TR Forums

Performance numbers produced by AMD internal testing show the performance benefits of Vulkan versus OpenGL implementation:

Up to 27% faster performance at 1920×1080 using Radeon Software 16.7.1 and DOOM Vulkan on Radeon™ RX480 than with Radeon Software 16.7.1 and DOOM OpenGL.1
Up to 23% faster performance at 2560×1440 using Radeon Software 16.7.1 and DOOM Vulkan on Radeon™ RX480 than with Radeon Software 16.7.1 and DOOM OpenGL.2

Disclaimer:
1. Testing conducted by AMD Performance Labs as of July 6th, 2016 on the AMD Radeon™ RX 480, on a test system comprising Intel i7 5960X CPU, 16GB DDR4-2666 Mhz system memory, Radeon Software Crimson Edition driver 16.7.1 and Windows 10 x64 using the game DOOM on the ultra preset. PC manufacturers may vary configurations, yielding different results. At 1920×1080, Radeon Software Crimson Edition 16.7.1 running DOOM OpenGL scored 106.40 and Radeon Software Crimson Edition 16.7.1 running DOOM Vulkan scored 135.65 on AMD Radeon™ RX 480, which is 27.5% faster performance. Performance may vary based on use of latest drivers. Test results are not average and may vary. RS-43

2. Testing conducted by AMD Performance Labs as of July 6th, 2016 on the AMD Radeon™ RX 480, on a test system comprising Intel i7 5960X CPU, 16GB DDR4-2666 Mhz system memory, Radeon Software Crimson Edition driver 16.7.1 and Windows 10 x64 using the game DOOM on the ultra preset. PC manufacturers may vary configurations, yielding different results. At 2560×1440, Radeon Software Crimson Edition 16.7.1 running DOOM OpenGL scored 68.51 and Radeon Software Crimson Edition 16.7.1 running DOOM Vulkan scored 84.34 on AMD Radeon™ RX 480, which is 23.1% faster performance. Performance may vary based on use of latest drivers. Test results are not average and may vary. RS-44

Source for the above: http://radeon.com/doom-vulkan/

Video from WCCFtechTv: Radeon RX 480 DOOM Vulkan vs OpenGL

Reproduced, RX 480 with Vulkan is close enough to GTX 1070 and GTX 980 Ti that it's laughably underpriced.

Topinio wrote:

Reproduced, RX 480 with Vulkan is close enough to GTX 1070 and GTX 980 Ti that it's laughably underpriced.

...if the only game you ever play is Doom.

I've noticed some gains with my lowly 660Ti using Vulkan API in Doom, but nothing earth-shattering. It appears to be benefiting weaker GPUs and CPUs at the cost of increasing system memory consumption. Doom eats-up an additional 1GiB of system memory but still remains under 5GiB of system memory consumption.

Topinio wrote:

Reproduced, RX 480 with Vulkan is close enough to GTX 1070 and GTX 980 Ti that it's laughably underpriced.

NVIDIA GPUs don't yet have Async Compute support which, according to Bethesda, will be available "soon".

Airmantharp wrote:

...if the only game you ever play is Doom.

Yeah, Doom thread and the results on that page are interesting in this context.

End User wrote:

NVIDIA GPUs don't yet have Async Compute support which, according to Bethesda, will be available "soon".

Sure, NVIDIA is worse than AMD here and any AMD card comes out well, e.g. the R9 390 being ~30% better than the GTX 970.

NVIDIA has hardware and driver issues with this game, and async compute, that AMD does not have; pretty sure I could predict how slated AMD would be were the situation reversed.

Last I saw (from the FAQ) was "We are working with NVIDIA to enable asynchronous compute in Vulkan on NVIDIA GPUs. We hope to have an update soon." - has there been an update that actually says the feature will be available soon (versus the update that says there will be an update)?

biffzinker wrote:

I also see it's using OpenGL 4.5 now instead of 4.3 for AMD Radeon cards.

Same thing for my NVIDIA equipped setup.

