PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Number crunching : Getting Prime GPU WUs (Message 60210)
Posted 2966 days ago by Profile Sorceress
To me this sounds like BOINC trying to divide processor time equally among projects. Those Genefers take quite a long time while, if I remember correctly, most of the other projects will take a lot shorter. So in order to meet the sharing of resources you've set, I'd expect BOINC to not download a PG workunit for a while until all the other projects are back to the share they'd need to have.


This has been going on for a long time, while this is the 3rd Genefer I've had in 3 weeks. Does Boinc try to micro manage processor time to that degree? I have had other project's long running GPUWU and didn't change the pattern I am seeing. If so, how do I erase the data base on how much time is alotted to each GPUWU and start with a clean slate? This doesnt seem to affect CPU WUs. I always have some for every project in my cache. It's just the GPU WU that are being affected. I have PG set @100% resources while the other GPUWU projects are set @50%.
2) Message boards : Number crunching : Getting Prime GPU WUs (Message 60205)
Posted 2966 days ago by Profile Sorceress
Great! Still doesnt answer my question. How do I get Boinc to download Prime GPU WU with having to force it to?


The full answer to that question is complicated and depends on exactly what WUs already have been sent to your computer.

Assuming that there's no tasks at all on your computer, you need to do all of the following:

1) First of all, make sure your computer is assigned to the proper BOINC venue, then go to the BOINC PrimeGrid preferences page here at PrimeGrid and make sure the "Use Nvidia GPU" or "Use ATI GPU" (as appropriate) is checked. If you ONLY want GPU tasks, then make sure "Use CPU" is unchecked, or if you want both CPU and GPU tasks then check the "Use CPU" box also.

2) For the individual subprojects, where it says LLR-CUDA (beta) uncheck this box. This isn't really available and may cause the scheduler to get confused and either not send you anything, or send you the wrong tasks.

3) Select the GPU tasks you want to recieve.

4) Even if you unchecked the "Use CPU" box, select at least one CPU task.

That should do it, assuming you don't have any other tasks already on your computer.

If you DO have tasks already on your computer, you can either be patient and wait until BOINC, in its infinite wisdom, deigns it to be the proper time to download something from PrimeGrid, or you can try to force it to do so. This isn't really very intuitive and I prefer to avoid ever running more than one project on a single computer. Besides setting "no new work" on all non-PrimeGrid projects, you'll need to raise your cache size to be larger than the amount of work currently remaining on your computer and/or abort some tasks in order to convince BOINC that it needs to download more work. Also, bear in mind that under many circumstances BOINC won't download anything at all if there's a task that is in the "high priority" state, i.e., BOINC thinks it's going to miss its deadline.

So much for the long, theoretical, tries to answer every possible way of answering your question. It's looks like you've run both a PPS sieve and GFN task today on your GPU, so I take it everything's now working the way you want it to work?


All this has been done, thanks. I had to force Boinc to download the last two GPUs from Prime. So no, it's not working right.

Let me try to clarify. Over a month period, I will always have 1-2 GPU WUs from all my GPU projects but Prime in my cache. Normally, when Boinc finishes a GPUWU, it will d/l a new one to replace the one finished. Not so with PG. I have to force Boinc to d/l PG GPUWUs most of the time or there aren't any. Once Boinc finishes a PG GPUWU it then d/ls only CPU WUs until I force it to d/l more GPUWUs. In the meantime, Boinc is cycling the other projects GPUWU on a regular basis. I do not understand why Boinc does this with PG? I process Collatz, Milkyway, Einstein, MooWrapper and Seti Beta GPUWUS besides PG and I always have GPUWUs from all of them but PG. Why is PG being treated differently? Is there a priority setting that I can change to make Boinc d/l GPUWUs from PG like it does the rest? PG was my first GPU project and so is special. I want it to have PG GPUWUs before I have anyof the others. I have noticed in the 'Event Log' that Boinc will routinely ask all my projects for GPUWUs but not PG. If there is no GPUWU in my cache for a GPU abled project, Boinc will d/l at least 1 or 2. WHY? (I hope this is making sense. :))

Example: Right now I have 1 Collatz, 3 Einstein, 4 Milkyway, 2 Moo Wrapper, 2 Seti Beta and 1 PG GPUWUs. When Boinc finishes the PG WU, it will not get any more GPUWUs until I force it to. But Boinc will replace the other GPUWUs as they finish without forcing it.
3) Message boards : Number crunching : Getting Prime GPU WUs (Message 60201)
Posted 2966 days ago by Profile Sorceress
If you have any PrimeGrid workuits suspended, you won't be able to get new PrimeGrid workunits. That holds true for any project, not just PG.


As far as I know, if you suspend any project's WU, Boinc will download another. That's why you set 'new new tasks' But no, Prime was not suspended as I stated. All other projects were.


Unless Berkeley very recently changed the behavior of the BOINC client, you are mistaken.

Absolutely, positively, without any doubt: If you have any individual PrimeGrid tasks suspended, you will not be able to download any tasks from PrimeGrid.

