GPU or CPU

[spikey_richie]

Hey gang, long time no post.

So I upgraded my rig and decided to run some BOINC on it, but for whatever reason it only sees 2 of my 8 cores... so I decided to move to F@H with -smp which works just fine.

Question though - I have an i7-2700k OC'd to 4.7ghz per core, and a GTS-450 GFX card. Should I fold on my GPU or CPU for better results?

[Dirk Broer]

That's either a mighty strange i7 or you have a too old BOINC version incapable of reporting your true CPU.
Try stock speed and start BOINC, post your first lines of the BOINC log while starting I'd say.
Then repeat the process for the OC'd settings.
As for crunching for best results: either both CPU and GPU or GPU alone for most projects.
If Folding sees your 4+4 cores (4 real, 4 hyper-threaded) it will be quite a cruncher @4.7Ghz

[NeoGen]

Hey spikey_richie
Long time no post indeed. Glad to see you around! :)

I would also check on the boinc manager, under the menu Tools => Computing preferences, on the processor usage tab look in the bottom and make sure that you have "On multiprocessor systems use at most 100% of the processors"
And also one thing I learned when I got my first >2 core machine was that in some projects I had to go to the account settings on the projects and set on the Computing Preferences the option "On multiprocessors, use at most" (and the number of cores). I'm looking at it right now on my BURP account settings and I have it set to 8 processors. Given the advances in technology I'm thinking I will soon have to increase this value if I get a new cruncher. :)

For Folding@Home, and pretty much all projects I think strong GPUs do laps around even the fastest CPUs, so if you really want you can set it for both CPU and GPU to get some extra points, but I myself would put the CPU crunching on something else.

That is a mighty machine you got there, push it to the maximum! :icon_mrgreen:

[Dirk Broer]

I would also check on the boinc manager, under the menu Tools => Computing preferences, on the processor usage tab look in the bottom and make sure that you have "On multiprocessor systems use at most 100% of the processors"
And also one thing I learned when I got my first >2 core machine was that in some projects I had to go to the account settings on the projects and set on the Computing Preferences the option "On multiprocessors, use at most" (and the number of cores).

Now that's some sound advice from NeoGen! I assumed you'd already checked that yourself, but you should never assume when giving advice...

[Brucifer]

I don't see any mention of which client you are using. Since the release of the v7.1.50 client I have been using that on my Fermi core GPU's. Also tried it on some smp dedicated cores. However I have just gone back to only using the GPU's. While the energy draw is higher on the gpu's they do put out a decent quantity of work. If you are one of the gents that is into the "science only" aspect, then run everything... Keep an eye on the f@h help forum as there is quite a bit of discussion there on hardware recommendations and requirements for certain projects and the bonus points with the SMP projects. The SMP high pointers were recently increased in hardware requirements which take a very serious server-grade system to run those projects. However they do put out massive points. At this point there aren't any bonus points for GPU's, but there is talk of it. To run f@h now for the points only thing, you have to have some serious hardware and also the bucks to put into your electrical bill.

My own primary motivation is the death of my mother-in-law due to complications encountered in her battle with Alzheimer's. Presently the Alzheimer's work is being done with the GPU's, and within that it falls into the cuda gpu's primarily -- at least as far as I've seen but I could easily be way wrong here. lol I have two GTX-560ti's and one GTX-460 GPU's that have been crunching mostly project 8031 work units that return 3843.40 points each. Now and then I also kick in a GTS-250 that is a really noisy thing, but I only run it when I feel I can tolerate the noise. lol -- that isn't too often. When I add the SMP cores to the GPU's, the heat generated really starts stressing my air conditioner, so I usually just run the gpu's. Then I run a few ATI gpu's on some rc5 as long as the heat doesn't get too bad. The ATI/AMD clients just aren't optimized on f@h at this point like they are on the cuda fermi units. That is another thing though that is being worked on and hopefully will yield better returns on wattage versus work_units completed. If I was a rich man, I'd put together a 32 core system and go after the big points, but I just can't see one or two of those systems in my budget in my retired life style and still be able to play with my other toys. :-)

So if you have cuda gpu's, then my recommendation to you would be to just run those on f@h unless you have the means to keep things cool. Of course f@h would like you throwing everything you have at it. lol

[spikey_richie]

BOINC client was boinc_6.12.34_windows_x86_64, I did as you suggested and after it has benchmarked it says something like available cores: 8, using: 2. I uninstalled it again so I can't give you the exact details...

For Folding, I'm using the v6.41 GPU3 client for GPU crunching, and/or the SMP2 v6.34 CPU client with -smp on.

I could crunch GPU and CPU, but it kills the temps so I'd rather do one or the other. I have the machine set to shutdown when any core hits 78°C, and with GPU & CPU going together it goes off pretty quick. I don't go about 75°C with an ambient of about 20°C and the fans set to auto @ 100% load. I'm slowly creeping up the AMD Users F@H leader board, but I'm only really crunching for 8-10 hours per day.

CPUz link is here: http://valid.canardpc.com/show_oc.php?id=2184848

[Brucifer]

to get the v7.5.1 client you need to go to the f@h form and the thread dealing with the open beta for v7. in there you can download the v7 client. You will get better points using that client than the 6.34 one.

