Nvidia's GPU market share hits 90% in Q4 2024 (gets closer to full monopoly)

[igormp]

Incompetence? What, like Nvidia doesn't have engineers who know how their GPUs work or something? I find that hard to believe.
Not wanting to mention it, hoping that no one would notice and they could just get away with it sounds more plausible.

At this point we can just guess, the link I have does give an interesting insight and some opinions of their of what may have caused it.
You're free to keep thinking that, of course, it could very well be the truth as well, we won't know for sure unless we get some insider knowledge.

So where's 56-70% of those dies, then?

I guess now you're conflicting yields with the fact that consumer blackwell has low production. The fab allocation for that node is mostly being used for their enterprise big chips (think H100/B100).

No, it doesn't. It's a x-300 die. Full dies are designated x-400.

That's not always the case. The GB206 only has 36 SMs, which the 5060ti is likely to use all of those. If it's using all the SMs out of the die, does it matter which name it's using?
As another counterpoint, the 3090ti, which used the full GA102, was a x-350 chip.

The problem is that it's not a refresh. It's something that they could release right at the start, but they don't because they want to look like the good guys who give you something more when they actually don't.

What if I said I could give you a chocolate bar for $2 right now? You'd be happy, right? What if I came back next week and said "you know what, I actually could have given you two chocolate bars for that price, but I didn't because I was a dick, but if you give me another $2, I'll give you those 2 chocolate bars and we call it even, ok?"

So what? It's still a refresh, that's common market practice, every company does so, and AMD has done so as others have already said with X3D and those XT chips.
I think it's moot to discuss that, it'll just devolve into "capitalism bad!" without any meaningful conclusion, so I'd say for us to put a stop on that specific point, agreed?

I'm simply disappointed because I find this practice dishonest and deceitful.

Take the 4080... It came with a cut-down die for $1200. A year later, the 4080 Super got released with the full die for $1000. Where were those full dies all that time? And why wasn't the 4080 $900 or something?

Same as above. 5950x (which I have) came out hella expensive, then the 5900XT came later way cheaper. Yet, I'm not mad at it, it's just how technology and the market goes over time. New products are expensive, both due to the lack of fabrication maturity and higher demand, then it becomes cheaper and better over time until a new product replaces it.

What's with the low availability? I was talking about yields. If only partially enabled dies get released, that suggests 0% yields... on the same f*ing node that AMD makes Navi 48 on! Does AMD have no problem with yields, while Nvidia's just magically improves just in time for a Super "refresh" (again, on the same node)? And the Easter bunny exists? What am I, 4, to believe shit like that?

Just wanted to reiterate that yields are kinda moot regarding this, consumer blackwell just has low production volume, period.

So what you're suggesting is that by making a much smaller quantity of chips, you have to sell all of them as one SKU not to segment the already low quantity of chips even further?
So that if Nvidia has one wafer allocated to GB205, then all of them have to be the same SKU because there's not enough chips in total, while if AMD has 10 wafers for Navi 48, they can afford to sell one wafer's worth as a non-XT and the rest as XT?

Not exactly like that, but your line of thought is on the right path.

If you make a wafer of GB205s with 140 total usable dies, 30 of which came with defects (assuming rough numbers that I gave before), you could make 110 "5070 Super" and only 10 5070 with another 20 5070 mobile, which would be weird given how the cheaper products often end up with higher volume of sales.
Or you could just laser all of them a bit, throw away chips that are below a certain threshold, and sell all of those as a 5070. Given you only have 120~140 total chips, throwing away some parts of it makes sense since your logistics becomes way simpler to sell a single product.

Now, if you 10x that, you suddenly have 1100 5070 Supers, 100 5070 and 200 5070 mobile. Now it starts to make more sense to think about introducing a second line and bother with the logistics and packaging for a different segment.

If so, that kind of makes sense. Still not any nicer to gamers, but oh well...

I mean, since the crypto bubble there has been nothing nice to gamers whatsoever lol

Why was the 3GB 1060 cut down from the 6GB 1060 then? Pretty deceptive to call both chips the same but 1 is weaker all over...

What does this have to do with the previous discussion? Yeah, the 1060 3gb was a die cut, and the naming was misleading, what else is there to say, or what does this has to do with yields/availability?

