gamegpuGears of War Ultimate Edition Benchmarks

[tential]

It really helps if you post fraps runs next to the CPU charts... or just figure out a way to bench it all. I mean if you're going to go through the work, you might as well be thorough so people don't call foul.

 

[Mahigan]

Soso, Dan Baker has insight? Great, maybe he can explain this:
2 threads - 57FPS:

4 threads - 87 FPS (54% gain)

8 threads - 92 FPS (5% gain and less work on each thread)

Looks like a perfect multi rendered engine.

/edit: In 1080p the frames jump from 57fps with 2 threads to 106fps with 4 threads. That is an ~86% increase. Hm.

2 cores plus HT = 4 threads... :/

 

[tential]

2 cores plus HT = 4 threads... :/

He posted the game using 8 threads... what more do you need? You said it couldn't be done and he's doing it....

 

[Mahigan]

He posted the game using 8 threads... what more do you need? You said it couldn't be done and he's doing it....

Yes and the 8 thread picture shows 4 cores with a lot of work and 4 cores with low utilization.

That's not multi-threaded rendering. Multi-threaded rendering evens the load across all cores. Let me show you.

Spoiler

His image actually proves my point. He just isn't knowledgeable enough to notice it.

His image shows 4 threads with a lot of usage and 4 others with significantly less. In other words, only 4 threads have rendering tasks assigned and the other 4 are swapping with the busy cores. So at any given time you only have 4 busy threads. Hence my comment about 2 cores with HT (or 4 cores w/o ht).

Review sites have noticed the same thing.

 

[USER8000]

I like how some of you conveniently forget that a number of UE3 games from the last few years show scaling upto a reasonable number of threads,ie,around 4. Even the engine used in BF3 could scale upto 4 threads.

That does not mean its a proper DX12 engine.

From PCWorld:

http://core0.staticworld.net/images...s_of_war_19x10_medium_dx12-100647959-orig.png

Sadly, the result looks like a flashback to the benchmark for 3DMark’s API Overhead test in DirectX 11. And no, I wasn’t hitting some odd frame rate cap, as lowering the resolution and image quality pushed performance up to 125fps. It makes you wonder what exactly lets Microsoft label it as a DirectX 12 game. I’ve asked Microsoft for clarification, but I have yet to hear back.

Microsoft’s Gears of War Ultimate Edition is touted as a DX12 game but the built-in benchmark mode makes it look more like a DX11 game.
To be fair to Gears of War, my testing was done solely in the game’s built-in performance benchmark. While multi-core efficiency is one of the feature achievements of DirectX 12, other aspects of the new API would give Gears of War the DirectX 12 check-off. All I know is the performance benchmark doesn’t seem to improve as you increase CPU cores.

Looks like the game really only really uses two cores for the internal benchmark.

 

[Mahigan]

I like how some of you conveniently forget that a number of UE3 games from the last few years show scaling upto a reasonable number of threads,ie,around 4. Even the engine used in BF3 could scale upto 4 threads.

That does not mean its a proper DX12 engine.

From PCWorld:

http://core0.staticworld.net/images...s_of_war_19x10_medium_dx12-100647959-orig.png




Looks like the game really only really uses two cores for the internal benchmark.

Exactly.

PcWorld is correct. The multi-threaded rendering feature is what helps AMD GCN overcome its API overhead bottleneck.

Without this feature, we're looking at a DX11 level title in terms of multi-threading.

I know that this doesn't appeal to certain folks wishes but that is what GoW delivers.

If any title is not representative of DX12, it isn't AotS or Hitman but rather GoW.

 

[tential]

You've convinced me.

 

[PhonakV30]

Soso, Dan Baker has insight? Great, maybe he can explain this:
2 threads - 57FPS:

4 threads - 87 FPS (54% gain)

8 threads - 92 FPS (5% gain and less work on each thread)

Looks like a perfect multi rendered engine.

/edit: In 1080p the frames jump from 57fps with 2 threads to 106fps with 4 threads. That is an ~86% increase. Hm.

Well, I trust this article more than yours :

http://www.pcworld.com/article/3039...es-you-really-need-for-directx-12-gaming.html

I told you GoW can not scale up to 8 Cores.

Edit : here True DX12 :

I couldn't believe myself when i saw Mercennarius's result :

source : http://forums.anandtech.com/showpost.php?p=38088143&postcount=1304

 

[Mahigan]

Dismissing Dan Baker is akin to dismissing John Carmack, Tim Sweeney or Johan Andersson. Dan is a very respected developer.

 

[Mahigan]

I have more to add,

If you simply port a DX9/10/11 engine to DX12 you get only one tweak and that's the general DXRuntime multi-thread optimizations which are the Explicit CPU/GPU synchronizations.

In order to port UE3 to DX12, properly, you would have to rewrite all of the shaders and at least multi-thread your shadow maps thus moving them away from the main thread.

The main thread would retain the main scene cmd lists while the worker threads would execute their shadow map cmd lists in parallel. That would at least save you some CPU time.

Another tweak would be to execute indirect. So you buffer multiple draw calls into an indirect draw call buffer and then execute them all with a single draw call.

Another improvement could be achieved by using DX12's shader cache (dev controlled).

Then of course you have the resource binding where you could bundle up to 1 million resources (ex textures) on NV hardware (or unlimited/limited by memory on AMD hardware aka full heap) thus making it simple when resources are required (ex texture streaming) on a parallel running copy thread on the GPU.

I mean there are a ton of optimizations that could have been used for GoW if it weren't build on an ancient DX9 UE3 engine and that's ignoring full multi-engine support.

So no, GoW UE is not at all a good representation of DX12. It's not even a good representation of DX11. It really is the worst possible example of DX12 imaginable (and I am not using that lightly).

Heck there aren't even Conservative Rasterization or ROVs (two NVIDIA DX12 FL 12_1 rendering features) tweaks used in the game.

End rant.

 

[sontin]

Yes and the 8 thread picture shows 4 cores with a lot of work and 4 cores with low utilization.

That's not multi-threaded rendering. Multi-threaded rendering evens the load across all cores. Let me show you.

[/Spoiler]

His image actually proves my point. He just isn't knowledgeable enough to notice it.

His image shows 4 threads with a lot of usage and 4 others with significantly less. In other words, only 4 threads have rendering tasks assigned and the other 4 are swapping with the busy cores. So at any given time you only have 4 busy threads. Hence my comment about 2 cores with HT (or 4 cores w/o ht).

Review sites have noticed the same thing.

Even on your AMD marketing slide the workload under DX12 is not equal. There is one main worker thread and distribution happens differently on the cores.

I dont have a real 8 Core processor. So the OS should schedule the main work on the real cores first. And it is doing it very good. :thumbsup:

BTW: In 720 with high settings HT provides additional >20% more Frames over 4 Cores alone. But i guess this is just normal for DX11 engines, too. :\

 

[Bacon1]

http://gamegpu.com/action-/-fps-/-tps/gears-of-war-ultimate-edition-test-gpu.html

Scroll down to CPU tests