Hi!

If I put this GTX 275 card in my old PC, which has only PCI Express 1.x x16 slot, does it still give me about 1 teraFLOPS of computing power? That card is developed for PCI Express 2.0 x16 slot. So, does it work slower in my computer? Does it not give me 1 TFLOPS of computing power?

Thanks!
____________

sorry to say, buddy, but u will not get 1 terraFLOPS anyway :-) my GTX275 OCed 702/1584/1260 give 760 giga only :-)

I'm not sure about PCIe,. what's your mobo model?
____________

sorry to say, buddy, but u will not get 1 terraFLOPS anyway :-) my GTX275 OCed 702/1584/1260 give 760 giga only :-)

I'm not sure about PCIe,. what's your mobo model?

So, what do you think? What RAC will I get with my Pentium D 920 CPU and Palit NE3TX275FHD94 GeForce GTX 275 on my PCI Express 1.x slot?

My Pentium D 920 usually gets RACs something like 300...500.
____________

PCIe 1 16x delivers plenty of speed for GPU-Grid. If you're loosing anything it's well below measurement error / accuracy (normal spread in WU times).

The GTX275 of a friend used to get ~20k RAC here, but due to the recent improvements to the client this value is going up. It has reached 24k by now and if I project speed based on 1 sample WU it would reach 34k RAC (seems a bit too high). Assuming the machine is crunching 24/7.

BTW: running a Pentium D is quite inefficient, as it's very slow for the power it consumes. Sadly your board doesn't support Core 2 CPUs. If I were you I'd stop crunching on the Pentium D to save energy cost. The GPU will more than make up for it ;)

MrS

ExtraTerrestrial Apes

agree 100%. GTX275 should work fine.

The only thing - my GTX275 gives my around 40k (http://boincstats.com/stats/user_graph.php?pr=ps3grid&id=34254), but not 34k :-)
____________

BTW, 4870 being cheaper then GTX275 gives almost twice more FLOPS - 1360 against 760
____________

skgiven,
Regarding Linux - there are tons of manuals on Linux installation (even in pictures). Nothing special, really ;-) It's "profitable" for the project (more and faster job done) and for users (more credits). Basically, it's

sh BOINC

and u'r done. But NP - I can make smth like our local manual, including OCing process.

If we are talking about "ideal world", it will be nice to have lower CPU usage. Again, I'm not complaining, I'm just reporting :-)

ExtraTerrestrial Apes,
OC'ed 4870 (850/900 @1.3375V) gives 112k in MW. IMO from hardware point of view ATI are more powerful. SDK AFAIK is not that great, but progressing and we can see 1st results - MW & Collatz. Hope more project will start to use the power of GPUs.

I decided to take both paths and "came alive" my previous rig :-) Astronomy was 1 of my favorite classes at school :-)
____________

now Travis cancelled that WUs which were really nice in terms of credits :-) Same time but twice more credits :-)

I'm moved to 0.22 which got bug fix for progress bar. But other then that 0.20 worked just fine.
____________

··· Hope more project will start to use the power of GPUs.···

There is a very recent new project. It works with ATI graphic cards. You can see it here.

Don't bother trying, it's not worth the effort!

The GTX275 of a friend used to get ~20k RAC here, but due to the recent improvements to the client this value is going up. It has reached 24k by now and if I project speed based on 1 sample WU it would reach 34k RAC (seems a bit too high). Assuming the machine is crunching 24/7.

You initially got a few errors, but after that you only had one error. So in the last 8days you completed 11 of the 12 tasks (+1 in progress). You are picking up full credit; so you are finishing the tasks in good time. Although your present RAC shows about 6 or 7K, you are actually getting about 10K per day on average. If it was running 24/7 I think it would be more like 17K, as is. I expect you are getting less because it is not running GPUGrid all the time, you may be running other GPU tasks or very heavy CPU tasks, or possibly with limited system RAM or drive space (you can check that yourself).
RAC is calculated over an extended period of time so it will take a while longer to balance out.
Your GTX275 should be able to get 30K per day at least, if your system is optimised to RUN GPUGrid tasks.

The amount of difference made by leaving your CPU free at all times for GPUGrid is minimal at best and would not out weigh the benefits I personally believe WCG is brining to humanity. Running bth works quite well.

If you really want to maximize GPUGrid over all else then install linux and run nothing else. Next most valuable increase would be to up your shader clocks, then up your cpu clocks, finally up clock on the memory of your 275. Not much increase from the last couple but that would be my approach (I've done lots of testing) if I wanted to see what was the absolute best I could do for GPUGrid.
____________
Thanks - Steve

So, should I not run World Community Grid when running GPUGRID?

The amount of difference made by leaving your CPU free at all times for GPUGrid is minimal at best and would not out weigh the benefits I personally believe WCG is brining to humanity. Running bth works quite well.

