Hmmm...CUDA not so fast when compared to the optimised AK SETI ap on my Q9450.

This wu.
23/12/2008 8:15:06 PM|SETI@home|Starting task 14no08ae.31662.224103.8.8.164_1 using setiathome_enhanced version 605
23/12/2008 10:08:03 PM|SETI@home|Computation for task 14no08ae.31662.224103.8.8.164_1 finished

Agreed, not as fast as v8 on mac either. Who ever updated the front page with 10x faster, obviously hadn't tried the app out. it's going to take 1 hour 20 minutes for a unit that would take 55-57 minutes with V8 Xeon.

---

I don't think the comparison was supposed to be against the opti app but the stock.

---

Agreed, not as fast as v8 on mac either. Who ever updated the front page with 10x faster, obviously hadn't tried the app out. it's going to take 1 hour 20 minutes for a unit that would take 55-57 minutes with V8 Xeon.

However, if devs would think *a bit* ahead, it would be possible to run AK V8s on CPUs and CUDA on GPU at the same time and that would be worthwile, even if CUDA app is a bit slower than AK V8.
Concerning speedup figures, of course, the 5x to 10x speedup was observed compared to the previous stock app.

[edit]a typo[/edit]

BR,

---

Well it sure as hell isn't 5 or 10 times faster than the stock app either. I ran a stock app unit first, 1400 seconds for 17.41 credits. On the Mac and with V8 Windows that would take 700 to 750 seconds.

So you reckon it will do short units in just over 2 minutes in real time?

Don't make me laugh.

---

The other thing I suspect, considering the power connectors on graphics cards, is that using the graphics card is not very energy efficient.

Will depend on the card.
They still seem to be keeping "The Truth About Graphics Power Requirements V2" pretty up to date with newer cards.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

You've got to laugh.

1 hour 19 minutes for 14.40 credits. So 4740 seconds compared to stock app of 1404 seconds. hmm, I wonder which is the one to use.

This is with a 8800GT. I'll let it finish another unit before I kill Boinc on Windows.

---

Do you have a free core for the CUDA app or are you running the CUDA app with updated priorities? I know my 9600GT it is much faster then my PentiumD, though that should really be expected.

The only problem for me is the CUDA app stalls all the time and I have to close BOINC and reopen it in homes that the CUDA app will start working again after a restart.

---

tfp, it was only the CUDA app running. All eight cores were totally free.

---

I want to say in the beta the CUDA app was doing one WU every 30 minutes or so for 65 point WU, which is much faster then a dual core 805 PD. Anyone with 4 or 8 cores in a box will do better with just running the normal app unless they can get a bunch of AP WU to run while running MB on CUDA or if they are on multiple projects.

To me the stalling out is annoying and I would agree this app was not ready for prime time. There things really didn't need the volume usage to find in my opinion. It will all get hammered out in time.

---

Well, once they're able to let us do both cpu and gpu MB it will be alot better. Post like this have me hopeful it won't be too long before we will be able to do just that, so we don't have to choose between one or the others.

---

I did'nt know a thing about cuda being out but I had this wingman with exeptional short timngs and I went to his results page and saw a lot of those since a few days. He did 20 times faster than before.Do'nt know how much that was in clock time.
So I went to the message boards and I found out.
Unforunately this result is purged already. Have no cuda wingmen now.

As stated before... It FULLY depends on the card as to how fast it is...
I can do that same work unit on a gtx 260 core 216 SC in 10 minutes...
Granted its not as fast as 8 cores but it is up to 12x faster compared to a single core (on specific wu's)

Just wait until you can crunch MB on CPU and GPU then you will be cruising.
Now if they had waited for this feature before releasing cuda...
~Bob

---

Do you Good Search for Seti@Home? http://www.goodsearch.com/?charityid=888957
Or Good Shop? http://www.goodshop.com/?charityid=888957

Well I just kicked th mac pro, trying cuda again. 23 minutes for a 58 credit unit. Not bad. I take back some of my earlier critic.

---

Basically I want to opt out of CUDA on SETI and just run the optimised AK app - but still use CUDA for GPUGRID. How can I do this? Can I still use the app_info.xml as previously used for opt AK app?

Basically I want to opt out of CUDA on SETI and just run the optimised AK app - but still use CUDA for GPUGRID. How can I do this? Can I still use the app_info.xml as previously used for opt AK app?

Yes, provided you supply the app as well.

The option to not use the GPU on Seti has arrived as well, in case you want to set it anyway. It's in the project preferences.

---

Basically I want to opt out of CUDA on SETI and just run the optimised AK app - but still use CUDA for GPUGRID. How can I do this? Can I still use the app_info.xml as previously used for opt AK app?

[edit]Beaten to it - again!!![/edit].

F.

---

The option to not use the GPU on Seti has arrived as well, in case you want to set it anyway. It's in the project preferences.

Thanks for that link Jord.

I too am better off with my Q9450 running MB than my 8600GT. It takes almost twice as long using CUDA as it does the AKV8 apps.

---

Calm Chaos Forum...Join Calm Chaos Now