• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

GeForce value over time and 40-series pricing.

you may never get a nvidia card to review ever again unless you play ball. With the price these things cost you have to be Linus to buy them all to review
also i'm sure those 10.000$ Louis vuitton bags are some editor review somewhere, if you don't care about money

Plus the only way to get a pre-release or launch review up is through Nvidia.

I think at some point both reviewers and consumers are going to make a stand to make change happen.
 
... despite it being 71.8% more expensive and having a 39.6% smaller die then the 3080
What the actual fuck does die size have to do with the price that NVIDIA has to pay TSMC for GPUs manufactured on a leading-edge node? Nothing, that's what.

I wish people would educate themselves before spouting such stupidity.
 
What the actual fuck does die size have to do with the price that NVIDIA has to pay TSMC for GPUs manufactured on a leading-edge node? Nothing, that's what.

I wish people would educate themselves before spouting such stupidity.
I tend to agree. Smaller die? Yes, but a wafer now costs $20,000 instead of $7,000.
 
I tend to agree. Smaller die? Yes, but a wafer now costs $20,000 instead of $7,000.
And NVIDIA doesn't care because its compute customers - its fastest-growing revenue segment - are happy to pay whatever NVIDIA charges for its new GPUs, because the reduction in compute time of said GPUs is worth far more than the asking price. Such is capitalism.
 
Last edited:
I find it funny that we're still extrapolating Nvidia Ada prices and spending time on discussing trends and stuff, when we have AMD heavily discounting RDNA 2 cards right now.

Conclusion: Who cares about the 4080 and 4090? Let them rot on store shelves. ;)
 
Another apologist thread suggesting via mental gymnastics and a vague realisation of what the pandemic and mining craze did to the pricing over the last few years.

That Nvidia are being fair.

My arse, your talking out of your arse.

Stop propping up shitty ideals to make right that which is absolutely wrong.

The last five years happened.

Don't imply tacit fairness where pure bullshit is enshrined.

Especially when Nvidia's MSRP is such a tangential pile of irrelevant turd, find a 4090 at MSRP.

Few 4080s are being sold at MSRP.

why is the one that was released in 2018 a $100 more than the one that was released in 2020 ?

whyisthat.jpg


There is no sane reason for that pricing, the card is 5 years old.
 
Last edited:
Stop giving Nvidia a pass on it's greed during the pandemic. They dedicated a portion of their production just to mining cards,
Which mining cards?
allowed the bulk sale of card directly to miners,
Nvidia doesn't make video cards. They "make" a small amount of reference ones, but that it. What exactly did they "allow"?
and jacked up MSRP of cards released later like the 3080 Ti to cash in as much possible.
That they did and mining was probably one of the reasons. But you have no proof it was the only reason.

Let's cool off a bit and separate facts from guesses, shall we?
 
Don't kid yourself if you think Nvidia doesn't control the flow. There was no shortage from the suppliers for them. I don't buy it. They are big enough to have everything to build cards with well in advance, they have the buying power, they have the presence. They controlled the flow of numbers out the door, they controlled what cards went where, they controlled who got what, which means they controlled the market. Profits up 300%? Can't say I am surprised.
 
What the actual fuck does die size have to do with the price that NVIDIA has to pay TSMC for GPUs manufactured on a leading-edge node? Nothing, that's what.

I wish people would educate themselves before spouting such stupidity.

Die size is everything when it comes to the price Nvidia pays.

You want an exact comparison on the latest node? Great, let's compare the 4080 to the 4090.

You are talking about 60% of the die size at 75% of the cost. Mind you that's before you consider that yields increase exponentially as you reduce die size. The 4080 isn't expensive because leading edge nodes are expensive, it's expensive because Nvidia thinks it can get away with it. I'd be surprised if the 4080 had even half the relative cost compared to the 4090 for Nvidia simply because of how much better that smaller chip will yield.

I tend to agree. Smaller die? Yes, but a wafer now costs $20,000 instead of $7,000.

As pointed out above, even relative to Nvidia products using the exact same node pricing doesn't add up.

