Nvidia Pascal Announcement

[Ext3h]

What you are describing is very similar to NVIDIA multirate shading tech in VRWorks already supported on Maxwell

You are right, it's not much different. Except that back then VRWorks only did multi-resolution, means different viewport, but only a single projection matrix. Going by the GL_AMD_vertex_shader_viewport_index extension, AMD features something just as capable as Nvidias fast geometry shader path, so there shouldn't be any actual difference in performance.

But you don't need to allow arbitrary quads for viewports in order to do lens matched shading. You can just subdivide into 12 or 16 rectangles right away, and just apply an adjusted projection matrix to the vertex dispatched by the GS to each single viewport. Apart from trading masking for rectangular viewports, that's exactly the same as what Nvidia now showed.

(Well, you can't divide into more than 9 viewports on Maxwell, because that's the limit for the viewport array size on Maxwell. Interesting enough, GCN allowed 16 viewports from the beginning, just as Pascal does now.)

 

[Benetanegia]

Tottentranz you might want to do some research about the guy that wrote that article pretty sure he is on Overclockers.net forums......

The guy doesn't even know how to read percentages...

He says:

According to techpower up at 1080p the GTX 1080 leads the 980 by 36% and at 1440p the GTX 1080 has a 40% lead over the GTX 980. Obviously the reviews ran the GTX 980 at stock clocks.

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/26.html

He's just doing 100%-60% = 40% faster.

 

[3dilettante]

The following shows a minor difference after toggling asynchronous compute:
http://www.computerbase.de/2016-05/geforce-gtx-1080-test/11/

 

[trinibwoy]

The following shows a minor difference after toggling asynchronous compute:
http://www.computerbase.de/2016-05/geforce-gtx-1080-test/11/

And here too though results are inconsistent.

http://www.techspot.com/review/1174-nvidia-geforce-gtx-1080/page5.html

 

[CSI PC]

You are right, it's not much different. Except that back then VRWorks only did multi-resolution, means different viewport, but only a single projection matrix. Going by the GL_AMD_vertex_shader_viewport_index extension, AMD features something just as capable as Nvidias fast geometry shader path, so there shouldn't be any actual difference in performance.

But you don't need to allow arbitrary quads for viewports in order to do lens matched shading. You can just subdivide into 12 or 16 rectangles right away, and just apply an adjusted projection matrix to the vertex dispatched by the GS to each single viewport. Apart from trading masking for rectangular viewports, that's exactly the same as what Nvidia now showed.

(Well, you can't divide into more than 9 viewports on Maxwell, because that's the limit for the viewport array size on Maxwell. Interesting enough, GCN allowed 16 viewports from the beginning, just as Pascal does now.)

So why the 80% improved speed between the old and new way when considering the two screens-lenses in VR that are actually different (for accurate perspective)?
I thought the discussion earlier concluded how the single pass render state can only be applied to Pascal and not Maxwell, not due to the viewports but the double or triple window/perspective view; both of these come back to the Polymorph 4 engine hardware.

Not sure how you can get those performance gains with the way your explaining *shrug*.
Thanks

 

[pjbliverpool]

GTX 1080: What's not being discussed.

Some very valid points:

1 - Obvious BS with "faster than 980 SLI" general claims.

2 - Initial "9 TFLOPs" number ninja-edited to 8.2 TFLOPs after the presentation and before the reviews

3 - GTX 1080 results absent from AOTS benchmark database

4 - "Async Compute" claimed everywhere, but zero performance gain observed from the only game that uses it (maybe AOTS for nvidia is still using the old dedicated nvidia path without async enabled, so there's some benefit of the doubt in here IMO)

5 - Rise of the Tomb Raider being benched everywhere in an admittedly (by the devs themselves) broken DX12 mode. Is every reviewer out there so damn ignorant regarding this case?.

6 - New SLI bridges are not compatible with the old ones and are not bundled with the new cards, cost $30 and are rigid. This means if you want to do SLI, pay another $30. Change motherboards with different spacing, pay another $30.

7 - Where is Doom's Vulkan mode? It was available for a live demo 2 weeks prior to the launch but it wasn't available for launch?
I wonder what the performance upgrades between IHVs will be for an API whose origin is a fork of Mantle...

8 - This one is the funniest:
When the Fury and Fury X came out, every reviewer tested with the factory-overclocked (and some even manually overclocked) 980 and 980 Ti cards because that's what they had in their hands. Come the time to review the GTX 1080, magically everyone has stock-clocked 980 and 980 Ti cards to compare to.