[spikey_richie]

to get the v7.5.1 client you need to go to the f@h form and the thread dealing with the open beta for v7. in there you can download the v7 client. You will get better points using that client than the 6.34 one.

v7.5.1 CPU (SMP) or GPU?

I've fixed BOINC with your advice, so I'm going to go back to that.



27/03/2012 21:11:10 | | Starting BOINC client version 6.12.34 for windows_x86_64
27/03/2012 21:11:10 | | log flags: file_xfer, sched_ops, task
27/03/2012 21:11:10 | | Libraries: libcurl/7.21.6 OpenSSL/1.0.0d zlib/1.2.5
27/03/2012 21:11:10 | | Data directory: E:\ProgramData\BOINC
27/03/2012 21:11:10 | | Running under account Richard
27/03/2012 21:11:10 | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-2700K CPU @ 3.50GHz [Family 6 Model 42 Stepping 7]
27/03/2012 21:11:10 | | Processor: 256.00 KB cache
27/03/2012 21:11:10 | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 syscall nx lm vmx tm2 popcnt aes pbe
27/03/2012 21:11:10 | | OS: Microsoft Windows 7: Home Premium x64 Edition, Service Pack 1, (06.01.7601.00)
27/03/2012 21:11:10 | | Memory: 15.92 GB physical, 16.92 GB virtual
27/03/2012 21:11:10 | | Disk: 74.53 GB total, 67.41 GB free
27/03/2012 21:11:10 | | Local time is UTC +1 hours
27/03/2012 21:11:10 | | VirtualBox version: 4.1.8
27/03/2012 21:11:10 | | NVIDIA GPU 0: GeForce GTS 450 (driver version 29610, CUDA version 4020, compute capability 2.1, 1024MB, 421 GFLOPS peak)
27/03/2012 21:11:10 | climateprediction.net | URL http://climateprediction.net/; Computer ID 1208959; resource share 50
27/03/2012 21:11:10 | | General prefs: from http://bbc.cpdn.org/ (last modified 23-Aug-2006 05:09:46)
27/03/2012 21:11:10 | | Host location: none
27/03/2012 21:11:10 | | General prefs: using your defaults
27/03/2012 21:11:10 | | Preferences:
27/03/2012 21:11:10 | | max memory usage when active: 8150.55MB
27/03/2012 21:11:10 | | max memory usage when idle: 14671.00MB
27/03/2012 21:11:10 | | max disk usage: 1.00GB
27/03/2012 21:11:10 | | max CPUs used: 2
27/03/2012 21:11:10 | | don't use GPU while active
27/03/2012 21:11:10 | | suspend work if non-BOINC CPU load exceeds 25 %
27/03/2012 21:11:10 | | (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
27/03/2012 21:11:10 | | Not using a proxy
27/03/2012 21:11:11 | climateprediction.net | Sending scheduler request: To fetch work.
27/03/2012 21:11:11 | climateprediction.net | Requesting new tasks for NVIDIA GPU
27/03/2012 21:11:12 | climateprediction.net | Scheduler request completed: got 0 new tasks
27/03/2012 21:11:12 | climateprediction.net | Project has no tasks available
27/03/2012 21:13:48 | climateprediction.net | update requested by user
27/03/2012 21:13:52 | climateprediction.net | Sending scheduler request: Requested by user.
27/03/2012 21:13:52 | climateprediction.net | Requesting new tasks for CPU and NVIDIA GPU
27/03/2012 21:13:53 | climateprediction.net | Scheduler request completed: got 0 new tasks
27/03/2012 21:13:53 | climateprediction.net | Project has no tasks available
27/03/2012 21:13:53 | climateprediction.net | General prefs: from climateprediction.net (last modified 27-Mar-2012 21:13:39)
27/03/2012 21:13:53 | climateprediction.net | Computer location: home
27/03/2012 21:13:53 | climateprediction.net | General prefs: no separate prefs for home; using your defaults
27/03/2012 21:13:53 | | Preferences:
27/03/2012 21:13:53 | | max memory usage when active: 8150.55MB
27/03/2012 21:13:53 | | max memory usage when idle: 14671.00MB
27/03/2012 21:13:53 | | max disk usage: 1.00GB
27/03/2012 21:13:53 | | Number of usable CPUs has changed from 2 to 8.
27/03/2012 21:13:53 | | suspend work if non-BOINC CPU load exceeds 25 %
27/03/2012 21:13:53 | | (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)


My computing prefs were capped at 2 cores /blush.

So... next question. If I go back to BOINC, which project do you want me to throw this beast at?

[Dirk Broer]

Use the beast (8 cores at 4.7Ghz:shocked:) at your own discretion at any project you prefer, we'll see you soaring by I guess....

Where's my dodeca-core 6 Ghz piledriver when I need it?:BangHead:

[spikey_richie]

I put it onto some Collatz last night, and I had a ton of failed units (GPU I assume)... I think it's related to the post regards CUDA not working properly with the latest nVidia drivers :(

[Dirk Broer]

Not only do they make crappy cards, they bring out the matching drivers as well..