which project for my 5870 GPU?

[Sunny129]

i typically use my 5870 2GB GPU for Milkyway@Home and SETI@Home (mostly Astropulse, but occasionally Multibeam) 24/7. the S@H project has been having significant issues for a month or so now and work production has been sparse, so i've kept my GPU on MW@H for the most part. MW@H is a nice fallback b/c it is generally a reliable project with lots of up time and hardly any down time...unfortunately this is about to change - the powers that be suspect that the server failure issue that's been going on for the last few days is a hardware issue. so in addition to the rare server issues the project is currently having (and resulting in intermittent work production), the MW@H server will be shut down tomorrow (indefinitely i guess) to try and solve the problem.

since i cannot cache anymore than 12 MW@H tasks at once, and since they only take 2:00 minutes on average to complete (depending on the type of task), i'll run out of MW@H GPU work in ~20 minutes. likewise, i don't have enough S@H Astropulse tasks cached (nor is the project producing new AP tasks fast enough to keep me from running out) to carry me through the duration of the S@H server issues. i'll probably try switching to S@H Mulitbeam first (even though S@H is having server problems too, Multibeam work production is always much higher than Astropulse work production). but if that can't carry me through rough times, i'll have to switch to another project or else my GPU will sit idle.

what other projects out there allow ATI GPU crunching (single precision or double precision, it doesn't matter)? i prefer astronomy/physics projects, but wouldn't mind taking up a number theory or biology/chemistry/anatomy project either...

TIA,
Eric

 

[wayliff]

looks like folding@home supports ATI 5000 series cards.

http://folding.stanford.edu/English/DownloadWinOther

 

[Sunny129]

i'll have to look into that...i've never participated in a non-BOINC project. is it that much more involving to get set up than a BOINC project is (seeing as how i already have BOINC installed)? also, what are some BOINC projects that can use a 5870?

TIA,
Eric

 

[wayliff]

I have had both side by side. F@home is very easy to install.
I will see if I can find other BOINC GPU projects.

 

[salvorhardin]

The only other projects that run on ati gpus are collatz and primegrid. collatz will give you more credit that primegrid. Right now I'm crunching collatz on my 5850 and averaging 150k a day. On my 4830 primegrid is my backup for milkyway. Both collatz and primegrid are math projects, so you might prefer f@h.

 

[wayliff]

found this link...
boinc.berkeley.edu/wiki/GPU_computing

however boinc is currently offline for maintenance

also looks like Collatz Conjecture also does GPU folding with ATI

 

[salvorhardin]

For f@h try the v7 beta client it unifies the cpu and gpu programs making it easier to install.

 

[Sunny129]

thanks so much for the suggestions guys. PPD isn't that important to me, so i guess it'll come down to which one is most interesting to me. if there are any other non-BOINC projects, i'd be happy to look into them too.

 

[Sunny129]

For f@h try the v7 beta client it unifies the cpu and gpu programs making it easier to install.

well i'd only be crunching F@H on the GPU...i'm assuming that, despite the fact that this installer will install both CPU and GPU applications, i don't have to actually use the CPU app?

 

[sangyup81]

I use Collatz as my backup to MW. Primegrid on ATI uses too much CPU for my liking

 

[Sunny129]

I use Collatz as much backup to MW. Primegrid on ATI uses too much CPU for my liking

how much CPU does the Primegrid ATI GPU app use?..and how about Collatz?

 

[salvorhardin]

well i'd only be crunching F@H on the GPU...i'm assuming that, despite the fact that this installer will install both CPU and GPU applications, i don't have to actually use the CPU app?

during setup you can choose what to run. just choose gpu and it will only run that. when crunching primegrid if my cpu cores aren't being used it uses a whole core. Since i'm doing correlizer on all 4 cores it limits primegrid to about 20% of a core. collatz uses a minimal amount, just to send/receive data.

 

[Sunny129]

hmm...well my decision just got more difficult. on the one hand, searching for primes seems more exciting & scientifically fruitful than trying to solve the Collatz Conjecture. on the other hand, i'm used to GPU apps like MW@H and S@H AP/MB using only 4-5% of a CPU core (as opposed to an entire core), and so Collatz sounds more appealing from that angle. i'm gonna look into F@H too and then weigh my options...

thanks again,
Eric

 

[Sunny129]

anyone care to give me a crash course on how to install Collatz for my ATI GPU? first of all, exactly which application am i supposed to download HERE?