FYI the $20,000 figure you are quoting is incorrect. $20,000 is for 3nm and that's only at the launch. Nodes costs decrease over time, which is why AMD isn't paying anywhere near the $16,900 for it's 5nm wafers. This is also ignoring yields, which are higher on TSMCs newer nodes than it's 7nm nodes. If the cost per wafer goes up but the yield increases those two factors can very well offset each other.

I have pointed this out elsewhere in the TPU forms (check those out for more details) as well but recent wafer price increases are not the largest price increases in GPU history. If 3nm only costs 20,000 at launch that would be a very tame sized increase over the price of 5nm. Historically GPU prices do no increase each time wafer cost increases and in some cases GPU price has actually decreased. Wafer cost is only a single variable, something many defenders of Nvidia's pricing point to despite this fact and historical data.

Which mining cards?


Nvidia doesn't make video cards. They "make" a small amount of reference ones, but that it. What exactly did they "allow"?

Nvidia controls the size of branding on the box, how petty they can be with GPU allotment, what their partners can call their cards, ect. We all remember the GPP and how that brought forth AIBs revealing just how controlling Nvidia is of anything with an Nvidia GPU in it. You are telling me they weren't aware that AIBs were selling cards directly to miners? No chance Nvidia wasn't aware and consenting.

That they did and mining was probably one of the reasons. But you have no proof it was the only reason.

Let's cool off a bit and separate facts from guesses, shall we?

Individually these actions would have been anti-consumer, together they prove that Nvidia had the appropriate Mens Rea that fits the idea that Nvidia was catering to miners.

Nvidia may have had more reasons of course but those are incidental to the conversation.
 
Last edited:
FYI the $20,000 figure you are quoting is incorrect. $20,000 is for 3nm and that's only at the launch.
Launch price of 5nm node was $16K. TSMC had at least one (I think 2) major price increase since, and Nvidia is using a custom 4nm node. I think that it is probably around $18K. However, reading into it, it looks like Samsung 8N may well have been between $7K and $9K per wafer.

It was a bit of a hyperbole to point out that a smaller die with more transistors isn't necessarily cheaper.

That being said, I did some calculations by actual specifications that consumers "should" care about (shader count, RAM bandwidth, VRAM quantity, TBP, etc) and found that there was no way in heck to assign a value to each of these that could account for the 4080 costing more than around $900.
 
Launch price of 5nm node was $16K. TSMC had at least one (I think 2) major price increase since, and Nvidia is using a custom 4nm node. I think that it is probably around $18K. However, reading into it, it looks like Samsung 8N may well have been between $7K and $9K per wafer.

It was a bit of a hyperbole to point out that a smaller die with more transistors isn't necessarily cheaper.

That being said, I did some calculations by actual specifications that consumers "should" care about (shader count, RAM bandwidth, VRAM quantity, TBP, etc) and found that there was no way in heck to assign a value to each of these that could account for the 4080 costing more than around $900.

The node Nvidia is using is essentially 5nm+, same as how TSMC's 6nm is just 7nm+. If it does cost more, it's likely a very small amount. Otherwise it's not really worth it. Akin to TSMC 6nm, the density increase is extremely small with the upside that it's compatible with 7nm. In essence Nvidia could have taped out for 5nm and the design would have been compatible with TSMC's 4nm.

Here's TSMC's page for the 5nm which references 4nm: https://www.tsmc.com/english/dedicatedFoundry/technology/logic/l_5nm

4nm doesn't have it's own section due to the reasons stated above and on TSMC's website.

Yes, the 5nm price estimate were in the $16,000 - $17,000 range. Current estimates at over 2 1/2 years later place that at around $10,000.

As you pointed out, even under the most generous circumstances it's hard to explain the price of the 4080.
 
Yes, because overpriced, repurposed defective dies moved so large volumes...
Nvidia controls the size of branding on the box, how petty they can be with GPU allotment, what their partners can call their cards, ect. We all remember the GPP and how that brought forth AIBs revealing just how controlling Nvidia is of anything with an Nvidia GPU in it. You are telling me they weren't aware that AIBs were selling cards directly to miners? No chance Nvidia wasn't aware and consenting.
Of course Nvidia was aware. Consenting? I'm going to go out on a limb and guess you don't have a shred of evidence of that.
Individually these actions would have been anti-consumer, together they prove that Nvidia had the appropriate Mens Rea that fits the idea that Nvidia was catering to miners.