If you really want to maximize GPUGrid over all else then install linux and run nothing else. Next most valuable increase would be to up your shader clocks, then up your cpu clocks, finally up clock on the memory of your 275. Not much increase from the last couple but that would be my approach (I've done lots of testing) if I wanted to see what was the absolute best I could do for GPUGrid.

If you are heavy into disease projects, or projects like GPU Grid with molecular modeling in mind, then, by all means consider running WCG on the CPU while running GPU Grid on the GPU ... the point is to advance the science, not to run up more points here vs. there ... There are about 62 projects that are considered to be "alive" at this time (with about 10 in long term halts, they show off line and have been for days to months) and they all could use your help ...

I have for years suggested, and still do, that people should attach to as many projects as they find interesting ... the credit will come but the point is to accomplish the work so the science can be advanced ... in my case, those projects that I am not as interested in as others, well I give very low shares to and sometimes turn off work for weeks to months at a time.

Anyway, my two cents ...

HTH, you can run WCG and GPUGrid.

Do you run GPUGrid tasks 24/7 or does the system get turned off regularly?

Do you run other GPU tasks?

I am asking because your RAC and your daily returns are low for that card.
The balance of your system is massively in favour of the GPU; you have a Very good GPU, but you have a rather average CPU (Pent D 2.8GHz). I don’t know how much RAM or Free HDD space you have, but that might be a factor too.

If I were you I would try using one CPU core for WCG, and see if that speeds up your GPUGrid tasks significantly. By significantly I mean your overall cross-project Boinc points rise as a result. If they do not, then there would be no point to set your CPU usage to 50% and you would be better off running 2 CPU tasks and 1 GPU task simultaneously.
If you had a 1.8GHz dual core system I would be fairly sure you should just run 1 CPU task, and if you had a 3.2GHz Core 2 Duo I would be fairly sure you should run 2 CPU tasks.

I just use 3 of my 4 Phenom II 940 cores for crunching CPU tasks, because I have 4 overclocked GT240’s in that box. The overall CPU use sits at about 91% so I am really only losing 36% of one CPU core. For those with 2 big cards (GTX260 or above) that have an i7 with 8threads running, it is clearly beneficial to just use 7threads for CPU tasks, leaving a thread free for the GPUs which gobble most of it up anyway. Remember, a good GPU can roughly do 10times the work of a high end CPU. Even one GT240 would do twice as much as an i7-980X, with its 6 cores (12threads). £50 vs £850!

Overall, what I do expedites the GPU tasks by a few percent each, and there are 4 GPU tasks running 24/7, so for me this makes sense, and I do get more world wide Boinc credits this way. That said, for me it is not entirely about Boinc credits; I don’t try to find ET or count Pi, I just run the projects I believe are useful to mankind. If I was just interested in points I would just have ATI cards and just crunch sETi, Collatz or MilkyWay tasks.
I was not convinced about Folding@home’s GPU utilization, and I had problems crunching their tasks, so I stick to GPUGrid for GPU tasks. WCG does good work, but they do employ some dubious tactics, which is why I still support other projects such as climateprediction, rosetta, docking, malariacontrol, simap, RNA World, poem, the lattice project, Ralph and of course ibercivis, which is a bit similar to the WCG in that they run several different projects.

My tuppence worth ;)

If you are heavy into disease projects, or projects like GPU Grid with molecular modeling in mind, then, by all means consider running WCG on the CPU while running GPU Grid on the GPU ... the point is to advance the science, not to run up more points here vs. there ... There are about 62 projects that are considered to be "alive" at this time (with about 10 in long term halts, they show off line and have been for days to months) and they all could use your help ...

I like them all. I think I have registered for at least 82 BOINC projects at the moment (maybe a world record).

I have for years suggested, and still do, that people should attach to as many projects as they find interesting ... the credit will come but the point is to accomplish the work so the science can be advanced ... in my case, those projects that I am not as interested in as others, well I give very low shares to and sometimes turn off work for weeks to months at a time.

Anyway, my two cents ...

After buying GTX275 I think I should concentrate on projects with CUDA support. Of course, I'm also running CPU-only projects (100% CPU load all the time), but now it's more important to have some work for my GTX275. :)

It may well be, everytime I want to go check I can't find the stat site that allows you to sort out users by how many projects they have joined. You have me beat and no way I will catch up because of all those projets that came and went while I didn't do BOINC at all for about 2 years ...

Well, there are a few out there ... GPU Gird is probably one of the best as pure CUDA ... (keep us ATI users out ... :)

It may well be, everytime I want to go check I can't find the stat site that allows you to sort out users by how many projects they have joined. You have me beat and no way I will catch up because of all those projets that came and went while I didn't do BOINC at all for about 2 years ...

http://www.boincsynergy.com/stats/boinc-individual.php?db1=80&db2=more

Well, there are a few out there ... GPU Gird is probably one of the best as pure CUDA ... (keep us ATI users out ... :)

It would be cool to have ATI-card also. I only have one PCI Express x16 slot, though. LOL.