DirectX 11 vs 12 (AMD test)

[phant]

 

 

SOURCE: http://www.hwupgrade.it/news/skvideo/directx-11-vs-12-per-amd-gli-incrementi-sono-piu-che-tangibili_56630.html

 

 

Bye

Phant

[HiJack]

Good times to come :thumbup:

[BitMaster]

Good times to come :thumbup:

 

DCS is just about to hop to DX11, we won't see DX12 in DCS anytime soon I bet.

 

What a pitty, things could be so smooth with all cores involved and a more clever "underneath" from the DX part.

 

Bit

[HiJack]

DCS is just about to hop to DX11, we won't see DX12 in DCS anytime soon I bet.

 

What a pitty, things could be so smooth with all cores involved and a more clever "underneath" from the DX part.

 

Bit

Why don't you create the same test report for DX9 to DX11? I'm sure there are some advantages there as well.

[BitMaster]

Why don't you create the same test report for DX9 to DX11? I'm sure there are some advantages there as well.

 

True, some Dx9-2-DX11 comparison for AMD and Nvidia.

 

 

Just read that AMD Mantle is a dead horse.... some revolutions ahead that turn things over.

[BitMaster]

Baseline from Guru3d report:

 

Nvidia is miles ahead in DX11 driver optimization ( multi thread vs. single thread capability only with AMD drivers YET )

 

I am tempted to say, the love of DCS for Nvidia may very well stay the way it is, but it's AMD'S fault because of a lesser driver.

 

Bit

[Brisse]

I hate these threads. They are everywhere and they are nothing more than advertisements, even though it's unlikely that the OP is affiliated with any of the companies involved.

 

Just read that AMD Mantle is a dead horse.... some revolutions ahead that turn things over.

 

Nope! Mantle is not dead. AMD has given Khronos Group access to the code and Mantle will become part of the "next generation OpenGL" aka. Vulcan. AMD promised that Mantle would go open source. This is them fulfilling that promise.

[emg]

Why don't you create the same test report for DX9 to DX11? I'm sure there are some advantages there as well.

 

These are powerpoint slides from AMD presentations.

Edited March 29, 2015 by emg

[HiJack]

Yes but my point was that we will most lightly see performance gain going from DX9 to DX11 also and keeping in mind that 100% of Ed's users are on DX11 supported hardware that is the best way to go. So performance charts like this but for DCS would be interesting to see. :thumbup: That is also why I commented that there is good times ahead with DX12 bringing another step in the direction of better graphics and at the same time giving a performance gain but ranting out at ED for not going to DX12 right away I feel is wrong.

 

[Tone71]

That chart actually shows a loss in average performance (in Lost Planet 2) with DX11 over DX9; I assume, however, that this is because of additional DX11 features in that particular game, e.g. tesselation.

 

Edit: And do we all really use DX11 supported hardware...?

[GeorgeLKMT]

Edit: And do we all really use DX11 supported hardware...?

 

DX11 is 5 years old. Chances are, if your GPU can't support DX11, it doesn't matter because it's probably old crap anyway. ;)

[Tone71]

DX11 is 5 years old. Chances are, if your GPU can't support DX11, it doesn't matter because it's probably old crap anyway. ;)

 

5 years already!? Wow, that went quick. In that case, good point, well made :)

 

Edit: From June last year http://forums.eagle.ru/showthread.php?t=126222&highlight=dx11. Not quite 100% but 96.77%.

Edited March 29, 2015 by Tone71

[SkateZilla]

The sad part is, AMD's DX11 Performance in their drivers is so bad in some titles, that the 1000+% Increases are from like 9 FPS to 60 FPS..

[7rooper]

I see why the thousands of increase in performance for AMD from DX11 to DX12.

We'll wait until nvidia release its numbers

[emg]

Sites like guru3d.com and hexus.net will do independent (as much as we can hope) benchmarking when able to.

