After a few rendered wus: 24500 sec instead of 31500, CPU Util about 15-16% (q6600 3.4 GHZ)

Impressing! Thanks for that increase (about 23-25%) for the gtx260

AiDec,

I took a look at your results and I see two things: most 6.45 WUs needed 33 - 37 ms/step, a few 32 or 31 and some over 40 ms. With 6.48 most WUs are between 24 and 27 ms/step. How did you arrive at 10%?

And your GPU clock is changing a lot in the recent WUs. Did you take this into account when looking at the results?

MrS

Yes.

Steps:
6.45 without OC (that`s why ca. 33-34 ms)
6.45 OC (28-30 ms)
6.48 OC (24-27 ms)

I think it`s little bit more clear now :).


And additional explanation (based on old 6.45):

If I`m crounching with 3x280 then steps approx(!): 1st-32ms, 2nd-34ms, 3rd-40ms. Also sometimes I`m using just 2xGPU or 1xGPU to crunch when I`m playing games and sometimes I`m setting SLi. Don`t try to look at all WU`s, just some of them are `clear crunched` and I think I know which of them I should use to check new app`s performance :).

---

9600gso stock, 6.3.10 client, q6600, 3 cpu tasks, one core reserved for GPU

6.45: 89ms/iter, 222.5 cpu sec
6.48: 86ms/iter, 61112 cpu sec

analysis: since I dedicated a core to the GPU, the minor GPU improvement only costs me a bit of heat. However, if I were using the last CPU core for something else, as one does in 6.3.17, 5% improvement in GPU performance for 300 times the CPU might not be worth it on low end cards.