Arguable. You can't compare the gain between AMD and Nvidia with something like Async. It's designed to boost efficiency when the architecture is being inefficient. If AMD sees a bigger boost, you can just as easily attribute that to being gaps in their scheduling as you can to them having a superior implementation. As a basic example, if a Fury X is running a particular benchmark at 80% of it's total output, it can schedule in 20% worth of compute in and become extremely efficient. Let's say you see a 20% increase in performance (which is good obviously). A GTX 1080 might be running that same benchmark at 90% of it's total output, only allowing room for an additional 10%. So when you swap between Async and not, the Fury X gains 20, while the 1080 only gets 10. But that's because the 1080 was effectively performing more efficiently before. Again this is super simplified, but with the nature of how Async works, saying one is better than the other -- or that Nvidia would even benefit from having ACEs like GCN does, isn't really an accurate. Also there were hardware changes to Pascal's scheduler to make these gains happen. The scheduling for the most part is still done in software, but there is a hardware element to it.
well... the link compares performance in dx12 with and without async... and with async amd have gains about 10% and nvidia 2% in 2k resolution thats reduced performance difference about fury x and gtx 1080... without async the difference are about 20% more with the 1080... and using async reduced to 12%... it´s so clear
Won't change the point of Denial lol. Async benefits cards with architecture designed to received multiple queues of tasks. While Nvidia GPU indeed will perform better when you send them one big queue with everything. Seriously both are arguable, and multiple queue of tasks require more work from the developement team, to fully utilize AMD card that otherwise underperform. The more i think of it, the more i'm 100% with Nvidia on that one: Developement should be able to be lazy and send your card whatever, lol.
It's really not that clear though. The point is, without Async the Fury X should technically be faster then it is. Async doesn't magically make a card go faster, it fills losses and boosts efficiency of what's already under the hood. Cards have a maximum theoretical performance. Nvidia's cards are closer to their theoretical maximum then AMD cards are. So when you turn Async on, AMD cards gain more benefit, because they weren't performing as optimally as they should be in the first place. Perfect example of this is a 980Ti vs a Fury X. Ti stock is around 6tflops, Fury X is around 8.5. The Fury X should smash the Ti in literally every game. It doesn't because it's not performing at 8.5, it's performing closer to 6. When you turn Async on, it's shooting up - reducing that gap, probably to around 8tflops. That's when the Fury X pulls ahead. When you clock a 980Ti at 1500mhz and also shoot the Ti's tflops to 8, they perform identically, even with Async completely disabled for the Ti. And this isn't to say that AMD's architecture is bad, it's not. It just going to benefit a lot more from Async then Nvidia's ever would. Nvidia could literally clone AMD's Async hardware to it's own and it still wouldn't match AMD's performance gain with Async.
Just saw it! €789 at online retailers. What a fricking rip-off! And they doing it so openly. Just sad.
give u a like. AMD is always look better in paper but not real life. async compute is just a short drama will end soon.
And the whole joke is just getting funnier and funnier. I'm just referring to prices Matt. Nothing to do with ya.
You realise US price doesnt include VAT? Europe has like 20% VAT. 699 x 1.20 = 840$. So it would equal to 840US which is not too far from your price isnt ?
US doesnt have VAT. Many places like newegg/amazon etc dont charge tax so we would pay 599/699 total depending on edition. Europeans are screwed when it comes to pricing.
Sweden's prices have been confirmed for a while now at a whopping 7699SEK (823€ / 923$) regardless of manufacturer. Suck on that :banana:
i understand your point of view... but i need disagree because look strange that amd not using all if in dx 11 vs dx 12 amd beat the same nvidia look in same test the furyx in dx11 makes 46fps in dx11... in dx12 without async does 52fps a gain about 15% and with async goes for 57fps more gains about 10% now when we looks nvidia gtx 1080 makes 62fps in dx11... in dx12 goes 63fps and with async usage 65fps... is a great difference... now if amd not using all potential why outperforms nvidia gains when using dx12 even without async? to me is more believable nvidia not using your potential and not amd
That's totally backwards. The card with the most improvement is clearly the one that had the most space for improvement. AotS in particular is likely to have had more work put in towards tuning for GCN, and considering how drawcall heavy a game it is, there's considerable CPU overhead with AMD's DX11 driver (exacerbated by GCN's inefficient command processor, that's been redesigned for GCN 3 (polaris and vega)). So AMD gets a bigger improvement under DX12, all that says is that DX11 was running worse. If NV gains nothing going to DX12 all that says is that the DX12 path is making just as efficient use of the hw as dx11 path was. Moreover, if you look at AotS and compare cards flop for flop Maxwell matches Fiji, despite not having asynchronous compute to help saturate the shader array. So I would even go as far as saying that as far as AotS is concerned, Async is allowing GCN to catch up with Maxwell's shader utilization.
It seems like there is a bigger gap in performance from the 1080 to the 1070 than it is from the 980 to the 970. On paper at least
