Tuesday, August 18th 2015
AMD GPUs Show Strong DirectX 12 Performance on "Ashes of the Singularity"
Stardock's "Ashes of the Singularity" may not be particularly pathbreaking as an RTS, in the Starcraft era, but has the distinction of being the first game to the market with a DirectX 12 renderer, in addition to its default DirectX 11 one. This gave gamers the first peak at API to API comparisons, to test the tall bare-metal optimizations of DirectX 12, and as it turns out, AMD GPUs do seem to benefit big.
In a GeForce GTX 980 vs. Radeon R9 390X comparison by PC Perspective, the game seems to perform rather poorly on its default DirectX 11 renderer for the R9 390X, which when switched to DirectX 12, not only takes a big leap (in excess of 30%) in frame-rates, but also outperforms the GTX 980. A skeptical way of looking at these results would be that the R9 390X isn't optimized for the D3D 11 renderer to begin with, and merely returns to its expected performance vs. the GTX 980, with the D3D 12 renderer.
Comparing the two GPUs on CPU-intensive resolutions (900p and 1080p), across various CPUs (including the i7-5960X, i7-6700K, i3-4330 dual-core, FX-8350, FX-6300, reveals that the R9 390X has a narrow performance drop with fewer CPU cores, and has slight performance gains with increasing number of cores. Find the full insightful review in the source link below.
Source:
PC Perspective
In a GeForce GTX 980 vs. Radeon R9 390X comparison by PC Perspective, the game seems to perform rather poorly on its default DirectX 11 renderer for the R9 390X, which when switched to DirectX 12, not only takes a big leap (in excess of 30%) in frame-rates, but also outperforms the GTX 980. A skeptical way of looking at these results would be that the R9 390X isn't optimized for the D3D 11 renderer to begin with, and merely returns to its expected performance vs. the GTX 980, with the D3D 12 renderer.
118 Comments on AMD GPUs Show Strong DirectX 12 Performance on "Ashes of the Singularity"
AMD's professional board figures aren't a straightforward extrapolation of market share as the consumer market largely is. Nvidia boards sell at vastly greater ASPs. To use the "MAC" (Mac Pro) example you cited:
The top Mac Pro comes standard with Dual FirePro D500's and can be upgraded to Dual FirePro D700's for $600.
The FirePro D500 is a custom part analogous to the HD 7870XT ( so sits between the Pitcairn-based FirePro W7000 and Tahiti PRO- based W8000) The W7000 is priced at around $600 each, the W8000 at ~ $1000 each
The FirePro D700 is a FirePro W9000 (also in a custom form factor as per the other FirePro D boards). The W9000 retails for around $3000 each......and Apple offers an upgrade to two of them - from boards priced at less than $1K, for $600
Given that Apple has to factor in their own profit and amortization from warranty replacements, how much do you think AMD's unit price contract is for these custom SKUs?
AMD basically purchased market share (and the cachet of marketing by being allied to the Apple Mac Pro). If AMD rely upon this business model, they build market share all the way to the poor house.
AMD wins market share. "Let me explain to you why there is nothing to see here, or even worst, why it is bad for AMD".
With NVIDIA's Gameworks, CDPR lost control over their Witcher 3 PC source code.
[INDENT]Our dear friends over at PCGamesHardware had an interesting interview with CD Projekt RED’s lead engine developer, Balázs Török, in which Balázs claimed that AMD users will be able to enjoy the new fur and hair GPU-accelerated effects, provided Nvidia lets them to.
When asked about the new hair and fur effects, Balázs said that – at this moment – these aforementioned effects run on AMD Radeon GPUs.
“At the moment it works, but if it can run at the end, the decision of Nvidia. What matters is the direction in which they continue to develop the technology and whether and what barriers they are installed. But I think it will also run on Radeon graphics cards.“[/INDENT]
So we have 3 GT 730s.
One is 96 Fermi cores, 128bit DDR3. <<<This one is not even DX12 yet.
One is 384 Kepler cores, 64bit GDDR5. <<< This one is the good one
And the last one is 384 Kepler cores, 64bit DDR3. <<< This one you throw it out the windows. 12.8GB/sec? Even minesweeper will have performance problems.
As for DX12, I don't think we can judge everything by one game. It's too early to say much about DX12 other than it potentially can offer some significant improvements. AMD's drivers are known to have more overhead than nVidia's and DX12 might make that less of a problem than it is now.
Honestly, I don't think anyone should get angry or lose sleep over this.
All I did was to point out the pricing of AMD parts used in the "MAC".
