AMD Mantle API [updating]

[Kaotik]

Speaking of which, so far Mantle's main advantage has been CPU centric, I still don't get why it's limited to only GCN? why can't it be extended to prior architectures?

Because GCN defines the minimum required feature set, it has to be set somewhere, they chose GCN?

 

[UniversalTruth]

Because GCN defines the minimum required feature set, it has to be set somewhere, they chose GCN?

No, it is because their finance department said so for improving the sales



Which is, of course, super duper annoying and it only gives the sign of the value of the brand which has always been lower than Nvidia's (because of many simple similar cases) Grrr

 

[Dave Baumann]

Its a full 3D API. Again, the draw call limitation is just one facet, a big one and one that Mantle set out to resolve, but this is full, low level API and there is more to explore with it yet.

 

[Davros]

any chance we could could get a cut down api for older cards where the drawcall stuff is done in matle and all the other stuff is d3d ?

 

[Marc]

Would love to be able to release a good benchmark externally, but the one we use internally for multiplayer is not something we can release as our "pseudo soldiers" run on the dedicated server that one sets up on a local machine and then connect to with the machine one wants to play and measure on.

Hi Repi,

I'm Marc, editor in chief of http://www.hardware.fr . Even if a dedicated server is needed, such a benchmark would be VERY useful to us in order to get deterministic results, a thing not possible in real multiplayer.

There is really no chance to release this, even under NDA ?

Thanks

 

[AlexV]

No, it is because their finance department said so for improving the sales



Which is, of course, super duper annoying and it only gives the sign of the value of the brand which has always been lower than Nvidia's (because of many simple similar cases) Grrr

UT you are really not getting the message about your signal-to-noise ratio. GCN is a rather different beast from ATI's prior art, your (conspiracy) theory is baseless.

 

[Marc]

Not really, they didn't specify how they tested multiplayer and most importantly: if the multiplayer was full 64 player server or empty. Affects CPU load massively.
Update: looks like they did specify that it was 64-player MP server which is good assuming it was full the entire time for all tests and they played for a quite long time as the workload is quite variable

I found their results very interesting but i don't understand how they achieve such an amount of reliable benchmark in such a short time in BF4 multiplayer.

 

[DavidGraham]

Its a full 3D API. Again, the draw call limitation is just one facet, a big one and one that Mantle set out to resolve, but this is full, low level API and there is more to explore with it yet.

Hence why I asked why not bring this one "facet" down to other architectures .. there is nothing wrong with partially supporting Mantle on hardware other than GCN, GCN gets the full package (CPU+GPU) ..others benefit too from the CPU centric stuff.

 

[Dave Baumann]

Hence why I asked why not bring this one "facet" down to other architectures .. there is nothing wrong with partially supporting Mantle on hardware other than GCN, GCN gets the full package (CPU+GPU) ..others benefit too from the CPU centric stuff.

Its not something that can be "separated", it is part and parcel of the API hence the entire API needs to be written for the architecture.

 

[repi]

Hi Repi,

I'm Marc, editor in chief of http://www.hardware.fr . Even if a dedicated server is needed, such a benchmark would be VERY useful to us in order to get deterministic results, a thing not possible in real multiplayer.

There is really no chance to release this, even under NDA ?

Thanks

Sorry it would be substantial amount of work to provide that and there is no possibility of giving out the dedicated server builds.

 

[DavidGraham]

more windows 8 testing NVIDIA DX11 vs AMD Mantle
http://gamegpu.ru/action-/-fps-/-tps/battlefield-4-mantle-vs-directx-11-test-gpu.html

 

[Ryan Smith]

Edit: N/M, beaten to it.

 

[itsmydamnation]

while its a datapoint a 4.9ghz 3970 in SP.......................... in a word, who cares? its good to know on a likely water cooled ~250-300 watt CPU that you can cover the DX overhead with raw CPU power. Also that they dont show any of the runs perf graphs means we cant get a good idea of mean/standard deviation/interframe variance......

look at the multiplayer benchmarks (why people play Bf4)
http://www.pcper.com/reviews/Graphi...Early-Performance-Testing/More-Mantle-Results
http://www.hardwareluxx.de/index.ph...nisse-im-kampf-mantle-vs-directx.html?start=1


hell even there own (translated by google) conclusion says the samething:

Probably just a slight increase due to the use pretty powerful CPU in our test stand, which in DirectX 11 mode allows the full potential of video cards. This review has been given a single test of the game and in the near future we will test MANTLE multiplayer graphics cards and processors, where possible, we will see a much larger increase in speed.

to me the gamer config that matters the most are the x770k/8350's (with no OC and modest OC's 4.3/4.4 etc) with 290's.

 

[Andrew Lauritzen]

Cool to see results coming in now. Mostly as expected, but some nice gains in BF4. There seems to be something weird going on with the specular in Mantle (extra bright and aliasy?), but perhaps that's related to the bug fix that was quoted.

gain of 24FPS on average ...gain of 15FPS on average

10-15fps more on the minimal..