 

[somethingsketchy]

What version of Windows (or Linux) do you have? 32bit or 64bit? Judging by what you have for a card, I'd say go with the Win##_2.09_ATI.zip (where you substitute ## for the bit version of your OS).

 

[Sunny129]

What version of Windows (or Linux) do you have? 32bit or 64bit? Judging by what you have for a card, I'd say go with the Win##_2.09_ATI.zip (where you substitute ## for the bit version of your OS).

well i'm running Windows XP 32-bit...but perhaps i should have been a bit more specific with my question...

1) obviously i need to choose one of the Win32 apps, but how do i know whether i should be running v2.06, v2.08, or v2.09?
2) what's the difference between say v2.09 w/ AMDCALxx.DLL files and v2.09 w/ ATICALxx.DLL files?

 

[sangyup81]

I believe you'd want the ATI one for the 5000 series. The 6000 series definitely uses the AMD one

 

[salvorhardin]

Just add the collatz project through boinc. since you were running milkyway you already have the dlls that you need (all I have installed is catalyst 11.2 and run both). when you create the project go to your account / collatz conjecture preferences and then untick use cpu and tick use ati gpu. you should be then set.

 

[Sunny129]

yeah i guess i just thought that i would need to manually install an optimized app for the GPU in addition to just attaching to the project through BOINC...but you were right, and now i'm crunching away on collatz and mini collatz tasks on the GPU.

another question for you Collatz GPU crunchers out there - does underclocking the GPU memory increase Collatz task run times? in Milkyway@Home, i can highly underclock my GPU memory without pentalty to task run times, thereby reducing GPU temps by 4-5°C. i don't have time to search the Collatz forums or experiment myself b/c its bedtime here, otherwise i would...i just figured someone would know if you can treat Collatz the same as MW@H in that respect...

also, does Collatz support checkpointing? or will tasks start over from the beginning if i suspend one mid-crunch and later resume it?

 

[petrusbroder]

Yeah, it does support check-pointing. If you stop in the middle of a WU you start again at the last checkpoint ...

 

[Sunny129]

ok, so i've determined that, unlike MW@H, Collatz is adversely affected by reducing the GPU's memory clock. at stock GPU clocks (850MHz core, 1200MHz memory), a Collatz 2.09 task takes ~24 minutes, but when i cut the GPU memory clock in half (850MHz core, 600MHz memory), a Collatz 2.09 task takes ~30 minutes. i haven't calculated the time difference with a mini_collatz 2.09 task yet, but i would imagine crunch times would increase ~25% there too...

mind you, i don't really need to downclock the GPU memory to reach acceptable GPU temps while crunching Collatz - i only really need to do that for MW@H, which stresses the GPU much more. this is only a dilemma when running both projects simultaneously (like i did last night for testing purposes) - either i downclock the memory so that temps stay where i want them while crunching MW@H, and Collatz run times suffer a bit...or i leave the GPU memory clock alone to get normal Collatz run times, and MW@H runs a bit on the hot side. fortunately my intentions are to have Collatz as a backup to MW@H (and not run them simultaneously)...and the MW@H project just went down for its maintenance/repairs, so i'll be able to run Collatz at the stock GPU clocks and run times won't suffer...

...of course this all boils down to coincidence if Collatz task run times are not standardized like MW@H task run times are.

 

[sangyup81]

thanks for the findings! now i'm thinking of overclocking my 5770's memory clock =)

 

[Sunny129]

well don't take my conclusions to heart just yet. again, i don't know if Collatz task run times are standardized (i.e. i don't know if Collatz v2.09 tasks should have very similar run times, or if they can vary significantly from one task to the next). so the fact the one collatz v2.09 task took ~24 minutes with a 1200MHz GPU memory clock while another took ~30 minutes with a 600MHz GPU memory clock could be purely coincidence unless someone else can confirm my results.

you should test my results using your 5770. to give you a reference point (despite having a different GPU than you), cutting my GPU memory clock in half (going from the stock 1200MHz to 600MHz) increased my Collatz run times by approx. 25%.