Hi all,
I am submitting 500 WUs on acemdlong, 53900 credits each.
The group is 3EKO_8LIG. It is the same protein target of my previous runs, a molecular chaperon (HSP90) involved in many cellular functions. I put an high concentration of the small ligand in order to improve the sampling.

Thanks to all for your computing time,

cheers,

Paola
~

I haven't received any 3EKO_8LIG from the long tasks' queue.
However, I've received a few from the short tasks' queue.
This one took 11 hours and 47 minutes to finish on my GTX 590, so I thought this is a long one, but I've received 14,250 credits for it, just like for the other (short) ones. (normally they take about 4 hours with the cuda3.1 application, and about 3 hours with the cuda4.2 application to finish)
My GPU wasn't downclocked, I've double checked it.

I am currently receiving those in the CUDA 4.2 short queue as well.

EDIT: Wow, um, that's a big disparity. My 680 is looking to complete one of those WUs in 2 hours.

Short queue here also on ACEMD2: GPU molecular dynamics v6.16 (cuda42)

GTX 670 w/ 301.42, Core@ 1249, Memory@ 3206
Win7 x64, BOINC 7.0.25, i7-920 @4.317, HT ON, 1 thread free

3EKO_35_7-PAOLA_3EKO_8LIGs-2-100-RND6955_0...7,454.10
3EKO_45_4-PAOLA_3EKO_8LIGs-2-100-RND1741_2...7,397.16
3EKO_25_2-PAOLA_3EKO_8LIGs-2-100-RND5744_1...7,533.46
3EKO_43_6-PAOLA_3EKO_8LIGs-1-100-RND0691_1...7,340.14

---

Thanks - Steve

Short queue cuda42
GTX560Ti CC2.1 890MHz, XP 64bit, no core free, swan_sync not set, process priority tammed to above normal
3 tasks ...3EKO_8LIGs... run times 11.750 - 12.900 s, GPU load 80 - 85%

Wow, um, that's a big disparity. My 680 is looking to complete one of those WUs in 2 hours.

What do you mean by big disparity?
Your GTX 680 is theoretically 66% faster than one GPU of my 590, so my runtimes should be around 3 hours 20 minutes.
My GTX 580 @ 850MHz completes one of these WUs around 2 hours and 20 minutes.

What am I missing here? 12 hours is a long time. Or am i not understanding something?

Paola - thanks for the new work and I would also like to say that it is nice to get feedback on how your research is progressing. From the crunchers side of the fence we sometimes wonder if what we are doing actually makes a difference and you letting us know that this new set represents refinements to a previous model is heartening. It tells us that you have already analyzed the results we helped generate and they are not just sitting in a database taking up space. Thaks again and keep up the good work!!!

Crunchers - just a couple observations ...

snip... This one took 11 hours and 47 minutes to finish on my GTX 590 ...snip

Wow, um, that's a big disparity. My 680 is looking to complete one of those WUs in 2 hours.

Let's drop the outlier 11 hr 47 min run out of the card comparison part of this discussion as clearly it is not representative of the normal runtimes for the 3EKO_8LIG 3EKO WUs.

...snip ... (normally they take about 4 hours with the cuda3.1 application, and about 3 hours with the cuda4.2 application to finish)

What do you mean by big disparity? Your GTX 680 is theoretically 66% faster than one GPU of my 590, so my runtimes should be around 3 hours 20 minutes.
My GTX 580 @ 850MHz completes one of these WUs around 2 hours and 20 minutes.

RV - You posted that a 590 takes 4 hours and then say it should only take 3 hours 20 minutes but then say if running cuda v4.2 it only takes 3 hours. No wonder there is some confusion :-)

Theoretical numbers aside, a 33% or 50% reduction on runtime is a big overall win for GPUGrid. I'm sure the difference would be greater if 5pot's numbers came from WinXP instead of Win7.

While the 580 is much closer there is still a substantial reduction in runtime.

---

Thanks - Steve

What am I missing here? 12 hours is a long time. Or am i not understanding something?

Oh sorry, I've misunderstood you :)
I thought that you mean disparity between the performance of GTX 590 and the GTX 680....