I feel dirty after reading the drivel in the link. I honestly don't understand your negativity towards Nvidia recently. The 1080 is a great card, enjoy it for what it is, I'm sure we're all hoping AMD will also come out with something competitive very soon. p.s. the compute performance of the 1080 at the rated boost speed is 8.88 TF, not 8.2.

 

[Deleted member 13524]

The following shows a minor difference after toggling asynchronous compute:
http://www.computerbase.de/2016-05/geforce-gtx-1080-test/11/

Isn't this within margin of error? The difference is less than 3%..

I honestly don't understand your negativity towards Nvidia recently.

My only bias is to be anti-circlejerking, which I've been seeing a lot.

The GTX 1080 is a great card. It's a new performance king. I wish I had one right now (though if I did I'd probably sell it because I could make a fortune on ebay... and then maybe I'd buy one when they actually became available).

However, there are a number of factors that should be mentioned. The new SLI bridges are crap that IMO exist to just take more money from customers. Non-overclocked Maxwell cards appearing in every review when apparently no one had them during the 390 and Fury releases seems downright dishonest. Talking so much about async compute in the presentation and then have no actual results to prove it is odd at best. Ninja-editing specs just before the review NDAs are lifted is dishonest. Tomb Raider being tested everywhere with a well-known-to-be-broken DX12 patch means either dishonesty or complete ignorance.

And when these points are brought-up, the circlejerk says "no one is mentioning these because they're all unimportant".

 

[3dilettante]

Isn't this within margin of error? The difference is less than 3%..

I cannot read German and don't feel like interpreting auto-translate.
The uncertainty in the results would come from methodology, sample size, and the variation between them.

Decent clustering can still provide confidence for small differences or would the following indicate that the AMD 7970 doesn't have it either:
http://www.anandtech.com/show/10067/ashes-of-the-singularity-revisited-beta/6

 

[lanek]

The following shows a minor difference after toggling asynchronous compute:
http://www.computerbase.de/2016-05/geforce-gtx-1080-test/11/

stupid question maybe, but how are they enable/disable Async compute under DX12 ? Is there a setting in the game for it ?

 

[3dilettante]

There's an ini setting for the application.

stupid question maybe, but how are they enable/disable Async compute under DX12 ?

The settings.ini file can be modified to change whether it is used.

 

[lanek]

There's an ini setting for the application.


The settings.ini file can be modified to change whether it is used.

Ook, i see now. Good to know.

 

[Ext3h]

So why the 80% improved speed between the old and new way when considering the two screens-lenses in VR that are actually different (for accurate perspective)?
I thought the discussion earlier concluded how the single pass render state can only be applied to Pascal and not Maxwell, not due to the viewports but the double or triple window/perspective view; both of these come back to the Polymorph 4 engine hardware.

Not sure how you can get those performance gains with the way your explaining *shrug*.
Thanks

80% in that synthetic demonstration with the high resolution mesh, when switching between two and a single render pass. And in that scenario, the vertex shader was most likely the heavy lifter, with essentially no post processing or any expensive operations in the fragment shader.

What we concluded earlier in the discussion was only that non-rectangular viewports were not possible. Not that you couldn't use more than one projection matrix. Only the non-rectangular viewport is a new feature of the hardware. (Well, and now allowing 16 viewports in the viewport array, instead of only 9.)

For lens matched shading, the performance gain from the reduced oversampling is really just another +30-40%, the other +60% are from using the 980 as a base for comparison. Makes for a total of "2x faster". Which by the way isn't saving more than the previous multi resulution approach - only the visual quality is better.

 

[CSI PC]

........ Non-overclocked Maxwell cards appearing in every review when apparently no one had them during the 390 and Fury releases seems downright dishonest. Talking so much about async compute in the presentation and then have no actual results to prove it is odd at best. Ninja-editing specs just before the review NDAs are lifted is dishonest.

And when these points are brought-up, the circlejerk says "no one is mentioning these because they're all unimportant".

