My first results in

Using your nVidia or AMD Graphics card for BOINC computation.
User avatar
rowpie
Peon
Posts: 239
Joined: Mon May 29, 2006 11:26 pm

#1 My first results in

Post by rowpie »

Ok so thats the first 2 done. Can anyone tell me if these look about right?

http://www.gpugrid.net/results.php?userid=11658

where do we find this seconds per step value that people are using to measure card speed?
sneakysaurus

#2

Post by sneakysaurus »

You can see the timings at the bottom of your task ID links:

http://www.gpugrid.net/results.php?hostid=19920

You seem to have a GTX 260 but your times to complete look a bit slow to me, 1 job took 60000s and the other 40000s.

How many BOINCS tasks are you running? 3 + 1 GPU , or 4 + 1 GPU. If 4 + 1 GPU that explains the slowness. Set your CPU allocation to 75% to fix that so you get back to 3 + 1.

I have an 8800gt with timings of about 50000s.

I would expect a GTX 260 to be less that 30000s, which is the main reason I am looking to buy one myself just now.

Anyone got some GXT260 results for rowpie to compare to?

On a similar GTX260 subject anyone know the bext deal on a 260 just now?

And is the 216 Maxcores version worth the extra cash with nay significant reduction in time to complete a GPU task?
sneakysaurus

#3

Post by sneakysaurus »

Just looked again does that beast of yours have 2 GTX260's? Very nice, but as I said they look a bit slow.

So similar question to before but this time are you running 2 + 2 GPU tasks? or 4 + 2 GPU tasks?
User avatar
rowpie
Peon
Posts: 239
Joined: Mon May 29, 2006 11:26 pm

#4

Post by rowpie »

that would actually be 8 plus 2 x gpu tasks. i think the seconds thing would have been completly thrown off by the fact that i was setting up gaming and swaping between sli and non sli mode due to this being my main gaming rig (was at a lan all day yesterday) it started on 2 fresh units last night so they should be finished when i get in so i'll force an update to send them in and we can see what the times are like.

do you still need to do a second monitor trick to get the most out of a two card setup in vista? (i remember reading about that a while ago) could explain why one was considerably slower.
User avatar
Megacruncher
G.L.S.B.
G.L.S.B.
Posts: 4699
Joined: Mon May 29, 2006 11:33 pm
Location: Edinburgh, Scotland
Contact:

#5

Post by Megacruncher »

Running 4+1 I'd expect to take 10hrs/36000s per WU on a 260.

I did read a review in PC Format magazine saying that there were problems with Hyperthreading & the i7 under Vista and their benchmarks were generally better with it switched off. Might be worth trying.
Willie the Megacruncher
Image
User avatar
rowpie
Peon
Posts: 239
Joined: Mon May 29, 2006 11:26 pm

#6

Post by rowpie »

well i dont think the monitors matter now.... time steps are close enough. i've changes the settings to 6 + 2 but im half way through 2 unites just now so there report wont be comparable. will see how the 2 after go however i noticed that they are the newest kind which are designed to work even better when all other cores are used... damn i could spend weeks optomising this thing (and ive not even overclocked it yet)
sneakysaurus

#7

Post by sneakysaurus »

I think even 6 + 2 is pushing it. I am not a big fan of hyperthreading, all it means is your are sharing resource accross cores. No matter what OS thinks you only have 4 cores.

To get max performance from your 260's 2 (or 3) + 2 is going to be your best bet. Although there are suggestions that the latest GPUGRID app appears to have sorted out the high CPU requirements, so 4 + 2 may work well here.

Unfortunately I think a lot of trial and error is required to get best balance.

The more I use GPUGRID the more I think dedicatiing lowish speed Dual Core CPUs to a GPU only box is best. e.g. a 1.8Ghz Duo with 2x260's would be good and leave dedicated high GHz Quads just to crunch normal projects. I certainly feel when running 3 + 1 on one of my 3Ghz Quads is a waste of a core, but I don't have much choice.
User avatar
rowpie
Peon
Posts: 239
Joined: Mon May 29, 2006 11:26 pm

#8

Post by rowpie »

yep i know what your saying. hyper threading has been improved a lot for the core i7 however i'll try with it off at some point. may need to sit down and plan what different setups to run the machine under and then compare its output directly to see how well it does.

been reading up on the hyperthreading bug in vista and i if i can see whats happening with threads changin core all the time.

i know what you mean about feeling like your wasting a core as i feel the same however to me this is a gaming rig and daily use box and a cruncher second. maybe one day i'll be able to afford dedicated crunchers but for now this will do.
User avatar
rowpie
Peon
Posts: 239
Joined: Mon May 29, 2006 11:26 pm

#9

Post by rowpie »

just for reference at any pont if i say 8 cores i know there is only 4 actual and 4 virtual. its just a bugger to trye using terminology to destinguise between them.

I tried to find out more details about other i7 machines in operation and everyone on boinc seems to be sticking with the 8 cores despite the hyperthreading bugs.

http://boincstats.com/stats/host_stats. ... st=0&or=16 -- some insane rac's on the 4ghz one.

the hyperthreading problems in vista seems to be mostly when your running two apps and it puts both threads on the one core as it can't tell the difference between real and virtual and this causes a slow down in day to day working. (only 11 core from 4 doing the work) in my case though as im permanently maxing out the threads they are all getting used anyway. so it couldnt exactly do it by mistake.

on the 6 + 2 i noticed that no single core hits 100% usage so it must be moving the work between all the cores a lot. this does probably cost me some overhead. i've messed about with the affinity to see whats going on (gave every process there own core - this maxed out core 1-6 and left 7 and 8 for a gpu each) and the gpu's definatly like having one thread each available. My max cpu usage for any one core is 12-13% of the 100 and both gpu processes like 8% steady each to supply them with information.

with me setting each process to a set core i get 92% usage. with them all set to use any i still get the same 92%. maybe with the hyperthreading i can afford to lose just 1 core from 4 to feed the gpu's instead of 2.

its all fun and games.
User avatar
rowpie
Peon
Posts: 239
Joined: Mon May 29, 2006 11:26 pm

#10

Post by rowpie »

ok when 3 posts in a row are your own is it sad? right anyways thats 2 new taks started on the 6 + 2 setup. interesting thing is that one is app 6.55 and the other is 6.56, the 6.56 app is only registering 1% cpu usage compared to 8% for 6.55.

i mean science is fun but can we start a boinc project to crunch these numbers for me as my head hurts trying to work this out lol
User avatar
Megacruncher
G.L.S.B.
G.L.S.B.
Posts: 4699
Joined: Mon May 29, 2006 11:33 pm
Location: Edinburgh, Scotland
Contact:

#11

Post by Megacruncher »

Your figures are soaring nicely whatever is happening.
http://ubt-seti.dyndns.org/tsbt/ubt_gra ... ame=rowpie of TSBT
You could be a millionaire before the year is out.
Willie the Megacruncher
Image
User avatar
ianmbaker2
Boinc Corporal
Boinc Corporal
Posts: 69
Joined: Sat Dec 29, 2007 11:51 am
Location: South Lanarkshire
Contact:

#12

Post by ianmbaker2 »

Your numbers on BoincStats are certainly moving up. The flash I saw in my rear view mirror was deffie aided by the 12 1/2 K a day for GPU grid. A big increase over my 3k from a 9800GT. You should overtake me in about 10 days at that rate.

And you had 7th best RAC in the team, from one box, today.

Well done.

Ian
Image
Post Reply Previous topicNext topic

Return to “Graphics Processing Unit (GPU)”