Topinio wrote:

Sure, NVIDIA is worse than AMD here and any AMD card comes out well, e.g. the R9 390 being ~30% better than the GTX 970.

NVIDIA has hardware and driver issues with this game, and async compute, that AMD does not have

As an NVIDIA owner I'm not complaining. I'm getting a 120-160 FPS @ 2560x1440 on Ultra when Vulkan is enabled. If that is with both "hardware and driver issues" and no async compute then I'm impressed.

End User wrote:

As an NVIDIA owner I'm not complaining. I'm getting a 120-160 FPS @ 2560x1440 on Ultra when Vulkan is enabled. If that is with both "hardware and driver issues" and no async compute then I'm impressed.

Gz (though I should hope so with a 4.7 GHz 4C Ivy and a brand new $700 GPU!)

Fact is, though, that the $450 GTX 1070 is beaten by 26% by a now $400 R9 Fury X in that game @ 2560x1440 on Ultra when Vulkan is enabled, so yes NVIDIA has issues.

Lower-end NVIDIA users are also not getting a Vulkan boost, especially those with 2GB cards.

I presume 970 owners like me are not getting benefits from Vulkan then, since Async compute is a big part of the performance boost and the Async promise on Maxwell is a lie?

Not that I'm complaining, I got 1440p Ultra settings to run at 85fps on my 85Hz monitor most of the time, probably helped by the fact I didn't like the lag caused by TX antialiasing modes so I was just running easy/cheap FXAA.

Chrispy_ wrote:

I presume 970 owners like me are not getting benefits from Vulkan then, since Async compute is a big part of the performance boost and the Async promise on Maxwell is a lie?

Not that I'm complaining, I got 1440p Ultra settings to run at 85fps on my 85Hz monitor most of the time, probably helped by the fact I didn't like the lag caused by TX antialiasing modes so I was just running easy/cheap FXAA.

You'll get a bump if you're using a low-end CPU... I've heard about some oddball Nvidia configs (i7 920 + Titan X, GTX 970 + AMD FX kit) that benefited substantially from Vulkan.

Chrispy_ wrote:

I presume 970 owners like me are not getting benefits from Vulkan then, since Async compute is a big part of the performance boost and the Async promise on Maxwell is a lie?

I guess not ... that CB article had improvements of the below at End User's settings from turning on Vulkan

Card	% gain
R9 Fury X	46
RX 480	40
R9 390	38
GTX 1070	2.7
GTX 970	0.83
GTX 980 Ti	0.47

Topinio wrote:

Chrispy_ wrote:

I presume 970 owners like me are not getting benefits from Vulkan then, since Async compute is a big part of the performance boost and the Async promise on Maxwell is a lie?

I guess not ... that CB article had improvements of the below at End User's settings from turning on Vulkan

Card	% gain
R9 Fury X	46
RX 480	40
R9 390	38
GTX 1070	2.7
GTX 970	0.83
GTX 980 Ti	0.47

I'll be honest, I'm somewhat puzzled by what I have seen on my system. I'm getting a 100% gain in performance when I switch to Vulcan from OpenGL 4.5. Using the same saved game point I see a jump from 80 FPS max to 160 FPS max. I've switched back and forth numerous times with the same results. Strange things are afoot at the Circle K.

Woo, 0.83% gain.

Concupiscence, I don't bother putting my system spec in my sig because there are too many of them and they change constantly (I test stuff as part of my job and there's enough spare kit that I end up taking home the offcast leftovers). Currently running a 4790.

When you see absolutely no improvement other than standard deviation, seems to me the feature is not on. They said Vulkan path gets disabled if you use any AA other than TSSAA. edit:typo

As for AMD side. Very very nice.

With GTX 980 ti, 6 core i7-4930k went from 54%(one thread maxxed) to 80% utilization(pretty even).

Topinio wrote:

End User wrote:

As an NVIDIA owner I'm not complaining. I'm getting a 120-160 FPS @ 2560x1440 on Ultra when Vulkan is enabled. If that is with both "hardware and driver issues" and no async compute then I'm impressed.

