SlickerVolunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 11 Jun 09
Posts: 2479
Credit: 702,712,176
RAC: 340,488
Optimized applications will be available for this project. They will require manual installation via the anonymous platform. Optimized applications currently in progress/completed are:
Win SSE
Win SSE2
Win x64 ATI/CAL/BROOK and SSE2
Edit: The Linux GPU (ATI and CUDA) applications are the only applications that still require manual installation. All other optimized applications are installed as standard applications once fully tested.
Edit: For those who missed the link on the home page, the power apps can be found on the Optimized Apps Download Page
/ 
Reply
Quote
SlickerVolunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 11 Jun 09
Posts: 2479
Credit: 702,712,176
RAC: 340,488
There is at least one application that people are running which is not an official one. Now, I could give a hoot whether you share it with others, but it pisses me off that those running it didn't first check whether that was acceptable.
Therefore, I am going to implement something that I hope will result in only approved applications being allowed to run.
Until then, the work generator is being shut down.
/ Reply
Quote
There is at least one application that people are running which is not an official one. Now, I could give a hoot whether you share it with others, but it pisses me off that those running it didn't first check whether that was acceptable.
GRMPF - now who is doing this again?
/ Reply
Quote
never mind, i think i know who got creative once again. :((
____________
/ Reply
Quote
Crunch3rVolunteer moderator
Project developer
Project tester
Send message
Joined: 30 Jun 09
Posts: 219
Credit: 4,553,141,487
RAC: 9,064,000
never mind, i think i know who got creative once again. :((
Now that would be interesting to know. If you know more than we do, please share.
____________
Team BOINC United.Join Science that matters.
/ Reply
Quote
never mind, i think i know who got creative once again. :((
Now that would be interesting to know. If you know more than we do, please share.
the hack i'll do - either slicker decides to come up with names or not.
____________
/ Reply
Quote
SlickerVolunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 11 Jun 09
Posts: 2479
Credit: 702,712,176
RAC: 340,488
FYI, there is no optimized app for the 64 bit versions as the 64 bit version is already optimized.
/ Reply
Quote
Hi!
Witch nVidia driver You recomended for 1.10 CUDA 32/64 app?
New driver give support for CUDA 2.3 - 190.38 WHQL
Old driver - 185.85 - CUDA 2.2
Older - 181.22 - CUDA 2.1
Also GT200 have improved Compute Capability over G92 and G80(1.3 vs 1.1 vs 1.0)
Now I have 182.50 (CUDA 2.1) but soon I will instal 190.38 (CUDA 2.3) - are they compatible with current app?
Is 64 bit CUDA version much faster than 32 bit CUDA?
Now WU takes 484s-529s on XPSP2-32bit.
Also I noticed that GPUGRID use much more GPU power then Collatz - GPUGRID makes 96C of GPU while Collatz only 81C :)
Collatz take 0.14CPU - is there way to increase this? Maby when increased it will be faster....
____________
POLISH NATIONAL TEAM - Join! Crunch! Win!
/ Reply
Quote
Does the 1.10 CAL version also need CAL/Brook 1.4 SDK and version 9.2+ of the Catalyst ?
/ Reply
Quote
SlickerVolunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 11 Jun 09
Posts: 2479
Credit: 702,712,176
RAC: 340,488
Does the 1.10 CAL version also need CAL/Brook 1.4 SDK and version 9.2+ of the Catalyst ?
Collatz v1.10 still uses the CAL/Brook 1.3 SDK and therefore requires version 8.12+ of the Catalyst drivers.
/ Reply
Quote
Ok... I tested...
Work's great with 190.38 WHQL :)
____________
POLISH NATIONAL TEAM - Join! Crunch! Win!
/ Reply
Quote
Ok... I tested...
Work's great with 190.38 WHQL :)
Same here, running the 190.38 Drivers on all NVIDIA Cards with no Problems here but had a Ton of them at the Grid Project. Nothing but errors and the Cards Downclocking on their own over there.
I haven't checked yet to see if they still Downclock running the Collatz Wu's yet though but they seem to run the Wu's just fine.
/ Reply
Quote
Am I missing something here, doing something wrong? I signed up today and ran 1 work unit using the default app. It ran for 2 hours and produced 73 credits. I then installed the sse2 optimized app, rebooted, and ran 5 more work units. They each took around 5 hours, and look like they're giving just 80 credits!!!
I was going to try the ati app next - but if that app is as good as the optimized sse2 app then it doesn't seem like a good idea!
