Question from a newbie with a new GTX650 added
31 Jan 2013 21:23:45 UTC
Topic 196781
(moderation:
I feel like such a newbie asking questions that are second-nature to 99% of you, but I'll ask again anyway.
I just installed a new GPU on my i5-2500 computer, an ASUS GTX650-E-1GD5, and it is already crunching a second BRP4 work-unit. It completed the first w.u. in just under 38-minutes.
Are the following observations nominal?
Under "Tasks" in the E@H task manager, it indicates that the w.u. is using "0.2 CPUs + 1 NVIDIA".
The ASUS package included a nice performance monitor, and here are some readings that may be helpful:
GPU temperature is nominally 53C, peaking as high as 55C.
GPU usage reads 92%
VID usage reads 0%
FB usage is 64% (I don't even know what "FB" means)
GPU clock - 1.07 GHz
MEM clock - 5.0 GHz
Proc clock - 2.14 GHz
Mem usage 283 MB
In the Win7 task manager, all 4 cores of the i5-2500 are running at 100% and well-within temperature specs (no higher than 46C).
If this all looks good, I'll order and install another GPU for use in my other (i5-2400) computer and then sit back knowing I'm being as useful as I can be to the project. Otherwise, please give me some pointers to improve things even more.
Thanks for all your help!
- Andy
Question from a newbie with a new GTX650 added
)
Hello Andy,
Yours is running as it should as far as the BRP cuda's and you are running the other tasks at the same time (GW s6)
The 0.2 is your settings on your account preferences page "GPU utilization factor of BRP apps"
If you want to see how fast it can run the cuda's you can set your preferences to only run BRP's and none of the other tasks after you finish the ones you have.
FB is the Framebuffer for your video output device that drives a video display from a memory buffer containing a complete frame of data.
And some of us overclock those clock settings which is pretty easy.
Your temps sound fine.......my 550Ti's overclocked and a 660Ti superclocked all run low to mid 60'sC running BRP cuda X2 along with a 2 core Test4Theory and at times I will add a LHC or Gamma Ray task on the quad-core hosts.
(set your preferences to "Use CPU - No" and check the Run only the selected applications Binary Radio Pulsar Search (Arecibo): Yes and uncheck the others)
I don´t have the 650, but
)
I don´t have the 650, but normaly with an I5 the gpu works faster/better if you leave at least a core free to feed it. My sugestion check your GPU usage with 100%,75%, 50% and 25% of max CPU usage and look what max you GPU usage. Each sistem is unique so you need to test.
Thanks for the quick and
)
Thanks for the quick and informative replies, MAGIC and Juan BFB.
I'm still a bit confused but things are starting to make sense to me. I changed my "GPU utilization factor of BRP apps" to 0.5 a couple of hours ago, and then again to 0.33. I noted that the completion times have increased for the BRP work-units, but that the total number of work-units per hour has increased.
Here's where I'm still confused:
The GPU only has 1-G of memory. Processing two work-units at a time used about 505-MB of that memory; three at a time is now using 741-MB. It would seem that attempting four at a time is pushing the memory limit, but MAGIC recommends setting the factor to 0.2 (five work-units at a time) which I don't understand. Please explain if you can.
I switched the number of CPUs to 75%, freeing up one core as a "feeder" for the GPU. What happened was odd, in that a couple of the CPU cores postponed the task they were working on and are now doing what are labeled "high priority" runs. (I have a *lot* of the GW-S6 work-units in my cache.)
I understand the need to experiment on my particular system, but I keep remembering that ominous "computer damage" warning about the GPU utilization-factor setting. :-) I'm still a little skittish about making changes unless I fully understand them. (And even then, I tend to make one change at a time, noting the differences, before I make another.)
I really appreciate all the help I am getting from your replies. Thanks again!
I have some 650's...I
)
I have some 650's...I actually found doing 0.5 / 2 at a time was best. 3 is possible, but I didn't notice any performance gain, and though your GPU has some free memory left running only 2, the rest of its resources will be near 100% running 2.
I think Magic either made a typo or had 0.2 stuck in his head from the "0.2 CPU" line in your post.
Yep.......mine are all set at
)
Yep.......mine are all set at 0.5 to run em at BRP X2
And no there is not an advantage to run more than 2 BRP's at a time with your GTX 650
And you also see that having all those Gamma Ray tasks and other tasks can make them switch to running at high priority.