Gz (though I should hope so with a 4.7 GHz 4C Ivy and a brand new $700 GPU!)

Fact is, though, that the $450 GTX 1070 is beaten by 26% by a now $400 R9 Fury X in that game @ 2560x1440 on Ultra when Vulkan is enabled, so yes NVIDIA has issues.

Lower-end NVIDIA users are also not getting a Vulkan boost, especially those with 2GB cards.

Oh no, I'm going to cry that I only get 100 FPS + on my 1070 in one game where an AMD card manages to match it. So sad. Guess I should just return it and buy an AMD card and only ever play that one game instead of the other hundreds in my Steam library.

Here's an idea, how about I load up one of several Flight Simulators I own and show you the FPS on my 1070 and you can compare that to any AMD card ever released including the Radeon Pro Duo. Preview: AMD cards SUUUUUUUUUUUUUCK at flight sim.

Vulkan doesn't help my system at all ... cuz my GTX 580 doesn't have it.

This might be the thing that pushes me to finally look at picking up a new GPU in earnest.

techguy wrote:

Oh no, I'm going to cry that I only get 100 FPS + on my 1070 in one game where an AMD card manages to match it. So sad. Guess I should just return it and buy an AMD card and only ever play that one game instead of the other hundreds in my Steam library.

Here's an idea, how about I load up one of several Flight Simulators I own and show you the FPS on my 1070 and you can compare that to any AMD card ever released including the Radeon Pro Duo. Preview: AMD cards SUUUUUUUUUUUUUCK at flight sim.

Do they? That's interesting, have you got any benchmark links from the last few years?

It's great that your GTX 1070 is good, is your view that that means the lack of Vulkan and async compute are non-issues for all people with NVIDIA cards? Those lower/older ones could use some help in situations where your new shiney still rocks, IMO it's an issue that this stuff isn't there (yet?).

Topinio wrote:

techguy wrote:

Oh no, I'm going to cry that I only get 100 FPS + on my 1070 in one game where an AMD card manages to match it. So sad. Guess I should just return it and buy an AMD card and only ever play that one game instead of the other hundreds in my Steam library.

Here's an idea, how about I load up one of several Flight Simulators I own and show you the FPS on my 1070 and you can compare that to any AMD card ever released including the Radeon Pro Duo. Preview: AMD cards SUUUUUUUUUUUUUCK at flight sim.

Do they? That's interesting, have you got any benchmark links from the last few years?

It's great that your GTX 1070 is good, is your view that that means the lack of Vulkan and async compute are non-issues for all people with NVIDIA cards? Those lower/older ones could use some help in situations where your new shiney still rocks, IMO it's an issue that this stuff isn't there (yet?).

Sadly, reviewers stopped using flight simulators for testing because
a) they're niche (usually sales of a particular version of any given flight sim will be in the tens, maybe hundreds of thousands vs. millions for AAA titles)
b) they're often very CPU-bound
I remember back in the day when Tom's Hardware used FSX as a benchmark in many hardware reviews. SimHQ was also a good site for sim related hardware testing a number of years ago, but they seem to be more of a news aggregator and forum host nowadays.

The best I can do for you is point you to the largest flight sim community on the net and their "FSXMark 11" benchmark thread which includes posts from hundreds of community members running a standardized test http://www.avsim.com/topic/329116-fsxmark11/ and the spreadsheet that accumulates the results (updates ceased some time ago, see thread for complete results) https://docs.google.com/spreadsheets/d/ ... edit#gid=0

One of the best examples I can point you to (from the spreadsheet) is line 28 which is the result of forum user "RyanF" and is the highest recorded result for an AMD graphics card user, a 6990 producing 42.9 FPS average. Now, I realize a 6990 is quite old at this point but it's relevant because it's compared to its contemperate cards from the time. Now look at line 26 from user "Superpilotv2" who is achieving 43.6 FPS with
a) a slower CPU (4.6GHz 2500k vs 5.1GHz 2600k
b) a MUCH slower, several generations older Nvidia GPU, a GTX 260

Again, see the thread for more results if you don't like this particular data point. Current hardware is achieving in excess of 60 FPS on the same test, over 70 from the fastest user (a well known forum member by the name of "Westman" who uses phase change cooling to OC his CPUs for this benchmark).

