The AMD Mantle Thread

[jj109]

http://www.youtube.com/watch?v=Ms16uGxQzSY

Full AMD stream directly from the source instead of the stuttering mess. Starts at 31:00

That's clearly running at 1920x1080 with TWO GPUs. I don't know who DamnedLife thinks he's fooling when the resolution is printed right on the video.

 

[DarkKnightDude]

I can't make out what the FPS is. So we're getting a separate program reader for mantle (graphs, fps, and stuff in game)?

 

[3DVagabond]

That's clearly running at 1920x1080 with TWO GPUs. I don't know who DamnedLife thinks he's fooling when the resolution is printed right on the video.

Where did you get that shot? When I watch it on YouTube the top of the screen is cut off and you can't see that info.

 

[jj109]

Where did you get that shot? When I watch it on YouTube the top of the screen is cut off and you can't see that info.

You have to full screen it and then do a screen cap once the header goes away.

Also, here it is with 2x MSAA, rest Ultra with 100% scaling on an overclocked 780 at 1080p. I was never CPU bottlenecked.

 

[bystander36]

Where did you get that shot? When I watch it on YouTube the top of the screen is cut off and you can't see that info.

It was in the video I watched. I saw the green meter pretty clearly, but the detailed info is hard to read.

 

[jj109]

Default Ultra with HBAO and 4x MSAA. Even more GPU bottlenecked.

Thank god there's a checkpoint right there. It took 15 minutes to get to that segment.

Let me go downclock and disable some cores and do another one 🙂.

 

[Gloomy]

That cannot be a 290X. It's impossible. I get 80 - 140 FPS in that section at 1080p.

 

[BallaTheFeared]

What are you getting at 1150\1500 in that exact spot?

 

[jj109]

Ok, hyper-threading disabled and 4 cores locked to 3.0 GHz (no turbo). I finally get CPU-bound when I turn OFF MSAA. I also changed the FOV to default, since that's a major CPU impact.

Should I down clock some more?

Edit: The yellow line is the CPU time, so if it's higher than the green, the game is CPU bound.

The mantle demo's perf graph is very strange. When it says avg 14 ms GPU time, their FPS counter shows 100+, which isn't possible. Maybe since they're using 2 GPUs, it's 7ms GPU time instead. That would make their demo CPU bound still.

 

[Gloomy]

What are you getting at 1150\1500 in that exact spot?

70-85 if I just spawn and move where he is, a bit higher if I kill a few enemies. He's right at the beginning of the checkpoint so there's a huge firefight, it's pretty variable.

If I clear out all the enemies like in the mantle demo, frame rates are anyone's guess... well over 100 most of the time. The demo is a joke, lol.

 

[VulgarDisplay]

It would be nice to know the exact settings they were using to make a proper judgment.

 

[jj109]

70-85 if I just spawn and move where he is, a bit higher if I kill a few enemies. He's right at the beginning of the checkpoint so there's a huge firefight, it's pretty variable.

If I clear out all the enemies like in the mantle demo, frame rates are anyone's guess... well over 100 most of the time. The demo is a joke, lol.

Well, at least we know whatever CPU they were running is fast as a i5 at 3.0 GHz for BF4, assuming they didn't fudge the settings. Now to restore my overclock profile...

 

[3DVagabond]

You have to full screen it and then do a screen cap once the header goes away.

Also, here it is with 2x MSAA, rest Ultra with 100% scaling on an overclocked 780 at 1080p. I was never CPU bottlenecked.

If you look at the demo the yellow line (cpu) is well below the green (gpu). In yours the yellow tends to run higher than the green. I would assume that means that your cpu usage is higher? Not sure if I'm interpreting that correctly.

 

[DarkKnightDude]

The mantle demo's perf graph is very strange. When it says avg 14 ms GPU time, their FPS counter shows 100+, which isn't possible. Maybe since they're using 2 GPUs, it's 7ms GPU time instead. That would make their demo CPU bound still.

