AMD Mantle: A Graphics API Tested In Depth

[TheGreatWarMage]

Just to ask, why was the 4170 picked?

4170 was either equal to or beat the 8350 in bf4, but in Thief lost by decent margins (mantle and non, and gtx). I expected a better margin to say the least.

"This puts more control of resources in the developer's hands, instead of the API's, allowing for better optimization."

I am wondering if Thief was designed with Mantle as a core, and bf4 just had it in mind. If this statement given above is accurate, and the scores more so, than this API shows much more merit than it is given. Look at the 4770k with the GTX, than the 8350 with Mantle, than the 8350 GTX. Those changes are huge. It is showing mantle cards steaming passed their scores with the GTX.

It seems Mantle is evening the playing field for the poorer made CPUs, while showing what they would be like if they were made up to scuff like the high end I7s

 

[vertexx]

In previous articles I have read, it was clear that Mantle provided the best performance boost in CPU bound scenarios, which plays out in your thief testing. in BF4, you kept raising the graphics quality, thus testing in GPU bound scenarios and largely invalidating the results. And Multiplayer pushes the CPU even more, so I agree with the masses here that you guys need to figure out some way to do multiplayer testing.

 

[AndrewJacksonZA]

Thanks for the research done Don. I appreciate some facts, even though it has deflated my fanboyism for Mantle a little bit. However, pairing this data with price data it can help sway purchasing decisions for some people, myself included.

One comment though, the graphs are quite difficult to read on a 4.5" cellphone screen, could you perhaps try putting a pattern on the DirectX or the Nvidia cards to make it easier to read please? Then there will be solid bars for Mantle, a gradient for DirectX and a pattern for Nvidia.

Thanks again Don, it was a nice read!

 

[somebodyspecial]

AMD's Mantle is available to users of certain Radeon cards, as are the first few titles with corresponding API support. We gathered up a number of CPUs and graphics boards, fired up Battlefield 4 and Thief, and set off on a benchmarking odyssey.

AMD Mantle: A Graphics API Tested In Depth : Read more

Can you retest the 290x/780ti in 1080p? It's clear from your results all of the cpus at the top are waiting on 780ti because you raised the resolution (for who I don't know? <4% of the planet?). It makes the benchmarks here pointless as you can't see the cpu's spread their wings. All of the scores are essentially the same and less than 4% run above 1920x1200 so what is the point (and most of those people have 2 or more cards to play above that according to steam survey)? 1080p would be more interesting and then you wouldn't have an FX4170, i3-3220, A10-7850 etc running like an i7-4770 due to gpu limits. i3-3220's don't score like 4770's unless you're graphics is holding you back. Drop the res and rerun at least the 780ti/290x tests in ONE of the games please. You pretty much killed the whole point of the high-end benchmarks.

On another note, I'm wholly unimpressed even with the current state of your results for high end. BF4 was supposed to be great on mantle, yet 290x still can't take out 780ti in their own $8Million optimized game. You get what you pay for, which is why 780ti comes at a premium I guess (they lose in star swarm now too). Dropping to 1080p would show this even more pronounced. I expected them to win their primary game across the board. With the results being all over the place (thief terrible with 4770+290x, with or without mantle it got smashed by 780ti, is mantle only good with AMD cpus? dx fares no better with Intel either), I'm left thinking this will be a LONG beta program.

It also seems AMD should have been spending some time/R&D on DirectX driver optimization, looking at their results. With so few mantle games even by end of year or next year they had better start doubling down on DirectX improvements IMHO. If they don't start concentrating on dx/opengl my next card purchase is NV by default just due to knowing they ARE working on them and they work across all games all the time. You can't just say MANTLE is in beta when the ALL INTEL chips scores so badly in DX too. Are their directx drivers going to crap on Intel for a reason? I get Mantle possibly being aimed at lowend or even at AMD cpus early on, but what they heck is wrong with the DX scores in thief? Why is i7-4770+290X in DirectX scoring 44fps avg while 4770+780TI is scoring 79.9fps? FX8350 outscores all Intel chips in DX. Flop in the NV 780ti though and all Intel chips rise to the top vs. AMD dx. The i3 on 780ti (59.3fps) actually beats 4770 on 290x (47.3fps) by a large margin. The DX 290x scores are awful on Intel. BF4 shows Intel+290x DX lower also but not as bad as thief. Has NV just really done their homework on DX, or does this just apply to these two games? I think you need to do a few more game tests in DX for 780ti vs. 290x to see if they are just completely ignoring INTEL+DX here or are they really behind now after the last few revs of NV concentrating on DX to offset mantle?