So the problem here is not one of architecture, but of software. AMD has stated they no longer support the world's most popular flight sim, Flight Simulator X. This has been well documented within the flight sim community. Nvidia isn't exactly optimizing for it either, but they at least offer SLI compatibility profiles for FSX and its successor, P3D and continue to ensure that these sims don't see performance degradations whilst maintaning compatibility. If you happen to be a flight simmer with an NV card and experience graphics-related issues and can demonstrate that they are the fault of the driver, if you submit a bug report to NV it will eventually get fixed. The same most certainly cannot be said for AMD users running flight sims.

There's an absolute library full of information on this subject at the various flight sim sites, it's hard to pull it together to give you a concise summary. You would have to do quite a bit of reading if you want to get a greater understanding of the issue and/or verify the veracity of these claims.

End User wrote:

biffzinker wrote:

I also see it's using OpenGL 4.5 now instead of 4.3 for AMD Radeon cards.

Same thing for my NVIDIA equipped setup.

Use to default to OpenGL 4.3 before the Vulkan patch for Radeon equipped system though.
Try this post: viewtopic.php?f=12&t=117859#p1305666

So one game Doom isn't relevant but a niche flight simulator is?

Do you need all that FPS so the propeller blur is just right?

Frugal wrote:

So one game Doom isn't relevant but a niche flight simulator is?

Do you need all that FPS so the propeller blur is just right?

Well, Doom is one game, flight simulation is an entire genre and there are multiple representatives within the genre, all with performance figures that demonstrate AMD's performance is inferior. I haven't even mentioned how poorly Bulldozer family CPUs perform in these sims. Modern OC'd Haswell and Skylake systems can hit average frame rates in the 60s during benchmarking, OC'd FX series chips are in the 20s! For the record, the MS Flight Simulator franchise at one point was the best selling PC "game" franchise in history, with over 20 million copies sold.

This is a rather nuanced discussion, given your flippant remark you don't appear prepared to participate. Flight Sim X is based on Microsoft's "ESP" game engine, as is its successor Prepar3d (P3D). This engine has its roots in a development philosophy which believed the "free lunch" of never-ending CPU performance increases by way of clock speed increases would mask any short-comings in engine design. Around the mid 2000's when it became apparent that this free lunch was in fact over, MS did not adapt and the last MS product to make use of the ESP engine was shipped in the form of FSX. This is still the most popular flight simulator in use today, despite its age. Successors (P3D) and also-rans (X-Plane) are slowly ticking up in marketshare. Given the design philosophy of the time, ESP is largely serial in nature, and not easily parallelized. Thus, its performance is largely dependent upon single-threaded CPU performance. Given the right settings and combination of detailed 3rd party add-ons, an ESP-based sim can bring any computer in existence to its knees. I've seen systems with 5GHz Haswell CPUs and volt-modded 980 Tis massively overclocked running this sim with framerates in the low teens, so when you make ignorant comments such as "do you need all that FPS so the propeller blur is just right" you demonstrate said ignorance.

Flight simulators can be used for entertainment, but most use them for flight instruction and as training aids. For reference, the U.S. Air Force uses an ESP-based sim, Lockheed Martin's P3D, to train their pilots. With that in mind, a fluid frame rate is absolutely crucial during critical phases of flight such as takeoff and landing. These also happen to be the most demanding scenarios for the simulator given the conditions, looking out of the virutal cockpit of a complex aircraft in the vicinity of a highly detailed airport with AI traffic is going to generate more of a workload for the game engine than any other scenario.