Didn't they say they were CPU bound with mantle even with a FX CPU clocked at 2 GHz? Impressive if you ask me.

 

[jj109]

If you look at the demo the yellow line (cpu) is well below the green (gpu). In yours the yellow tends to run higher than the green. I would assume that means that your cpu usage is higher? Not sure if I'm interpreting that correctly.

No, the GPU time graph on their demo is screwed up. They're getting 100+ fps while at 14 ms GPU avg time, which means that they either have 1400ms in one second or it's wrong.

I'm going with "wrong," and that they had 7ms GPU time with ~10 ms CPU time.

CPU time tracks GPU time very closely since it's a waste of resources to draw more frames than the GPU can handle (if GPU time > CPU time)

Didn't they say they were CPU bound with mantle even with a FX CPU clocked at 2 GHz? Impressive if you ask me.

Did they say that for this demo or is that just imagination running wild?

 

[zlatan]

That's clearly running at 1920x1080 with TWO GPUs.

If the configuration has Kaveri APU and a Radeon dGPU than this is normal, because the iGPU is an active reasource for some stuff.

 

[DarkKnightDude]

Did they say that for this demo or is that just imagination running wild?

I was referring to one of their slides for the StarSwarm thing where they claimed they were GPU bound, my mistake. I think this BF4 demo was with a Kaveri APU with an 290X.

 

[3DVagabond]

I was referring to one of their slides for the StarSwarm thing where they claimed they were GPU bound, my mistake. I think this BF4 demo was with a Kaveri APU with an 290X.

Yes, it was the StarSwarm demo that they said they were still GPU bound with the 8350 @ 2GHz.

 

[jj109]

If the configuration has Kaveri APU and a Radeon dGPU than this is normal, because the iGPU is an active reasource for some stuff.

Well, judging by the "local resources" text on the upper right, it's either a 290 or 290X.

Edit: Ok, here it is in i3-4340 mode:

About 10-15 ms CPU time. Not bad for a i3. It's hard to get a consistent scene with enemies and not get blown up trying to do the screenshot.

 

[BallaTheFeared]

Its probably a lower end AMD CPU with a 290x, no real increase in GPU performance but for one game you don't feel like an idiot pairing a $700 GPU with an awful CPU.

 

[ShintaiDK]

This is why PR material cant be trusted.

 

[desprado]

We dont know any specs yet but seem like amd is hinting that it will increase CPU performance by mantle.

 

[jj109]

Another one in i3-4340 mode.

Not bad at all. Avg of 10.42ms CPU even with that huge screenshot spike.

Compare to Mantle on Mystery APU + 290X at 9.6ms CPU time.

And with that I'm going back to my quad core.

 

[ph2000]

Wasn't the PR claim that Mantle took 2 man-months to implemnt in BF4? I guess they only have 1 guy working on it now?

wasnt that oxide ?

 

[krumme]

wasnt that oxide ?

Yes. But it was only for the cpu part without any gpu optimization.

There is no way a serious mantle implementation, using all the tech decribed as eg. Johan at apu13, only cost 2 man months for an engine like bf4. He never spoke of 2 man-month either. And thats for up to 45% performance.

Lonbjerg is just low level trolling and having fun. Its rethorical questions.

What remains of fact is, we have oxide showing an impressive engine running, a meager 2 month after they were introduced to mantle. Performance on the application level is up 300% for mantle compared to dx, on what is probably a state of the art RTS engine. The engine uses no lead thread for a starter. Thats with or without mante, and its far beyond any engine until now. Its build to be state of the art at dx.

2 man-month to get 300% performance on top of that - using only the cpu part - is stellar.

What stands is, if this is doable in 2 month - if it was 2 or 10 man months - its the best investment a developer can do, and the most easy decision to take at least on the RTS side. Its a no brainer. Its obvious different for non RTS games. You dont get 300% performance in your engine for 20K usd for FPS. lol.

Infraction issued for callout.
-- stahlhart