This is probably true.
DX12 due to Microsoft pushing it is much more prevalent than Vulkan, sadly.
Keysplayr
Elite Member
- Jan 16, 2003
- 21,219
- 56
- 91
stormkroe
Golden Member
- May 28, 2011
- 1,550
- 97
- 91
https://www.computerbase.de/2016-07...md-nvidia/#diagramm-doom-mit-vulkan-2560-1440
1070 - 88
RX480 - 79
RX480 is just 10% behind 1070. Where 1060 is going to end up?
I have no idea if somebody imagined RX480 to get that close to 1070 a couple of weeks ago.
When informed people dig down why this is the case there would be no other reason to buy 1060 over RX480 only if that person is a fan.
littleg
Senior member
- Jul 9, 2015
- 355
- 38
- 91
Surely that depends on how widespread Vulkan adoption is. It's far too soon to tell at the moment but as of right now most games are being coded for DX12 (of the advanced APIs).
Vulkan may see good takeup or it may not, i'm not sure that you can base a purchasing decision just on that factor until we have more information.
stormkroe
Golden Member
- May 28, 2011
- 1,550
- 97
- 91
So this is specifically a DOOM observation? Or do you mean in general that you think 480 will be within 10% of a 1070? If so, when?
sze5003
Lifer
- Aug 18, 2012
- 14,320
- 683
- 126
I think it means dx 12 games depending on how they are developed and if they can benefit from vulkan. But of course no one can see the future.
It happened in one game it can happen in many future games.
RX480 has better designed for DX12 and Vulkan.
Just go and read the thread. All the points why it might be are discussed over there.
10% slower is roughly what you'd expect if you just looked at peak theoretical compute for each card:
Interesting to see how Vulkan Doom performance correlates closely to theoretical peak compute:
- 970 -- 3.4 TFlops vs. 60.6 FPS
- 390 -- 5.1 TFlops vs. 77.4 FPS
- 480 -- 5.2 TFlops vs. 79 FPS
- 980ti -- 6.1 TFlops vs. 85 FPS
- 1070 -- 5.8 TFlops vs. 88 FPS
- Fury X -- 8.6 TFlops vs. 110 FPS
https://www.computerbase.de/2016-07...md-nvidia/#diagramm-doom-mit-vulkan-2560-1440
http://www.pcper.com/reviews/Graphi...-RX-480-will-launch-199-more-5-TFLOPS-compute
http://techreport.com/review/28513/amd-radeon-r9-fury-x-graphics-card-reviewed/4
Actually we can, seeing how DX12 & Vulkan were built around GCN, so we are finally getting to the point where the console wins have helped in GCN cards' performance getting better over time. In fact gen1 based 7970 has had significant gains, going forward the Polaris based GPU's & consoles might be a deadly mix for Nvidia.
This is not to say that Nvidia will fall behind, but unless Volta has more GCN (specific) elements it'll age worse than Maxwell. Also the new(er) titles will be near optimized for GCN4 but specific features that boost performance, like async, will still take a bit of time to get fully utilized since the architecture isn't fully known/exploited to its fullest.
sandorski
No Lifer
- Oct 10, 1999
- 70,862
- 6,396
- 126
With Vulcan and especially DX12 here and being used going forward, I suspect that even if the 1060 is faster now it certainly won't be a year from now. It could be 30+ % slower by then. Even in comparison of the 480 4gb v 1060 6gb this is likely to be the case. 1060 3gb is already a bad choice as it's likely some current games will be forced to use lower settings out of the gate.
Fury X has seriously prodigious compute throughput. Matching or exceeding even a 1080. For DX11, it kind of went underutilised, many complained it was bottlenecked, but it's really coming into its own in Vulkan/DX12 games. AMD probably overspecced Polaris 10 for compute intentionally as well. Given a 32 ROP - 256 bit part, P10 could have done just as well with 32 CU in DX11. But those extra 4 CUs will make the difference in beating GP106 going forward, despite the huge clock speed deficit.