NV has the money to waste on a Gsync failure (if not-so-free sync actually works at some point), but AMD can't afford to blow money on a Mantle failure, due to dx12, opengl or just plain having it ignored if they don't write checks repeatedly to get it used like with EA. It would have been cheaper to just write some great drivers for dx/opengl. Quarterly report later today from AMD (after midnight now).

 

[abundantcores]

User testing around the Internet shows Intel are at least as fast in Mantle as AMD CPUs.

Its obvious there is something very wrong with your testing.

[Answer by Cleeve:]

Actually, nobody on the internet has really tested anything but Radeon R9 290s when it comes to Mantle. Instead of assuming it always works in every situation like everyone else, we actually tested it, not only with the 290, but with a wide range of CPUs and GPUs. We did this with feedback from AMD.

Mantle is not the perfect, slick API that people assume it is. AMD stresses that its considered a beta and is not yet a final product.

I believe that our test results are quite accurate. The problem is that the API, something the developer considers unfinished at this point, probably still needs work and has room to grow. And I'm sure it will get better over time.

But pretending it works 100% perfectly doesn't help anyone. That's not even something that AMD would try to say at this time as it has obvious issues. They are working on it. But there is a value to in-depth testing and to sharing the inconsistencies.

i7 with 4 cores and 8 threads right here http://forums.overclockers.co.uk/showthread.php?p=26614553#post26614553

Using your settings.

Min 62 FPS
Avr 82 FPS

here is one with a 4770K @ 4.5Ghz

Avenged7Fold: 290X @ 1300/1604, 4770K @ 4.5Ghz - Mantle FPS: Min 71.8 / Avr 88.5. http://forums.overclockers.co.uk/showpost.php?p=26380265&postcount=291

In CF i7 again with Mantle:LtMatt: 290P @ 1250/1625, 2700K @ 5.3Ghz - Mantle FPS: Min 109.5 / Avr 142.5
http://forums.overclockers.co.uk/showpost.php?p=26367850&postcount=186

Those two ^^^ are actually on the highest possible settings.

I know a lot of people using i7's with mantle, in Thief and BF4, some with CF 290's, they all report FPS gains with Mantle over DX, especially when in CF. in BF4 averaging 170 FPS with 120 FPS minimums. over 50% higher than they do in DX.

I don't know where you have gone wrong. but everyone on our forum is utterly confused by your i7 results.

http://forums.overclockers.co.uk/showthread.php?t=18611853

One other thing, Mantle vs DX in BF4 (Single player) is utterly meaningless, if anything it shows you don't understand Mantle at all.

BF4 in Single player has almost 0 work done on the CPU, which is why all the CPU performance is virtually the same.

Only in Multiplayer maps does the CPU matter, you don't even need to have any players on the map, the difference in DX and Mantle in Multiplayer is massive, go to Shanghai, empty server, CN base to Flag B, you will see a huge difference, especially in CF.
Its so easy to find where Mantle matters, its child's play, its blatantly obvious.
I just don't get how this generation of reviews can understand so little about what they are doing, and make so many mistakes and have so many apparent problems along the way,
not just here, almost anywhere (in English speaking Nations) Mantle reviews are either completely out of kilter with user experience, or tested in a way where Mantle has no effect, testing BF4 in single player only, never testing in Multi GPU, others use massively overclocked i7 CPU's on mid range GPU's, the theme is always the same, using completely pointless testing they conclude "Mantle has little to no impact" its almost as if they are all reading from the same deliberately ignorant script.
its almost laughable, if it was not so despairing.

