It's not just AMD+Mantle vs. Nvidia. This situation is quite GPU limited and the tested graphics cards doesn't offer the same performance. Reference GTX 780 Ti is about 10-15 % faster than R9 290X (at default modes). PCGH tested 13% OCed GTX 780 Ti vs. 4% OCed R9 290X. It means, that this GTX would be ~20-25% faster than the R9 (in a common set of games used in most reviews). Mantle reduced it to 6 %.
AMD Mantle API [updating]
- Thread starter MarkoIt
- Start date
in PCGH test, they was use 2 different setups for the 780TI, one is called max: they target 95°C temperature + higherTDP limit.
This allow the gpu to go way higher than his stock speed .. on 1920x1080, there rating is only of 7% more for the 780TI, and 2% at 1440p. The Phantom gpu is effectively a lot stock overclocked anyway.
Now i will not too much base myself on BF hardline right now, i still dont know how they arrive to pull 30fps more of the 290x under DX11 with the Bf4 engine ( i have watch different review of the 780TI of BF4, in windows 8.1, with latest drivers, we are far of this ). For be honest, i dont know why PCGH do this test with it, knowing the state of the game ( for have play it ... )
Last edited by a moderator:
Perhaps having standarized API overhead benchmark for DirectX ala OpenGL API test wouldn't be such a bad idea. Otherwise can imagine sometime in the future IHVs made claim that their close to metal API is XYZ% faster than DirectX XX, when actually significant portion of DirectX deficiencies comes from their part of its implementation.
Why? What the point of using AMD DirectX11 performance in comparisons when its a Mantle enabled game with a Mantle enabled gpu?
I can see it being an issue if you are talking DX11 performance on non Mantle AMD cards or non Mantle PC games. But DX11 performance when the configuration (software + hardware) allows for Mantle is hardly evidence of potential performance issues when the configuration does not.
I don't see the point of devs and the AMD driver team spending resources on trying to up DX11 performance when Mantle is readily an option. Unless you have a circumstance where Mantle use leads to an over all uneven experience, all you end up doing is catering to a situation where someone has a Mantle supported card and Mantle supported game but doesn't want to make use of the Mantle API for no rational reason.
Last edited by a moderator:
The point of comparing Mantle and DX11 on the same GPU is that it allows you to compare 2 APIs, all other things equal. Shocking, I know.
It is fine to add a caveat that this doesn't put DX11 performance of that GPU in the best possible light, but it's still a very interesting comparison. You can't have it both ways: "Look how amazing Mantle is compared to DX11, but don't you dare actually comparing them."
Even without game specific optimization, you could expect/hope a mature API like DX11 to not perform atrociously.
It is fine to add a caveat that this doesn't put DX11 performance of that GPU in the best possible light, but it's still a very interesting comparison. You can't have it both ways: "Look how amazing Mantle is compared to DX11, but don't you dare actually comparing them."
Even without game specific optimization, you could expect/hope a mature API like DX11 to not perform atrociously.
The point of comparing Mantle and DX11 on the same GPU is that it allows you to compare 2 APIs, all other things equal. Shocking, I know.
It is fine to add a caveat that this doesn't put DX11 performance of that GPU in the best possible light, but it's still a very interesting comparison. You can't have it both ways: "Look how amazing Mantle is compared to DX11, but don't you dare actually comparing them."
Even without game specific optimization, you could expect/hope a mature API like DX11 to not perform atrociously.
But that's only true for that 1 manufacturer with that specific device driver. Afterall, the equivalent Nvidia GPU seems to be trouncing AMD in the DX11 path. Surely that has to at least point to weakness in AMD's DX11 drivers and not in DX11 API. ?
YOu should maybe watch different DX11 games and benchmark whre Nvidia is totally dont trouncing AMD .. ( starting by the last skydriver or even my 2x HD 7970's in CFX, trounce easely 2x 780TI overclocked ( over 8K points higher of 2x 780TI with same CPU in line ) 4930K@4.5ghz ) ( and this is without cfx working well in the last combined score ( i got the same result with or without cfx enabled in the last test ( 0.1fps difference in combined test ( classic bug since firestrike have been released ) ... I dont think it use mantle anyway.. ( lets hope Nvidia fix thére things, because this is really embarassing for them )
The problem with BF4, is i can use one map in multiplayer and got 10fps more that any 780TI with a single 290x, or i can use another and get the invert result.. In reality, you need to watch, what level we are speaking about, the setting, and do a mix of both results for get an idea ..
