• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Polaris architecture – GCN 4.0

Teh 300 serious I would expect, the Fury parts may be a good hint that the fastest Polaris will at least be close enough and so much cheaper the FuryX becomes pointless
 
Teh 300 serious I would expect, the Fury parts may be a good hint that the fastest Polaris will at least be close enough and so much cheaper the FuryX becomes pointless

This^

The Polaris parts available this year I think will give the GTX 1070 a lot of serious competition.

Going on the specs the 1070 is going to be 3/4 the performance of the 1080. As the 1080 is not that much ahead of the old GM200 cards, this puts the 1070 well within reach of the upcoming Polaris cards.
 
Also you seen this about Pascal's Async Compute: http://wccftech.com/nvidia-gtx-1080-async-compute-detailed/

Turns out it doesn't have hardware support still, and its Ashes performance was just down to brute force Tflops. So Vega looks scarier vs big Pascal now we know that.

And just in general Polaris/Vega vs Pascal in DX12 games going forward is going to be tough on Nvidia.

Tom Peterson seems to have said otherwise in the pcper interview unless i have misunderstood something
 
Tom Peterson seems to have said otherwise in the pcper interview unless i have misunderstood something

Well unless Nvidia come out with "it's turned off in the drivers still" then the results from Ashes are conclusive.

If it's working correctly, and they gain no performance from async, then Pascal can't have hardware support.
 
Well unless Nvidia come out with "it's turned off in the drivers still" then the results from Ashes are conclusive.

If it's working correctly, and they gain no performance from async, then Pascal can't have hardware support.

I very much doubt the developers have had time to enable async compute for Pascal given you can't even buy Pascal yet. Its a;most certainly running the same code path as the 980ti so is meaningless.
 
I very much doubt the developers have had time to enable async compute for Pascal given you can't even buy Pascal yet. Its a;most certainly running the same code path as the 980ti so is meaningless.

If that turns out to be the case, fair enough.

But as it stands right now, there's no quantitative evidence Pascal has hardware async compute, only Nvidia's word. And they said ambiguous things about Maxwell too.
 
Last edited:
If that turns out to be the case, that's fair enough.

But as it stands right now, there's no quantitative evidence Pascal has hardware async compute, only Nvidia's word. And they said ambiguous things about Maxwell too.

When you say things like "hardware async compute" it is clear you don't really understand the complexity of what this actually means. It is complex, and AMD's marketing is doing a very good job of telling people that their way of supporting DX12 multi-engines is the only way, which is only adding to the confusion.
 
Nvidia know what they are doing and if Pascal had hardware support i think they would be shouting about it as loud as they could. They have said there support is improved which sounds like they improved the software side. AMD shout as loud as they can about it and if Nvidia could do it better you can bet they would be screaming about it instead of it's improved. AMD support it fully and i still think Nvidia don't.
 
Last edited:
Nvidia know what they are doing and if Pascal had hardware support i think they would be shouting about it as loud as they could. They have said there support is improved which sounds like they improved the software side. AMD shout as loud as they can about it and if Nvidia could do it better you can bet they would be screaming about it instead of it's improved. AMD support it fully and i still think Nvidia don't.

There are writeups that explain it - and AoTS is now showing an uptick going from DX11 to 12 for the 1080 at some resolutions when running with async on vs off
 
AMD have a really good chance to regain some much needed market share with Polaris

Nvidia seem to have scored an own goal with the pricing of the 1080

Yea this launch might not be Nvidia's finest but there brand power is at a level that they can afford a mistake or 2. If AMD play there cards right they can come in and take advantage.
 
There are writeups that explain it - and AoTS is now showing an uptick going from DX11 to 12 for the 1080 at some resolutions when running with async on vs off

It certainly looks better but the 1080 is better full stop. Lets see AMD's next gen in comparison. Roy said if dx12 and vr are important people should be excited (was something along these lines) This suggest's to me they are not to confident about dx11 but in dx12 they are. I am done with hype trains though so it could be all bs.
 
Last edited:
I very much doubt the developers have had time to enable async compute for Pascal given you can't even buy Pascal yet. Its a;most certainly running the same code path as the 980ti so is meaningless.

No those fantastic driver engineers at Nvidia have only had the last 6 months to do it on Maxwell 28nm/Maxwell 16nm. They havent done it. They have never come out and said we have hardware asynchronus support. Read the PC per stuff and watch the interview. They have improved the pre-emptive stuff but thats it. Pascal has NO Async Hardware Support regardless of what you are saying. Pre-emption and brute force is the way they are handling it and when they have it in Volta they will sure as hell tell us.
:)
 
When you say things like "hardware async compute" it is clear you don't really understand the complexity of what this actually means. It is complex, and AMD's marketing is doing a very good job of telling people that their way of supporting DX12 multi-engines is the only way, which is only adding to the confusion.

Fair enough. I'll rephrase it to; Cannot gain meaningful performance from performing asynchronous compute tasks, merely can accept/interpret such tasks correctly in their GPU pipeline.
 
Last edited:
Back
Top Bottom