I will say tho, this is by far the best Mantle test from English speaking Nations to date. you did use weaker CPU's to clearly show the CPU overhead advantage that Mantle has over DX. your review is pretty good overall.

It would have been better, more complete with Multi GPU testing, and be it AMD not giving you the right feedback or you not not trying hard enough to to work out your problems with the i7, those results are not the norm, Hyper Threading works perfectly in Mantle, your results are confusing to a lot of those using i7's with Mantle, please look again at it. ask AMD about it. It would also be interesting to know what they have to say about it.

When I get home in a few days I will do a proper Mantle review, show how, where and why.

 

[cleeve]

I don't know where you have gone wrong. but everyone on our forum is utterly confused by your i7 results.

Well, I am, too. That's why I reported them. I will point out that everyone on your forum seems disinterested in the 95% of results that are consistent and have wholeheartedly concentrated on the anomaly. I do find it interesting that you automatically assume that the software is 100% perfect and that I *must* be doing something wrong, though. Is it impossible for the software to be imperfect?

It's really, really easy to immediately blame the tester and test conditions and invalidate the results out of hand when something doesn't go as planned. Then again, that's where you can learn some interesting nuances and see if everything might not be as perfect as we'd like. What you might want to ask yourself is how I got those strange results on the exact same test systems that delivered the consistent Thief results that showed Mantle with an obvious and consistent advantage. If our lab conditions are so flawed, do we throw those results out, too?

And you know what? As a publication, it would be way easier to hide and bury anomalies and pretend they don't exist for the sake of showing consistent results. But that's not what a good lab tech does, though, even though good lab techs who report inconsistency tend to be the ones who get invalidated by passers by. But when you encounter something odd, and retest it, and its repeatable, it's worth reporting. It gives the devs and hardware manufacturers feedback, and shows that maybe everything isn't 100% OK.

It's not that simple and its not cut and dry. I guarantee I'm testing in a lab environment with far more checks and balances and careful setup than you gents on an internet forum. Is everyone whose posting their results using the same drivers as I did, the same game version, and the same benchmark run, etc, etc? Of course they aren't.

You have to ask yourself those questions. I know considering the complex implications of imperfection is harder than pointing a finger and saying it sucks... just like it's harder for me to share results that aren't ideal.

 

[Fredvdp]

I have tested the difference between Direct 3D 11 and Mantle in a single player segment of Battlefield 4. I have an AMD Radeon R9 270X and a Phenom II x4 965 BE (bottleneck), and tested the game on high settings.

The difference is ridiculous. I got 35-40 fps in Directx 11, and 60-80 fps in Mantle. I tested the exact same area in both cases.

 

[utengineer]

There was a link to a Polish website (I think it was polish) that showed CPU and GPU benchmarks for multiplayer. Most people didn't think it was all that reliable, but it did show that CPU scaling was much more pronounced in multiplayer. I didn't think they (EA) used a different texture set or models for multiplayer or anything though... not sure on that one.

The problem is that every single game is different so you can't reliably benchmark multiplayer rounds. In the olden days you could generally record a session of muliplayer on your own dedicated server and then play it back so you could get repeatable results, but since EA seems to hate PC gamers you can't do cool stuff with games anymore.

Thanks for your response. I understand every game is different. Let me provide another example. When Unreal Tournament 3 released, my PC could not run the Single Player Campaign, but the Multiplyer worked fine. Visually, it appeared the Single Player mode had all the stunning graphics, but the Multiplayer appeared to be using lower grade graphics. I assumed this is why I could run MP and not SP.

You recall Crysis 3 releasing Hi-Res texture updates after the game released? This appeared to only enhance the Single Player mode scenes and maps.

 

[abundantcores]

I don't know where you have gone wrong. but everyone on our forum is utterly confused by your i7 results.

Well, I am, too. That's why I reported them. I will point out that everyone on your forum seems disinterested in the 95% of results that are consistent and have wholeheartedly concentrated on the anomaly. I do find it interesting that you automatically assume that the software is 100% perfect and that I *must* be doing something wrong, though. Is it impossible for the software to be imperfect?

