• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

NVIDIA 4000 Series

How come anyone still believes any of the rumours?
believes no, comments on yes
the only thing I believe is Nvidia getting its bean counter out to apply maximum raking over consumers in literally anyway it can

even Kopite 'WCCFtech reporter alt profile' 7Kimi says he's disappointed in RDNA3. WTF does that even mean?
 
I think probably RDNA3 single die graphics cards this year. MCM RDNA3 cards in the 1st half of 2023.

That doesn't mean that they won't be a big improvement, particularly for ray tracing.

Did the rumours suggest the single die cards would use 5nm or 6nm EUV?
 
July release if these rumours are to be believed ... so in 2 months time i can grab one of these apparently :cry: That 3090Ti didn't have a long time at the top if so ;) but in all seriousness i doubt we will see it come July , more likely officially announced in July with cards starting to drop in September which would be 2 years on from 3000 series but we shall see ....
 
Weren't all the performance rumours about the RTX 3000 series completely wrong?

Nobody knew about the doubling of Single Float performance, for example.

I doubt anyone thought the RTX 3080 would have 10GB of VRAM. Or that the RTX 3090 would have 24GB.

How come anyone still believes any of the rumours?

Also, isn't the RTX 4000 called 'Ampere Next'? Which strongly implies an improvement of Ampere built on 5nm, rather than a new architecture.

The trouble with fancy sounding codenames, is that tend to lead to a lot of hype and misinformation.
Nvidia generally leaves the SKU specs till the last minute but their strategy has been to give you as little performance as they can get away with and I think that was the plan for ampere before they got wind of AMDs resurgence.

If it wasn't for AMD the 3080 would have been on the 3070s GA104 and priced at $699 but their hand was forced which is why we got the 10gb 3080 on the more expensive 102 die which left AIBs unable to manufacture it for the MSRP.
 
Last edited:
I am 50/50 on this.

It could be tory style leaks where Nvidia are deliberately leaking to gauge public feedback on proposals. Or its people with fake news speculating hoping they are close enough to get credibility in the future.
 
Nvidia generally leaves the SKU specs till the last minute but their strategy has been to give you as little performance as they can get away with and I think that was the plan for ampere before they got wind of AMDs resurgence.

If it wasn't for AMD the 3080 would have been on the 3070s GA104 and priced at $699 but their hand was forced which is why we got the 10gb 3080 on the more expensive 102 die which left AIBs unable to manufacture it for the MSRP.

I don't have enough knowledge to informedly agree or disagree with that, so my opinion is probably not worth much, but I find that sort of thing quite… if not probable in the specific case at hand, then at least realistic in the general sense of today's climate. In short, companies act that way. People act that way. The use of mathematical optimization in real life, with people trying to control and manage everything and milk everything to the last drop, rather than just living their lives. So yeah, patterns of conduct like that are quite realistic, quite likely to keep appearing more and more often and dominating the social landscape. Why give more if you can give less, why take less if you can take more. And completely managed, controlled communication and image, nothing authentic any more. Still, I hope perhaps things aren't going to be as bad as that.

***

On a different note, when do people believe 4070 and 4060 will come out? Probably no chance of 4080 with an MSRP in the eight- or nine-hundred range?
 
Any updates on this info?
"Samsung has announced that construction of a new production line has started and that the company expects the plant to be operational in the second half of 2022"

Link here:

Based on that, I wouldn't expect an early launch of the RTX 4000 series.
 
I'd guess that they will use whichever is cheaper, has plenty of capacity and good yields. Nvidia said that they had failed to predict the demand for Ampere, I'd expect them to avoid underproducing GPU dies again, if possible.
 
Last edited:

 
Nvidia used TSMC's foundries to fab the compute / deep learning focused GA100 Ampere chip, on their 7nm process.

It's possible that some of the 'leaks' we are seeing are related to upcoming compute / deep learning chips.

