Would you like to have GPU accelerated compiling tools?

[killer89]

I'm working with Resident Evil: Zero's Training Facility.
Compile takes (full compile) overall 12 minutes.

Yep, the map is complex as hell.
(managed to reduce the compile time from 20 to 12 minutes.)

Compiling it with VBCT also seems to be couple of minutes faster than in hammer. :v:

[jbeans]

I doubt I need more power than six cores, but it would be cool.

[TerabyteS]

[QUOTE=applemaster;23624508]I doubt I need more power than six cores, but it would be cool.[/QUOTE]

You don't understand. A GPU would be 20 times faster than 6 cores on this task.

[jbeans]

I do understand. And could we run CPU and GPU?

[robmaister12]

[QUOTE=applemaster;23625094]I do understand. And could we run CPU and GPU?[/QUOTE]

Technically you could, but you would need two instruction sets, one for the GPU and one for the CPU. You would pass one calculation to a CPU thread, and another one to an OpenCL/CUDA thread. However, you would actually be slowed down towards the end of the compile when your CPU is crunching one calculation while the GPU has already finished another 20. Then you have to sit and wait for the CPU to finish that last calculation.

[Magman77]

Like OpenCL styled so it compiles on GPU and CPU?

Fuck yes.

[BmB]

[QUOTE=robmaister12;23626931]Technically you could, but you would need two instruction sets, one for the GPU and one for the CPU. You would pass one calculation to a CPU thread, and another one to an OpenCL/CUDA thread. However, you would actually be slowed down towards the end of the compile when your CPU is crunching one calculation while the GPU has already finished another 20. Then you have to sit and wait for the CPU to finish that last calculation.[/QUOTE]

Chances are the time that/those calculation(s) take would be more than made up for by the time saved not having to do them on the GPU so it could finish the rest faster.

[ph:lxyz]

My map is badly optimised (mostly because it is a Spacebuild map with RP city planets)

I'm thinking of editing vvis.exe, then compiling it on my Linux box which has 8 cores and a GTX 285, then somehow make OpenCL work too so that it uses all of them. Then, during the compile phase (on my 1-core Windows PC) I can call a script which SSH's into my Linux box, copies the files over and runs VVIS on all CPUs (or 7 of them) and the GPU, then sends the results back to the Windows box.

Is this a good idea? Or a waste of time?

[Deadchicken]

That sounds fantastically complex, what will it accomplish?

[Wolfie13]

[QUOTE=ph:lxyz;24458806]My map is badly optimised (mostly because it is a Spacebuild map with RP city planets)

I'm thinking of editing vvis.exe, then compiling it on my Linux box which has 8 cores and a GTX 285, then somehow make OpenCL work too so that it uses all of them. Then, during the compile phase (on my 1-core Windows PC) I can call a script which SSH's into my Linux box, copies the files over and runs VVIS on all CPUs (or 7 of them) and the GPU, then sends the results back to the Windows box.

Is this a good idea? Or a waste of time?[/QUOTE]

It's not VVIS that needs this optimisation, it's VRAD.