 

[lilhasselhoffer]

…do you actually not understand why there are several absolutely innocuous (by big corpo standards) explanations for this other than “NV bad” that have to do with yields improving over time (as they do)? You realize that even the highly profitable enterprise segment did NOT get full AD103 cards, right? Like, straight up, the yields were THAT poor. The 4080S is a result of said yields improving and NV having enough full chip stockpile to actually generate enough cards to sell. As for why the chose to refresh the 4080 instead of funneling those chips to enterprise, well, that’s quite simple - Blackwell was already selling there and as for older Ada chips - most customers were interested in AD102 products by that point. The 4080S exists as a consumer product literally because it could not sell effectively anywhere else, hence the price cut too, just to empty the stockpile.


Welcome to… *checks* …more than a decade ago. I hope your nap under a dune somewhere in Sahara was nice. But yes, it IS their worst launch in quite a while. They just have the luxury of not really caring. I would not too if I were them.

So I realize that I'm jumping back, but the discussion of AMD being a paper launch and seeing this has me asking a fundamental question. Just how bad are the actual yields?

I ask because November 12th was when Nvidia stopped making the higher end 4000 series GPUs (PC Mag article). If we assume that they started at low yield in October, before a November cut-off with sufficient confidence in production October-November (low yield), November-December-January-February-March. So...4 months of full production and a start-up month. Despite this, places that had demand for hundreds of cards got a couple. If you then divide that by days, it's basically like their target was measured in cards per day rather than hundreds of cards...as would have and come anywhere near meeting demand.

On the other hand, there were reports of dozens of 9070 and 9070xt cards. Likewise, AMD did not officially kill production of the 7000 series cards, yet somehow managed a release in March that had enough volume to deal with some of the demand. There are plenty of people here who wanted a card day one and are crying over the situation...but if Nvidia launched this bad with more than 4 months of build they yields were either utterly trash, in their infinite wisdom they decided to switch the consumer lines without a stable output, or they are making almost no volume of consumer cards. That good old Nintendo style artificial scarcity is trash.

So...to your point. Is this 4 months of incompetence, 4 months of a yield so low that they should not have cut production of old product, or some other flavor of gross incompetence? I ask because a surface level read of your points seems rational...until you dig 1 centimeter below the surface and find the poop nugget that was the launch is not covering a golden core, it's covering a bunch of corn.

 

[igormp]

So I realize that I'm jumping back, but the discussion of AMD being a paper launch and seeing this has me asking a fundamental question. Just how bad are the actual yields?

I ask because November 12th was when Nvidia stopped making the higher end 4000 series GPUs (PC Mag article). If we assume that they started at low yield in October, before a November cut-off with sufficient confidence in production October-November (low yield), November-December-January-February-March. So...4 months of full production and a start-up month. Despite this, places that had demand for hundreds of cards got a couple. If you then divide that by days, it's basically like their target was measured in cards per day rather than hundreds of cards...as would have and come anywhere near meeting demand.

On the other hand, there were reports of dozens of 9070 and 9070xt cards. Likewise, AMD did not officially kill production of the 7000 series cards, yet somehow managed a release in March that had enough volume to deal with some of the demand. There are plenty of people here who wanted a card day one and are crying over the situation...but if Nvidia launched this bad with more than 4 months of build they yields were either utterly trash, in their infinite wisdom they decided to switch the consumer lines without a stable output, or they are making almost no volume of consumer cards. That good old Nintendo style artificial scarcity is trash.

So...to your point. Is this 4 months of incompetence, 4 months of a yield so low that they should not have cut production of old product, or some other flavor of gross incompetence? I ask because a surface level read of your points seems rational...until you dig 1 centimeter below the surface and find the poop nugget that was the launch is not covering a golden core, it's covering a bunch of corn.

Yields are one thing, production volume is another.
I do believe the GB202 has bad yields, but that's only used for the 5090.
The rest of the lineup should have good yields and be really close to their Ada equivalents. The issue with those is the sheer lack of production. Reminder that the same node for the consumer blackwell is also used for the enterprise products, and you can guess which ones brings more profits.
And that even though Nvidia has stopped production of 4000 series, they did not stop producing Ada entirely, it likely is still being produced (at a lower volume) for the enterprise market as well.

 

[lilhasselhoffer]

Yields are one thing, production volume is another.
I do believe the GB202 has bad yields, but that's only used for the 5090.
The rest of the lineup should have good yields and be really close to their Ada equivalents. The issue with those is the sheer lack of production. Reminder that the same node for the consumer blackwell is also used for the enterprise products, and you can guess which ones brings more profits.
And that even though Nvidia has stopped production of 4000 series, they did not stop producing Ada entirely, it likely is still being produced (at a lower volume) for the enterprise market as well.