The power of 64 Compute Units. Jesus Christ. AMD was serious when they said they designed GCN for TFLOPS/mm. I mean, 980 Ti can be overclocked to reach the Fury X's peak compute, but at the total expense of power efficiency over Fiji.
On paper Fury X is absurd but reality disappoints with performance in actual games. I wonder what is holding it back. Synthetics show Fury X's ROPs on par with Nvidia. It seems performance is a bit better with Vulkan/D3D games but still beneath the specifications. Maybe it is drivers as others say. If that's the case it is sad to see hardware let down by software.
Anyway, I hope the GTX 1060 is cheap. I'd like something small & quiet to replace my 7950. Looking toward 480 at the moment but may as well wait it out.
Anyway, I hope the GTX 1060 is cheap. I'd like something small & quiet to replace my 7950. Looking toward 480 at the moment but may as well wait it out.
That's not true, with Hawaii they had 64 ROPs & Fiji had the same number, now the number of CUs went up from 44 to 64 likewise the TMUs from 176 to 256 that alongside the massive BW from HBM. Some have speculated for a long time that ROPs have held the Furies back making it an unbalanced design, just like the RX 480, so 96 or even 128 ROPs would have taken the Fury X to another level however it's all conjecture now, since we can't have that hypothetical GPU for testing.
I was referring to this graph
From this synthetic benchmark we could conclude: while Fury X has same number of ROPs as Hawaii each is much more efficient. Secondly, total ROP throughput was on par with Nvidia's offerings. More ROPs really couldn't hurt but if it is a bottleneck it is not evident from synthetics.
From this synthetic benchmark we could conclude: while Fury X has same number of ROPs as Hawaii each is much more efficient. Secondly, total ROP throughput was on par with Nvidia's offerings. More ROPs really couldn't hurt but if it is a bottleneck it is not evident from synthetics.
Last edited:
I guess I should've worded that better, anyway whilst it's true that the Fury X (ROP) is more efficient than Hawaii a lot of those gains can be attributed to the architectural changes that GCN3 brought, HBM helped as well.I was referring to this graph
From this synthetic benchmark we could conclude: while Fury X has same number of ROPs as Hawaii each is much more efficient. Secondly, total ROP throughput was on par with Nvidia's offerings. More ROPs really couldn't hurt but if it is a bottleneck it is not evident from synthetics.
Now looking at the graphs, Fury X should've dominated 980Ti, based on a number of synthetic tests, but it didn't in real (DX11) games. IMO this is where the ROP bottleneck comes into play, as I stated & it's all conjecture atm but Fury X did seem unbalanced to me, as compared to Hawaii. So even if the 64ROP were supposedly on par or even better than the 980Ti, the rest of the GPU was held back by the (relatively) small number of ROPs on the Fury X & hence didn't perform as well it it could.
Car analogy ~ it's like expecting a (superior?) V8 with messy valve timings to beat a fine tuned V6 using shear HP alone. I'm not an expert on cars but I'd assume the result would be similar to how the Fury X fared against 980Ti.
Last edited:
It's too bad Fiji only has 4GB, if it had 6 or 8GB, I would have bought it instantly on release. :/
ASUS Geforce GTX 1060 Strix:
http://imgur.com/a/J2U3F
Especially those games clearly sponsored by AMD...
https://twitter.com/idSoftware/status/752609704795398144
...Using special shaders only on their hardware to 'boost' performance:
http://gpuopen.com/gcn-shader-extensions-for-direct3d-and-vulkan
http://imgur.com/a/J2U3F
And as we know, people buy GPUs based on a single game performance.
Especially those games clearly sponsored by AMD...
https://twitter.com/idSoftware/status/752609704795398144
...Using special shaders only on their hardware to 'boost' performance:
http://gpuopen.com/gcn-shader-extensions-for-direct3d-and-vulkan
Last edited:
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter