IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
Cry more. AMD has earned every bit of criticism and scrutiny through their own actions, from people who have been buying their products from the beginning.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
AMD Vega and Zen Gaming System Rocks Doom At 4K Ultra Settings
- Thread starter HardOCP News
- Start date
Cry more. AMD has earned every bit of criticism and scrutiny through their own actions, from people who have been buying their products from the beginning.
Dayaks
[H]F Junkie
- Joined
- Feb 22, 2012
- Messages
- 9,774
It is not just % faster over previous it is also efficiency gain as well. Nvidia with less TFLOPs still had much greater efficiency resulting in better performing cards in general. Vega TFLOPs increase alone without efficiency increase would be a letdown. So a 40% increase in TFLOPS and a 30% increase in efficiency would give you a 1.3 x .4 = 52% performance over Fiji, hype train, hype train, hype train (new song)
The math should be 1.3*1.4 = 1.82 or 82% increase.
JustReason
razor1 is my Lover
- Joined
- Oct 31, 2015
- Messages
- 2,483
You know how much I hate your driver bickering, more so because you leave out any real info as to the real issue. But in any case with CF there was a fix and why a great portion of users DIDNT have an issue.
http://www.radeonpro.info/
The setting DFC was what a great deal of the AMD user base needed and for those of us that used Radeonpro we had it.
Simple truth is that Radeonpro was a FREE program that was a great boon for gamers. It had the dynamic V-sync that many bitched about Nvidia having but AMD not. Yet save one article, most never mentioned Radeonpro in their analysis of the CF stutter issue. And that one article ( I think Toms ) only did after a rash of posters mentioned it.
Seriously: Every poster in any forum that goes through the trouble to pretend to be the bastion of knowledge and yet never mention the totality of facts should be ashamed. We here are OCers and we SHOULD know every resource with which to attain maximum performance. It is obvious some of you do not and therefore should refrain from commenting further.
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
You had to run third-party apps *to hope* that stuttering issues were resolved. Also, you mention V-Sync on, which is a no-go for many games. FreeSync and G-Sync aren't all the rage for nothing.
Your emotional attempt to cover for AMD's deficiencies here is silly. The reality, plain and clear, is that AMD did not do the groundwork to ensure that Crossfire was a seamless technology and delivered on the promise of nearly doubling performance without side effect. Nvidia didn't quite do that either, but they did a far better job, and they did it by planning and developing for that result *years* before AMD was called to the plate over their glaring shortcomings.
JustReason
razor1 is my Lover
- Joined
- Oct 31, 2015
- Messages
- 2,483
At first with your earlier idiotic post, I thought about responding but decided you weren't worth the effort. But seeing you felt the need to prove your namesake here is not ironic... Nothing about my post is emotional, just a statement of all the FACTS, something a great deal of you don't do. My point was on someone acting as if there was no solution at all, in spite of the FACT there was. Albeit not ideal for some it was a solution none-the-less. And to the V-sync no go, those gamers tend to be your highest-framerate-possible group, and at high frame rates >120 the frame stuttering was nearly next to non-existent as far as perceptibility.
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
Just keep rolling with it. For many/most gamers, playing games where performance *mattered*, V-Sync is a no-go, especially at 60Hz, where stuttering is at it's worst.
And that outlines the point: using a third-party utility that may work sometimes for a few people is not a 'solution'. Period. And that's on AMD.
The only thing that needs boost is your reading comprehension. Those #s are for a Fury X.
Attachments
![77r[1].png](https://cdn.hardforum.com/data/attachment-files/2017/01/71432_77r1.png)
The only thing that needs boost is your reading comprehension. Those #s are for a Fury X.
Well going by Doom Vulkan they aren't getting what you are saying they have 40% more tflops and getting 30% more performance....... At least that is what it is looking like, remember Polaris's less performance per transistor than Granada/Hawaii?
Dirt 2? Considering you haven't used crossfire since 2009 (4870/4870X2) your opinion is invalidated.
Bankie
2[H]4U
- Joined
- Jul 27, 2004
- Messages
- 2,469
Honestly I think NVIDIA should get a pass on the time it's taken for them to optimize for Vulkan; IIRC parts of Vulkan are based on Mantle giving AMD a lead in development time AND there is only one game that benefits from it (DOOM) with hardly any new games coming to it in the near future (I did a quick search and didn't see any new titles). With VR being the "cool" thing at the moment it's understandable to me that NV would concentrate on it until Vulkan has gained more traction.
lol, so in essence Vega is going to kick some serious ass! Catch trains on fire
All right then, good catch so let say Vega is only 10% more efficient:
PerfVega = TFLOPSvega/TFLOPSFiji X EFFvega/EFFFiji
PERF = 12.5/8.6 X 1.1/1 = 1.6 or 160% over Fiji - Math does not lie unless you do it wrong.