I think they will probably use this approach again for the next series. So, compute chips will be probably be higher density + larger die size and fabbed on TSMC's 4nm process.

EDIT - Nvidia has officially confirmed that their H100 compute GPU will be fabbed on TSMC's 4nm EUV process, link here:

Another thing to mention, is that TSMC's fabrication process costs are reportedly increasing in 2023.

Hypothetically, Samsung's 5nm EUV (5LPE or 4LPE) process should allow Nvidia to more than double the transistor density of Ampere consumer GPUs (fabbed on Samsung's 8nm).

The transistor density of the GA102 (Samsung fabbed) chip was 'just' 45.1M / mm²
The transistor density of Samsung's 5nm EUV process is upto 134.9 MTr/mm2

Info here:
 
Last edited:
The 4070 should still release this year, no? The only question is if the MSRP moves to $599 or will they keep it $499? It was already fake, so it's hard to believe given inflation that they'd just pretend it's like that again.
 
Nvidia used TSMC's foundries to fab the compute / deep learning focused GA100 Ampere chip, on their 7nm process.

It's possible that some of the 'leaks' we are seeing are related to upcoming compute / deep learning chips.

I think they will probably use this approach again for the next series. So, compute chips will be probably be higher density + larger die size and fabbed on TSMC's 4nm process.

EDIT - Nvidia has officially confirmed that their H100 compute GPU will be fabbed on TSMC's 4nm EUV process, link here:

Another thing to mention, is that TSMC's fabrication process costs are reportedly increasing in 2023.

Hypothetically, Samsung's 5nm EUV (5LPE or 4LPE) process should allow Nvidia to more than double the transistor density of Ampere consumer GPUs (fabbed on Samsung's 8nm).

The transistor density of the GA102 (Samsung fabbed) chip was 'just' 45.1M / mm²
The transistor density of Samsung's 5nm EUV process is upto 134.9 MTr/mm2

Info here:


You can't compare theoretical maximum for one product to actual on another.
You use the actual density of a Samsung 8nm GPU with a density of 46.1/mm2 and then using theoretical maximum for Samsung 5nm. The Samsung theoretical maximum for 8nm is 61/mm2

So an actual GPU made on 8nm ended up at just 75% of theoretical density of the node. So applying that to 5nm gives 100/mm2 actual density on Samsung 5nm
 
The 4070 should still release this year, no? The only question is if the MSRP moves to $599 or will they keep it $499? It was already fake, so it's hard to believe given inflation that they'd just pretend it's like that again.
Considering the impending cost of living crisis, I don't think Nvidia has much room to move the MSRP upwards. Won't stop them from trying though.


Prolly fake as I don't see why they'd release a 4090Ti so early. USually released a year after the other SKU's. Never know though.
Production and sampling happens quite a few months before release (especially if COVID is still restricting supply chains). It very well could be for the 4090ti, while the GPU's release date is still set for next year.
 

Prolly fake as I don't see why they'd release a 4090Ti so early. USually released a year after the other SKU's. Never know though.
Nvidia could just drop the 4090 and instead release it as a ti then that way they could price it at $2000 instead of $1500 which would be easier to market than releasing a 4090 with a price increase.
 
I think probably RDNA3 single die graphics cards this year. MCM RDNA3 cards in the 1st half of 2023.

That doesn't mean that they won't be a big improvement, particularly for ray tracing.

Did the rumours suggest the single die cards would use 5nm or 6nm EUV?


The more rumours come out the more "downgrades navi31 gets"

First it was an MCM GPU massive beast the came the downgrades. Cut down to 1 GPU with 15k cores and 500mb infinity cache, then cut down to 12k cores and 380mb cache and then supposedly only launch q1 2023.

It's just had another spec update, instead of 32gb vram it's been cut down to 24gb, but there is an upgrade too - 256bit bus to 384bit so bandwidth now estimated at 800GB/s
 
Back
Top Bottom