It's really, really easy to immediately blame the tester and test conditions and invalidate the results out of hand when something doesn't go as planned. Then again, that's where you can learn some interesting nuances and see if everything might not be as perfect as we'd like. What you might want to ask yourself is how I got those strange results on the exact same test systems that delivered the consistent Thief results that showed Mantle with an obvious and consistent advantage. If our lab conditions are so flawed, do we throw those results out, too?

And you know what? As a publication, it would be way easier to hide and bury anomalies and pretend they don't exist for the sake of showing consistent results. But that's not what a good lab tech does, though, even though good lab techs who report inconsistency tend to be the ones who get invalidated by passers by. But when you encounter something odd, and retest it, and its repeatable, it's worth reporting. It gives the devs and hardware manufacturers feedback, and shows that maybe everything isn't 100% OK.

It's not that simple and its not cut and dry. I guarantee I'm testing in a lab environment with far more checks and balances and careful setup than you gents on an internet forum. Is everyone whose posting their results using the same drivers as I did, the same game version, and the same benchmark run, etc, etc? Of course they aren't.

You have to ask yourself those questions. I know considering the complex implications of imperfection is harder than pointing a finger and saying it sucks... just like it's harder for me to share results that aren't ideal.

Don't just quote one tiny part of what I said and then respond to that text out of context, there is more text in my post which makes it pretty clear that I respect your testing.

I will say tho, this is by far the best Mantle test from English speaking Nations to date. you did use weaker CPU's to clearly show the CPU overhead advantage that Mantle has over DX. your review is pretty good overall.

It would have been better, more complete with Multi GPU testing, and be it AMD not giving you the right feedback or you not not trying hard enough to to work out your problems with the i7, those results are not the norm, Hyper Threading works perfectly in Mantle, your results are confusing to a lot of those using i7's with Mantle, please look again at it. ask AMD about it. It would also be interesting to know what they have to say about it.

Why is it even relevant that your testing in a lab environment? if your lab environment produces different results than what users can expect then your lab is mute, it says nothing about reality, users don't game in a lab.
So is there a reason your 'Lab' result do not tally with users results?

Users use the latest version of the game and drivers, that's it. don't you do the same in your Lab?

Do you have anything to say about testing Mantle BF4 in Single player vs Multi player maps?

 

[cleeve]

Don't just quote one tiny part of what I said and then respond to that text out of context, there is more text in my post which makes it pretty clear that I respect your testing.

I responded to your entire post, which I read. I simply anchored to that statement because it made sense to do so. I did notice that you've tried to be gracious, so please don't be defensive, i'm not attacking you. Just trying to impart another point of view.

Why is it even relevant that your testing in a lab environment? if your lab environment produces different results that what users can expect then your lab is mute, it says nothing about reality, users don't game in a lab.
So is there a reason your 'Lab' result do not tally with users results?

Users use the latest version of the game and drivers, that's it. don't you do the same in your Lab?

Its relevant because it means I'm taking care to use the same drivers, software versions, hardware settings, and benchmark run. I'm also using a benchmark run directly from the game, so it's valid, but does that mean that it's the same one you are using? Are those test conditions invalidated when someone gets a different result using a different part of the game with different settings, hardware, software? Do you see how futile it is to run something from home under completely different conditions and declare "TEST INVALID, I GOT 126 FPS!!!!"

As for drivers, the article mentions the long-term nature of these tests, and that we ran some post-14.6 driver runs with better results. 14.7s just got released, but it takes us weeks to finish a long-term investigation like this. Things will change over time, that's inevitable - and it's good! We can provide a snapshot, but I can't write up a couple weeks of testing the second an update comes out, not properly anyway.

I think there may be more to it than you realize, that's all. We're all interested in the best possible information, so we're on the same side.

As for multiplayer, I've covered that in previous responses. I don't believe the environment is consistent enough to effectively separate mantle as a variable.

