L3DT Performance

[Hypnotron]

Generating these very large maps is a multi day affair on my old Pentium 4 2.5Ghz with 1.5GB RAM.

I bought this cpu in in Oct 2002... 3.5 years old. Time to upgrade? Anyone make huge mosaic maps in l3dt and know if modern cpu's can significantly cut down on the time to process these maps.

And frankly, i'm not even interested in cutting down the tmes by half... id like to have a machine that could cut the times down to 1/4 of what it currently takes. But I know there arent any 10GHz(or rated 10Ghz) machines out there even after nearly 4 years!

What about multip cpu + multicore? Can l3dt do operations in parallel (and the types of operations it does seem very conduscive to parallelization).

-Mike

[Aaron]

Hi Mike,

L3DT doesn't currently take advantage of multi-core systems, but I will do more on this in future releases. I think I can 'fairly easily' parallelise the perlin, attributes map, normals map, and texture map algorithms. However, the design/inflate, fractal, water-mapping, and shadow-casting algorithms can't be readily decoupled, so they may remain single-threaded ops. There is an additional complication in ensuring safe disk-paging for multi-threaded operations, which will require some extra work in the base classes.

Cheers,
Aaron.

[Kiraya]

Hi Mike,

L3DT doesn't currently take advantage of multi-core systems, but I will do more on this in future releases. I think I can 'fairly easily' parallelise the perlin, attributes map, normals map, and texture map algorithms. However, the design/inflate, fractal, water-mapping, and shadow-casting algorithms can't be readily decoupled, so they may remain single-threaded ops. There is an additional complication in ensuring safe disk-paging for multi-threaded operations, which will require some extra work in the base classes.

Cheers,
Aaron.

take a look at openmp. it can save you a lot of work.

[Aaron]

Hi Kiraya,

Thanks for the suggestion. OpenMP looks interesting.

Cheers,
Aaron.

[trollfiddler]

3.2Ghz dual core machine with 2Gb of memory takes me 6 minutes to go through a 32 x 32 map using all the defaults. I've also got dual 7800GTX cards using SLI, so I don't know if the on-screen rendering of progress is speeded up any by that.

Anyway, it's faster than several days :-)

The majority of time is spent in the water table section, which is frustrating, as most of my maps have very little water.

[Aaron]

Hello,

On my AMD XP2600+ (2.1GHz) it takes ~7.5 minutes to make a 32x32 map (2048x2048 HF) with all the defaults. This is only 25% longer than trollfiddler's, so it looks like dual-core doesn't help much.

However, I think Hypnotron is talking about really big maps, like 64k x 64k or greater. In these cases, several-day-long calculations are de rigueur.

I've also got dual 7800GTX cards using SLI, so I don't know if the on-screen rendering of progress is speeded up any by that.

Very nice, but that shouldn't make much difference in L3DT.

Cheers,
Aaron.