I would say the reference-to-reference comparison should only be with same manufacturer and not competing ones; primarily because AMD/NVIDIA have a different approach to the behaviour-thresholds for the variables involved for headroom; either from say OC on a reference card or say custom AIB.
Reference-to-Reference is only a baseline to see the improvement on previous generation, and even that is not fully clear as at times either company has a situation where a specific model-card is limited in some way and requires AIB to improve on that.
That said I do like it when a site also adds a custom AIB design from previous generation by same manufacturer; pcgameshardware used all references and also a MSI Lightning 980ti.
Anyway usually it is only the early adopters who buy a reference design over the superior custom AIB, and this can be applied to both companies.
Cheers

 

[CSI PC]

80% in that synthetic demonstration with the high resolution mesh, when switching between two and a single render pass. And in that scenario, the vertex shader was most likely the heavy lifter, with essentially no post processing or any expensive operations in the fragment shader.

What we concluded earlier in the discussion was only that non-rectangular viewports were not possible. Not that you couldn't use more than one projection matrix. Only the non-rectangular viewport is a new feature of the hardware. (Well, and now allowing 16 viewports in the viewport array, instead of only 9.)

For lens matched shading, the performance gain from the reduced oversampling is really just another +30-40%, the other +60% are from using the 980 as a base for comparison. Makes for a total of "2x faster". Which by the way isn't saving more than the previous multi resulution approach - only the visual quality is better.

They did it also with multiple VR demos and not just synthetic.
Cheers

 

[CarstenS]

GTX 1080: What's not being discussed.

Some very valid points:

1 - Obvious BS with "faster than 980 SLI" general claims.

2 - Initial "9 TFLOPs" number ninja-edited to 8.2 TFLOPs after the presentation and before the reviews

3 - GTX 1080 results absent from AOTS benchmark database

4 - "Async Compute" claimed everywhere, but zero performance gain observed from the only game that uses it (maybe AOTS for nvidia is still using the old dedicated nvidia path without async enabled, so there's some benefit of the doubt in here IMO)

5 - Rise of the Tomb Raider being benched everywhere in an admittedly (by the devs themselves) broken DX12 mode. Is every reviewer out there so damn ignorant regarding this case?.

6 - New SLI bridges are not compatible with the old ones and are not bundled with the new cards, cost $30 and are rigid. This means if you want to do SLI, pay another $30. Change motherboards with different spacing, pay another $30.

7 - Where is Doom's Vulkan mode? It was available for a live demo 2 weeks prior to the launch but it wasn't available for launch?
I wonder what the performance upgrades between IHVs will be for an API whose origin is a fork of Mantle...

8 - This one is the funniest:
When the Fury and Fury X came out, every reviewer tested with the factory-overclocked (and some even manually overclocked) 980 and 980 Ti cards because that's what they had in their hands. Come the time to review the GTX 1080, magically everyone has stock-clocked 980 and 980 Ti cards to compare to.

I find your lack of difference... sad.

ad1: That's marketing. I'm sure with the recent number of titles not decently supporting multi-GPU, you won't have to use very obscure tests in order to find a mix of tests that support this claim. Full disclosure: I find this claim also ridiculous.

ad2: That number was pointed out initially by considerate press to be too high.

ad3: Maybe people using the gog.com version? (me) Or are considerate enough, not to have their results that were generated when the NDA was still up uploaded somewhere? (me). Maybe.

ad4: Not everywhere.

ad5: For completeness sake, ROTTR is a popular one among other DX12 titles.

ad6: Why would new bridges have to be compatible with old ones? Which card has 3 SLI fingers so that would be of any use? And you still can use your dear old bridges, if you like.

ad7: Coming hopefully? As stated by id software it would. Remember how long it took to first implement Mantle in Battlefield? December turning february (or was it the last day of jan?). Or the birth pains of AotS? Might have something to do with the fact that Doom would be the 2nd game only support it in the first place?

ad8: MSI 980 Ti Lightning, at ~1430 MHz with a 33% overclock the fastest air cooled 980 Ti you can buy. Compare for yourself.
http://www.pcgameshardware.de/Nvidi...8/Specials/Benchmark-Test-Video-1195464/2/#a1
(Ashes, as well as ROTTR also in there)

 

[xEx]

18% more clocks and 15% more performance?

 

[xEx]

Do we have a word on when are the customs cards coming out?

 

[CarstenS]

Seems like ppl believe same day as FE:
https://forum.beyond3d.com/posts/1915242/

Which is weird, because one of the reasone cited for the price premium was TTM.

 

[xEx]

I heard a rumor that Nvidia wont allow customs cards until they sell X amount of FEs. Just a rumor in a forum tho.

 

[Razor1]

nah doesn't look like that now as they will be released at the same time.