[HiJack]

Anyway, I think we should wait for the release version of DCS World 2 before drawing any conclusions whether we need the power of DX12 to run this sim. :thumbup:

[phant]

DirectX 12 Asynchronous Shaders

 

 

Bye

Phant

[BitMaster]

......and keeping in mind that 100% of Ed's users are on DX11 supported hardware that is the best way to go. ....

 

 

 

Actually not, afaik my 980 is a DX12 card, so is any 970 and 9xx card coming to the market.

 

 

Yes, I read that Mantle goes into openGL/GC after I made that post.

[Brisse]

Actually not, afaik my 980 is a DX12 card, so is any 970 and 9xx card coming to the market.

 

He's not wrong though, because the 9xx series is backwards compatible so technically they are both dx11 and dx12 cards, and thus included in the "100%" that HiJack mentioned.

 

dx12 is a bit different from previous iterations though. It will most likely work with older cards that wasn't specifically designed as dx12 cards from the start. That means there will be advantages by using dx12 even if you have hardware that only officially supports dx11.

[SkateZilla]

AMD already announced GCN cards are DX12 compatible minus some small features.

 

That being said, AMDs new CPUs will be 16 cores.

[BitMaster]

AMD already announced GCN cards are DX12 compatible minus some small features.

 

That being said, AMDs new CPUs will be 16 cores.

 

 

so I can spawn 8 iterations of DCS on a single server with enough RAM :)

 

16:2=8

 

Time flies and DCS's core seems to be glued to the spot with 1 core ( I disregard the sound core ).

 

 

For gaming I rather had a 4 core CPU at 7GHz than a 16 core CPU at 4-5GHz, honestly.

 

Bit

[sobek]

Who knows what comes after EDGE. If they can ever manage to make the physics engine run on a separate thread, that alone would be huge.

[SkateZilla]

so I can spawn 8 iterations of DCS on a single server with enough RAM :)

 

16:2=8

 

Time flies and DCS's core seems to be glued to the spot with 1 core ( I disregard the sound core ).

 

 

For gaming I rather had a 4 core CPU at 7GHz than a 16 core CPU at 4-5GHz, honestly.

 

Bit

 

Doubt it, as you dont have a gpu for each instance :-)

 

But AMD claims DX12 + 16 cores = Amazing GFX object draws.

 

 

As for AMD Zen x86, the top end is rumored to be 16 x86 Cores, 4 cores per module, 4 modules, Dedicated L2 cache per core, Dedicated L3 cache per module, ReDesigned Front end abd FPU.

 

Compared to BD, PD, SR, and EX that has shared L2 cache per module abd Slow L3.cache, a under powered front end and under performing FPU that couldnt feed all 8cores fluidly.

[Sgt_Baker]

Regarding the last few, very interesting, posts: I often contemplate the relative lack of parallelism in games. Coming professionally from the exact opposite end of the spectrum (oodles of data, faster = bigger profit, plus a huge dose of R&D), one develops something of a "1000-core stare".

 

During the development of UltraMFCD, solely for reasons of curiosity and having a nice source of 2GB/s of raw data, I developed a pixel shader which would encrypt the raw video streams coming out of DCS with a pseudo-OTP system. Even with both encryption and decryption of this absurd code-toy running in parallel with DCS, the sim's FPS only dropped by 20% when running on 2048 gfx cores. A very useful bit of research which will be used elsewhere.

 

Now, this brings me to my principal question: Why has the games industry been so slow to adopt a parallel-coding paradigm? For an industry whose principal metrics include "FPS" ("do everything you're already doing, just faster"), one would have imagined they'd be a the front of the proverbial queue.

 

The mind boggles.

 

Regards,

 

Baker

Edited April 2, 2015 by Sgt_Baker

[phant]

 

Why has the games industry been so slow to adopt a parallel-coding paradigm?

 

Because is very difficult to do properly.

 

Bye

Phant