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

gpuerrilla · 22 Aug 2020 at 12:07

Its one of two things @Stu999 - either AMD are giggling as they are poised to do some disruption knowing they have bridged the gap in performance, or they are sweating on it hoping nvidia fail on both counts of power requirements and pricing.

Satchfanuk · 22 Aug 2020 at 12:09

Calin Banc said:
RT co-processor.

If you want to keep believing that rumour, despite multiple sources saying it doesn't exist, then you go right ahead

Stu999 · 22 Aug 2020 at 12:09

Th0nt said:
Its one of two things @Stu999 - either AMD are giggling as they are poised to do some disruption knowing they have bridged the gap in performance, or they are sweating on it hoping nvidia fail on both counts of power requirements and pricing.

Yeah, it’s definitely hard to guess which way it is going to go. Very interesting times though.

Satchfanuk · 22 Aug 2020 at 12:12

Participant said:
I think it says that nVidia are prepared to issue a card that will enable 4K 120Hz ultra.

I think it's more likely to be because Nvidia are wringing its neck in desperation to maintain the overall performance crown but you do you

CAT-THE-FIFTH · 22 Aug 2020 at 12:15

FoxEye said:
Yeah there has to be some reason for changing from xx80 Ti to xx90.

I don't see that being a throw-away decision by marketing. I see that as being a calculated change because of ... something.

It also wouldn't surprise me if the GA102 is relatively small(under 500MM2),so Nvidia makes even more margins!

Or maybe the RTX3090 has that RT conprocessor Coreteks was talking about.

Satchfanuk · 22 Aug 2020 at 12:22

CAT-THE-FIFTH said:
It also wouldn't surprise me if the GA102 is relatively small(under 500MM2),so Nvidia makes even more margins!

Or maybe the RTX3090 has that RT conprocessor Coreteks was talking about.

GA102 is 627mm2, so not exactly huge but not small by any means.

https://www.techpowerup.com/gpu-spe...GA102 GPU uses the,mapping units and 96 ROPs.

CAT-THE-FIFTH · 22 Aug 2020 at 12:34

Satchfanuk said:
GA102 is 627mm2, so not exactly huge but not small by any means.

https://www.techpowerup.com/gpu-specs/nvidia-ga102.g930#:~:text=NVIDIA's GA102 GPU uses the,mapping units and 96 ROPs.

How did TPU get that number without any leaked die shots?

Satchfanuk · 22 Aug 2020 at 12:38

CAT-THE-FIFTH said:
How did TPU get that number without any leaked die shots?

I think there's a certain amount of guesswork involved in that spec sheet... it says 'speculation' in the notes at the bottom of the page :rolleyes:

humbug · 22 Aug 2020 at 12:44

Satchfanuk said:
GA102 is 627mm2, so not exactly huge but not small by any means.

https://www.techpowerup.com/gpu-specs/nvidia-ga102.g930#:~:text=NVIDIA's GA102 GPU uses the,mapping units and 96 ROPs.

7552 Shaders. vs 4,352 on the 2080TI, that's +73%, that should make it about ~50% faster than the 2080TI.

Also. TDP, 400 Watts, vs 250 Watts 2080TI.

Satchfanuk · 22 Aug 2020 at 12:51

humbug said:
7552 Shaders. vs 4,352 on the 2080TI, that's +73%, that should make it about ~50% faster than the 2080TI.

Also. TDP, 400 Watts, vs 250 Watts 2080TI.

Assuming yields allow all those shaders to be active

LePhuronn · 22 Aug 2020 at 12:55

Satchfanuk said:
Did Jensen just have a lot of scrap metal lying around the house that he wanted to get rid of?

He's used wood screw before

Satchfanuk · 22 Aug 2020 at 12:56

LePhuronn said:
He's used wood screw before

Ah those were the days

At least now, when it comes to upgrading, you won't have to worry about selling it on second hand, just take it to the local scrap merchants and weight it in

humbug · 22 Aug 2020 at 13:01

Satchfanuk said:
Assuming yields allow all those shaders to be active

They will...

humbug said:
7552 Shaders. vs 4,352 on the 2080TI, that's +73%, that should make it about ~50% faster than the 2080TI.

Also. TDP, 400 Watts, vs 250 Watts 2080TI.

Ok, lets says there is no "80 CU RDNA2" lets say its 64 CU, AMD have done 64 CU's many times before lets assume that's it.

5700XT = 40 CU @ 1900Mhz

Big Navi 64 CU's = +60%
The PS5 runs at 2.23Ghz: vs the 5700XT that's + 17%

Lets round Big Navi up to 2.3Ghz because if the PS5 can do 2.23Ghz i'm sure Big Navi can do at least 2.3Ghz.

The 2080TI is 30% faster than the 5700XT.

Big Navi +60% Shaders + 21% clocks = 5700XT + 81% at 0.7 Scaling = +57% actual.

So a 64 CU 2.3Ghz Big Navi would be about 30% faster than a 2080TI, GA102 would be 20% faster again.

This assuming 0 IPC gain RDNA2 vs RDNA1.

Poneros · 22 Aug 2020 at 13:07

humbug said:
They will...

Ok, lets says there is no "80 CU RDNA2" lets say its 64 CU, AMD have done 64 CU's many times before lets assume that's it.

5700XT = 40 CU @ 1900Mhz

Big Navi 64 CU's = +60%
The PS5 runs at 2.23Ghz: vs the 5700XT that's + 17%

Lets round Big Navi up to 2.3Ghz because if the PS5 can do 2.23Ghz i'm sure Big Navi can do at least 2.3Ghz.

