The Witcher 3 driver

http://www.geforce.com/whats-new/articles/the-witcher-3-wild-hunt-is-your-system-ready

Nvidia will launch their W3 driver shortly before the game goes live on Tuesday.

 

Another GameWorks game. Can't wait!

 

It wasn't that Nvidia "couldn't get a deal", it's that Nvidia wasn't willing to do it for peanuts, and wanted to retain rights. Same thing with Intel. Console companies don't want to work with those two anymore because they won't work cheap, cheap, cheeeeeeap, coupled with some of the stuff Nvidia pulled with Microsoft and the original XBOX.

 

What happened to the original xbox?

 

Long story short.....
Nvidia wanted to raise the pricing of the chips it was supplying to Microsoft for the xbox.
Nvidia stating that it was losing money selling the chips at the original agreed upon pricing deal with Microsoft.

result: Nvidia lost the dispute in court and was forced to continue supplying Microsoft with the chips at the original agreed upon pricing.


Note: Of course that is my "VERY SIMPLIFIED" version, I'm not going back and reading / quoting all the details.
Just do a google search for Nvidia Microsoft xbox pricing dispute if you want to know all the details.

 

Well that is good enough for me, too bad Nvidia didn't go bankrupt.

 

nV Admin in comments claim that W3 is not CPU-intensive game.

http://www.geforce.com/whats-new/ar...-hunt-is-your-system-ready#comment-2026525266

It means that lots of CPU can be consumed by AMD **** drivers for massive overhead.
FFS.

 

The drivers don't have any massive CPU overhead. Maybe they don't profit form multi-threading as much as the NVidia drivers do, but there is no noticeable higher CPU load, if it is not caused by something ****ed up in the Engine, like PhysX for example.

 

NVIDIA drivers in general have twice the draw calls / second with the benchmark test with DX11. If you don't think twice as quick is much of a difference I don't know what is.
MT DX11 is another story.

 

But the drivers do not have CPU overhead, which is they do not cause a higher CPU load. In any case the overall CPU load should be less, if you can put fewer draw calls through.

 

Are you drunk or from AMD?

 

Can you do anything but repeat NVidia fanboy BS?

Draw Calls take CPU load, more Draw Call have more CPU load. The AMD driver cannot utilize multiple cores to increase the Draw Call count, ergo the CPU load on all but one core must be lower.

Edit: Here you can see what I mean (scroll down to the CPU usage): http://www.anandtech.com/show/8962/the-directx-12-performance-preview-amd-nvidia-star-swarm/4

NVidia has a higher load on all cores in DX11, because they can distribute the Draw Calls evenly, while AMD has it distributed only over one and a half core. The overall CPU usage is lower, but the Draw Call count is also lower. The AMD driver cannot put as many Draw Calls through, because they cannot make full use of the CPUs.

 

^^^^
This

My anger isn't that performance stinks for AMD only. It stinks for all vendors.

Nvidia is shooting money at companies (not devs) in order to bribe them but reduce the quality of their titles.

 

It's not only the cheap factor. Nobody had anything close to HSA and a gpu architecture as programmable as GCN. Modern consoles can use the GPU as part of the CPU almost.

What BoMby says is essentially correct, but with wrong words. What I bet he means is the the AMD driver does not tax the CPU (which is correct), but it underperforms in CPU/GPU tasks (which is also correct).

Overhead does not mean "heavy". In this case looks like poor optimization that makes the driver not to be able to "draw" fast for the GPU. Overhead means wasted resources.

 

ummm....

you are aware that there is only one submission thread, for both AMD and nvidia.

All those extra threads are just worker threads that have to return data to the single submission thread.

They dont make the submission thread run any faster.

just the way dx11 works.

This is the biggest advantage of DX12: asynchonous submission.

 

I don't know how exactly NVidia build their Wonder driver, but it probably is something like a simple receiver thread, immediately distributing the serial data it gets to multiple threads, which then do the "heavy lifting".

 

If it was that simple, it would have been duplicated the moment it was done.

 

I guess the real problem is keeping the correct order when sending it to the GPU from the multiple worker threads? Or maybe they have some major problem with their code structure, so they would've to completely rewrite everything?

 

again, they are only send to the GPU from a single submission thread.

That thread compiles and orders the batches from the worker threads.

DX12 allows multiple submission threads (asynchonous submission). This is the reason why the batch submissions are 10x dx11 numbers in DX12/Mantle/Vulkan

 

I made you a picture, to show you how I think it currently works:



This is pure speculation of course.