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

Stanners · 4 Mar 2020 at 15:06

humbug said:
No i don't, i'm still interested to know what @LambChop would do.

You know the answer to that one surely

humbug · 4 Mar 2020 at 15:07

Stanners said:
You know the answer to that one surely

Yeah i think i do....

gpuerrilla · 4 Mar 2020 at 15:08

subbytna said:
Can you imagine the price of that titan though. £2.5k incoming!

Something I could only dream of owning! My limit would be about £500 for a component im afraid.

Stanners · 4 Mar 2020 at 15:08

humbug said:
Yeah i think i do....

We could be wrong!

gpuerrilla · 4 Mar 2020 at 15:11

Ahh is he another Greggo? Well that's a rhetorical question then.

humbug · 4 Mar 2020 at 15:18

Stanners said:
We could be wrong!

I have been wrong many times... it would be a pleasant surprise if i am, tho for as much as i would like @LambChop to say yes he would and then hold him to that when AMD pull one out of the hat i doubt either will happen.

D.P. · 4 Mar 2020 at 16:00

humbug said:
Edit: if RDNA2 has Variable Rate Shading it will be performing better per clock - per shader, and RDNA is already the same IPC as Turing without that.

I see this mentioned a lot and it just highlights a fundamental misunderstand of VRS.

VRS is not free performance. It requires explicit developer support and is not a uniform performance enhance under most uses as it will depend on surface rendering details (e.g., a smooth textured uniform lit surface can be rendered at a lower rate than a complex surface).

So far veyr few games support this, where it is supported performance gain are typically 2-4%. Turing already supports VRS, so any game that will have VRS added in the future will automatically help Turing and Ampere.

There also isn't a difference in implementations so it is not like AMD's VRS will magically be better than nvidia's The performance gains will be near identical.

as for IPC, that i impossible to compare between generations let alone between IHVs and in general for GPUs is utterly meaningless.

humbug · 4 Mar 2020 at 16:05

D.P. said:
I see this mentioned a lot and it just highlights a fundamental misunderstand of VRS.

VRS is not free performance. It requires explicit developer support and is not a uniform performance enhance under most uses as it will depend on surface rendering details (e.g., a smooth textured uniform lit surface can be rendered at a lower rate than a complex surface).

So far veyr few games support this, where it is supported performance gain are typically 2-4%. Turing already supports VRS, so any game that will have VRS added in the future will automatically help Turing and Ampere.

There also isn't a difference in implementations so it is not like AMD's VRS will magically be better than nvidia's The performance gains will be near identical.

as for IPC, that i impossible to compare between generations let alone between IHVs and in general for GPUs is utterly meaningless.

VRS does not require developer support, its determined at the hardware level with the DX12 API.

D.P. · 4 Mar 2020 at 16:46

humbug said:
VRS does not require developer support, its determined at the hardware level with the DX12 API.

No, this is completely wrong

https://hexus.net/tech/news/graphics/128588-microsoft-adds-variable-rate-shading-support-directx-12/

VRS allows developers to selectively reduce the shading rate in areas of the frame where it won’t affect visual quality, letting them gain extra performance in their games

https://devblogs.microsoft.com/directx/variable-rate-shading-a-scalpel-in-a-world-of-sledgehammers/

Our VRS API lets developers set the shading rate in 3 different ways:

Per draw

Within a draw by using a screenspace image

Or within a draw, per primitive

The Developer has to actively select which draw call or even which triangles will be shaded at which rate.

humbug · 4 Mar 2020 at 17:13

D.P. said:
No, this is completely wrong

https://hexus.net/tech/news/graphics/128588-microsoft-adds-variable-rate-shading-support-directx-12/

https://devblogs.microsoft.com/directx/variable-rate-shading-a-scalpel-in-a-world-of-sledgehammers/

The Developer has to actively select which draw call or even which triangles will be shaded at which rate.

Read this...

https://www.pcgamesn.com/microsoft-variable-rate-shading-directx-12-nvidia-intel-gpu

DX12 makes it agnostic, its at the API level. its developer input only in the sense of enabling it..

TNA · 4 Mar 2020 at 18:20

No time to read those links, but wondering who is wrong? It is always fun to see D.P. get owned. Not so much with humbug, he gets owned quite often

humbug · 4 Mar 2020 at 18:31

TNA said:
No time to read those links, but wondering who is wrong? It is always fun to see D.P. get owned. Not so much with humbug, he gets owned quite often

TNA · 4 Mar 2020 at 20:12

humbug said:

233 · 4 Mar 2020 at 20:56

anyone else think AMD are just a bit too quiet

i suspect their going to turn round and give Nvidia the biggest bitch slap imaginable, GPU equivilant of Zen inbound

or then again that might be intel.

regardless of the outcome would be epic to see green and red fighting it out for top spot if only to bring the pricing back to something approaching almost sane levels. Maybe intel is going to become the go to budget GPU option for performance/£ in 2020

gpuerrilla · 4 Mar 2020 at 21:12

Too many flops from the gpu wing over recent years means I no longer buy the pre-hype - just wait for the actual before weighing it up. If they can produce anything like the cpu model then it would be awesome for all consumers having concrete choice regarding performance whichever side you chose.

Gerard · 4 Mar 2020 at 21:23

Th0nt said:
Too many flops from the gpu wing over recent years means I no longer buy the pre-hype - just wait for the actual before weighing it up. If they can produce anything like the cpu model then it would be awesome for all consumers having concrete choice regarding performance whichever side you chose.

Intel probably thought the same after amd rolled out bulldozer and its iterations, rested on their laurels then got pumped hard. Same thing could easily happen in the gpu space.

gpuerrilla · 4 Mar 2020 at 21:29

Gerard said:
Same thing could easily happen in the gpu space.

I always believe it could its a long shot though as the budget between them both is rather large. Also nvidia will have seen how ryzen has wobbled intel over the past few years so its not like they can have an excuse.

hyperseven · 4 Mar 2020 at 21:37

Th0nt said:
I always believe it could its a long shot though as the budget between them both is rather large. Also nvidia will have seen how ryzen has wobbled intel over the past few years so its not like they can have an excuse.

Yep, they 'wobbled' Intel so much it caused them to have a record Q4 in 2019

pmc25 · 4 Mar 2020 at 21:41

hyperseven said:
Yep, they 'wobbled' Intel so much it caused them to have a record Q4 in 2019

From sectors AMD does not compete in, or is only just beginning to.

Their desktop business might as well be in ruins by the end of this year.

Also, looking at the details of that DOE supercomputer contract that was announced today with Zen 4 and likely the 5nm shrink of Arcturus, both Intel and NVIDIA might be in for a barren stretch in the very lucrative exascale space. It appears Zen 4 will be an even bigger jump than Zen 2 or Zen 3, given what they're doing with Infinity Fabric.

hyperseven · 4 Mar 2020 at 22:22

pmc25 said:
From sectors AMD does not compete in, or is only just beginning to.

Their desktop business might as well be in ruins by the end of this year.

Also, looking at the details of that DOE supercomputer contract that was announced today with Zen 4 and likely the 5nm shrink of Arcturus, both Intel and NVIDIA might be in for a barren stretch in the very lucrative exascale space. It appears Zen 4 will be an even bigger jump than Zen 2 or Zen 3, given what they're doing with Infinity Fabric.

Well, that has filled my speculation and nonsense quota for today

'might', 'appears', 'likely'...really. Nice to dream