21
$\begingroup$

I saw in the news recently that NVIDIA has placed limits on the hash rate for mining Ethereum cryptocurrency. This is purportedly to get more GPUs into the hands of gamers instead of crypto miners.

What is the advantage to NVIDIA to doing this? Why are they not aiming to simply maximize the demand for their product?

$\endgroup$
9
  • 33
    $\begingroup$ Cryptocurrency is a current hype. After the hype, NVIDIA still wants to have a market, which they won't have if gamers were forced to move on and away from NVIDIA. $\endgroup$
    – Polygnome
    Mar 1 at 12:34
  • 6
    $\begingroup$ I think it gets clearer when you also consider that nVidia is a software technologies vendor as well - CUDA, RTX, DLSS and other software technologies are reasons why non-crypto customers choose nVidia over their competitors. Adoption of nVidias software howeer requires minimum amount of market share for proper cards (RTX and DLSS need new scarce 3xxx series cards) and if nVidia can't deliver enough of them, they'll fail to form a competitive moat with RTX/DLSS and other AI technologies which will hurt them long term. $\endgroup$
    – Mavrik
    Mar 1 at 18:36
  • 1
    $\begingroup$ Granted, their only other competitor is AMD - and their cards are better/more sought after for mining. $\endgroup$
    – Keltari
    Mar 1 at 20:32
  • $\begingroup$ The common-sense reason is fairly obvious, so I assume you want to know the economic explanation? $\endgroup$
    – user253751
    Mar 2 at 10:06
  • $\begingroup$ I do not think NVIDIA is being nice to its core community just for the sake of being nice, and it is not obvious to me where the profit incentive is. $\endgroup$
    – Zaz
    Mar 2 at 21:20

4 Answers 4

12
$\begingroup$

This topic is explained by market segmentation with price discrimination in monopolies.

This is widely used in, for example, price discounts for senior/retired people.

By targetting different market segments with different prices you can maximize profits because their price sensitivity is different (i.e. miners are likely to buy at higher prices, while regular consumers will sharply reduce their willing to buy at those same prices).

This is true as long as the following conditions are met:

  • Consumers can't easily switch between segments (i.e. a retired man can't suddenly turn into a child)
  • Consumers can't (easily) resell their goods to the other segments
  • It's a monopoly (or oligopoly). In a perfect market situation, your competitors will offer cheaper prices for your more costly segment.

This is the same reason why consoles used to be region locked. Otherwise it would be convenient to travel to the cheapest region, and then bring those products back to your home country for resell.

In order to enforce these requirements, GPUs used for gaming can't be used for mining.

Additionally there could be other reasons. For example NVIDIA wants to heavily push Raytracing because it gives them an edge compared to the competition.

However games will not adopt Raytracing if a large portion of their users don't have capable cards. Right now the number of titles supporting RT is low. And those who do, it is just an optional gimmick that doesn't add much to the experience; and the first gen RT cards (i.e. RTX 2080) aren't fast enough for a raytracing-rich experience.

$\endgroup$
2
  • 2
    $\begingroup$ While this is a reason that NVidia has historically disabled things not needed for gaming (like intentionally dramatically limiting double-precision math speeds) on SKUs for the gaming market, what's at play with the mining market is more what bta's answer discusses. Miners buying up GPUs and drying up the supply has been a major problem for the consumer PC market (and especially the gaming market) for a while now and NVidia is trying to address that, as the gaming market has been the main driver behind GPU sales since GPUs have existed. $\endgroup$
    – reirab
    Mar 1 at 20:40
  • 4
    $\begingroup$ I used to work for a manufacturing company. They had a product with roughly thirty options which all cost the customer extra. Only three of those required hardware changes to the product, and the remaining options were all software switches (not even differing firmware). Even then, the hardware differences were only really required due to export controls. $\endgroup$ Mar 2 at 1:43
17
$\begingroup$

Limiting the hash rate is maximizing the demand for their product.

Historically, gamers have been crucial to Nvidia's success. Their demands drive innovation, and they're the ones willing to spend top dollar to buy expensive new flagship products with the latest features. The war between ATI (now AMD) and Nvidia for market share among gamers has been raging for decades.

One of Nvidia's other big customer segments is the scientific computing community. Their graphics cards are used to accelerate all sorts of compute-heavy workloads, from machine learning to self-driving cards to simulating the inner workings of a black hole.

Both of these key market segments are really hurting right now due to crypto miners snatching up all of the available graphics cards. By making their cards less attractive to miners, they're showing that they still care about their core customers and aren't abandoning them over the latest fad. It makes businesses more confident that they can build their products on Nvidia technology without worrying about whether they can source the hardware to build their products.