So is flight simulation a niche compared to AAA multi-million sellers? Sure. Does that make it less relevant commercially? From one perspective when measuring direct software sales only. However, given how much money has been invested into hardware platforms for this simulation (take a look at some of the $100,000+ home cockpits or multi-million $ setups in use by the USAF or private flight training institutions) the total market for this particular niche is actually much larger. Not to mention the add-on sales. 3rd party developers of high fidelity sceneries charge $20-$100 a piece, and the same goes for individual aircraft. Some people have spent more in flight sim add-ons than most people will spend on software in their entire lives. I know I have thousands invested, and my collection pales in comparison to some.

So a crummy game engine, many fewer users and it needs to be fast because training.

Maybe they should try changing the API or buy a plane instead?

Frugal wrote:

So a crummy game engine, many fewer users and it needs to be fast because training.

Maybe they should try changing the API or buy a plane instead?

Again demonstrating ignorance. Why don't you go look up how much planes cost?

How's that near 1080 level performance 480 working out for you? Still can't get an answer out of you about that one... Gee, I wonder why...

techguy wrote:

Topinio wrote:

techguy wrote:

Oh no, I'm going to cry that I only get 100 FPS + on my 1070 in one game where an AMD card manages to match it. So sad. Guess I should just return it and buy an AMD card and only ever play that one game instead of the other hundreds in my Steam library.

Here's an idea, how about I load up one of several Flight Simulators I own and show you the FPS on my 1070 and you can compare that to any AMD card ever released including the Radeon Pro Duo. Preview: AMD cards SUUUUUUUUUUUUUCK at flight sim.

Do they? That's interesting, have you got any benchmark links from the last few years?

It's great that your GTX 1070 is good, is your view that that means the lack of Vulkan and async compute are non-issues for all people with NVIDIA cards? Those lower/older ones could use some help in situations where your new shiney still rocks, IMO it's an issue that this stuff isn't there (yet?).

Sadly, reviewers stopped using flight simulators for testing because
a) they're niche (usually sales of a particular version of any given flight sim will be in the tens, maybe hundreds of thousands vs. millions for AAA titles)
b) they're often very CPU-bound
I remember back in the day when Tom's Hardware used FSX as a benchmark in many hardware reviews. SimHQ was also a good site for sim related hardware testing a number of years ago, but they seem to be more of a news aggregator and forum host nowadays.

The best I can do for you is point you to the largest flight sim community on the net and their "FSXMark 11" benchmark thread which includes posts from hundreds of community members running a standardized test http://www.avsim.com/topic/329116-fsxmark11/ and the spreadsheet that accumulates the results (updates ceased some time ago, see thread for complete results) https://docs.google.com/spreadsheets/d/ ... edit#gid=0

One of the best examples I can point you to (from the spreadsheet) is line 28 which is the result of forum user "RyanF" and is the highest recorded result for an AMD graphics card user, a 6990 producing 42.9 FPS average. Now, I realize a 6990 is quite old at this point but it's relevant because it's compared to its contemperate cards from the time. Now look at line 26 from user "Superpilotv2" who is achieving 43.6 FPS with
a) a slower CPU (4.6GHz 2500k vs 5.1GHz 2600k
b) a MUCH slower, several generations older Nvidia GPU, a GTX 260

Again, see the thread for more results if you don't like this particular data point. Current hardware is achieving in excess of 60 FPS on the same test, over 70 from the fastest user (a well known forum member by the name of "Westman" who uses phase change cooling to OC his CPUs for this benchmark).

So the problem here is not one of architecture, but of software. AMD has stated they no longer support the world's most popular flight sim, Flight Simulator X. This has been well documented within the flight sim community. Nvidia isn't exactly optimizing for it either, but they at least offer SLI compatibility profiles for FSX and its successor, P3D and continue to ensure that these sims don't see performance degradations whilst maintaning compatibility. If you happen to be a flight simmer with an NV card and experience graphics-related issues and can demonstrate that they are the fault of the driver, if you submit a bug report to NV it will eventually get fixed. The same most certainly cannot be said for AMD users running flight sims.

