The AMD Mantle Thread

[cytg111]

They could split Direct3D in two.

DirectShell, the top layer compatible with most hardware, similar to todays version.

DirectCore would be the low level layer, similar to Mantle. It would allow vendor specific extensions and/or extensions injection by the driver. DirectShell would be build on top of DirectCore.

Vendors would be able to provide special versions of DirectShell optimised for a specific hardware type.

Yes, this would be a nightmare to support, but hey ...

Once in a while you have to clean up, if mantle can provide the claimed improvements, then I'd argue its time to take out the trash. Sure, for directx it would require some extra support, but give it a few years and you could problary shut down the old codepath. How else would you go about innovating your api to leverage new hardware ?

 

[cytg111]

To be honest I would quite like Microsoft to make Mantle an official "DirectShell"-style API, with support for multiple GPU vendors. Mantle already uses Microsoft's HLSL to write shaders- why not make it an official part of DirectX 12, and get NVidia and Intel on board too?

+1

 

[krumme]

There is a lot of misunderstanding of Mantle only beeing for cpu perf. Here is what Johan says there is of GPU optimizations:

Can optimize with pipeline state knowledge
Can optimize across all shader stages

Driver stages:
Gives driver a lot more knowledge and flexibility
App can avoid expensive/redundant transistions such as eg. surface decompression

Exposing existing GPU functionality eg.:
Quad and Rect-lists
HW specific MSAA & and dept data access
Programmable sample patterns

Asynch dma transfers
Asynch compute together with graphics
Multiple compute kernels colaborating
Compute as frontend for gfx pipeline

And by getting better cpu utilization also helps gpu by:
Less brute force rendering
Improve culling

https://www.youtube.com/watch?v=N_6CAneoW-0&feature=youtube_gdata_player
GPU part Start at 18:00

 

[sontin]

As for "it only improves AMD processors"- sorry, but this is complete and utter nonsense. The largest improvement in framerates, a 58% improvement, came from the combination of an Intel 3970X and Crossfire 290X. Go get some new talking points.

In 1080p in a game which runs fine on a single card and even archives in the same situation >100FPS using a GTX770 SLI and Intel CPU system.

It's 10% with a 290X in 1080p without AA. Let see 4K numbers. I mean with all the hype we should see the 58% even there.

 

[96Firebird]

Driver delayed? I was looking forward to review site results today, guess I'll have to wait and keep reading the same talking points over and over here...

Now let me ask this question. Do you think the review site results will be similar to AMD/DICE's results? Better? Worse?

 

[TerryMathews]

Posted by HFR reviewer....and a HFR forum member..

Driver delayed? I was looking forward to review site results today, guess I'll have to wait and keep reading the same talking points over and over here...

Now let me ask this question. Do you think the review site results will be similar to AMD/DICE's results? Better? Worse?

See above. Mid afternoon

 

[Shivansps]

The Freesync thread is that way ---->

As for "it only improves AMD processors"- sorry, but this is complete and utter nonsense. The largest improvement in framerates, a 58% improvement, came from the combination of an Intel 3970X and Crossfire 290X. Go get some new talking points.

wake me up when a dev cares about 290x crossfire performance.

 

[Noctifer616]

Now let me ask this question. Do you think the review site results will be similar to AMD/DICE's results? Better? Worse?

I am really interested in extreme scenarios like running 4 290X, and what would actually be better, an AMD 8 core or and Intel 4 core (8 threads). Assuming there would a CPU bottleneck to begin with.

 

[cytg111]

wake me up when ...

If only it was that simple.

 

[wand3r3r]

Mantle's main drawback is that it is a vendor specific windows 3d api. This is a purely rational claim and I don't think that it should be considered trolling.

I agree that's a drawback, look at a few pages of the thread though, it's not rational debate that I'm referring to.

 

[tonyfreak215]

The thing is that many of the people slating Mantle are doing so purely because it is an AMD initiative and they want it to be a failure.

That's the funny thing. DICE went to each of the GPU manufactures to create Mantle. AMD was the only one who was up for it.

 

[Grooveriding]

I am really interested in extreme scenarios like running 4 290X, and what would actually be better, an AMD 8 core or and Intel 4 core (8 threads). Assuming there would a CPU bottleneck to begin with.

An Intel chip will be better because they are simply faster in most games with their superior IPC. The reason the FX8350 does better than any i5 Intel chip is because BF4 actually takes advantages of multiple cores/threads, which most games do not.

Take a look at the valley benchmark thread on this forum. An overclocked 4960X bottlenecks 3 way Titans at 1080p. It would do the same for 4 way 290X at 1080p as well as even at 1600p I would guess. If you were chasing 120FPS at 1440p with such a setup, then I could see some benefit from Mantle for such an extreme rig. Otherwise I doubt it.

The benchmark from DICE is the one that interests me most. They show a 25% performance gain on an FX8350 & 7970 system in Battlefield 4 with Mantle, http://cdn.overclock.net/1/1e/900x900px-LL-1e8cb8de_wbbXob.jpeg. The FX8350 = i5 4670K in the game....

...so there could be some real tangible gains there for a lot of setups. I have a 4670K + 7950 system which if that pans out should see a big boost in performance.

As far as high end multi-GPU systems, I think the big Mantle gains will be had if you are gaming at 120hz. If you're on a high resolution 60hz monitor with multiple 290/290x and a fast CPU, I doubt it will make a difference.

A game I'd really like to see Mantle used in is Starcraft 2. Even with huge GPU power that game will give you low FPS with a lot going on at once. It is poorly threaded like all Blizzard games and gobbles up CPU. The same is true for World of Warcraft in a high activity circumstance, another game it would be interesting to see Mantle used in.