But I never have that problem when I run 2 BRP's (0.5) and the 2 cores T4T and even add a few days worth of GRays at the same time on a quad-core other than it will slow down the cuda's a bit.
When I want the cuda's running full speed on any of mine I set it at 0.5 and BRP's only and I load mine set at 10 days since they are finished so fast (about 95 per day are loaded set at 10 days)
Never fails and my processors are not the newest and fastest either.
http://einsteinathome.org/account/82814/computers
(sorry about my typo's....I was looking at new cards and ram and psu's at the same time for one of my other hosts upgrades)
The 650 may be the one I get since it is so cheap right now ($109) so are the 750watt psu's
Alec and Magic: Okay -- it
)
Alec and Magic:
Okay -- it all makes sense now, and I think I have everything adjusted properly.
I returned the GPU utilization factor to 0.5, and I hope I did the logical thing by also returning the maximum CPU-count back to 100% to use all four cores. It looks (and feels) right, and it's nice to see 4 CPUs and 2 of the GPUs working away on the numbers.
I was amazed to learn that my order for another GTX650 already has a tracking number. By the middle of next week I'll have both of my i5 computers doing BRP tasks as well as the others.
Kudos and thanks! to all who responded and helped me to undertand how it all goes together. It's great to be a part of such a worthwhile project, and a terrific community.
-- Andy
Hi Andy, Sorry I couldn't
)
Hi Andy,
Sorry I couldn't answer before this but I've had an injection in my eye today and it takes a few hours before I can see enough to read stuff again :-).
From lots of experience with quite a few 650s you have the perfect settings now - 2x GPU tasks and all cores running CPU tasks. With the 650s, it seems to make no difference if you leave a CPU core free so I always run all cores. If you run a 650 3x, you actually seem to lose a little so certainly not worth it. You would need a much more powerful GPU with more GPU RAM to even think of running 4x or higher.
When you install in the second host, the GPU times should be the same. Small variations in the CPU power like yours will make no difference tp GPU crunch times. The CPU tasks will probably take slightly longer however.
Many thanks to you for being willing to upgrade your hosts like this. The project really does appreciate all the help it can get and is very grateful for your contribution.
Cheers,
Gary.
OK with my card it's the 2GB
)
OK with my card it's the 2GB card. 1 task runs roughly 33 mins while 2 tasks runs roughly 56 mins each and 3 tasks are around 95 mins to 100 mins. While at alot of times there is either little to no gain running 3 task at a time. I have found running only 2 task at a time is better while using my computer such as web, emails and other light stuff.
I plan on adding another 650Ti to my desktop within the next month or so. For the price and everything I do "play games at times and run E@H" the GTX 650 is a really great card.
Another 6 or so hours of traveling and I hope to be at my new home and will get my desktop re-set up and working away again, but for now it's ZZZ's time :)
PC setup MSI-970A-G46 AMD FX-8350 8 core OC'd 4.45GHz 16GB ram PC3-10700 Geforce GTX 650Ti Windows 7 x64 Einstein@Home
Hey, another relative newb
)
Hey, another relative newb jumping in on this. After following this thread, I've now got my 4GB 680 running two BRP apps at once and all 8 cores on my i7-3770 running. Should I also put % CPU time at 100 as well? This may be a stupid question, but if all cores are devoting 100% to BOINC projects, how will the computer perform any other tasks (web browsing, etc.)?
RE: Hey, another relative
)
If you use your computer and it goes over 25% usage on CPU Boinc will suspend working but not so on GPU work. While web-ing and what not mostly will not be affected while Boinc is working however flash based pages may load a bit slower but over all you can still do things. The refresh on screen may be a bit slow as well but will still work. Changing to 100% core's will make use of all cores and changing the CPU time can make more heat on CPU's and may also slow response time down a bit but you can still do online stuff. You can always try a setting and if it does not work well for you you can change it back. If you plan on gaming you should suspend Boinc or have it setup to suspend on game opening.
Boinc does not take over the computer and make it useless while it is running. It can slow down a computer but everything else should be fine for ya.
PC setup MSI-970A-G46 AMD FX-8350 8 core OC'd 4.45GHz 16GB ram PC3-10700 Geforce GTX 650Ti Windows 7 x64 Einstein@Home