No 7 nm for Nvidia's upcoming Ampere gaming GPUs

[Redaktion]

Instead of making the jump to 7 nm, Nvidia will apparently be using Samsung's 10 nm process for the next gen Ampere gaming GPUs. If this is indeed the case, AMD's second gen 7 nm RDNA2 architecture could allow the red team to almost match Nvidia's high-end Ampere models.


https://www.notebookcheck.net/No-7-nm-for-Nvidia-s-upcoming-Ampere-gaming-GPUs.457329.0.html

[S.Yu]

That would be very disappointing.

[Anonyneko]

I guess that would mean laptop cards are even further away and won't be much of an improvement over the current ones, which weren't much of an improvement over last gen...

[william blake]

different tech is always cool but why not 7nm samsung?

[JohnMac]

Not really surprising considering AMD is still about on par with Pascal and Intel has nothing to show outside entry level laptop GPUs.

They really have no reason to push performance to the absolute limit when they'll still be the best on a cheaper node.

[deksman2]

Not really surprising considering AMD is still about on par with Pascal and Intel has nothing to show outside entry level laptop GPUs.

They really have no reason to push performance to the absolute limit when they'll still be the best on a cheaper node.

Actually, its because NV couldn't secure any 7nm capacity from TSMC.
AMD has been buying the leftover capacity from Apple and some other companies that gained priority.

Oh and, AMD's Navi is achieving same performance as comparable NV gpu's with half the die-size.
When undervolted (voltage optimised), Navi is consuming far less power than stock overvolt (which AMD does intentionally to max out the number of functional dies - so it doesn't represent 'true efficiency').

Navi was a pretty big jump in efficiency from pure GCN uArch (mostly because they removed compute hw and increased gaming relevant hw)... and AMD is now going to release RDNA 2 which touts ANOTHER 50% gain in performance per watt on the pretty much same node.

uArch to uArch efficiency gains like that (of 50%) are definitely possible without changing the node.
It will be interesting to see where AMD's performance goes from this point on though (they already use Infinity Fabric for data centres MI60 and MI100 - its only a matter of time before AMD uses it on GPU's).

AMD's primary focus has been the mid-range up until now - that's where majority of the gaming market share lies (and Navi did manage to take a significant chunk of it back from NV).


For reference sake: AMD managed to increase Vega iGP performance in Zen 2 mobile chips by about 56% from previous Vega iGP found in Zen 1 and Zen + APUs.

They also mentioned they will incorporate these modifications into future Navi gpu's (whether that means upcoming RDNA 2 or not remains to be seen).

At any rate, I wouldn't discount AMD at all.
The high end isn't really too important as only a minor niche of people will be able to afford it, however, AMD did say they will provide high end gaming GPU's this year with RDNA2