Nvidia may have had more reasons of course but those are incidental to the conversation.
All you have proven is that you can pick, choose and misrepresent to make your point. Pretty much the opposite of what I have suggested.
 
Last edited:
Yes, because overpriced, repurposed defective dies moved so large volumes...

You are going to need to provide a source for that. I remember rumors suggesting this but it was never verified that mining GPUs could have never became gaming GPUs. The rumors also claimed that they would ease GPU prices and supply, which never happened.

Of course Nvidia was aware. Consenting? I'm going to go out on a limb and guess you don't have a shred of evidence of that.

All you have proven is that you can pick, choose and misrepresent to make your point. Pretty much the opposite of what I have suggested.

Given that we know Nvidia controls everything tightly, Nvidia being aware of the mass sales of GPUs to miners is consent.

In my last comment I detailed their ability to commit such a act, of which the evidence is strong. Saying I provided no evidence is functionally incorrect, proving one has the mental state to commit an action is one of the most important things prosecution has to do in court.

Your logic seems to be that only direct evidence is admissible but even in court that is not the case even in criminal court where the bar is higher. In civil court (which is the closest equivalent) the bar for conviction is a preponderance of the evidence (or 51% likely or greater).

If we demonstrate that Nvidia, whom had vast control over their product and partners, did nothing when prices skyrocketed while also providing evidence to show they had the mental state, motive, and past behavior to directly or indirectly allow this to happen we've just demonstrated, with a preponderance of the evidence, that Nvidia is complicit in the price increases and is culpable as a result.
 
Historically GPU prices do no increase each time wafer cost increases and in some cases GPU price has actually decreased.
Historically doesn't mean shit because older nodes didn't require vastly expensive and slow EUV machines. Stop comparing apples to oranges.
 
Historically doesn't mean shit because older nodes didn't require vastly expensive and slow EUV machines. Stop comparing apples to oranges.

You already made this point earlier to which I replied with this:
Die size is everything when it comes to the price Nvidia pays.

You want an exact comparison on the latest node? Great, let's compare the 4080 to the 4090.

You are talking about 60% of the die size at 75% of the cost. Mind you that's before you consider that yields increase exponentially as you reduce die size. The 4080 isn't expensive because leading edge nodes are expensive, it's expensive because Nvidia thinks it can get away with it. I'd be surprised if the 4080 had even half the relative cost compared to the 4090 for Nvidia simply because of how much better that smaller chip will yield.

Of which you never replied to.

Past nodes have been vastly more expensive than: https://www.techpowerup.com/301393/...-cpus-gpus-to-be-more-expensive?cp=1#comments

I commented on that very article upon this very topic:
A 25% increase is actually pretty small. Here is that compared to other nodes provided on the chart in this article:

90nm to 40nm -> 30% price increase
40nm to 28nm -> 15% price increase
28nm to 10nm -> 100% price increase
10nm to 7nm -> 66.66% price increase

To say that such a modest price increase in the cost of wafers means GPU prices will increase belies the fact that we've had far more costly wafer price increases in the past yet GPU prices either did not increase or the price increase was very modest.

Using wafer pricing to justify increasingly high GPU costs is misleading because it's only a single factor in the cost of a GPU and ignores other factors, like how costs are spread across an increasingly large customer base.

It wasn't until Turing that GPU prices exploded and that was clearly motivated by greed and not node costs given Nvidia was using a much cheaper Samsung 8nm while AMD was selling their cards for less while using the more expensive TSMC 7nm.

We've seen price hikes of GPUs for several generations in a row now, the xx80 class GPU is now more than DOUBLE what it was with the 980.

If GPU prices do increase it's due to greed, plain and simple. Nvidia and AMD have already jacked GPU prices up enough to cover the next 20 years of wafer price increases.

Mind you the numbers above only include those from the provided graphs. In comparison to the cost increases of prior nodes, 5nm is normal and 3nm is below normal.

