraphrase what I said previously: without sufficient non-gaming demand, prices must regress to some degree.

Reply on news.ycombinator.com