The 2080TI is 30% faster than the 5700XT.

Big Navi +60% Shaders + 21% clocks = 5700XT + 81% at 0.7 Scaling = +57% actual.

So a 64 CU 2.3Ghz Big Navi would be about 30% faster than a 2080TI, GA102 would be 20% faster again.

This assuming 0 IPC gain RDNA2 vs RDNA1.

See this if it helps (5700XT):

Satchfanuk · 22 Aug 2020 at 13:10

humbug said:
They will...

Ok, lets says there is no "80 CU RDNA2" lets say its 64 CU, AMD have done 64 CU's many times before lets assume that's it.

5700XT = 40 CU @ 1900Mhz

Big Navi 64 CU's = +60%
The PS5 runs at 2.23Ghz: vs the 5700XT that's + 17%

Lets round Big Navi up to 2.3Ghz because if the PS5 can do 2.23Ghz i'm sure Big Navi can do at least 2.3Ghz.

The 2080TI is 30% faster than the 5700XT.

Big Navi +60% Shaders + 21% clocks = 5700XT + 81% at 0.7 Scaling = +57% actual.

So a 64 CU 2.3Ghz Big Navi would be about 30% faster than a 2080TI, GA102 would be 20% faster again.

This assuming 0 IPC gain RDNA2 vs RDNA1.

Strange thing is that TPU says that the 3080 = 5120 shaders and 3090 = 5376 shaders?

Unlike the fully unlocked GeForce RTX 3090, which uses the same GPU but has all 5376 shaders enabled, NVIDIA has disabled some shading units on the GeForce RTX 3080 Ti to reach the product's target shader count. It features 5120 shading units, 320 texture mapping units, and 96 ROPs.

https://www.techpowerup.com/gpu-spe... unlocked GeForce,mapping units, and 96 ROPs.

humbug · 22 Aug 2020 at 13:18

Poneros said:
See this if it helps (5700XT):

The 5700XT is very power efficient when tuned, those of who own one and have tuned it know that but other than that i don't see what i can do with this info? AMD have pushed the 5700XT to near its limits to get that extra few % for a lot more in power, this is what they do...

I think one thing we do know is we cannot compare the 5700XT power to RDNA2, the XBox Series X APU has 52 CU's running at 1900Mhz and an 8 core 16 thread Zen 2 CPU, the whole package power is 170 Watts. that's + 30% more CU's at the same clock speed and a Ryzen 3700X for about the same power as a 5700XT. so RDNA2 is much more efficient, probably about "50% better Performance per watt"

humbug · 22 Aug 2020 at 13:22

Satchfanuk said:
Strange thing is that TPU says that the 3080 = 5120 shaders and 3090 = 5376 shaders?

https://www.techpowerup.com/gpu-specs/geforce-rtx-3080-ti.c3581#:~:text=Unlike the fully unlocked GeForce,mapping units, and 96 ROPs.

Don't know then, unless that's an older publication.

I think Nvidia are worried about AMD, that's why we no longer have a 3080TI which that reports to be, now we have a 3090 with 7552 Shaders, i believe Nvidia's top card will be named 3090 and it will have 7552 Shaders because if they don't AMD WILL beat them.

BigBANGtheory · 22 Aug 2020 at 13:23

humbug said:
They will...

Ok, lets says there is no "80 CU RDNA2" lets say its 64 CU, AMD have done 64 CU's many times before lets assume that's it.

5700XT = 40 CU @ 1900Mhz

Big Navi 64 CU's = +60%
The PS5 runs at 2.23Ghz: vs the 5700XT that's + 17%

Lets round Big Navi up to 2.3Ghz because if the PS5 can do 2.23Ghz i'm sure Big Navi can do at least 2.3Ghz.

The 2080TI is 30% faster than the 5700XT.

Big Navi +60% Shaders + 21% clocks = 5700XT + 81% at 0.7 Scaling = +57% actual.

So a 64 CU 2.3Ghz Big Navi would be about 30% faster than a 2080TI, GA102 would be 20% faster again.

This assuming 0 IPC gain RDNA2 vs RDNA1.

which imho is the Navi 22 and puts it squarely against the 3080 which if they can sell for £650 mission accomplished

Navi 21 then rolls out with 80 CU to challenge the 3090 (maybe not beat it) and sells for £1k unless they use lots of HBM2e which case it will be more

Satchfanuk · 22 Aug 2020 at 13:30

humbug said:
Don't know then, unless that's an older publication.

I think Nvidia are worried about AMD, that's why we no longer have a 3080TI which that reports to be, now we have a 3090 with 7552 Shaders, i believe Nvidia's top card will be named 3090 and it will have 7552 Shaders because if they don't AMD WILL beat them.

I don't believe the 7552 shaders. It's a higher number than the A100 and I can't see that being true. TPU have the usual 'speculation' disclaimer all over those pages so I guess they don't know the actual numbers yet.

humbug · 22 Aug 2020 at 13:36

BigBANGtheory said:
which imho is the Navi 22 and puts it squarely against the 3080 which if they can sell for £650 mission accomplished

Navi 21 then rolls out with 80 CU to challenge the 3090 (maybe not beat it) and sells for £1k unless they use lots of HBM2e which case it will be more

If AMD come out with an 80 CU Big Navi Nvidia are beaten.

Satchfanuk said:
I don't believe the 7552 shaders. It's a higher number than the A100 and I can't see that being true. TPU have the usual 'speculation' disclaimer all over those pages so I guess they don't know the actual numbers yet.

If Nvidia don't have a 7552 shader GPU AMD will demolish them.