I assume you mean .25 or .2 or .15 and not fifteen and I assume you are referring to a setting for GPU utilization factor? If so, you choose a value so that the reciprocal gives the number of concurrent tasks to run on the GPU. For example, a value of 0.2 would attempt to run 5 concurrent tasks. You can't just blindly choose a big number of concurrent tasks and expect it to work. There are three separate factors for three separate app classes and each app would have different memory requirements so you may very well need to plan for different values depending on which app (or apps) you wanted to end up running.
I notice you are currently running just about all apps as you have tasks on board for everything except the GW search (S6 CasA). [There is a new beta-test app for the GW search which is looking quite promising.] If you continue to mix your current ones, BRP4G, BRP5 and FGRP3, you need to be careful as the FGRP3 GPU tasks need a lot more memory than the other two. On your 2GB Pitcairn series GPU you should be able to run up to 4 concurrent tasks for BRP4G or BRP5 but if you try to run more than 2 FGRP3 you may well experience troubles at some point.
Another thing that is going to affect performance is the fact that your motherboard PCIe slot is probably only PCIe1, judging by the fact that you have a Q6600 CPU. EAH GPU apps use lots of PCIe bandwidth so your crunch times will be about a quarter to a third longer than if the slot was PCIe2 or 3. This is going to limit your ability to "max out your GPU" as you put it. Another factor you will need to be careful about is the number of 'free' CPU cores to reserve for GPU support duties. Overall performance may be able to be improved by a bit of experimenting with this.
By the way, does "max out" simply mean showing something like 95%+ GPU utilization or are you actually wanting to max out your credit return? :-). If you care to explain a bit more about your setup and what you want to achieve, it would be easier to give more specific recommendations.
I just want the gpu to be best utilized . if I bave a .5 cpu X .25 gpu I can run several but the cpu is running poem and rosetta . so generally I have just one core running this project . . I have seen the card run 80 + percent and thats the reason I got it is to crunch as I really have on board video
The .5 CPU you mentioned is actually per GPU task. So to run 4 GPU tasks you will need a minimum of 2 'free' CPU cores, otherwise GPU performance, in both runtime and loading percentage take a big hit.
You should easily be able to "max out" your GPU with some experimentation - task choice and number of concurrent tasks, running your other projects at the same time, or not etc. will reveal the best performance from your card.
You just have to be willing to put in the time to find the best configuration for your machine, task type, and individual project priority. Only you, will find whats best for you :-) but running 4x on the GPU you really ought to have at least 2 cores free.
I have 4 Q6600 machines and still think them to be eminent crunchers. If you wanted my honest opinion, I would suggest you 'free' all cores and just run the GPU flat out!.. Although that doesn't help your other projects.
The best thing for you do do is set it at GPU utilization factor of BRP apps 0.5 and run it for a day or 2 and see what the time is and the daily RAC
You have a quad-core so it depends on what else you want to run as far as CPU tasks.
If you leave one core free that will be the best way to test to see if 2 tasks at a time is your best setting or is you may want to try the .33 to run GPU tasks X3
Binary Radio Pulsar Search (Perseus Arm Survey) is the on to run the tests on.
my mother board is PCIe 16 x 1.1 v [ ASUS P5G41T-M LX ] right now I have 2 cores running .5 x .25 - and I have a good 80 + usage . which is great . but when its running one or 2 gpu units its underutilized . I wish their was a varable setting to keep the GPU running heavily
my mother board is PCIe 16 x 1.1 v [ ASUS P5G41T-M LX ] right now I have 2 cores running .5 x .25 - and I have a good 80 + usage . which is great . but when its running one or 2 gpu units its underutilized . I wish their was a varable setting to keep the GPU running heavily
I'm not sure I fully understand you but I *think* you are saying that there are times when your GPU has 4 concurrent tasks running and there are times when you only have one or two. You wish to have the GPU always running 4 concurrent tasks. If this is correct your solution is quite easy.
You appear to be running three separate GPU apps, Binary Radio Pulsar (Arecibo) - BRP4G, Binary Radio Pulsar (Perseus Arm Survey (PAS)) - BRP5, and the Fermi Gamma Ray Pulsar - FGRP3. The first two of these have small enough memory requirements and are quite mature applications so you should have no trouble running 4 concurrent GPU tasks by setting the GPU utilization factor of BRP apps to 0.25. The FGRP3 app is a new app that has larger memory requirements and needs a lot more CPU support than the others. There is a separate GPU utilization factor for this app and if you have that factor at 1, other GPU tasks will stop when one of these is running. These tasks need a lot of memory and I found you can only reliably run two concurrent on a 2GB GPU without risking running out of memory.
Your best option would be to set the FGRP3 GPU utilization factor to -1 which will prevent the scheduler from sending any more of these tasks to you. When your current FGRP3 GPU tasks have finished (or when you abort them) your GPU will then be able to always run 4x with the BRP tasks. Whilst you wont receive further FGRP3 GPU tasks, the -1 setting could still allow FGRP3 CPU tasks to run on your 2 available CPU cores. I see you are attached to other projects so I guess you may not want CPU tasks here. If so you could deselect the FGRP3 app entirely using those settings where you tick the box for the apps you want. Just untick the Gamma Ray Pulsar search.
gpu usage question
)
I assume you mean .25 or .2 or .15 and not fifteen and I assume you are referring to a setting for GPU utilization factor? If so, you choose a value so that the reciprocal gives the number of concurrent tasks to run on the GPU. For example, a value of 0.2 would attempt to run 5 concurrent tasks. You can't just blindly choose a big number of concurrent tasks and expect it to work. There are three separate factors for three separate app classes and each app would have different memory requirements so you may very well need to plan for different values depending on which app (or apps) you wanted to end up running.