Production volume is entirely based upon yields. You can theoretically have less production than yields....but that would mean you are sitting on a volume of chips that aren't in a card. That makes absolutely no sense. It makes more sense that you cannot assemble, because you have no chips.

My point regarding this is not to waffle on the production lines and minutea. My problem is that if you are a GPU manufacturer, and you cut your core line, you need to have a replacement. The remaining 4000 series is the low end stuff...which doesn't touch the 5090, 5080, and 5070 that are released. So...the remaining question is how Nvidia is came to the paper launch they did. Four months of yield issues is...not really acceptable if you kill the stuff that is currently yielding, and thus assembling, well. So, you've got three answers:
1) Nvidia is incompetent, and the low yields are a function of them launching badly with a line they knew was low yielding, and thus is a blatant anti-consumer cash grab.
2) Nvidia is malicious, and knows they can charge anything because AI makes their GPUs a vestigial component they can basically force almost any price out of.
3) Nvidia's yields are not really an issue, and claiming the 5000 series had any direct impact from the yields is not honest because they simply give not a single crap about consumer...so they are greedy.

None of those three answers fall under the umbrella of "regular non-malicious production issues" which offer an excuse for the 5000 series launch. Something I might agree with you on prima facia, but once you actually consider the 4 month gap is no longer a viable or morally reasonable thought process.

 

[igormp]

Production volume is entirely based upon yields. You can theoretically have less production than yields....but that would mean you are sitting on a volume of chips that aren't in a card. That makes absolutely no sense. It makes more sense that you cannot assemble, because you have no chips.

That's for the overall line. You can have a order of 10k wafers/month, and segment those in different lines of chips, such as GB202 up to GB207, GB100s and GH100s. Then you can split your wafer allocations into whichever products you want.

My problem is that if you are a GPU manufacturer, and you cut your core line, you need to have a replacement. The remaining 4000 series is the low end stuff...which doesn't touch the 5090, 5080, and 5070 that are released. So...the remaining question is how Nvidia is came to the paper launch they did.

They are using their fab allocation for more GH100 and GB100 dies, that's heir core line. You're mistaken if you think the lower end parts (AD102/GB202 and below) are their "core lines".

3) Nvidia's yields are not really an issue, and claiming the 5000 series had any direct impact from the yields is not honest because they simply give not a single crap about consumer...so they are greedy.

Yields are only an issue for the bigger dies, such as the GB202 and the x100 ones.

Nvidia simply did not allocate enough wafers for their consumer lineup, simple as that, hence why the low volume.

 

[oxrufiioxo]

Nvidia simply did not allocate enough wafers for their consumer lineup, simple as that, hence why the low volume.

I think what most don't realize is that shareholders can actually sue if Nvidia doesn't prioritize high margins products here in the great US of A.

The only way they could have actually avoided this is spinning off the gaming segment of their business as it's own company which isn't going to happen.

Right now TSMC has a monopoly though so they'd also likely have to use a different node which nobody offers currently both intel and Samsung are way behind so gamers need at least 2 things to happen the AI bubble bursting and another silicon maker to actually compete with TSMC before things are not shit again.

 

[Frick]

I think what most don't realize is that shareholders can actually sue if Nvidia doesn't prioritize high margins products here in the great US of A.

In this case maybe but it would have no legs to stand on.

 

[oxrufiioxo]

In this case maybe but it would have no legs to stand on.

It's actually happened before and the shareholders won.

Let's be real though a company is going to priortize high margins regardless of the threat of legal actions. It's just another thing here in America that companies have to be worried about.

 

[Scattergrunt]

They did go full into damage control once this happened, and the whole product brought tons of chaos, but I don't think this is something equivalent to "5070 = 4090", instead I think of it as something closer to Intel's CPUs degrading. Still bad, but for different reasons.

Not to jump into this debate as its not my expertise of knowledge but as far as I understand the debate, I don't think I'd use the Intel CPUs degradation as an example. There's a lot of evidence pointing to the fact Intel at least suspected this was a issue long before they actually addressed it, going off of some of the sources of GN and commentary I've heard buzz around that time anyway

I mean I'm not gonna defend the 970, the situation should of been handled way better, and the response by NVIDIA was pitiful. Though to be fair, I can't tell you for a fact if NVIDIA was intentionally trying to obfuscate this fact like Intel seemed like they were.

Intel at least gave extra warranty on the CPUs tho lol

It was a bandaid over a giant gushing bleeding mark and their stocks took a tank for it for a reason. They got a rightful kick in the nuts, and now they made a improved generation with interesting prospects for the future, even if its not a interesting buy for gamers right now.

 

[Frick]

It's actually happened before and the shareholders won.