The advantage of Mantle in BF4, @ 1440p, is whatever map i use, i have an incredibly high low minimum framerates, and i got more framerates in average even if is not a big improvement( 10fps ) with my 2x 7970's that i have normally ( full ultra setting + 4xMSAA ).. ( And im sorry, but some light effects are just better in Mantle than in DX11 ( fires effects, explosion and no bug in Lockers when you are outside in the snow with the HDR reflections ( just an example )..
The last improvement is Mantle allow to framepacing directly from the game engine, not by the driver.. you have even the choice of the level of framepacing ( the hardcore version decrease much the frame per second than the second )...
Seriously if Nvidia had not CUDA for support them, what they could do... we have a 5.7Tflops SP 290x vs a 780TI-Titan in the the 4.x Tflops SP ... we have a W9100 who is able to provide 2Tflops DP vs the best dual cores gpu of Nvidia who is not even to provide 1.5Tflops DP .. imagine a W9100x2 in a single board for DP.. this card could provide more DP Tflops than a Titan in SP Tflops with a single core or the Teraflops SP of a 7970Ghz.. ( 4.0+ Tflops DP )..I Hope Nvidia will start to react and stop to push their limit for nothing ...
If you dont use CUDA today, i dont see any professional user buy an Nvidia GPU's .. Or you will need 2 Nvidia gpu#s for attain the same level of performance of AMD GPU's in computing task ? good luck .. ( or even in SP lol ).. Maya have switch to his own computing engine ( who is better on AMD gpu's, ), Adobe use OpenCL now, Autocad still use CPU ( shame ), 3Dmaxs inititial render and blender still use Nvidia ( but the raytracing plugin with AMD mental ray are faster on both ( still get some problem with blender for compile ) ... 4D is going to OpenCL, most high end simulation pc are going to different use who require more AMD .. why they still try to use Nvidia GPU's is over me .. Mathematic and simulations are mixed... but ..
The problem with BF4, is i can use one map in multiplayer and got 10fps more that any 780TI with a single 290x, or i can use another and get the invert result.. In reality, you need to watch, what level we are speaking about, the setting, and do a mix of both results for get an idea ..
The advantage of Mantle in BF4, @ 1440p, is whatever map i use, i have an incredibly high low minimum framerates, and i got more framerates in average even if is not a big improvement( 10fps ) with my 2x 7970's that i have normally ( full ultra setting + 4xMSAA ).. ( And im sorry, but some light effects are just better in Mantle than in DX11 ( fires effects, explosion and no bug in Lockers when you are outside in the snow with the HDR reflections ( just an example )..
The last improvement is Mantle allow to framepacing directly from the game engine, not by the driver.. you have even the choice of the level of framepacing ( the hardcore version decrease much the frame per second than the second )...
Seriously if Nvidia had not CUDA for support them, what they could do... we have a 5.7Tflops SP 290x vs a 780TI-Titan in the the 4.x Tflops SP ... we have a W9100 who is able to provide 2Tflops DP vs the best dual cores gpu of Nvidia who is not even to provide 1.5Tflops DP .. imagine a W9100x2 in a single board for DP.. this card could provide more DP Tflops than a Titan in SP Tflops with a single core or the Teraflops SP of a 7970Ghz.. ( 4.0+ Tflops DP )..I Hope Nvidia will start to react and stop to push their limit for nothing ...
If you dont use CUDA today, i dont see any professional user buy an Nvidia GPU's .. Or you will need 2 Nvidia gpu#s for attain the same level of performance of AMD GPU's in computing task ? good luck .. ( or even in SP lol ).. Maya have switch to his own computing engine ( who is better on AMD gpu's, ), Adobe use OpenCL now, Autocad still use CPU ( shame ), 3Dmaxs inititial render and blender still use Nvidia ( but the raytracing plugin with AMD mental ray are faster on both ( still get some problem with blender for compile ) ... 4D is going to OpenCL, most high end simulation pc are going to different use who require more AMD .. why they still try to use Nvidia GPU's is over me .. Mathematic and simulations are mixed... but ..
Last edited by a moderator:
DavidGraham
Veteran
Except that Mantle has yet to provide any.
True, this should knock quite a few percentage off NV results, but the fact still remains, DX11 performance is no way near the expected level on AMD hardware.
Again, not everybody who owns AMD actually know about Mantle let alone have the knowledge to activate it,it's quite natural that only a fraction of AMD owners actually do, not everybody is a tech savvy. So what do you think is happeneing to these owners? Ignorance in this case is a curse, they would be getting lackluster performance for no reason.
We have StarSwarm, and we can hope FuturMark actually implements a Mantle renderer in 3D Mark, that would be quite awesome.
firstminion
Newcomer
That's what he said.
True. I hope Dave is reading this and will act accordingly.
When Mantle is mature and support is added prior to a game shipping it will surely become the default for AMD hardware.
