[B2.0] Ugly shadows when computed in Ultra quality

[niarfman]

The game is more optimized for Nvidia, the studio works with Nvidia cards essentially.

Hopefully it works fine with Radeon too

No framerate issue ingame

[niarfman]

I have done the following test : My configuration (i7 2600k - 8GB - Radeon 7950) I removed the Radeon and put a very weak GeForce 8400GS (Cuda 1.1 compatible).

After more than 10 minutes and less than 25% accomplished, I stopped the computing.

So I can conclude :

• GPU is used either if ATI or Nvidia. The more your GPU is fastest will be the computing
• The computing is optimized for Nvidia. With a same level card, you will gain time
• Prefer a strong ATI card than a weak GeForce

Ps : Yes I know these tests are useless ... but I like to understand completely ...

[om23]

yo,
tell me CPU% and GPU% usage while rendering the shadows.


ps.: my thought about the speed:

mania studio use nvidia, ok, but it doesn´t mean maniaplanet using cuda!
gtx is openCL capable.
all doesn´t matter, this is not the point.
MP is made for dual core CPU.
one core is rendering something and the other core is sending data to the gpu.
I see here a problem, manaiplanet is not optimized for multicore CPU and multi GPU.
even single GPU is not optimized, bcs the gpu is not full loaded.

I think there is no way to get this optimizations in 2013, maybe 2014, but probably 2020 with ManiaPlanet² and TM³

edit: check this for openCL comparsion
http://clbenchmark.com/result.jsp?dev-config=single
http://clbenchmark.com/result.jsp?dev-config=homogen <- multi gpu

[niarfman]

Hello

While computing shadows, the CPU utilization is around 20% of the 8 cores and the GPU is around 60% in my case (i7 + Radeon 7950) :


The CPU detailed usage :

[TMarc]

I think there is no way to get this optimizations in 2013, maybe 2014, but probably 2020 with ManiaPlanet² and TM³

We already have ManiaPlanet 2.0: ManiaPlanet 1.0 was Canyon only

[niarfman]

And this kind of optimization is not so simple. You must separate threads manually if you want to multi-thread them. It's not always possible. And dividing a thread to split it and reconstruct the thread at the end is ...

IMO it's not the most important think to optimized for now. Maybe later

[om23]

niarfman, is this on my canyon map or another one?
about the hardness of doing optimization, I think it depends on your skills how hard it is
my opinion: other programs with openCL support can use multi gpu

TMarc, yeah my fault. do u think with valley and stadium there will be maniaplanet 3.0 released?

[niarfman]

22% CPU and 52% GPU with your map

The hardness depend on skills and also on libraries optimizations. If OpenCL includes them, it fine. But rewriting the code to fit OpenCL needs some work and at this time this is not the main goal

[om23]

22% CPU and 52% GPU with your map

hmmm, this looks better as on my system with 45%cpu and 25%gpu....

nevermind

[niarfman]

Ok, the GPU is more used in my case but the benefit is less than yours.

Maybe Cuda use less the GPU but more efficiently

Afterall the Core 2 Quad is not an i7 but you are able to compute as fast as my configuration. The proof your graphic card is better for kind of computing ^^

Please, also use HWINFO in order to compare efficiently.