Any idea why the optimized app is worse than the deafult here? (I did run cpuz to confirm that sse2 was correct)
/ Reply
Quote
Any idea why the optimized app is worse than the deafult here? (I did run cpuz to confirm that sse2 was correct)
I guess you are talking about your Core i7 with Vista64?
Collatz benefits really a lot from a 64Bit application. So going with the SSE2 compile reverts you back to 32Bit and you end up slower.
Furthermore the SSE2 application may require some handtuning to gain speed. Currently it is just created by setting a compiler switch. But in view of the other improvements on the horizon, that handtuning may have to wait a bit.
/ Reply
Quote
Any idea why the optimized app is worse than the deafult here? (I did run cpuz to confirm that sse2 was correct)
I guess you are talking about your Core i7 with Vista64?
Collatz benefits really a lot from a 64Bit application. So going with the SSE2 compile reverts you back to 32Bit and you end up slower.
Furthermore the SSE2 application may require some handtuning to gain speed. Currently it is just created by setting a compiler switch. But in view of the other improvements on the horizon, that handtuning may have to wait a bit.
That's the one. Thanks for that - I'll go back to the standard app on the i7.
/ Reply
Quote
SlickerVolunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 11 Jun 09
Posts: 2479
Credit: 702,712,176
RAC: 340,488
Gipsel is correct. The 64 bit app is already twice as fast as the fastest 32bit app (and possibly faster). The 32 bit versions either have to emulate 64 bit storage which uses twice as many 32 bit variables to hold the same amount of data. That means twice as many shifts, adds, ands, etc. to do the same amount of work.
/ Reply
Quote
I'm running the collatz_1.10_windows_x86_64__cal optimized app on 2x ATI 4850X2 cards. For me on an I7 920 at stock on Windows Server 2008 it takes 2 hours on the CPU and 2.5 hours using the optimized app on the ATI's.
The messages tab in BOINC shows it's using the optimized app from the app_info. What am I doing wrong :)
/ Reply
Quote
SlickerVolunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 11 Jun 09
Posts: 2479
Credit: 702,712,176
RAC: 340,488
I'm running the collatz_1.10_windows_x86_64__cal optimized app on 2x ATI 4850X2 cards. For me on an I7 920 at stock on Windows Server 2008 it takes 2 hours on the CPU and 2.5 hours using the optimized app on the ATI's.
The messages tab in BOINC shows it's using the optimized app from the app_info. What am I doing wrong :)
Looking at results like
this I'd say that even though it may think it is running, it really isn't. I find it very strange that there is no output and yet it thinks it was successful.
Until you get actual output, it won't matter how long it takes to run.
1. I'd start with shutdown and power off and then reboot. Once CAL code starts erroring out, it may not start working again w/o a reboot.
2. What driver version are you using? I assume you have renamed the 3 ATIxxxxx64.DLL files to AMDxxxxx64.dll (e.g. calrt, caldd and one other I can't think of off the top of my head)
3. Make sure the catalyst control center is NOT running. Rumor has it that Brook/CAL code and CCC don't get along.
4. If you haven't already, change the app_info.xml so that avg_cpus is 1.0 instead of 0.5. That will limit the concurrent WUs to 8 instead of 16. 16 is a waste since only 2 will run at a time. The other 14 will sit idle. BOINC isn't smart enough to know that the ATI app uses the ATI card and there's only 1 of those (or X2 = 2 in your case). The default -n2 command line option runs 2 WUs at a time on the one device. The other 6 sit idle.
/ Reply
Quote
I changed the avg_cpus to 1.0 instead of 0.5 but only 1 Wu is still being run on my i7 with a 4870 & 4850 X2 installed in the Box. I run 9 Milkyway's normally in it so the Cards are being seen by the System ???
Hmmmmm, was running 1 AQUA MT Wu & Suspended it, 5 more Collatz Wu's started up ...
/ Reply
Quote
SlickerVolunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 11 Jun 09
Posts: 2479
Credit: 702,712,176
RAC: 340,488
I changed the avg_cpus to 1.0 instead of 0.5 but only 1 Wu is still being run on my i7 with a 4870 & 4850 X2 installed in the Box. I run 9 Milkyway's normally in it so the Cards are being seen by the System ???
Hmmmmm, was running 1 AQUA MT Wu & Suspended it, 5 more Collatz Wu's started up ...
Using the official boinc clients, ATI = CPU as far as BOINC is concerned. Since Aqua MT = multiple CPUs, I can see how that would happen.
/ Reply
Quote