Back after some inactivity, a few questions regarding GPU and CPU
28 Apr 2019 0:15:48 UTC
Topic 218747
(moderation:
Hello!
I've been spending the last year (or more) crunching for World community grid, and decided it was time to branch out a little. I have a few questions.
1. I have an older Toshiba laptop with an i5-3317U processor. In theory, this mean that it can crunch with both the CPU and Intel HD Graphics. Which would be better science wise? I'd rather not use both, as the CPU throttles quite heavily when both it and the GPU are in use and it does produce quite a lot more heat.
2. On a similar note. I have a Lenovo x1 Extreme on order. I'd like to maximize it's lifespan as well as contribute here. It does have an Nvidia GTX 1050TI, so I'll probably only be using that for crunching so it doesn't get too hot, but are there any particular laptop coolers folks recommend for crunching?
3. This leads me to my final question. Having a look through account settings, it doesn't appear I can disable the Intel GPU for use on a specific machine. E.g. if I only want the Intel GPU to be used on the toshiba laptop, but not on the Lenovo one (I'm not entirely sure if the Nvidia GPU can be used on top of Intel) or how much extra heat that will produce.
Am I missing something here?
Any other hints or tips folks can provide to get me started, apart from the very excellently written topic in the forum regarding new / existing users returning?
On average, how much memory, both in disk size and ram, does a task take up? I have a Ryzen machine here and Climateprediction was using a ton of virtual memory. I'm guessing that won't be an issue here, but just double checking.
any help appreciated!
Thank you!
wolfman1360 wrote:3. This
)
https://einsteinathome.org/account/prefs/project
There are four different preference sets available (generic, home, school, work). You can use those venues to save different settings for four different kind of computation scenarios. Then you can mate each of your hosts individually to any of those four preference sets by choosing a corresponding 'Location' at each host details page.
Current RAM or VRAM requirements per task (approximately):
FGRPB1G: 800-900 MB
FGRP5: 580 MB
O1OD1E: 170 MB
O2AS20-500: 340 MB
Each task will require only a few MB or maybe couple of dozen MB on disk. Not much anyway. Einstein is NOT one of those projects that heavily loads your disk.
Thank you so much for this
)
Thank you so much for this information.
I'm blind and using a screen reader here, so exploring more I was able to find Generic along with all of the other profiles. I'm assuming by clicking on a computer name I can set which profile it runs off of?
Am I to assume the higher ram projects are meant for higher end CPUs? I have a Ryzen 7 1800x here that's currently crunching some O2AS20-500. It looks like it's going to take just about a day to complete. I also have my RX570 crunching on FGRPB1G. Is this the only GPU project that currently receives work? I see nothing else regarding GPU, so I'm assuming FGRP5 and O1OD1E are all run on various CPUs (the latter lower end Atoms, which I also own a few of, as well as Android devices)?
Sorry if I'm overcomplicating all of this.
Regarding my AMD GPU. Is there any real benefit to running this one in particular with 2 WUs? That would mean 12% or so CPU would be reserved for it if I'm reading that right, not including any other CPU windows needs in the background. Would I see any performance gains or should I leave it as is? Right now, I believe I have the CPU set to 93% usage so the GPU has enough to run optimally. It looks like it's taking just about 30 minutes to complete each WU.
Again, I may be overthinking this. I'd rather be safe than sorry though.
wolfman1360 wrote:exploring
)
Yes. That 'Location' setting can be found at bottom of host page. For example one of your hosts:
https://einsteinathome.org/host/12775393
'Location' is an identical type of drop-down-menu as is the box for choosing a preference set. To summarize... Location = Preference set. Your host will get information about the new location only after Boinc makes a contact with project server. It would be a good idea to hit "Update" for Einstein on the Boinc client.
O2AS20-500 is currently the slowest CPU app. Those tasks will take the most time to complete. I don't know anything about how Ryzen's will behave when running many concurrent tasks or even maximum possible, but if you're running plenty of them then run time per task might get slower. That's how it has been with older platforms.
I think it's basically up to user what kind of run times he likes to see. I mean, as far as I know, none of the Einstein work is chopped to different size of tasks based on the host speed. GPU and CPU tasks from one app might have been chopped that way in the past to keep run times more reasonable between those different targets. But purely within an application... server doesn't divide the machines in any groups based on their speed. All traditional computers (Windows, Linux, MAC) are at the same line. Then it's just up to individual machine specs and computational power how fast a task will run.
"Smart phones" (Raspberry Pi, Andoid etc.) is a different group, but currently those hosts can get only BRP4 work. Traditional computers can't get work from BRP4.
O1OD1E was "Engineering" beta test app and had both CPU and GPU work, but this app is somewhat taken back to shipyard and at least partially shut down. Some hosts have stopped receiving tasks from it though some still seem to receive at least GPU tasks.
FGRPB1G is the only GPU app on the normal stuff. All other apps are for CPU.
Sorry, I don't know how that GPU will perform. It would be best to try out both scenarios. Changing from 1x to 2x can mean additional fan noise and heat. How well tasks would validate... that is also something to keep an eye on.