Increasing the available supply of cards for non-miners also gives Nvidia a chance to steal normal users from their competitors whose cards are still monopolized by miners and aren't available for purchase. This is a relatively easy way to potentially pull in a large number of gaming users, a demographic that tends to have a lot of brand loyalty. If ATI limits the hash rate and Nvidia doesn't, then they risk losing a lot of users.

It's not really about how many cards they sell. Their number of unsold cards is currently as close to zero as it will ever get. Limiting the hash rate results in the same overall number of cards being sold, but satisfies a greater number of customers. Those customers represent a diverse range of market segments and use cases. Allowing a relatively small, highly specialized user group (crypto miners) to eat up the bulk of the company's output is not a good strategy for ensuring the company's long-term growth and success. Should crypto go out of vogue, they'd lose the lion's share of their sales and the bulk of their customer base would have already moved on to other vendors who cared enough to ensure product was available for them.

$\endgroup$
3
  • $\begingroup$ Hi, welcome to Economics:Stack Exchange. Please consider improving the answer by adding references from reputable and scholarly sources. As many other science stacks do, we require formal proofs, statistical evidence or links to external sources for answers making claims which are not common knowledge. Unsourced material can be edited or deleted. For more details see our help center and FAQ on community standards for answers $\endgroup$
    – 1muflon1
    Mar 1 at 21:35
  • 2
    $\begingroup$ Comments on Stack Exchange aren't really the place for your rants about a product or practise, simply because the question mentions them. $\endgroup$
    – Nij
    Mar 2 at 1:50
  • 3
    $\begingroup$ I think your statement about number of customers is key. Or, as you say, diversification. Therefore, robustness of market. $\endgroup$
    – Pablo H
    Mar 2 at 13:01
8
$\begingroup$

Disclaimer: I am not at all familiar with the hardware industry; in fact I believe this answer makes most of my points but is better tailored for the specific industry in question.


Market share

A possible explanation is that Nvidia has a market share to maintain. If few computers are equipped with their new graphics cards and few people play games on them, it is possible that designers will move on to more available graphics cards. This could hurt Nvidia in the future.


Vertical discrimination

Another possible explanation is vertical discrimination/screening by quality. The idea is that if a company sell two products that are imperfect substites of each other, and one is clearly of better quaility, the company can put a high price on a premium product (designed specifically for ethereum mining) and a lower price on the inferior product (designed for gaming). The more price sensitive consumers will pick the later, enabling the company to practice second degree price discrimination.

For a nice example and discussion see McAfee's Damaged Goods paper. From the introduction

The 486SX processor of Intel Corporation was initially produced in a curious way. Intel began with a fully functioning 486DX processor, then disabled the math coprocessor, to produce a chip that is strictly inferior to the 486DX but more expensive to produce. Nevertheless, in 1991, the 486DX sold for \$588, and the 486SX for \$333, a little over half the price of the chip that is less expensive to produce (Frenkel, 1991).
We will argue in this paper that this is not an isolated incident, and that many manufacturers intentionally damage a portion of their production. The obvious reason for doing so is to permit price discrimination. By producing an inferior substitute, the manufacturer can sell to customers who do not value the superior product so much, without decreasing demand for the superior product very much.

Note: In the comments under this answer people more knowledgeable about the industry write that this is an oversimplification of 486_X story. Could well be!


Brand image

Many people have concerns about the detrimental environmental effect of cryptocurrency mining, thus it is possible that the above is also a PR move to protect Nvidia's image.

$\endgroup$
6
  • $\begingroup$ It could also be to save the industry from dying in face of constrained supply. OP is perhaps right that NVIDIA can simply increase the supply enough to be a monopolist. But (using some industry specific knowledge) the production expansion is difficult and time consuming. In the meanwhile, NVIDIA does not want the gaming industry to die. So it decides to ration the demand so both get part of the pie, which it can later exploit (as you said by vertical discrimination) by expanding output. $\endgroup$
    – Dayne
    Mar 1 at 7:03
  • 9
    $\begingroup$ The case of the 486DX/486SX is not as straight-forward as McAfee makes out. When you produce large chips (such as the 486DX), sometimes a few gates don't work; normally, the manufacturer has to discard this chip. Now, with the 486DX, if that error was isolated to the floating point section, what they could do is shut down that section, thus creating a 486SX - hence, turning what would be a total loss into a product that still could be sold. Now, the number of 486SX's they got was insufficient to meet demand, so they did turn some full 486DX's into SX's - I don't know the exact numbers. $\endgroup$
    – poncho
    Mar 1 at 14:23
  • 3
    $\begingroup$ More 486SX fun: machines with 486SX would often have a vacant socket for a "487 math co-processor". But the 487 was a actually a 486DX, which would disable the original 486SX (source) $\endgroup$
    – Jonathan
    Mar 1 at 15:35
  • 3
    $\begingroup$ For what it's worth, what is mentioned regarding the 486 is completely commonplace in graphics processors. It's way more expensive to make separate chip designs for different steppings of GPUs, so they sell the ones where everything is working as the high-end chips and, on the ones where something isn't working, they cut off the power supply to that part of the chip and sell it as a lower-end chip. This both increases yields and reduces manufacturing and design cost. It's standard procedure for complex processor manufacturing where yields are often a problem. $\endgroup$
    – reirab
    Mar 1 at 20:33
  • 1
    $\begingroup$ The thing with disabling part of the chip and selling it as a lower-priced SKU is very common to this day, for CPUs as well. e.g. a 20-core server CPU might be a 28-core die had defects in up to 8 of the cores, so they fused them off. Once yields improve, then yes they will still do that for some perfectly working 28-core dies. (There are often 3 sizes of die, low/ medium/ high core count. Or for desktop, dual vs. quad core until recently we have 4, 6, and 8 core CPUs. Or Apple's 7 vs. 8 GPU-core M1 chips.) $\endgroup$ Mar 2 at 6:26
1
$\begingroup$