There's an absolute library full of information on this subject at the various flight sim sites, it's hard to pull it together to give you a concise summary. You would have to do quite a bit of reading if you want to get a greater understanding of the issue and/or verify the veracity of these claims.

The flight simulator genre is pretty much "professional-only". It was killed by FPS shooter craze during late 1990s and rise of gaming consoles that have internet access. The big gaming publishers and developers move onto into more lucrative genres.

Both Nvidia and AMD don't care for this genre as far as optimizations and driver support is concerned. It is very low on their priority list. There's just no economic incentive in it.

Also, according to this http://steamspy.com/app/379720

there's WAY more people running FSX (just FSX, not even talking about P3D or X-Plane) than there are running Doom.

this https://en.wikipedia.org/wiki/List_of_b ... g_PC_games plus this http://steamspy.com/app/314160 equals about twice the user base, and the first number is an extremely conservative estimate as it's based on a quote from 2009.

Krogoth wrote:

The flight simulator genre is pretty much "professional-only". It was killed by FPS shooter craze during late 1990s and rise of gaming consoles that have internet access. The big gaming publishers and developers move onto into more lucrative genres.

Both Nvidia and AMD don't care for this genre as far as optimizations and driver support is concerned. It is very low on their priority list. There's just no economic incentive in it.

I don't think it's accurate to say there is "no economic incentive" for IHVs to optimize their drivers for flight sims, the professional market can be extremely lucrative. See Tesla, Quadro, etc. Very high margins in those products, much higher than Geforce. More important than optimization though is ensuring the sim can still run on today's driver, and Nvidia at least does this. This falls under the same category of support so it's really splitting hairs at this point.

techguy wrote:

Krogoth wrote:

The flight simulator genre is pretty much "professional-only". It was killed by FPS shooter craze during late 1990s and rise of gaming consoles that have internet access. The big gaming publishers and developers move onto into more lucrative genres.

Both Nvidia and AMD don't care for this genre as far as optimizations and driver support is concerned. It is very low on their priority list. There's just no economic incentive in it.

I don't think it's accurate to say there is "no economic incentive" for IHVs to optimize their drivers for flight sims, the professional market can be extremely lucrative. See Tesla, Quadro, etc. Very high margins in those products, much higher than Geforce. More important than optimization though is ensuring the sim can still run on today's driver, and Nvidia at least does this. This falls under the same category of support so it's really splitting hairs at this point.

Flight simulators pale in comparison to other professional-tier applications like number crunching, rendering farms, CAD and imaging. Besides, flight simulators in the professional world are coded on non-x86 hardware and custom graphical stuff. They aren't being done on your average Quadros and FirePro stuff.

techguy wrote:

Also, according to this http://steamspy.com/app/379720

there's WAY more people running FSX (just FSX, not even talking about P3D or X-Plane) than there are running Doom.

this https://en.wikipedia.org/wiki/List_of_b ... g_PC_games plus this http://steamspy.com/app/314160 equals about twice the user base, and the first number is an extremely conservative estimate as it's based on a quote from 2009.

Current FSX community is much smaller then the current Doom 2016 base. It is one of the classic examples of a vocal, but tiny niches. They both pale in comparison to MOBAs like DOTA2 and LOL.

Core i7 4790K overclocked to 4.6 GHz
16 GB of DDR3 overclocked to 2200 MHz
Windows 10 x64 Professional
Samsung 850 EVO 250 GB (Windows 10)
2x 120 GB SSDs in RAID-0 (this is where DOOM Lives)
R9 280/7950 GPU overclocked to 1100 MHz, and GDDR5 overclocked to 1725 MHz (6900 MHz effective) voltage is at 1.231 for VDDC

DOOM at Ultra Presetting with film grain turned off
Screen resolution is 1920x1080 with the refresh rate at 80 Hz

OpenGL 4.5
120 FPS Max
66/72 FPS averaging

Vulkan
165 FPS Max
89/95 FPS averaging