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

pete910 · 27 Sep 2020 at 22:04

Navip said:
Any rumours of AMD supporting VRR on new cards? I assume they will have HDMI 2.1
I have C9, would like to have options, don't want to be locked to Nvidia only cards, since LG refused to add freesync on 2019 models

AMD have supported VRR since their 2xx series cards, Not working on LG earier models it;s LGs issue. it works on the CX range(2020).

SmokSmog · 27 Sep 2020 at 22:13

Humbug is adding percentages, someone please shoot him.

Increasing clocks by 31,5% and IPC by 10% isn't 41,5% speed bump but 44,65% ( 1,315x1,1=1,4465)

10% ipc 50% shaders 23,5% clocks = 2,03775 ( 103%)

10% ipc 100% shaders 16% clocks = 2,552 (155%) Nvidia devastation

gpuerrilla · 27 Sep 2020 at 22:20

SmokSmog said:
10% ipc 100% shaders 16% clocks = 2,552 (155%) Nvidia devastation

<Jensen November 2020>
Quick lets launch these 30 series cards, forget the watts, heat, crashes we dont have time!
</leather>

TNA · 27 Sep 2020 at 22:22

humbug said:
How about AMD put 12GB of VRam on it and sell it for £400? Would that make more sense?

How about £600? AMD are not a charity you know?

chris85oc · 27 Sep 2020 at 22:26

Remember. If/when the hype train crashes, you’ll end up in disappointment. And it won’t even be AMDs fault this time but random ‘techtubers’ + random forum fanboy users.

4K8KW10 · 27 Sep 2020 at 22:28

chris85oc said:
Remember. If/when the hype train crashes, you’ll end up in disappointment. And it won’t even be AMDs fault this time but random ‘techtubers’ + random forum fanboy users.

We'll see. Some promises are flying very generously, though

https://videocardz.com/newz/amd-navi-21-to-feature-80-cus-navi-22-40-cus-and-navi-23-32-cus

gpuerrilla · 27 Sep 2020 at 22:30

chris85oc said:
Remember. If/when the hype train crashes, you’ll end up in disappointment. And it won’t even be AMDs fault this time but random ‘techtubers’ + random forum fanboy users.

Sound a bit bitter there chris? At least your showing your true colours..

g67575 · 27 Sep 2020 at 23:03

Newb question - What happens when I watch a 8K / 4K video on my 4K monitor, with a desktop resolution of 1080p?

EDIT - Youtube reports the viewport resolution as 1080p and the 'current / optimal res' as 8k. So, I think it downscales the internal resolution of the video to whatever you set the desktop resolution to.

I suppose you can scale up 4k desktop resolution to avoid this problem, but I found this desktop res caused to many problems to be worth it.

Ideally, choosing full screen in videos would change the desktop resolution automatically (if needed), as can be done in games in full screen mode...

Kirby Wurm · 28 Sep 2020 at 00:07

SmokSmog said:
Humbug is adding percentages, someone please shoot him.

Increasing clocks by 31,5% and IPC by 10% isn't 41,5% speed bump but 44,65% ( 1,315x1,1=1,4465)

10% ipc 50% shaders 23,5% clocks = 2,03775 ( 103%)

10% ipc 100% shaders 16% clocks = 2,552 (155%) Nvidia devastation

Scaling isn't linear with the number of shaders I would guess. Certain workloads perhaps but wouldn't more ROPs be required to give a more consistent performance increase in all scenarios?

SmokSmog · 28 Sep 2020 at 01:41

Kirby Wurm said:
Scaling isn't linear with the number of shaders I would guess. Certain workloads perhaps but wouldn't more ROPs be required to give a more consistent performance increase in all scenarios?

I just calculated humbug ramblings, nothing more , I'm aware that scaling isn't linear.

Scougar · 28 Sep 2020 at 02:52

Is that 64 figure the ROPS?