Linux/SteamOS to get Mantle port
http://www.pcworld.com/article/2364...ux-and-steam-boxes-with-its-mantle-tools.html
http://www.pcworld.com/article/2364...ux-and-steam-boxes-with-its-mantle-tools.html
Oh, absolutely. The Mantle vs DX11 comparison is just one sample point. Without more of them, it's impossible to draw a single conclusion and yours may very well be the right one.
But to proclaim, as one individual did, that it's unfair to even make the comparison is just stupid. There's a huge difference between making a comparison and drawing a conclusion.
Why is that stupid? You can make the comparison all you want, but it doesn't tell you anything other than on particular titles Mantle runs better than DX11 on AMD hardware and DX11 runs better on Nvidia cards.
It's hard to quantify the data or draw general conclusions on how performance differs on a strictly technical level. nvidia is far more motivated to push it's DX11 path to the point of easily justifying the effort to cater specifically to Mantle supported titles in an attempt to make the gap as small as possible. Meanwhile, AMD is encouraged to do the exact opposite. AMD wants you to use Mantle when supported, a wider gap pushes penetration not dissuades it.
If you dont even attempt to parse out opportunities of obvious bias in your data, then you can't trust conclusions that simply ignore those opportunities.
If I wanted to draw some general assertions about DX versus Mantle on AMD hardware, I would look at two sets of data. That being Mantle versus Nvidia DX11 and AMD DX11 versus nVdia DX11 using non Mantle titles. If you know how mantle generally trends against Nvidia Dx11, you should have a pretty good ideal of how Mantle performs against AMD DX11. All without having to depend on data points where Mantle may have influenced the performance of the AMD DX11 path.
Last edited by a moderator:
one game is pre mantle, the other supports mantle but both were tested running DX11
I don't see a drastic change in relative performance, but the Nvidia cards gained a little ground it seems (compare 7970 to 680, 570 to 6970)
I was watching a live interview with Richard Huddy on pcper.com yesterday, he claimed Mantle will be "open" probably next year, AMD will allow others (like NV and Intel) to use and adapt the API with no license fees, but wants to keep control over the API,
also he said Nvidia GPUs should run Mantle well, and that Mantle started like 3 years ago but development 18 months or something, the video should be on youtube soon
I argue that even this by itself is an interesting conclusion. You're trying to make the case that not knowing at all is better than knowing. Yes, I think that is stupid, even if the results confirm the hypothesis.
But let's imagine for a second that this comparison has resulted in the opposite conclusion: DX11 on a Mantle game runs better on AMD than Nvidia. That would have been even more interesting. But you would have never known. And the thing is: I'm sure that there are going to be games out there where this actually will be the case.
I am not saying the comparison is completely useless. I am saying its limited in academic terms and in terms of influencing a gpu purchase unless you don't want to make use of Mantle as an api.
David made a comment that AMD should work on reducing the gap between AMD DX11 versus NV DX11 for Mantle games. I took it as the gap is larger than typical on Mantle supported games.
If true, just looking at DX11 performance on Mantle games with Mantle supported gpus won't tell you much in terms of general AMD DX11 performance on those cards. Supporting Mantle is unduly affecting the AMD DX11 path, so while its good to use the comparison to enlighten us to that fact. It does little else.
In terms of trying to buy the right gpu for you. Unless you want competitive DX11 performance on AMD Mantle supported cards with Mantle games, why would you care how the AMD DX path performs. You have to be hostile towards Mantle as an api for that to matter. And because the presence of Mantle support negatively affects AMD DX11 performance, you are better off looking at non Mantle games when trying to compare nVidia DX11 versus AMD DX11.
Yes, in terms of general gpu interest, the comparison has a place in the discussion. In terms of affecting purchases, it should have no bearing unless you have some prejudice that isn't performance related against the Mantle api.
We're in total agreement then!
Seem to be. I think I was more focus on the comparison in terms of using them to based purchasing decisions. While I concede the comparison itself have value in B3D discussion.
The discussion stemmed from David Graham's assertion that AMD DX11 performance should be of concern to consumers in mantle enabled games. I guess you could argue that until mantle is at a proper release stage and is enabled as default by the game (or AMD drivers/settings wizard software) then this does matter to the consumer. The other concern was how the optimisation of the DX11 path would suffer for non mantle supporting AMD hardware (legacy). As pointed out by HMBR, there doesn't seem to be a glaring issue here in at least one example. This does need to be considered as more mantle games are released, however, this older hardware will usually have a different optimisation path anyway. So unless AMD are reducing man hours on legacy path optimisation to instead improve mantle, there shouldn't be an issue.