I notice you are currently running just about all apps as you have tasks on board for everything except the GW search (S6 CasA). [There is a new beta-test app for the GW search which is looking quite promising.] If you continue to mix your current ones, BRP4G, BRP5 and FGRP3, you need to be careful as the FGRP3 GPU tasks need a lot more memory than the other two. On your 2GB Pitcairn series GPU you should be able to run up to 4 concurrent tasks for BRP4G or BRP5 but if you try to run more than 2 FGRP3 you may well experience troubles at some point.
Another thing that is going to affect performance is the fact that your motherboard PCIe slot is probably only PCIe1, judging by the fact that you have a Q6600 CPU. EAH GPU apps use lots of PCIe bandwidth so your crunch times will be about a quarter to a third longer than if the slot was PCIe2 or 3. This is going to limit your ability to "max out your GPU" as you put it. Another factor you will need to be careful about is the number of 'free' CPU cores to reserve for GPU support duties. Overall performance may be able to be improved by a bit of experimenting with this.
By the way, does "max out" simply mean showing something like 95%+ GPU utilization or are you actually wanting to max out your credit return? :-). If you care to explain a bit more about your setup and what you want to achieve, it would be easier to give more specific recommendations.
Cheers,
Gary.
I just want the gpu to be
)
I just want the gpu to be best utilized . if I bave a .5 cpu X .25 gpu I can run several but the cpu is running poem and rosetta . so generally I have just one core running this project . . I have seen the card run 80 + percent and thats the reason I got it is to crunch as I really have on board video
The .5 CPU you mentioned is
)
The .5 CPU you mentioned is actually per GPU task. So to run 4 GPU tasks you will need a minimum of 2 'free' CPU cores, otherwise GPU performance, in both runtime and loading percentage take a big hit.
You should easily be able to "max out" your GPU with some experimentation - task choice and number of concurrent tasks, running your other projects at the same time, or not etc. will reveal the best performance from your card.
You just have to be willing to put in the time to find the best configuration for your machine, task type, and individual project priority. Only you, will find whats best for you :-) but running 4x on the GPU you really ought to have at least 2 cores free.
I have 4 Q6600 machines and still think them to be eminent crunchers. If you wanted my honest opinion, I would suggest you 'free' all cores and just run the GPU flat out!.. Although that doesn't help your other projects.
Gavin.
The best thing for you do do
)
The best thing for you do do is set it at GPU utilization factor of BRP apps 0.5 and run it for a day or 2 and see what the time is and the daily RAC
You have a quad-core so it depends on what else you want to run as far as CPU tasks.
If you leave one core free that will be the best way to test to see if 2 tasks at a time is your best setting or is you may want to try the .33 to run GPU tasks X3
Binary Radio Pulsar Search (Perseus Arm Survey) is the on to run the tests on.
my mother board is PCIe 16 x
)
my mother board is PCIe 16 x 1.1 v [ ASUS P5G41T-M LX ] right now I have 2 cores running .5 x .25 - and I have a good 80 + usage . which is great . but when its running one or 2 gpu units its underutilized . I wish their was a varable setting to keep the GPU running heavily
RE: my mother board is PCIe
)
I'm not sure I fully understand you but I *think* you are saying that there are times when your GPU has 4 concurrent tasks running and there are times when you only have one or two. You wish to have the GPU always running 4 concurrent tasks. If this is correct your solution is quite easy.
You appear to be running three separate GPU apps, Binary Radio Pulsar (Arecibo) - BRP4G, Binary Radio Pulsar (Perseus Arm Survey (PAS)) - BRP5, and the Fermi Gamma Ray Pulsar - FGRP3. The first two of these have small enough memory requirements and are quite mature applications so you should have no trouble running 4 concurrent GPU tasks by setting the GPU utilization factor of BRP apps to 0.25. The FGRP3 app is a new app that has larger memory requirements and needs a lot more CPU support than the others. There is a separate GPU utilization factor for this app and if you have that factor at 1, other GPU tasks will stop when one of these is running. These tasks need a lot of memory and I found you can only reliably run two concurrent on a 2GB GPU without risking running out of memory.
Your best option would be to set the FGRP3 GPU utilization factor to -1 which will prevent the scheduler from sending any more of these tasks to you. When your current FGRP3 GPU tasks have finished (or when you abort them) your GPU will then be able to always run 4x with the BRP tasks. Whilst you wont receive further FGRP3 GPU tasks, the -1 setting could still allow FGRP3 CPU tasks to run on your 2 available CPU cores. I see you are attached to other projects so I guess you may not want CPU tasks here. If so you could deselect the FGRP3 app entirely using those settings where you tick the box for the apps you want. Just untick the Gamma Ray Pulsar search.
Cheers,
Gary.