 

[abundantcores]

Don't just quote one tiny part of what I said and then respond to that text out of context, there is more text in my post which makes it pretty clear that I respect your testing.

I responded to your entire post, which I read. I simply anchored to that statement because it made sense to do so. I did notice that you've tried to be gracious, so please don't be defensive, i'm not attacking you. Just trying to impart another point of view.

Why is it even relevant that your testing in a lab environment? if your lab environment produces different results that what users can expect then your lab is mute, it says nothing about reality, users don't game in a lab.
So is there a reason your 'Lab' result do not tally with users results?

Users use the latest version of the game and drivers, that's it. don't you do the same in your Lab?

Its relevant because it means I'm taking care to use the same drivers, software versions, hardware settings, and benchmark run. I'm also using a benchmark run directly from the game, so it's valid, but does that mean that it's the same one you are using? Are those test conditions invalidated when someone gets a different result using a different part of the game with different settings, hardware, software? Do you see how futile it is to run something from home under completely different conditions and declare "TEST INVALID, I GOT 126 FPS!!!!"

As for drivers, the article mentions the long-term nature of these tests, and that we ran some post-14.6 driver runs with better results. 14.7s just got released, but it takes us weeks to finish a long-term investigation like this. Things will change over time, that's inevitable - and it's good! We can provide a snapshot, but I can't write up a couple weeks of testing the second an update comes out, not properly anyway.

I think there may be more to it than you realize, that's all. We're all interested in the best possible information, so we're on the same side.

As for multiplayer, I've covered that in previous responses. I don't believe the environment is consistent enough to effectively separate mantle as a variable.

My point was if your lab environment is like that of a user then it should not be different to a user, so it is relevant, obviously you had problems other aren't, I think its something that needs investigating further.

As for multiplayer, I've covered that in previous responses. I don't believe the environment is consistent enough to effectively separate mantle as a variable.

If any thing using an empty MP map is more consistent than trying to play single player, an empty MP map never changes and your not being shot at.

Its also very easy to test for that consistency, run the same test several times. its very simple, spawn at the CN flag, start the bench and move from there to Flag B, do it over and over again in the same way, its very consistent and doing that shows a huge difference between Mantle and DX, especially the more GPU power you put into it, or less CPU power.

These are not exactly that, they were done a while ago on a P-II x6 CPU and 7870XT, both with the same medium setting and similar, they had players on the map. so as I say not exactly what I'm talking about. I did testing in an identical sense but did not record those. in a few days I can shot you exactly what I'm talking about.

DX https://www.youtube.com/watch?v=3WFqX5k_GgA&list=UU-G7rC9knVxOW9FZWiew3rg

Mantle https://www.youtube.com/watch?v=pcEEuIvN8FI&list=UU-G7rC9knVxOW9FZWiew3rg

Keep your eye on the FPS, there is a very significant difference.

When I get back I will give you proper identical runs on empty servers with a 290 and an FX-8350

 

[mapesdhs]

I have tested the difference between Direct 3D 11 and Mantle in a single player segment of Battlefield 4. I have an AMD Radeon R9 270X and a Phenom II x4 965 BE (bottleneck), and tested the game on high settings.

Not convinced that really means much on an IPC-starved CPU like the Ph2.

Ian.

 

[cleeve]

My point was if your lab environment is like that of a user then it should not be different to a user, so it is relevant, obviously you had problems other aren't, I think its something that needs investigating further.

Of course it is! And we plan to re-investigate Mantle as more titles get introduced and it gets updated over time.

Its also very easy to test for that consistency, run the same test several times.

I've done that. I'm not happy with the results, and the margin of error bothers me when we're trying to isolate a separate variable like Mantle that might cause a very low impact.

As you said, though, what's the point of results if it doesn't reflect what real people will actually experience?
How valid are tests where we have to search for a map that nobody else is playing in?