Clock speed will play the major role but AMD will also increase efficiency as well. Looking real good and thanks for the correction - Vega if it is clocked at 1500mhz+ should be a great performer.
I think the programmers at ID software should get the credit on that one. Basically able to get more of the full potential out of AMD hardware, Nvidia hardware was already pretty much max out, hence the smaller gain.
Doom Vulkan uses AMD intrinsic shaders, which at the time nV didn't expose for their hardware, so that is what we are seeing there.
Given the relatively bad Ogl performance to begin with, its just like Dx11 to 12 but even more pronounced by adding shader intrinsic on top of the driver overhead issues.
Given the relatively bad Ogl performance to begin with, its just like Dx11 to 12 but even more pronounced by adding shader intrinsic on top of the driver overhead issues.
Stryke1983
Limp Gawd
- Joined
- Dec 1, 2011
- Messages
- 271
Did I seriously just read posts by someone trying to claim that any stuttering with Crossfire (not that SLI didn't have issues as well) in the past was effectively due to user error? Just off the top of my head I know several major tech websites, including the very one we are posting on, that did articles about the problems with multi GPU systems and specifically the stuttering that hit Crossfire especially hard a while back.
What's next? Am I going to have someone telling me that the Bulldozer chips were the best chips for gaming? The Titan GPU line is great value for money? The 290X was famous for it's low noise and heat?
What's next? Am I going to have someone telling me that the Bulldozer chips were the best chips for gaming? The Titan GPU line is great value for money? The 290X was famous for it's low noise and heat?
What's going to kill Vega is random over-speculation of what its capabilities are (or aren't). All we can assume is it's going to perform at what was displayed in two games. Everything else (better drivers = better FPS; under clocked because it was thermally throttled, etc.) are all speculations that may or may not turn out to be true. Reliance on these speculative items to boost (no pun intended) capabilities of the final product may end up killing it when it finally releases. We have approximately six months before it comes out and, within that six months on par with the 1080 is going to turn into 'it's a Titan-killer'; 'it's going to compete with Volta, all iterations'; 'nVidia is going to be crushed by the Vega star'; etc.
We need a healthy AMD. I bought a 1070 because AMD has nothing new in that region. I'd like to see AMD have a performance card I can purchase again. Over-hyping doesn't do AMD any favors.
We need a healthy AMD. I bought a 1070 because AMD has nothing new in that region. I'd like to see AMD have a performance card I can purchase again. Over-hyping doesn't do AMD any favors.
Just to add,
worth noting the ROPs situation has improved for AMD due to the coherent link with the new L2 cacher structure.
Appreciate you probably know Razor but just adding it to the conversation as it is something that has been missed by quite a few sites.
Cheers
Yeah and also they do not have the same level of Vulkan extensions (true low level functions) that AMD does, and this is where the real benefit comes from with Doom.
Cheers
those theoratical tflops you are talking about applies to both fury and vega. So you are back to square one when it comes to effective performance increase over fury. Your argument will be valid if he was comparing it to nvidia. Cheers.
I understand that driver updates have played a major role in performance of XFire setups. I just refuse to believe that out of a couple of games Dirt 2 would be the only one stutter free back then.
Pure luck was not the determining factor in my experience. The logical conclusion in this case is user error.
Well so you are saying OpenGL drivers do not extract the full potential from Nvidia hardware? Meaning Nvidia drivers have been over rated through out the years? Even Vulkan does better in Doom for Nvidia (thanks to Developer and AMD in making Mantle with other developers). Or is it that Nvidia and their OpenGL/Vulkan drivers do indeed extract most of the performance out of Nvidia hardware and there is really not much more you can gain from it? -> That is most likely the real answer. AMD hardware just has more potential then being used previously with OpenGL.
Also looks like some here think that AMD will extract less performance then before from the theoretical max TFLOPS rating with all the hardware upgrades (man AMD must be stupid or something). If the card is a 12.5 TFLOP card and has 4096 shaders - it will need to run at around 1500mhz. At 1500mhz it should perform roughly 50% better over Fiji and with better efficiency it should do better. Will the software/drivers extract that from the hardware is probably only thing that would really prevent a rather stellar performing card but even that should be short live in my opinion. Will memory bandwidth become an issue since it should be the same as Fiji other then latency will be less due to the faster running HBM speed? With better cache setup especially for rasterization that should not be the case. We will see how all of this pans out over time.