To understand NVidia's position, it helps to understand a bit of the manufacturing background.

NVidia's GPU's are very high-tech products. NVidia itself does not manufacture them. NVidia is known as a "fabless" vendor; it doesn't have the "fabs" (factories) in which chips are made.

Instead, NVidia relies on TSMC and Samsung. These companies operate the fabs that are shared by many fabless vendors. This makes sense, because a fab is incredibly expensive nowadays. TSMC for instance is spending 44 billion USD in 2022 alone to build new fabs. They can finance this because NVidia has promised to spend billions with TSMC. Still, it takes years to build a new fab.

In the short term, it means that NVidia bought a fixed amount of capacity in the TSMC fabs. NVidia can choose how to use this fixed capacity in order to optimize revenue.

We can now apply a simple model, where NVidia has two products ("crypto" and "gamer"). Let's assume first that these products are equally complex to produce. Both products also have a fairly standard demand curve: as the price goes up, demand goes down. But the curves do not need to be identical.

In this case, we expect NVidia to select a price for both products such that the marginal profit for both chips are identical: producing one extra crypto chip will yield N dollars of profit, but this would mean producing one less gamer chip at a cost of the same N dollars.

This strategy works when there are two demand curves, for two independent markets. But this requires an artificial distinction to be made. Without the distinction, there is one product, with an aggregate demand curve. NVidia would end up selling more chips to the lower end of the crypto market, and less to the higher end of the gamer market.

I disagree with the other answer which suggests the 486 example. The reason is that the 486SX and the 486DX were competing in the same market. A single individual buyer might reasonably choose between the two. In comparison, NVidia's gamer and crypto markets have virtually no overlap.

The "market share" argument is fairly irrelevant too. GPU purchases are not strongly influenced by past purchases. Dell will happily switch from AMD to NVidia or back based on the deal they're offered today.

$\endgroup$
7
  • $\begingroup$ "I disagree with the other answer which suggests the 486 example." I don't know, maybe you are right! "NVidia's gamer and crypto markets have virtually no overlap." So what is your answer to the actual question, why did NVidia limit the hash rate, if the purpose was not to separate these two markets? $\endgroup$
    – Giskard
    Mar 1 at 16:15
  • 2
    $\begingroup$ Also, this answer seems to suggest the market share argument is relevant for something called Raytracing. $\endgroup$
    – Giskard
    Mar 1 at 16:18
  • 4
    $\begingroup$ Sorry, but your last point completely ignores that nVidia is also a technology vendor, not just a hardware vendor. Their newest processors feature hardware acceleration for raytracing (RTX) and AI technologies like DLSS. Those technologies ensure longer-term dominance of nVidia in the market (for historic example, see nVidia CUDA) and will fail to do so if the market share of new cards doesn't reach the required number for software (gaming) developers to implement it. Abandoning gaming markets for crypto is a textbook case of optimizing for short-term gain while damaging long-term prospects $\endgroup$
    – Mavrik
    Mar 1 at 18:32
  • $\begingroup$ Market share is definitely a factor in GPU competion both historically and presently. Having sufficient share that developers will target/optimize around your hardware is a massive advantage. So much so that in the past both AMD and NVidia have paid game developers for this prioritized optimization. That is to say nothing for the the influence of word of mouth or a tendency for consumers to stick with what they perceive as a reliable brand. $\endgroup$ Mar 1 at 18:46
  • $\begingroup$ While market share isn't everything, it is nevertheless still quite relevant, both in the gaming market and in the GPGPU (i.e. using GPUs as computational accelerators) market, though moreso in the latter. In my opinion (having been programming GPUs since around the time CUDA first became available,) CUDA is technologically superior to OpenCL, but using it means that you're locked into NVidia. If ability to source NVidia chips is a concern, then people will use OpenCL (or some other API/language) instead to keep their options open. $\endgroup$
    – reirab
    Mar 1 at 20:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.