What company?

 

[80-watt Hamster]

What company?

The inciting case was Dodge v. Ford.

 

[oxrufiioxo]

The inciting case was Dodge v. Ford.

Which is even more crazy that Dodge won because a big part of it was them not wanting to lower their prices it would be the equivalent of Nvidia having a 10% stake in AMD and them not wanting them to target market share and still winning.


Gotta love the USA. Land of the free unless you want to benefit the consumer.

 

[Scattergrunt]

Gotta love the USA. Land of the free unless you want to benefit the consumer.

Land of the 'free', full of 'liberty' and 'free speech'.

Remember kids, "We the People," are always the losers in the big American game.

 

[lilhasselhoffer]

That's for the overall line. You can have a order of 10k wafers/month, and segment those in different lines of chips, such as GB202 up to GB207, GB100s and GH100s. Then you can split your wafer allocations into whichever products you want.

They are using their fab allocation for more GH100 and GB100 dies, that's heir core line. You're mistaken if you think the lower end parts (AD102/GB202 and below) are their "core lines".

Yields are only an issue for the bigger dies, such as the GB202 and the x100 ones.

Nvidia simply did not allocate enough wafers for their consumer lineup, simple as that, hence why the low volume.

Stop. Listen. I know you want to explain, but you don't get it.

If your supposition is that all of the capacity is being transferred, then Nvidia is knowingly catering to someone other that consumer GPUs, and thus is malicious because they are only serving whomever pays them the most at any time.
If they are having issues with yield, as you stated "is something non-malicious" then they are incompetent in getting the lines up and running.
There is no middle ground or alternative...you either aren't making them or aren't capable of making them. Is it better to be stupid or anti-consumer?


If you want to argue that the issues are not malice, and not incompetence, but innocuous inefficiency, and then argue that they are diverting production only to their highest profit margin business...then you don't understand the words coming out of your own mouth. Malice to GPU consumers is not inherently evil, but telling us everything is fine, it's non-malicious malice is pretty....special.

 

[Onasi]

If your supposition is that all of the capacity is being transferred, then Nvidia is knowingly catering to someone other that consumer GPUs, and thus is malicious because they are only serving whomever pays them the most at any time.

…how is a for-profit company working to maximize profits malicious? What?

There is no middle ground or alternative...you either aren't making them or aren't capable of making them. Is it better to be stupid or anti-consumer?

How are they anti-consumer? Why are some people in this thread so stubbornly stupid on this one point - you and me are NOT their primary customers anymore. Simple as that. They are serving a more profitable customer base. Is Ferrari being anti-consumer by not releasing a budget grocery-getter? Is Scania by not building mopeds?

If you want to argue that the issues are not malice, and not incompetence, but innocuous inefficiency, and then argue that they are diverting production only to their highest profit margin business...then you don't understand the words coming out of your own mouth. Malice to GPU consumers is not inherently evil, but telling us everything is fine, it's non-malicious malice is pretty....special.

There. Is. No. Malice. Corporate entities and their business decisions are amoral for the most part. As in - that’s not part of the equation. Again - consumer market woes are just collateral. In fact, arguably an irrelevant one since gaming GPUs are mostly luxury toys, while NVidias enterprise customers might be argued to be more relevant and useful for society as a whole. See, I too can make nonsensical emotional arguments!

tl:dr - You not being able to get a GPU to play vidya on is not important and NV is acting like any well run company in their shoes would.

 

[lilhasselhoffer]

…how is a for-profit company working to maximize profits malicious? What?


How are they anti-consumer? Why are some people in this thread so stubbornly stupid on this one point - you and me are NOT their primary customers anymore. Simple as that. They are serving a more profitable customer base. Is Ferrari being anti-consumer by not releasing a budget grocery-getter? Is Scania by not building mopeds?


There. Is. No. Malice. Corporate entities and their business decisions are amoral for the most part. As in - that’s not part of the equation. Again - consumer market woes are just collateral. In fact, arguably an irrelevant one since gaming GPUs are mostly luxury toys, while NVidias enterprise customers might be argued to be more relevant and useful for society as a whole. See, I too can make nonsensical emotional arguments!

tl:dr - You not being able to get a GPU to play vidya on is not important and NV is acting like any well run company in their shoes would.

Again, stop. This is a thread about....