 

[caswow]

An Intel chip will be better because they are simply faster in most games with their superior IPC. The reason the FX8350 does better than any i5 Intel chip is because BF4 actually takes advantages of multiple cores/threads, which most games do not.

Take a look at the valley benchmark thread on this forum. An overclocked 4960X bottlenecks 3 way Titans at 1080p. It would do the same for 4 way 290X at 1080p as well as even at 1600p I would guess. If you were chasing 120FPS at 1440p with such a setup, then I could see some benefit from Mantle for such an extreme rig. Otherwise I doubt it.

The benchmark from DICE is the one that interests me most. They show a 25% performance gain on an FX8350 & 7970 system in Battlefield 4 with Mantle. The FX8350 = i5 4670K in the game, so there could be some real tangible gains there for a lot of setups. I have a 4670K + 7950 system which if that pans out should see a big boost in performance.

As far as high end multi-GPU systems, I think the big Mantle gains will be had if you are gaming at 120hz. If you're on a high resolution 60hz monitor with multiple 290/290x and a fast CPU, I doubt it will make a difference.

so you think downsamling or ssaa etc. cant be considered for people with 60hz monitor to absorb the overhead.

 

[OCGuy]

I think the messages we are hearing are a little all over the place.


Buy AMD APUs due to their very strong GPU and Great Value - but then show APU + 290 as a Mantle selling point


4k is the wave of the future, and 290x is the 4k king currently - but then use 290x Xfire bottlenecked @ 1080p for the big Mantle FPS increase

Devs want Mantle because it makes porting from consoles quick and easy - Months of man hours put in to optimize BF4 for Mantle, not to port it



What I believe the reviewers will do is give us the scenarios where I think Mantle will shine. (AMD did give us one of these, with the 8350 + 7970).

People have the perception that early i5/7 and obsolete AMD CPUs are all they need, when in fact they are showing their age. It has been easy to just upgrade your GPU and not buy a whole new mobo/CPU for quite some time.

I can see Mantle extending the life on systems that will email and surf the web for another decade +, and are only really slowing down in demanding games with multi-GPU or high-end single GPU combos.

I would highlight the 7970 case, and leave the odd pairings alone.

 

[NTMBK]

wake me up when a dev cares about 290x crossfire performance.

For the love of god. Someone pipes up with "it only helps low end rigs!" and someone else points out that it performs well on an insane Crossfire rig. Someone else comes out with "but nobody uses a rig like that!" so someone else points out that it also helps low end rigs. And then we get back to "it only helps low end rigs!"...

 

[sandorski]

...
Devs want Mantle because it makes porting from consoles quick and easy - Months of man hours put in to optimize BF4 for Mantle, not to port it....

Dice essentially did Port it. Not from Console to PC, but from DX to Mantle.

 

[Grooveriding]

so you think downsamling or ssaa etc. cant be considered for people with 60hz monitor to absorb the overhead.

No. You're creating a highly GPU limited situation by forcing to cards to render huge resolutions using SSAA or downsampling.

Whereas if you're running native 1080p or 1440p with a tri/quad GPU setup trying to hit 120FPS you're going to run into CPU bottlenecks trying to get the cards to open up.

 

[96Firebird]

See above. Mid afternoon

Which means no review sites will have reviews ready by today. Which is what I said...

Edit - It looks like Warsam71 said that review sites are using the BETA drivers, but that contradicts what I've read elsewhere. Either way, I am eagerly awaiting impartial results.

 

[OCGuy]

Dice essentially did Port it. Not from Console to PC, but from DX to Mantle.

But Xbox One's API is very similar to DX.....so won't that always be the case?

 

[ASM-coder]

For the love of god.

It's come down to unsubstantiated opinion(redundancy intentional,)
with attempts at verification using a very small set of benchmarks.
And like statistics, apparently these benchmarks can be used to prove anything.

I am so ready for the AMD drivers.

 

[sandorski]

But Xbox One's API is very similar to DX.....so won't that always be the case?

I dunno, maybe.

 

[TerryMathews]

Which means no review sites will have reviews ready by today. Which is what I said...

Edit - It looks like Warsam71 said that review sites are using the BETA drivers, but that contradicts what I've read elsewhere. Either way, I am eagerly awaiting impartial results.

There's no reason to think that. Benchmarking doesn't take that long.

 

[psolord]

I am getting 25.60fps on Star Swarm/Attract/Extreme/1080P/D3D on my Primary rig (cpu @ 4Ghz).

Unfortunately the bennchmark crashes if MSI Afterburner is running so I cannot see the gpu load. I did see one of my CPU cores maxed out though, so we have a cpu limited scenario here, possibly meant to be, since the Mantle equivalent should be much faster, when available.

 

[96Firebird]

There's no reason to think that. Benchmarking doesn't take that long.

Lol.

 

[blackened23]

There's no reason to think that. Benchmarking doesn't take that long.

Uh. Context? Benchmarking by who? You? Or a review website?

I think it does take that long. For you it may not take that long, but for a review website interested in giving impartial results - there's a lot of pain associated with that from what i've read. There's a lot of stuff that happens to ensure results are reproducible and applicable, not to mention that the same scenes have to be hands down perfect across multiple runs. Across a long time span of 15+ minutes per tested scene.

Benchmarking done by a website isn't simply a matter of doing a 30 second run and saying screw it, that's it, from what I understand. Especially with the new boost algorithms of GPUs - there's a lot more pain associated with benchmarking than you seem to be aware of. At least if you want to provide credible results. Now there are websites that aren't interested in giving credible results. But for those that do. Benchmarking isn't a 5 minute affair as you think it is. Benchmarking Mantle and comparing it to similar nvidia hardware isn't going to be a 5 minute affair. Even ensuring the game scenes tested are exactly the same across multiple runs and a long time span (due to boost for both AMD/NV) takes a darn long time.