AMD Mantle API [updating]

Should be awesome for laptops, where you don't have ridiculously overclocked CPUs.
 
Last edited by a moderator:
The settings for the BF4 testing ostensibly keep GPU performance from being a limiter, but it still looks like diminishing returns at the upper FPS range on a 4770K that shows no signs of sweating.
There seems to be something that isn't budging much on AMD's side of the equation.

It is quibbling over milliseconds at this point, but the frame timings really show that while Kaveri isn't a terrible CPU, it's a stark comparison with the Intel solution that is more consistent and performant even with D3D.
 
New Star Swarm Build: Better Mantle Performance
http://oxidegames.com/2014/02/04/new-star-swarm-build-better-mantle-performance/


We just deployed a new build of our Star Swarm stress test that significantly improves the demo’s performance using AMD’s Mantle API. We suggest that those of you interested in benchmarking and performance numbers re-run your Mantle scenarios; you may be surprised at the results.

Did we crack some secret code or find a crazy new optimization? No, nothing so spectacular. The truth is that we made a mistake in our haste to deploy the build that stripped out the activation process. We didn’t follow our normal release process, and missed how a minor change in that build had disabled some of the Nitrous engine’s multi-threading features when using Mantle. Unfortunately we didn’t notice that at first, as nobody was running Mantle last week due to the beta driver being delayed.

It’s all fixed now, so Mantle users should see a noticeable boost in performance on most configurations (this fix doesn’t have a huge effect on powerful-CPU/slow-GPU systems). We’ve also addressed the “gray screen” issue that was affecting some non-English Windows systems.

We appreciate everyone’s patience and feedback as we hunt down some of these hard-to-reach bugs for this alpha test of the Nitrous Engine.


:D


Ok, people. It's time for new bench numbers!

That's swell, but Steam doesn't seem to know that is should be updating Star Swarm.
 
The smarmy smilie noise post has already been done.

I'm uncertain as to whether this is because of the Mantle driver.
Look at how much the 780 Ti benefits in D3D when changing from Kaveri to Haswell, which gives a measure of just how powerful the CPU is.
Then see how small the gain is for the transition D3D to Mantle on Haswell.
We know how sizeable the CPU side of AMD's D3D drivers are, given how ditching Kaveri does so well.

If the overall rendering methods are equivalent, Mantle doesn't create peak performance that's not there.
AMD's big cards have had difficulty distinguishing themselves at modest settings against more svelt Nvidia GPUs. Is this another instance where the architecture's own overheads keep it from spooling up with a lighter workload?
The 290X is more of GCN, but it isn't all that better.
 
Again, NVIDIA was not really that faster .. however after the patch they received significant performance gains, due to newly added optimizations that were absent since launch. In fact as I recall it .. NVIDIA was also the faster one back in Battlefield 3 days (FrostBite 2), the engine just seemed to favor them more.


Found this interesting AMD slide :

Mantle-Battlefield-4-Performance-635x434.jpg

Z
 
Last edited by a moderator:
Techreport's choice of game settings reads like a guide on what to use to minimize Mantle's benefit. Only High settings, no resolution scaling, only FXAA, single player. Perhaps Johan can chime in on this.
 
Found this interesting AMD slide :
Different cards, settings, and the results are consistent with AMD's statements of largest benefits in CPU-bound scenarios and not wholly inconsistent with TR's testing that tried to measure a CPU-bound scenario.

Techreport's choice of game settings reads like a guide on what to use to minimize Mantle's benefit. Only High settings, no resolution scaling, only FXAA, single player. Perhaps Johan can chime in on this.

Mantle's benefits, per AMD, are greatest when not GPU limited and when the CPU is more modest.
Reducing settings reduces the chance that you are GPU limited, but as I mentioned AMD tends to lean heavily on higher settings and resolutions to better leverage parts of the architecture that don't bottleneck at lower settings. This is why at times Nvidia cards manage to keep up in benchmarks until the resolution hits fill rate or bandwidth hard, or just caps out a limited frame buffer.
 
Different cards, settings, and the results are consistent with AMD's statements of largest benefits in CPU-bound scenarios and not wholly inconsistent with TR's testing that tried to measure a CPU-bound scenario.
Exactly, In fact if Mantle allowed AMD to pull ahead of NVIDIA's D3D (after the patch).. they would have chosen a different card than a regular 780, like a Titan or even a 780Ti. clearly they didn't. In fact the regular 780's performance is not that far behind. and we can easily extrapolate 780Ti's performance from that.
 
Techreport's choice of game settings reads like a guide on what to use to minimize Mantle's benefit.
TR's settings are what the game defaults to on most configs... hardly unreasonable. As noted, the best "benefit" is at low settings when maximally CPU bound, but how interesting is the fact that you can get 5ms/frame there instead of 10 (or whatever)? If you want a best case Mantle benefit benchmark, that's basically what Star Swarm is. Frankly if they wanted to minimize Mantle's benefit they wouldn't have benchmarked on AMD CPUs at all ;)

Besides I'm not sure what you're complaining about... TRs results are completely in line with everyone else. i.e. 10% ish when GPU bound.
 
I assume he meant the lack of multi player results, considering bf4 is predominantly a multiplayer game and that is where the main benefit to mantle can be found.

Sent from my phone so please excuse my post if it makes no sense.
 
Is there more in-depth information on the technical aspects of the Mantle driver and the hardware it utilizes that says this?

Mantle may be relying on some of GCN's changed memory addressing and software control to implement its low-level control.
The CPU-side optimizations work by putting more onus on the developer, but there may be a system component to it.
GCN is much more flexible with setup and control, while the VLIW designs are more restricted and may not have designs validated for the lower level interaction offered to programs under Mantle.

Some of the shortcuts Mantle relies on don't match up well with AMD's older GPUs. I'd argue that there might be more commonality with low-level access to graphics features, since there are elements that haven't evolved as much with regards to texture and ROP features compared to the execution and software control portions of the design. That wouldn't help with the CPU-side stuff assumed to be more portable.


the reason why older AMD cards are not supported by Mantle has nothing to do with GCN even though people keep repeating this over and over across tech forums. AMD made it clear a few months ago that Mantle is NOT tied to GCN and support for older hardware and even nVidia hardware can be added in the future.

proof: http://mygaming.co.za/news/wp-content/uploads/2013/11/22-Mantle-on-other-architectures.jpg
 
the reason why older AMD cards are not supported by Mantle has nothing to do with GCN even though people keep repeating this over and over across tech forums. AMD made it clear a few months ago that Mantle is NOT tied to GCN and support for older hardware and even nVidia hardware can be added in the future.

proof: http://mygaming.co.za/news/wp-content/uploads/2013/11/22-Mantle-on-other-architectures.jpg

Mind you, saying it's not tied into Mantle doesn't automagically make it viable on VLIW4.
 
excuses, excuses :rolleyes: there can always be written an API which works close to the metal with VLIW4 and VLIW5... this discussion is nonsensical...

I see you haven't learned anything in the past few years either, besides continued non-informed trolling..

AMD has had CTM, aptly called AMD Close To Metal for years already.
 
Back
Top