As Jim said, this isn't a PrimeGrid-specific behavior; this is how BOINC works and holds true for all projects.


Great! Still doesnt answer my question. How do I get Boinc to download Prime GPU WU with having to force it to?
4) Message boards : Number crunching : Getting Prime GPU WUs (Message 60197)
Posted 2966 days ago by Profile Sorceress
If you have any PrimeGrid workuits suspended, you won't be able to get new PrimeGrid workunits. That holds true for any project, not just PG.


As far as I know, if you suspend any project's WU, Boinc will download another. That's why you set 'new new tasks' But no, Prime was not suspended as I stated. All other projects were.
5) Message boards : Number crunching : Getting Prime GPU WUs (Message 60194)
Posted 2966 days ago by Profile Sorceress
I am having trouble getting Prime GPU WUs but not any other project's GPU WUs. I have to set all my projects to 'no new task' but Prime, then suspend all work units to get Prime to download any GPU WUs. Normally I have 2-4 GPU WUs for all GPU projects regulary, but none for Prime. I just get Prime CPU WUs. I have to force Boinc to get Prime GPU WUs or it won't get any. Anyway to set Prime first priority for GPU WUs then the rest?
6) Message boards : Generalized Fermat Prime Search : Generalized Fermat Prime Search (Message 59894)
Posted 2975 days ago by Profile Sorceress
Yes, depending on which model of the 8600 you had, the 620 is likely at least twice as fast, maybe a bit more.

A word of caution for the future, however. In Nvidia's numbering scheme, the last two digits, i.e. the "20" in "620" are more important than the first digit (i.e., the "6") in gauging how powerful the card is. For example, a GTX 480 is a lot faster than a GTX 560. The first digit is the generation of the card, and, for sure, each successive generation is somewhat faster and more energy efficient than the previous generation, but the difference between the high-end and low-end cards within each generation is huge.

EDIT: Getting back on topic, with the screen saver off, how's the GPU doing with GeneferCUDA?


Thank you for the info. I am hoping to upgrade to a much better computer with a good vid card this spring. I am waiting on a settlement that is taking it's sweet time. I wanted something that would work well in the interum. The 620 is holding it's own so far. I just got another Genefer WU so I will let you know how it does this time around. It did finish the last one with credits though it was bit late. :) I had several other CUDA WU time out though.Couldnt be helped.
7) Message boards : Generalized Fermat Prime Search : Generalized Fermat Prime Search (Message 59715)
Posted 2983 days ago by Profile Sorceress
I did have the screensaver running which is now shut off.


That is likely what was causing it to be so slow. My guestimate is that it should be able to run one of those work units in about 50 hours -- about the same as your CPU could do.


By slow, which specs should I be looking at for speed? The GT620 has a core clock of 700mhz and the GT560ti has a 900mhz core. Dosen't seem thats much of a difference. The memeory speeds are quite different though.


Scott's answer is correct -- it's the number of shaders that's most important. The 620 has 48 shaders, a mid-range card has about 400-500 shaders, and the top of the line have (I think) about 1500 shaders. A "shader" on a GPU is roughly analogous to a core on a CPU: the more shaders there are, the more calculations can be done simultaneously. It's the ability to run many calculations simultaneously that makes GPUs so fast, so the more shaders there are, the faster the GPU can crunch.


The gt620 has 96 stream processors. I can't find any info that says anyting about shaders. I understood that the stream processors were the GPUs. Yes it is on the low end but quite a step up from the gtx 8600 I had. It's the best I can afford at this time.
8) Message boards : Generalized Fermat Prime Search : Generalized Fermat Prime Search (Message 59634)
Posted 2985 days ago by Profile Sorceress
Indeed, the GT 620 is a very slow GPU -- not just for double precision programs like Genefer, but for all programs. It's one of the "value" or "entry level" GPUs in the 600 series.

That being said, a 250 hour run-time for that work unit is surprisingly long -- the fastest GPUs can complete those about 100 times faster.

I'm wondering if there's something that's somehow interfering with the GPU and keeping it from running full speed?


I did have the screensaver running which is now shut off. By slow, which specs should I be looking at for speed? The GT620 has a core clock of 700mhz and the GT560ti has a 900mhz core. Dosen't seem thats much of a difference. The memeory speeds are quite different though.

I have disabled the Genefer WUs.
9) Message boards : Generalized Fermat Prime Search : Generalized Fermat Prime Search (Message 59624)
Posted 2985 days ago by Profile Sorceress
I have a Genefer 1.07 CUDA32_13 running on my comp for 174hrs with 73hrs left. Is this a Generalized Fermat Prime Search WU? If so, why did I get it? I have set the long search to disable. Setting preference is for short task only! The WU was due on 11/16 at midnight so it will be long pass that when it's finished. And all my other GPU WU for other projects are now behind as well. What good is setting my preferences if the project ignores them. I have been gone for 4 days or I would have aborted this WU. I will really be pissed if this WU fails <angry face>
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 3.81, 3.94, 4.05
Generated 21 Jan 2021 | 14:12:55 UTC