## Die Suche ergab 6 Treffer

14. Jun 2009 20:24
Forum: Archiv
Thema: Theorie 3 Aufgabe 1 / B-Splines im Skript
Antworten: 2
Zugriffe: 305

### Theorie 3 Aufgabe 1 / B-Splines im Skript

Hallo, also bevor ich hier stundenlang losrechne: Soll ich bei der ersten Aufgabe auf dem dritten Theorieblatt wirklich alle Basissplines ausrechnen? Wenn ich das richtig verstanden habe ändern sich 6 der Basissplines beim ersten Einfügen von xi=3? Weiterhin ist die Definition der Basissplines im Sk...
11. Mai 2009 18:37
Forum: Archiv
Thema: Exercise 1
Antworten: 10
Zugriffe: 1735

### Re: Exercise 1

OK. Anyhow, i wanted to compare just the kernels... I managed to overlook the cudaThreadSynchronize while searching the Dev Guide for Sync, i'll try that. And then, finally, get some times with copying/setup :-) EDIT: Ok, now everything makes sense. GPU is reasonable fast and my implementation is no...
11. Mai 2009 18:07
Forum: Archiv
Thema: Exercise 1
Antworten: 10
Zugriffe: 1735

### Re: Exercise 1

Well, my intension was to compare only the kernels of my implementation and cublas. As far as i know, a syncthreads cannot be called from the host. I guess i'll just measure the time with copying...
10. Mai 2009 23:48
Forum: Archiv
Thema: Exercise 1
Antworten: 10
Zugriffe: 1735

### Re: Exercise 1

I forgot to mention, after the computation by the graphics card i always compare the result with the cpu computation, and (i'm going to check that routines again) the differences were moderate (maximal difference in one entry 1e-4), so either my comparison function is incorrect or the results are re...
7. Mai 2009 21:04
Forum: Archiv
Thema: Exercise 1
Antworten: 10
Zugriffe: 1735

### Re: Exercise 1

I have a question about time measerument: i tested my implementation and cublas with the cutTimer. No matter how big i choose the matrix dimension, say, 3200x3200 i get (kernel) times like 0.07ms. I sure believe CUDA can be fast, but that seems improbable. I use the timer like this: cutStartTimer(ti...
22. Apr 2009 20:54
Forum: Programming Massively Parallel Processors
Thema: Exercise 0
Antworten: 2
Zugriffe: 616

### Re: Exercise 0

Are we allowed to assume the matrix dimensions as mulitples of BLOCK_SIZE (thus not really arbitrary)?