Again though, wafer pricing is by far not the only variable determining card pricing. The larger the GPU market has grown, the more R&D costs are spread out. There's also the fact that TSMC's 5nm yield's better than TSMC's 7nm. 3nm might have even higher yields give the need for less patterning. They mean even less to AMD, who demonstrated that a 16 core CPU costs less than half of that same CPU but monolithic. Those saving only scale up compared to a monolithic design as you increase the total die area. I can only imagine the kind of yield and cost savings AMD is going to get from the 7000 series being that GPUs are larger than CPUs. Just a few other things to consider in addition to wafer pricing.
 
I can only imagine the kind of yield and cost savings AMD is going to get from the 7000 series being that GPUs are larger than CPUs.
+1 to this. Even if the main die is quite large, the savings will be considerable, as yields decrease exponentially with die size.
 
You are going to need to provide a source for that. I remember rumors suggesting this but it was never verified that mining GPUs could have never became gaming GPUs. The rumors also claimed that they would ease GPU prices and supply, which never happened.



Given that we know Nvidia controls everything tightly, Nvidia being aware of the mass sales of GPUs to miners is consent.

In my last comment I detailed their ability to commit such a act, of which the evidence is strong. Saying I provided no evidence is functionally incorrect, proving one has the mental state to commit an action is one of the most important things prosecution has to do in court.

Your logic seems to be that only direct evidence is admissible but even in court that is not the case even in criminal court where the bar is higher. In civil court (which is the closest equivalent) the bar for conviction is a preponderance of the evidence (or 51% likely or greater).

If we demonstrate that Nvidia, whom had vast control over their product and partners, did nothing when prices skyrocketed while also providing evidence to show they had the mental state, motive, and past behavior to directly or indirectly allow this to happen we've just demonstrated, with a preponderance of the evidence, that Nvidia is complicit in the price increases and is culpable as a result.
You play a fine game of relying on rumors and "facts", I'm gonna leave you to it.
 
You play a fine game of relying on rumors and "facts", I'm gonna leave you to it.

I mean, you are replying to my post which asked you for the source of the rumor you posted.

That's some irony right there. Of course you are going to leave, you can't prove that rumor.
 
I mean, you are replying to my post which asked you for the source of the rumor you posted.

That's some irony right there. Of course you are going to leave, you can't prove that rumor.
Well, if you must insist...

"Unlike the fully unlocked GeForce RTX 3090 Ti, which uses the same GPU but has all 10752 shaders enabled, NVIDIA has disabled some shading units on the CMP 90HX to reach the product's target shader count." Disabled shaders - you don't do that with working dies.

Furthermore: https://www.techpowerup.com/gpu-specs/?mfgr=NVIDIA&generation=Mining GPUs&sort=name
Mining-oriented SKUs dating back to Pascal. Thus, not something concocted during the past mining craze years.

Want a more direct confirmation CMP did not choke GPU supplies? Here: $24m revenue when a card sells for $8k means a grand total of 3,000 units sold.

As for the rest of your claims, I'm going to draw you a picture and apply Occam's razor.

4dummies.jpg

If I were a miner with $$$ to burn, the most straightforward way for me to get my hands on video cards would be from an AIB partner. Your theory where Nvidia somehow forces not one, but all partners to sell to miners to which Nvidia itself has no direct relation is... shall we say, way more far fetched and therefore way more improbable.
 
Unlike the fully unlocked GeForce RTX 3090 Ti, which uses the same GPU but has all 10752 shaders enabled, NVIDIA has disabled some shading units on the CMP 90HX to reach the product's target shader count." Disabled shaders - you don't do that with working dies.
3090 was the same.
 
3090 was the same.
No, it wasn't. 3090 has 10,496 working shaders, CMP 90HX only has 6,400. Clearly a salvaged part, with almost half the shaders disabled, I'd say.
 
No, it wasn't. 3090 has 10,496 working shaders, CMP 90HX only has 6,400. Clearly a salvaged part, with almost half the shaders disabled, I'd say.
I just meant that some were disabled - doesn't mean "can't make a gaming card"

Heck, they even made a 3070 ti with the GA102, only 6,144 shaders.
 
Back
Top