I swear to god people if I hear any more FPS deltas I'm going to start banning people repi already warned you guys! This is Beyond3D, I want to see frame times (in ms) And I definitely don't want to see any "minimum frame rate" measured over some interval nonsense. Frame time distributions everyone, say it three times

In case anyone encounters the same thing, Mantle requires Intel integrated GPU to be disabled in device manager to work. Star Swarm would crash every time on Mantle version otherwise.

Oh dear, that (and the lack of clarification to date on Mantle's interaction with WDDM) makes me slightly nervous about what they are doing in their driver :S Guess we'll have to see.

Looking forward to the techreport article.

 

[Alexko]

while its a datapoint a 4.9ghz 3970 in SP.......................... in a word, who cares? its good to know on a likely water cooled ~250-300 watt CPU that you can cover the DX overhead with raw CPU power. Also that they dont show any of the runs perf graphs means we cant get a good idea of mean/standard deviation/interframe variance......

look at the multiplayer benchmarks (why people play Bf4)
http://www.pcper.com/reviews/Graphi...Early-Performance-Testing/More-Mantle-Results
http://www.hardwareluxx.de/index.ph...nisse-im-kampf-mantle-vs-directx.html?start=1


hell even there own (translated by google) conclusion says the same thing:



to me the gamer config that matters the most are the x770k/8350's (with no OC and modest OC's 4.3/4.4 etc) with 290's.

I think the SLI of GTX 780 Tis was somewhat CPU-bound here, but since Mantle doesn't support Crossfire yet, no comparison was possible. I would argue that most gamers are likely to have mid-range or high-end Core i5s rather than big i7s or FX CPUs. HyperThreading is rarely helpful in games anyway, so there's little incentive to go for i7s if games are the target.

The interesting thing about Mantle, in my opinion, is the future: CPUs get maybe 20% faster every generation, while GPUs get something like 50% faster in the same time frame. That means current games are likely to become more and more CPU bound on future hardware. And future games, without Mantle, would be limited in their ability to push CPU loads. In other words, I'd expect Mantle's ability to reduce draw call overhead to bring bigger gains in the future, without even considering further engine/driver/API improvements.

 

[itsmydamnation]

to me the 860/2600/3770/4770 aren't really I7's.. they are all I5's....... real I7's are LGA1366 and above damnit!!!
I have a 3770K @ 4.4 (noctura D14) and to me thats a good upper bound for your "average serious gamer".

 

[Marc]

Sorry it would be substantial amount of work to provide that and there is no possibility of giving out the dedicated server builds.

I understand, thanks for the answer

 

[heli]

more windows 8 testing NVIDIA DX11 vs AMD Mantle
http://gamegpu.ru/action-/-fps-/-tps/battlefield-4-mantle-vs-directx-11-test-gpu.html

As far as I can tell GameGPU is using MSI Afterburner:

(Google translate)
"All the cards were tested for maximum graphics quality program MSI Afterburner."

Like FRAPS, Afterburner is a well known program to benchmark DX and OGL, but how accurate is it to benchmark Mantle? I am surprised it even produces numbers for Mantle. I don't have the hardware to test myself, for those who do, do FRAPS and Afterburner show FPS even under Mantle? If they do, how are they different to BF4's own performance monitor?

I think all future games that have Mantle support should have in-game performance tool like BF4 to avoid complication, otherwise AMD need to release their own benchmark tool before third party programs start to support Mantle path officially.

I agree nowadays reviewers need to show frame time, seeing only FPS doesn't really tell the whole story. A frame time graph is super useful. Readers should also educate themselves to look beyond FPS.

 

[Nemo]

I disabled ULPS and check this.
DX11 - Shangie - looking at tower from the sea - 64 man server = 65fps.
MANTLE - same location and conditions = 143fps.
My CPU bottleneck is removed! My 290x's are now scaling perfectly!
W00t!!!!

http://www.overclock.net/t/1429303/amd-mantle-discussion-thread/1350#post_21708645

 

[DavidGraham]

while its a datapoint a 4.9ghz 3970 in SP.......................... in a word, who cares? its good to know on a likely water cooled ~250-300 watt CPU that you can cover the DX overhead with raw CPU power.

Well, for starters single player is typically GPU limited at Ultra settings, you'd hardly gain any frames from say overclocking a good CPU from 2.0 to 4.0GHz. see this Anand test for example:
http://www.anandtech.com/show/7728/battlefield-4-mantle-preview

I posted this bench because it is not the first one to show significant Geforce gains in BF4.

Like FRAPS, Afterburner is a well known program to benchmark DX and OGL, but how accurate is it to benchmark Mantle?

I think this point deserves further investigation.

Meanwhile HOCP has posted their results, showing an interesting mix of CPU/GPU limitation in multi-player
http://www.hardocp.com/article/2014...mance_preview_in_battlefield_4/3#.Uu-spLQvkgQ