DX https://www.youtube.com/watch?v=3WFqX5k_GgA&list=UU-G7rC9knVxOW9FZWiew3rg

Mantle https://www.youtube.com/watch?v=pcEEuIvN8FI&list=UU-G7rC9knVxOW9FZWiew3rg

Nice videos, and obviously suggests that there is more to the story. The 7970 has 3GB of RAM and its a fast GPU, the next step down from Hawaii. After seeing the 270X results, introduces questions like, is it more to do with RAM like AMD said, or maybe the power of the GPU? Perhaps a 3GB card with enough raw power can make better use for Mantle? Perhaps the specific multiplayer test has something to do with it, or perhaps the newer driver and update?

I never said this was the end of the story, far from it. It's still in what AMD considers to be a beta phase and things will likely improve. More questions, more tests to do over time. If I can find a way to run multiplayer maps without people popping in and out it would be nice, maybe I'll ask Dice to set up a test server for me. We'll be creating unrealistic results with only a single player on a multiplayer map, though.

But I will point out that none of this invalidates the anomalous results we saw, or proves that the lab setup was anything but perfect. It simply shows that different results can be achieved with different variables.

 

[abundantcores]

My point was if your lab environment is like that of a user then it should not be different to a user, so it is relevant, obviously you had problems other aren't, I think its something that needs investigating further.

Of course it is! And we plan to re-investigate Mantle as more titles get introduced and it gets updated over time.

Its also very easy to test for that consistency, run the same test several times.

I've done that. I'm not happy with the results, and the margin of error bothers me when we're trying to isolate a separate variable like Mantle that might cause a very low impact.

As you said, though, what's the point of results if it doesn't reflect what real people will actually experience?
How valid are tests where we have to search for a map that nobody else is playing in?

DX https://www.youtube.com/watch?v=3WFqX5k_GgA&list=UU-G7rC9knVxOW9FZWiew3rg

Mantle https://www.youtube.com/watch?v=pcEEuIvN8FI&list=UU-G7rC9knVxOW9FZWiew3rg

Nice videos, and obviously suggests that there is more to the story. The 7970 has 3GB of RAM and its a fast GPU, the next step down from Hawaii. After seeing the 270X results, introduces questions like, is it more to do with RAM like AMD said, or maybe the power of the GPU? Perhaps a 3GB card with enough raw power can make better use for Mantle? Perhaps the specific multiplayer test has something to do with it, or perhaps the newer driver and update?

More questions, more tests to do over time. If I can find a way to run multiplayer maps without people popping in and out it would be nice, maybe I'll ask Dice to set up a test server for me. We'll be creating unrealistic results with only a single player on a multiplayer map, though.

Looking forward to that.

Having used a 7870 (270X) in BF4 with Mantle and got it working (as you can see in the vid) I think I may know what the problem is, you are right, its to do with Vram, but nothing to do with performance, if AMD were to be completely honest they might tell you that actually the old HD 7K. (now the 270/80/X) are not fully Mantle compatible, they have a different architectural memory management to Hawaii.

What I found is if you turn Texture Quality down in BF4's settings those older 2 or 3GB cards actually work perfectly in Mantle, or at least mine did and I have had success in advising others to do the same with older AMD cards, (tho not always)
If you have high Texture quality what tends to happen is the vRam fills up over time, eventually resulting in a stack overflow causing stuttering, for what ever reason the Vram is never cleared on the older cards, the higher the texture settings the faster the Vram fills.

Hawaii GPUs don't fill the Vram over time, its something that one can observe happening with older AMD cards if one runs BF4's OSD

 

[Onus]

I have tested the difference between Direct 3D 11 and Mantle in a single player segment of Battlefield 4. I have an AMD Radeon R9 270X and a Phenom II x4 965 BE (bottleneck), and tested the game on high settings.

Not convinced that really means much on an IPC-starved CPU like the Ph2.

Ian.

Seems to me this is exactly the issue Mantle is intended to address.

 

[BigGreenEgger]

I have a 280x and i7 2600k. I see very similar FPS in mantle or directx. What i can see big time is CPU utilization. My main measurement is heat. I OC the 2600k to about 4.4. With directx i sweat. With mantle, i'm comfortable. The downside is that i use DXTory for screen capture but it only hooks directx calls.

