AMD Navi 23 ‘NVIDIA Killer’ GPU Rumored to Support Hardware Ray Tracing, Coming Next Year

Lokken86 · 21 Sep 2020 at 01:22

Wrinkly said:
Got a Asus TUF 3080 OC on order as I want RT. I just can't see AMD's gen 1 RT catching Nvidia's gen 2.

That's the whole point of my question.

g67575 · 21 Sep 2020 at 01:48

Read this, it explains how Microsoft / AMD have been able to get about the same peformance from the Series S GPU (RDNA 2, 20 CUs), as the last gen Xbox One X (GCN, 40 CUs):
https://forums.overclockers.co.uk/t...gpus-rdna-2-cus-perform-much-better.18899482/

I looks like there is a large performance improvement from GCN GPUs (such as Vega / Radeon VII) to RDNA 2.

DDH · 21 Sep 2020 at 01:53

humbug said:
Surprisingly no, AMD has spent more $

Where is that data from?

This site begs to differ
https://www.macrotrends.net/stocks/charts/AMD/amd/research-development-expenses

Nvidia
https://www.macrotrends.net/stocks/charts/NVDA/nvidia/research-development-expenses

DDH · 21 Sep 2020 at 01:55

Lokken86 said:
Who'd get one if it was equal to a 3080 in rasterisation but only Turing level RT performance? If yes how much less would you want to pay for that?

This is pretty funny, when ampere is Turing levels of RT performance

Rroff · 21 Sep 2020 at 01:57

DDH said:
This is pretty funny, when ampere is Turing levels of RT performance

In games that make heavier use of RT the uplift over Turing is considerable - it is when you have heavy use of older non-RT rendering in play and only a small number of effects using RT that the gains are less significant.

humbug · 21 Sep 2020 at 01:57

Forbs.

https://www.forbes.com/sites/greats...he-table-with-intel-corporation/#869e32235951

DDH · 21 Sep 2020 at 01:58

g67575 said:
Why are ppl so obsessed with RDNA 2 being 80 CUs?

That probably wouldn't double the performance of say a GPU like the 40 CU 5700 XT. We know that because doubling the shader count alone did not double the performance of the RTX 3080 vs the RTX 2080 TI.

Instead, there is around a 1/3 increase in performance vs the RTX 2080 TI. That's because increasing the core count will only boost the TFlop count, not other areas like Texture rate and Pixel rate.

Assuming the Xbox Series X GPU die size is 170-200mm² (not the whole APU), we could see a RDNA 2 GPU with double the die size so around 400mm².

The question is, would increasing the die size of a Series X like GPU by 100% give a similar boost in performance? We know the Series X GPU is much more powerful than the Series S and the whole APU is about 89.4% larger (both have practically the same CPU).

Same guy who called ampere on samsung said biggest navi has 80cus. 80cus would be 5120 cores, so very similar to the 3090 in that respect. Personally id like them to go higher. MI100 is rumored to have 128

DDH · 21 Sep 2020 at 02:00

Rroff said:
In games that make heavier use of RT the uplift over Turing is considerable - it is when you have heavy use of older non-RT rendering in play and only a small number of effects using RT that the gains are less significant.

which games?

LeMson · 21 Sep 2020 at 02:03

DDH said:
which games?

Pong

g67575 · 21 Sep 2020 at 02:58

Who wants to 'bet'. I think an RDNA 2 GPU with 60 CUs will have 60-66% better performance than the GCN based Radeon VII.

This is mostly based on what I've read here:
https://www.theverge.com/2020/9/17/...xt-gen-gaming-performance-gpu-memory-hardware

Richdog · 21 Sep 2020 at 03:30

DDH said:
This is pretty funny, when ampere is Turing levels of RT performance

Lolwhut? Ampere is no improvement whatsoever in RT performance over a
Thuring? How can people write this drivel with a straight face? :eek:

Grim5 · 21 Sep 2020 at 03:41

So Ampere has shown that current games doesn't scale well after 5000 cores, wonder how that will affect big Navi

Freddie1980 · 21 Sep 2020 at 04:12

DDH said:
which games?

Quake 2 which is fully path traced.

Ampere has lots more RT performance then Turing but the issue is the pipeline is shared with everything else so when mixed with traditional rendering techniques things get congested and the GPU can't fully utilise all the extra cores.

Wrinkly · 21 Sep 2020 at 05:34

DDH said:
which games?