Last edited:
No that is not what he is saying
Shader intrinsics are lower level then the LLAPI's, the code unlike GLSL or HLSL, aren't complied by the graphics driver, the code is run straight on the GPU and each IHV have different shader interesics based on their architecture. This is why OGl doesn't have shader intrinsics, the OGL API doesn't allow for this, Vulkan does. So pretty much what you have are the OPTIMAL performance for different shaders on a per IHV level. Since nV didn't expose their shader intrensics at the point where Doom was being developed, ID never did a seperate Vulkan path on nV's cards now not sure if they were added afterwards, but I doubt they did, because later interviews after launch, ID did mention they were still working with nV on this, but nothing was stated after that.
Vulcan is an open source cross platform low level API. There is nothing more to AMD than there is to Nvidia when it comes to working with it.
HLSL for DirectX 11 and DirectX 12
NVIDIA provides a mechanism for using the intrinsics from HLSL in DirectX 11 and DirectX 12. Make sure to check out Alexey’s article on how to access the intrinsics using our NVAPI library.
GLSL for OpenGL and Vulkan
Our drivers expose the warp shuffle and warp vote intrinsics as a series of OpenGL GLSL extensions (GL_NV_gpu_shader5, GL_NV_shader_thread_group, GL_NV_shader_thread_shuffle) in addition to the cross-vendor Khronos OpenGL ARB extensions (GL_ARB_shader_group_vote, GL_ARB_shader_ballot)
Most differences between those are minor (“SubGroup” vs “Warp”, “Invocation” vs “Thread”); notable however is that the ARB extensions support implementations with maximum warp widths of 64, whereas the NVIDIA extensions assume a maximum warp width of 32 threads. This is mostly transparent to a shader developer, except that ballotARB returns the bit mask as a 64-bit integer, unlike ballotThreadNV, which returns the bitmask as a 32-bit integer.
For reference, here is how we roughly implement the cross-vendor intrinsics in terms of our native hardware functionality:
uint64_t ballotARB(bool value)
{
return uint64_t(ballotThreadNV(value));
}
float readInvocationARB(float data, uint index)
{
return shuffleNV(data, index, 32);
}
float readFirstInvocationARB(data)
{
// Do ballotThreadNV(true) to get bitmask for active threads
// Then find lowest set bit indicating first active thread
// Use then as index for shuffleNV
return shuffleNV(data, findLSB(ballotThreadNV(true)), 32);
}
The GLSL extensions discussed above are already supported by our Vulkan drivers via VK_NV_glsl_shader, which makes them available for experimentation today!
In parallel, we are also working with the respective Khronos working groups in order to find the best way to bring cross-vendor shader intrinsics already standardized in OpenGL over to Vulkan and SPIR-V. We are furthermore working on Vulkan and SPIR-V extensions to expose our native intrinsics, but prioritized the cross-vendor functionality higher, especially since there is notable overlap in functionality.
Last edited:
Open source Closed source doesn't matter in this case
And that is a copy and paste from nV's blog or dev site, saw that a while back, but having OGL in the middle muddle things up, Ogl will not allow the developer to set their own critical paths of when what gets done at what time. The API just doesn't have that type of control, intrinsics are available with DX11 and 12 too, but both of those particularly DX11 also has the same or similar problems for OGL.
I'm talking about the shader compiler and how it optimizes vs using code that is closer to metal, and how hand optimization works without a compiler in the middle by the way of an API. The moment you have an abstraction layer vs coding directly to metal, you already have perfomrance loss due to latency. Granted the latency isn't much but when you many shaders the performance loss is magnified.
Having critical functions done at the right time is important to keeping performance up, and with the compiler and an API in the middle, you will not get this all the time, hence where intrinsic shaders come into play. And as already know different architectures need shaders to be structured differently to get the most of any specific architecture, this is the same with shader intrinsics too.
PS the page you linked too was 2 months after Doom 2016 was released
so... . Yeah I don't think Doom is using nV intrinsic at all at this point.
Last edited:
Intrinsic functions gives direct access to the hardware - Nvidia has released that ability in OpenGL via extension and Vulkan. Or are folks saying here AMD drivers are superior to Nvidia's . Now ID software via interview, at least how I remembered it said they were working with Nvidia (right after Vulkan launch) to optimize Nvidia hardware like AMD hardware. That looks like it happen with about a 5-10% increase in performance in Doom later on. In other words Nvidia hardware is maxed out in Doom, AMD? Probably so as well.