If you want to have a good cry about it, be my guest - but between whines, maybe you can explain how AMD's FirePro market share is growing (in a fashion) - largely, as you've already said, because of Apple, yet AMD still bleeds red ink.Nice AMD supplied PPS. Why not use the latest figures that show that AMD's pro graphics have slipped back to 20%?So, a large chunk of AMD's pro graphics market is predicated upon a single customer getting boards at knock down pricing - around 1/10th of retail MSRP. As I said, AMD (or anyone for that matter) can grow market share if they offer deals like that - and let's face it, AMD have been infire sale mode for FirePro for some time. I'm also pretty sure if AMD offered Radeons at $5 each, they'd quickly gain a massive market share gain - but it doesn't mean **** all if it isn't sustainable. Not every entity can look forward to an EU bailout.
wccftech.com/amd-radeon-r9-390-390x-not-rebadges-power-optimization/Really how many ppl care about those cards being rebrands or not? No one cares if r5 240 or what ever is rebrand or 230. They are low end cards with very low power draw. No one that cares about performance buys them.
First DX12 gaming benchmark shows R9 290X going toe-to-toe with a GTX 980 Ti.
This gets more hilarious by the day, I'm literally getting upset AMD make no money!
1080p
290X
48
390X
53
980
50
980 Ti
50
1080p Heavy
290X
40
390X
46
980
44
980 Ti
43
Point is, don't start that hype train rolling just yet. Might turn in to a train wreck of disappointment like it has many times before. I guess though if people haven't learned by now they never will.
From www.techpowerup.com/180675/amd-tressfx-technology-detailed.html
"Technically, TressFX is a toolset co-developed by AMD and Crystal Dynamics, which taps into DirectCompute"
Should least be happy that TressFX uses a standard which everyone including NVIDIA was crying for.
I'm aware of the technical aspect. The problem was excessive tessellation that doesn't substantially improve the graphics appearance. The workaround was to re-use the same AMD driver side tessellation override feature for Crysis 2 NVIDIA patched.
The difference with Ashes of Singularity vs Witcher 3 is the source code availability for AMD(Red Team), NVIDIA (Green Team) and Intel (Blue Team). All interested IHVs can contribute to the same source code without being blocked an exclusivity contract.
Witcher 3 XBO/PS4 builds uses TressFX instead of NVIDIA's Hairworks.
Unreleased Witcher 3 PC build has TressFX enabled, but blocked NVIDIA exclusivity contract. Witcher 3 PC build with TressFX would have benefited lesser NVIDIA GPU cards.
PS; I own MSI 980 Ti OC with my MSI 290X OC.The pattern is similar to 3DMarks API overhead test results.It could also indicate AMD's DX11 driver is sub-par relative it's TFLOPS potential.
please don't feed into the hypetrain
en.wikipedia.org/wiki/Graphics_Core_Next
GCN 1.0 is 11_1:
Oland
Cape Verde
Pitcairn
Tahiti
GCN 1.1 is 12.0:
Bonaire
Hawaii
Temash
Kabini
Liverpool
Durango
Kaveri
Godavari
Mullins
Beema
GCN 1.2 is 12.0:
Tonga (Volcanic Islands family)
Fiji (Pirate Islands family)
Carrizo
Only NVIDIA's GM2xx chips are 12.1 compliant (which the GTX 980 Ti is). Even Skylake's GPU is 12.0.
So if the game supports feature level 12.1 and it is using it on GTX 980 but using 12.0 on 290X, it's not an apples to apples comparison. We'd have to know that both cards are running feature level 12.0.
AMD GCN cards can handle more resources as well so its not always going to be an apples to apples comparison.
Feature level: Maxwell 12_1, GCN 12_0.
CR (Conservative Rasterization) feature
Readcommunity.amd.com/message/1308478#1308478
Question:
I need for my application that every drawing produce at least one pixel output (even if this is an empty triangle = 3 identical points). NVidia have an extension (GL_NV_conservative_raster) to enable such a mode (on Maxwell+ cards). Is there a similar extension on AMD cards
Answer (from AMD):
Some of our hardware can support functionality similar to that in the NVIDIA extension you mention, but we are currently not shipping an extension of our own. We will likely hold off until we can come to a consensus with other hardware vendors on a common extension before exposing the feature, but it will come in time.
For ROV feature
AMD already supports Intel's "GL_INTEL_Fragmented_shader_ordering" in OpenGL.
From twitter.com/g_truc/status/581224843556843521
It seems that shader invocation ordering is proportionally a lot more expensive on GM204 than S.I. or HSW.
The change from DX11 to DX12 should be pretty swift for developers because it is nothing monumental. I believe DX12 automatically falls back to 11 and 10 depending on hardware capabilities.
AMD Mantle and PS4's lower level APIs has set the ground work for DX12.
Better examples of low-level API's would have been Glide and Metal, but most people block those out of their memory because it was a frustrating time in the PC world.
Well known 3D engines already has DX12 version e.g. Epic's Unreal Engine 4.9, Crytek's CryEngine, Unity , Square Enix's Luminous Engine and 'etc'.