Nvidia Post-Volta (Ampere?) Rumor and Speculation Thread

[no-X]

That article is 3 months old. How could it be tested using latest drivers?

 

[Deleted member 2197]

Since there was a new patch released (bug fixes) a few hours ago maybe a performance review will popup, though I doubt it. Most reviews seem to be a little earlier than ComputerBase review.

 

[DavidGraham]

New That article is 3 months old. How could it be tested using latest drivers?

Believe it or not, the is as fresh as it comes for this title.

 

[xpea]

According to a new DigiTimes report, Nvidia will be among one of the three major clients that will leverage TSMC's CoWoS packaging this year, with the other two being Xilinx and HiSilicon.
https://www.tomshardware.com/news/n...smcs-cowos-packaging-for-next-generation-gpus

Nothing news, Nvidia is using TSMC CoWos since 2017 in Volta
https://www.nvidia.com/en-us/data-center/v100/

MEMORY
CoWoS Stacked HBM2

These guys really write for nothing. so easy... let me tweet some exclusives too...

Spoiler: Nvidia Hopper CoWoS performance leak (Please Click!)

A "secret industry source" (= I put it straight out of my a*s) told me that next gen datacenter Hopper is a chiplet MCM design put in a gigantic CoWoS package with 96GB of HBM2e. This monstrosity will deliver 10 times CDNA2 performance. Yeah no less!
PS: Subcribe to get the follow up story

 

[Kaotik]

Nothing news, Nvidia is using TSMC CoWos since 2017 in Volta
https://www.nvidia.com/en-us/data-center/v100/


These guys really write for nothing. so easy... let me tweet some exclusives too...

Spoiler: Nvidia Hopper CoWoS performance leak (Please Click!)

A "secret industry source" (= I put it straight out of my a*s) told me that next gen datacenter Hopper is a chiplet MCM design put in a gigantic CoWoS package with 96GB of HBM2e. This monstrosity will deliver 10 times CDNA2 performance. Yeah no less!
PS: Subcribe to get the follow up story

Actually since 2016, GP100 used CoWoS too

 

[DegustatoR]

Believe it or not, the is as fresh as it comes for this title.

They do these tests each month actually, the last one is a week old:

https://www.computerbase.de/thema/grafikkarte/rangliste/#diagramm-performancerating-2560-1440

RDR2 specifically got a speed increase on Turing with the first couple of patches and a couple of NV driver updates. The renderer is obviously heavily GCN optimized so NV h/w suffered a bit at launch. Seems rather in line with your average performance expectations now though.

 

[CarstenS]

Those benchmarks seem to be identical to the ones in the earlier article, just not rounded to whole numbers.

 

[DegustatoR]

Those benchmarks seem to be identical to the ones in the earlier article, just not rounded to whole numbers.

It's very possible that there hasn't been any performance changes since around Dec-Jan in RDR2.

 

[techuse]

Seems Turing has improved in RDR 2. it would be nice if Pascal didnt perform like crud in so many modern games.

 

[Remij]

Something's happening on the 19th

https://twitter.com/x/status/1237657045022781440

 

[Pinstripe]

https://twitter.com/x/status/1237748855426310144

https://twitter.com/x/status/1237749126059528192

Hmm...

 

[McHuj]

I assume the ~140% perf over 2080Ti is in raster performance and not RTX performance? If it includes RTX performance, that is very disappointing.

I think Ampere needs to double RTX performance (and even that would be insufficient at the sub x70 tiers)

 

[techuse]

https://twitter.com/x/status/1237748855426310144

https://twitter.com/x/status/1237749126059528192

Hmm...

Very believable IMO. Hard to see overall performance being more than 50% faster as a best case.

 

[Freemantle]

That would be a very minor upgrade if true, looking at the most recent RTX 2080 Super review over @ TPU shows the full 3072 core TU104 already being anything from 12 to 18% slower than a reference 2080 Ti depending on the resolution used:

https://www.techpowerup.com/review/palit-geforce-rtx-2080-super-gaming-pro-oc/27.html

Based on that chart showing the new 3072 core GA104 apparently offering around 95% of the performance of the 2080 Ti.

 

[DegustatoR]

That would be a very minor upgrade if true

Impossible to say without knowing which products will go into which price ranges.

 

[Freemantle]

Impossible to say without knowing which products will go into which price ranges.

Yeah that's true, if it powers the xx70 tier or even xx60 then it's pretty nice. Of course other metrics like power consumption and presumedly improved RTX performance come into play.

 

[Deleted member 2197]

This information comes from Twitter user CorgiKitty, who is relatively new to the leaks scene, so we might need some additional salt this time.

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

 

[DavidGraham]

This unknown twitter guy keeps backtracking on his statements, obviously he knows very little, if the chip just taped out how he would know the other chips/full lineup? The clocks, performance target?

Also I thought NVIDIA stated that majority of their chips is going to be from TSMC and not Samsung? If the gaming chips is from Samsung then that statement is obviously wrong.

Hard to see overall performance being more than 50% faster as a best case.

Maybe for the 3080 class, but definitely not for the 3080Ti.

 

[Rootax]

I don't think they will go all out before seeing rdna2/big navi anyway.

 

[techuse]

This unknown twitter guy keeps backtracking on his statements, obviously he knows very little, if the chip just taped out how he would know the other chips/full lineup? The clocks, performance target?

Also I thought NVIDIA stated that majority of their chips is going to be from TSMC and not Samsung? If the gaming chips is from Samsung then that statement is obviously wrong.

Maybe for the 3080 class, but definitely not for the 3080Ti.

For equivalent tiers. 3080 being 50% faster than 2080ti is a pipe dream IMO.