. Now ID software via interview, at least how I remembered it said they were working with Nvidia (right after Vulkan launch) to optimize Nvidia hardware like AMD hardware. That looks like it happen with about a 5-10% increase in performance in Doom later on. In other words Nvidia hardware is maxed out in Doom, AMD? Probably so as well.Intrinsic functions gives direct access to the hardware - Nvidia has released that ability in OpenGL via extension and Vulkan. Or are folks saying here AMD drivers are superior to Nvidia's
In Vulkan the driver is completely bypassed*compiler, in OGL and DX11 that still isn't the case, the functions will work in OGL and DX11 but but the driver still sets everything up.
The compiler just doesn't translate the code to machine language, it also predicts what instructions should be sent to the GPU at what time.
So lets say a game was made on Xbox one or PS4 with intristics (AMD intristics) the developer now doesn't have to worry about it running poorly on GCN on the pc side of things, because the API's in Xbone and PS4 are already LLAPI's.
Ok, so that is reason why Vulkan performs up to 10% better now with Nvidia hardware. Gotcha. Thanks
Still performs the same as OGL, which it should perform better than OGL if using intristics.
http://wccftech.com/evga-gtx-1050-ssc-sapphire-nitro-oc-rx-460-triple-gaming-showdown/
Now the mins do go up for nV hardware so that shows that the LLAPI does have some benefits but average frame rates we don't see the increase which is where intrinsics should show a benefit.
The above graphs are rather meaningless due to the setting that is used. You will need the higher settings in Doom. Anyways my 1070 does much better in Vulkan ~10% over OpenGL. Is that because of CPU also? Could be. Now AMD did not expose the hardware in OpenGL thus the rather big improvement that is seen. Bottom line is AMD hardware is very capable when used, looks to be much harder to extract then Nvidia hardware.
These are low end cards tested lol, don't think they can do much more with the settings to hit 60fps
And this is the latest Doom patch and drivers the review was done just two days ago
TheBlueChanell
Supreme [H]ardness
- Joined
- Apr 15, 2005
- Messages
- 4,658
Just putting in my 2c. Long time user of both camps, right now I prefer AMD's drivers as well and their team has been on fire as far as releases / optimizations go the last ~6 months or so. I'm hoping the QT4 interface makes its way to the Linux side of things very soon.
Vega is the first hardware the Linux driver team has had unbridled access to pre-release and am really looking for to what kind of performance we get with AMDGPU/AMDGPU-Pro at launch.
I think people should really check what the complete functions both Nvidia and AMD offer in Vulkan; specifically the 5 AMD support compared to the 2 Nvidia has with regards Vulkan and how close to the metal they are.
Also there is this article that is highly relevant from Nvidia released ironically around the same time as the one linked earlier: https://developer.nvidia.com/reading-between-threads-shader-intrinsics
Of interest is in the article I linked from Nvidia they state:
Ironically they targeted the cross-vendor aspect 1st rather than their own native low level intrinsics.
Just to add, the article linked by imhotep directly mentions the Nvidia expert who did the article I linked as a source of more-further information.
Anyway we are digressing away from the point that AMD gets their results (irrespective of Nvidia) from the extensive shader extension they have in Vulkan for Doom, and this provides a lot more benefits than Async Compute.
Cheers
Last edited:
OGL and DirectX have a way of being supportied by the way of CUDA. "None of the intrinsics are possible in standard DirectX or OpenGL. But they have been supported and well-documented in CUDA for years. A mechanism to support them in DirectX has been available for a while but not widely documented."
Top of the page in this link https://developer.nvidia.com/unlocking-gpu-intrinsics-hlsl
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
If people saying Vega is a new uarch and not GCN. How will the GCN Shader Extensions then work in Vulkan? The answer is they wont.
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
This is AMD's lot in life; it's rare for them to ship bad graphics hardware (ahem, HD2900/HD3800-series), but man o man is the software rarely on point. It is nice to see their recent improvements, though!
KazeoHin
[H]F Junkie
- Joined
- Sep 7, 2011
- Messages
- 9,004
It seems like Vega is a 'true' GCN 2. not a 'GCN 2.0'. It uses the same instructions and such, but AMD seems to imply that it is a much better iteration of GCN.
Last edited:
As they did with Polaris, and we see how that turned out.
Look at this point there is very little AMD has stated outside of primitive shaders I don't see many changes for gaming. And if its the case, which it seems to be for the major features for Vega to use primitive shaders. Its going to be one cold year for Vega after launch if its full performance isn't attainable without primitive shaders.