 

[mapesdhs]

Seems to me this is exactly the issue Mantle is intended to address.

I'd rather they just developed a better CPU. Atm there's nothing in AMD's lineup that appeals to me.

Ian.

 

[childofthekorn]

With the growing number of titles utilizing mantle and the outreach their getting from their competitors that are interested in adding their own hardware I can't wait to see what this achieves. Although the gains aren't anything to praise as legendary still getting a few FPS is something to consider as good. How the hell is a free (for the consumer) FPS boost, no matter how small, going to be considered a bad thing? I'll be more than excited once AMD releases this for other brands so I can stop hearing the "AMD going proprietary" BS.

 

[somebodyspecial]

Seems to me this is exactly the issue Mantle is intended to address.

I'd rather they just developed a better CPU. Atm there's nothing in AMD's lineup that appeals to me.

Ian.

And while they're at it, better R&D on DirectX/OpenGL that effects nearly all games. Clearly DX has issues as I noted before. There probably going to get worse though, stock of 80c in the last hour after posting a loss and weak outlook for Q3. Still trying to dig through it all to find out what really didn't the damage and if the rumor of piled up stock of gpus was one of them after mining crashed.

 

[somebodyspecial]

With the growing number of titles utilizing mantle and the outreach their getting from their competitors that are interested in adding their own hardware I can't wait to see what this achieves. Although the gains aren't anything to praise as legendary still getting a few FPS is something to consider as good. How the hell is a free (for the consumer) FPS boost, no matter how small, going to be considered a bad thing? I'll be more than excited once AMD releases this for other brands so I can stop hearing the "AMD going proprietary" BS.

Only Intel asked to look (turned away multiple times), and from their own comments they'll pass now and concentrate on DX12. NV never wanted it. So they are both out. PS4 devs said they have no need for it as PS4's api is good enough. MS won't ever use it and will wait for dx12 for xbox1. I see no outreach that happens unless AMD writes a check causing the outreach (dice/ea $8mil payment as an example). It's like Intel with baytrail, they gave away 1.15Billion in socs, to make 50mil in revenue. That's PAYING you to use their chip. Hence the 1.1B loss on mobile this Q accelerated from last Q of ~950mil loss. Heck they may hit 5B losses on mobile this year. AMD has to pay people to use mantle, as a dev gets ZERO extra money from users buying a "mantle game" and instead only gets extra WORK.

It's considered a bad thing, when you took a FEW titles and gave a FEW fps, instead of just doing DirectX/OpenGL drivers massively like NV which affects ALL games for all your user base. All money spent on mantle does nothing for users of amd cards who can't use it, and it does nothing for games that do NOT use mantle at all. ALL of the money spent here should have went to better drivers for dx/opengl. You can't fall massively behind in ALL games for a few mantle games to run OK (which they still lose as shown here, 780ti tops list in both mantle titles). How good is mantle if you still can't win in your two OPTIMIZED games? Star Swarm is a loss now too. What was the point with results like these?

http://www.geforce.com/drivers/results/74636
What they needed to be doing is this every driver rev:
GeForce GTX 700 Series (Single GPU):

Up to 64% in Total War: Rome II
Up to 25% in The Elder Scrolls V: Skyrim
Up to 23% in Sleeping Dogs
Up to 21% in Star Swarm
Up to 15% in Batman: Arkham Origins
Up to 10% in Metro: Last Light
Up to 8% in Hitman Absolution
Up to 7% in Sniper Elite V2
Up to 6% in Tomb Raider
Up to 6% in F1 2013


GeForce GTX 700 Series (SLI):

Up to 71% in Total War: Rome II
Up to 53% in Sniper Elite V2
Up to 45% in Aliens vs. Predator
Up to 31% in Sleeping Dogs
Up to 20% in CoD: Black Ops 2
Up to 10% in Hitman Absolution
Up to 9% in F1 2013
Up to 7% in Far Cry 3
Up to 6% in Metro: Last Light
Up to 6% in Batman: Arkham Origins

IF NV keeps spending here massively, AMD will be worthless. The two worst games got 6%.
"Performance - Introduces key DirectX optimizations which result in reduced game-loading times and significant performance increases across a wide variety of games"

This was the driver that handed them BF4 and StarSwarm wins (thief too I guess).

http://www.geforce.com/drivers/results/75991
And now out of beta, note the list they say that gets >10% perf improvements. 337.88. Also note it affects nearly every card in use today (probably all in use, everything from the last decade). This is when a few fps in a VERY FEW games is a bad thing. Especially when the competition is doing it for everything and not just a few fps.