EastCoastHandle · 28 Sep 2020 at 03:18

Latest rum's

Navi 21 - 80CUs, 2.05-2.2GHz = 21-22.5TFLOPS
Navi 22 - 40CUs, 2.5GHz = 12.8TFLOPS
Navi 23 - 32CUs
Navi 31 - 80CUs with identical configuration as Navi 21; RDNA 3???
Van Gogh APU - 1 SA with 8 RDNA CUs = 8 CUs
Rembrandt APU - 2 SAs with 6 RDNA CUs each = 12 CUs

4K8KW10 · 28 Sep 2020 at 05:41

Scougar said:
Is that 64 figure the ROPS?

If the ROPs are modified and made faster, yes, they could stay only 64.

Nablaoperator · 28 Sep 2020 at 06:06

Theres no simple way to make ROPs faster.. it will be a huge breakthrough if that happens. its a fixed function pipeline..
theoretically 1 ROP can only output 1 pixel per clock.. because pixels get overwritten hundreds of times per frame we get much lower FPS
maybe they can improve pipeline prediction to improve effective throughput.. but as noted that will be a big breakthrough, both xbox/ps would have been bigging this up in their marketing collateral by now..

i believe atleast 80 ROPs... expect 96, anything over that and i will be amazed

Nablaoperator · 28 Sep 2020 at 06:33

Kirby Wurm said:
Scaling isn't linear with the number of shaders I would guess. Certain workloads perhaps but wouldn't more ROPs be required to give a more consistent performance increase in all scenarios?

We have a much simpler way to estimate targeted performance:
50% improvement per W .. RDRs leaks are suggesting 60%
old navi tdp 225W
navi 21 rumoured tdp 300W
so navi 21 would be 2.1x (300/225*1.6) old navi (RX 5700 xt) fps throughput

since these are commercial targets i am assuming these translate to end performance and not how fast instructions are being processed, as many such instructions have to be thrown out and don't influence the final frame

Troezar · 28 Sep 2020 at 06:44

Nablaoperator said:
We have a much simpler way to estimate targeted performance:
50% improvement per W .. RDRs leaks are suggesting 60%
old navi tdp 225W
navi 21 rumoured tdp 300W
so navi 21 would be 2.1x (300/225*1.6) old navi (RX 5700 xt) fps throughput

since these are commercial targets i am assuming these translate to end performance and not how fast instructions are being processed, as many such instructions have to be thrown out and don't influence the final frame

On that basis I'm thinking a smidgen below the 3080's performance assuming no secret sauce that boosts FPS further.

Nablaoperator · 28 Sep 2020 at 07:14

Troezar said:
On that basis I'm thinking a smidgen below the 3080's performance assuming no secret sauce that boosts FPS further.

That would be the best outcome... from a queue mgmt perspective

Jacky60 · 28 Sep 2020 at 07:33

james.miller said:
uuh, what?

I think you need to step away from the keyboard lad, before you bust a blood vessel. Where'd that aggression come from? Somebody **** in your cornflakes this morning?

'petulant' reaction aside, we have nothing but rumours. rumours built on rumours, no less. I don't need evidence that a rumour isnt true because that's not how it works, is it? but you know that of course.

sorely require? jog on? WTF is wrong with you?

The guy's seriously unwell, seems he comes here to vent all his frustrations with everyday life on people in the graphics card forum. It is quite funny how he gets apoplectic about the most trivial things

.

Kirby Wurm · 28 Sep 2020 at 08:13

SmokSmog said:
I just calculated humbug ramblings, nothing more , I'm aware that scaling isn't linear.

My comment as more aimed at humbug, apologies for not making that clear. Not that I don't wish it was true!

Nablaoperator said:
We have a much simpler way to estimate targeted performance:
50% improvement per W .. RDRs leaks are suggesting 60%
old navi tdp 225W
navi 21 rumoured tdp 300W
so navi 21 would be 2.1x (300/225*1.6) old navi (RX 5700 xt) fps throughput

since these are commercial targets i am assuming these translate to end performance and not how fast instructions are being processed, as many such instructions have to be thrown out and don't influence the final frame

Interesting. A month to go until we get some solid info, it'll be a long wait!

Troezar · 28 Sep 2020 at 08:23

Kirby Wurm said:
Interesting. A month to go until we get some solid info, it'll be a long wait!

Indeed. I wonder how many times RTG Paul can spin the same information "exclusive" between now and then?