If you don't get it, then let me short track this. Capitalism, unfiltered, is an exercise in the malicious acquisition of resources. This is why in pure capitalism monopolies exist, and why protected capitalism has government and regulation to prevent anything other than limited monopolies (IE, patents and trade marks). Maliciousness is confused with evil, but it is not. Intent to do harm, to enrich ones self, is not evil. The fact that you equate the two as acceptable is...kinda silly. I'd laugh, but too many people believe that any pure system is good or bad. Communism is maliciously bringing everyone to have what they need, and penalizing anyone who has ambitions at more. This is why I subscribe to protected capitalism as the thoroughly middle of the road success it's been...the only one that has sustained itself for centuries.


Nvidia is foregoing their GPU market. They are doing so, because the money is better. Their malice is to the market they are leaving, because it isn't as profitable. Again, not evil, just an intent to do harm. Remember, the more harm they do the more profitable they can be...unless a competitor exists and forces them to be better because they have to compete.


I...do find it funny that you want to pretend Nvidia is a person. It has wants and desires...but it does not. It's an entity which has stockholders. The stock holders want to extract as much money as possible from the market, largely without a desire to keep things going. Gains today that are 1% higher are better than 0.5% gains for four years. The malice there is in effectively viewing your consumers as wallets only. I don't think Nvidia is evil, as you want to propose. I think that the malicious approach to extracting as much capital as possible is going to be a failure in the long run. That is why most companies fail...because malice in moderation is a means to grow, malice unchecked is a ticket for customers to choose any replacement they can, and a complete lack of malice will inevitably lead you to fiscal ruin as somebody will always want what you have.

 

[igormp]

If your supposition is that all of the capacity is being transferred, then Nvidia is knowingly catering to someone other that consumer GPUs

Yes.

and thus is malicious because they are only serving whomever pays them the most at any time.

That's not being malicious, but rather something they HAVE to do by law, as others have already stated.
Companies always serve whomever pays them the most, that's how they profit.

If they are having issues with yield, as you stated "is something non-malicious" then they are incompetent in getting the lines up and running.

For their bigger dies that's indeed the case. But I don't see it as incompetence, it's just hard and something that's more up to TSMC to solve.
AMD has no product of the same magnitude as those huge dies being made nowadays. I don't think Intel has any either.

Malice to GPU consumers is not inherently evil, but telling us everything is fine, it's non-malicious malice is pretty....special.

At this point it just sounds like entitlement from your side.
Why should enterprise offerings be reduced to attend the demand of consumer GPUs that provide less profits?
Quoting myself here:

I think it's moot to discuss that, it'll just devolve into "capitalism bad!" without any meaningful conclusion

If you don't get it, then let me short track this. Capitalism, unfiltered, is an exercise in the malicious acquisition of resources.

just as I was going to press the "post" button, you sent that. Seems like I was on the right path lmao
The conversation totally derailed into off-topic IMO, this has nothing to do with yields or production capacity anymore.

 

[Onasi]

@lilhasselhoffer
That’s a very cool TED talk on the evils of late stage capitalism and the suffering of common man unable to play CP2077 with RT in his free time due to malicious business decisions regarding selling GPUs from a GPU making company, but I have no idea what that has to do with, like, anything up to this point and I humbly request you to cease and desist and either actually discuss tech (though probably in another thread, this one has lost the plot completely) or, alternatively, head to Reddit since they love takes like this out there.

 

[Rover4444]

…how is a for-profit company working to maximize profits malicious? What?


How are they anti-consumer? Why are some people in this thread so stubbornly stupid on this one point - you and me are NOT their primary customers anymore. Simple as that. They are serving a more profitable customer base. Is Ferrari being anti-consumer by not releasing a budget grocery-getter? Is Scania by not building mopeds?


There. Is. No. Malice. Corporate entities and their business decisions are amoral for the most part. As in - that’s not part of the equation. Again - consumer market woes are just collateral. In fact, arguably an irrelevant one since gaming GPUs are mostly luxury toys, while NVidias enterprise customers might be argued to be more relevant and useful for society as a whole. See, I too can make nonsensical emotional arguments!

tl:dr - You not being able to get a GPU to play vidya on is not important and NV is acting like any well run company in their shoes would.

NVIDIA probably should go back to gamers at some point though. The entire LLM side of the AI industry buys NVIDIA because of CUDA momentum while AMD offers a product more suited for that purpose for half the price, the imagegen side requires NVIDIA because it's actually good but they're just glorified render farms after all's said and done, the former is under immense pressure from more efficient models that don't require as many high-end GPUs as a necessity and upcoming stuff that uses unified memory instead of HBM... Idunno, the writing seems to be on the wall. The only thing that's keeping everything from crashing down is AMD not giving a shit about ROCm devs and RISC-V not being good enough.