 

[de5_Roy]

it was nice of you to use 1440p res. for the high end gfx cards. most of the "mantle's revolutionary improvements" i've seen so far are with these cards playing the games at 1080p, which i found kinda pointless. even many non-mantle games will be cpu limited at 1080p with these cards, albeit with 60+ fps.

i wonder what test platform and driver version amd used to diagnose the 4GB vram issue. what if 16GB ddr3 1866 (or faster memory) kit was used?

in this page:
http://www.tomshardware.com/reviews/amd-mantle-performance-benchmark,3860-4.html
is the last benchmark using 1440p resolution or 1080p resolution.?

are thief and bf4 sp memory bound? i am just thinking up unusual causes.

 

[somebodyspecial]

it was nice of you to use 1440p res. for the high end gfx cards. most of the "mantle's revolutionary improvements" i've seen so far are with these cards playing the games at 1080p, which i found kinda pointless. even many non-mantle games will be cpu limited at 1080p with these cards, albeit with 60+ fps.

i wonder what test platform and driver version amd used to diagnose the 4GB vram issue. what if 16GB ddr3 1866 (or faster memory) kit was used?

in this page:
http://www.tomshardware.com/reviews/amd-mantle-performance-benchmark,3860-4.html
is the last benchmark using 1440p resolution or 1080p resolution.?

are thief and bf4 sp memory bound? i am just thinking up unusual causes.

You can clearly see in BF4 it's completely GPU limited using 1440p (all the cpus score ~same). Why test for a res <2% of the world uses? They should test 1080p where most of us play (not me I'm in the 1920x1200 camp but the bulk 96% are below 1920x1200). Most people above 1920x1200 are using 2 cards or more and we're still talking the other half of that 4% using above 1920x1200 (about 1-2% using 1440p, and like I said most use 2 cards+). In MP BF4 would not be playable much maxed on even 780ti at 1440p. Mins are already 44 with SP. The guy running 1080p with the same setup will likely keep killing you over and over while you try to figure out how he's doing it.

 

[tomfreak]

You guys need to test MP... I know its hard and blah blah, but you can just add error margin to results and just describe the differences if you guys notice any.

You guys are testing a grippy car in just a drag race. That's unfair. You need to add cornering as well.
Cheers!

I hear you. But in this case it's not about difficulty, it's about adding unnecessary variables.

The point of benchmarking is to target specific data. For this Mantle article, that data is the graphics API in BF4, not SP vs. Multiplayer performance in BF4.

In this particular case, testing MP wouldn't make sense. it would just muddle the data with a less-consistent test case.

How many of ur work mates own a Bf4? Get ur mates up, rent a private server, have a tank or vehicle on a screen, blow things up, shooting. rise and repeat doing the same thing again on the different hardware setup? That should be pretty close to consistency.

 

[cleeve]

How many of ur work mates own a Bf4? Get ur mates up, rent a private server, have a tank or vehicle on a screen, blow things up, shooting. rise and repeat doing the same thing again on the different hardware setup? That should be pretty close to consistency.

In a perfect world!

I don't have an infinite budget though, We're talking a *lot* of manpower and manhours to do it right on multiple, identical configurations.

Wish I could! But I just don't have those resources.

 

[ArmedandDangerous]

Even if it's eclipsed by DirectX 12, there's every reason to believe that Mantle is what spurred Microsoft into motion on its next-gen graphics API

Except the fact that nvidia has been working with and on DX12 for over 4 years already.