I'm buying in as I feel RT is usuable this gen, at least at 1440p.

https://www.eurogamer.net/articles/digitalfoundry-2020-nvidia-geforce-rtx-3080-review?page=6

Code:

Control: 1440p, DX12, High, High RT, TAA

   2080     35
   2080Ti   46
   3080     70

Metro Exodus: 1440p, DX12, Ultra, Ultra RT, TAA

   2080     51
   2080Ti   66
   3080     90

Battlefield 5: 1440p, DX12, Ultra, Ultra RT, TAA

   2080     70
   2080Ti   92
   3080    115

Quake 2: 1440p, RTX, Vulkan, Max Settings

   2080     31
   2080Ti   41
   3080     62

Battlefield 5 is an odd one as they drastically cut the amount of RT in game since launch.

Troezar · 21 Sep 2020 at 07:19

The biggest issue with this discussion is we don't know what we don't know. I will bet that whatever the "leakers" are saying it's wrong and there will be a surprise or two.

As for RT, it's definitely the future but still looks a generation or two away for me. Not 5 mins ago the big arguements in here were over a few FPS difference when we were already well over 100 FPS. Now all of a sudden less than 60 FPS which was previously "unplayable" is now acceptable because "shiny".

The same goes for vendor specific technology. I'm not buying on the basis that something might be implemented if developers decide to and could be of variable success. That has pretty much failed every time, SLI, Crossfire, PhysX... currently ray tracing and DLSS seem very sparse in their support. This should improve over time hopefully.

Richdog · 21 Sep 2020 at 07:34

Troezar said:
I will bet that whatever the "leakers" are saying it's wrong

...because everything the leakers say is always wrong? :confused:

chroniclard · 21 Sep 2020 at 07:40

Troezar said:
The biggest issue with this discussion is we don't know what we don't know. I will bet that whatever the "leakers" are saying it's wrong and there will be a surprise or two.

As for RT, it's definitely the future but still looks a generation or two away for me. Not 5 mins ago the big arguements in here were over a few FPS difference when we were already well over 100 FPS. Now all of a sudden less than 60 FPS which was previously "unplayable" is now acceptable because "shiny".

The same goes for vendor specific technology. I'm not buying on the basis that something might be implemented if developers decide to and could be of variable success. That has pretty much failed every time, SLI, Crossfire, PhysX... currently ray tracing and DLSS seem very sparse in their support. This should improve over time hopefully.

Also not fussed with RT or DLSS style gimmicks. I just want solid 1440p+ performance.

(Which at the moment the 3080 is king of)

Chuk_Chuk · 21 Sep 2020 at 08:06

g67575 said:
That probably wouldn't double the performance of say a GPU like the 40 CU 5700 XT. We know that because doubling the shader count alone did not double the performance of the RTX 3080 vs the RTX 2080 TI.

Instead, there is around a 1/3 increase in performance vs the RTX 2080 TI. That's because increasing the core count will only boost the TFlop count, not other areas like Texture rate and Pixel rate.
.

Grim5 said:
So Ampere has shown that current games doesn't scale well after 5000 cores, wonder how that will affect big Navi

Different architects. Just because something doesn't scale well on Ampere doesn't mean the same will be true for Navi.

Watching Nerdtechgasms video on the Ampere architect. It seems that Nvidia is going the way of GCN with a more compute focussed card. That is why you will see a greater speed up on compute tasks with Ampere, than on gaming tasks.

It is interesting how paths cross. Nvidia going for a more compute focussed card and AMD going for a more gaming focussed card.

BigBANGtheory · 21 Sep 2020 at 08:31

RavenXXX2 said:
Fairly sure Microsoft and Sony bankrolled RDNA development. I think we are in for something special come the end of October.

yes indeed a smart move and partnership by AMD if as people are saying they are able to use the IP in dGPUs

Grim5 · 21 Sep 2020 at 08:42

Chuk_Chuk said:
Different architects. Just because something doesn't scale well on Ampere doesn't mean the same will be true for Navi.

Watching Nerdtechgasms video on the Ampere architect. It seems that Nvidia is going the way of GCN with a more compute focussed card. That is why you will see a greater speed up on compute tasks with Ampere, than on gaming tasks.

It is interesting how paths cross. Nvidia going for a more compute focussed card and AMD going for a more gaming focussed card.

I saw this video where this guy compares his XC3 3080 to his 2080ti

He found that Horizon Zero Dawn doesn't work with Ampere so his 2080ti is 20% faster in that game, it seems to support the theory that games must be developed for that architecture