I just wanted to ask which Cuda version is recommended for my top notch highend card, 2.3 or 3.2 ? Also, is there any specific driver which I should use? Atm I'm running 331.65

---

I just wanted to ask which Cuda version is recommended for my top notch highend card, 2.3 or 3.2 ? Also, is there any specific driver which I should use? Atm I'm running 331.65

For your GeForce G210 I'd use the Cuda23 app, for drivers there, isn't that much of a difference for that app, for the Cuda32, Cuda42 and Cuda50 apps there was a large slowdown for legacy GPU when you installed drivers later than 301.xx,
we believe Microsoft/Nvidia introduced virtualisation into the drivers after that, that legacy GPUs don't support, hence the slowdown, Cuda23 apps weren't effected.

Claggy

Thank you, Claggy!

I was running the 3.2 version with my current driver and it felt rather slow to me. Ofc I'm aware that this card is not really fast but still it felt slower then in the old days (I was away from S@H for quite some time). As you explained, the combination of a rather new driver and version 3.2 might be a problem. I'm back now at 2.3. Only for a few hours now, so I can't say if it's really faster or not but the gpu gets hotter, so that may be a sign for it running faster also.

---

You're done a Shortie with the Cuda32 and the Cuda23 app, while not exaxtly the same AR, they should be close enough:

The Cuda32 app took 1 hours 39 min 15 sec:

http://setiathome.berkeley.edu/result.php?resultid=3617543008

The Cuda23 app took 1 hours 32 min 39 sec:

http://setiathome.berkeley.edu/result.php?resultid=3617684001

Claggy

[quote]for the Cuda32, Cuda42 and Cuda50 apps there was a large slowdown for legacy GPU when you installed drivers later than 301.xx,
Claggy

I was thinking of looking up this info. Thank you for mentioning it. I got 331.82 uninstalled and 301.42 up and running.

---

for the Cuda32, Cuda42 and Cuda50 apps there was a large slowdown for legacy GPU when you installed drivers later than 301.xx,
Claggy

I was thinking of looking up this info. Thank you for mentioning it. I got 331.82 uninstalled and 301.42 up and running.

You're not running a legacy GPU!!!!!!!!!!!!!!!!!

GTX460 is a Fermi, it is not a legacy GPU!!!!!!!!!!!!

So no large slowdown on your GPU!!!!!!!

Claggy

Ok, now I'm confused. Can we control which cudaXX we get? If so would someone point me to the url about it?

I have a G210 on order to replace a ATI that wasn't a high enough version. So I am very interested in the question.

Thanks,
Tom

---

A proud member of the OFA (Old Farts Association).

Ok, now I'm confused. Can we control which cudaXX we get? If so would someone point me to the url about it?

I have a G210 on order to replace a ATI that wasn't a high enough version. So I am very interested in the question.

Thanks,
Tom

Just read the "Optimised Applications and Other Binaries, 2nd edition - Read Only" pinned near the top of this forum section. ;-)

Cheers.

Ok, now I'm confused. Can we control which cudaXX we get? If so would someone point me to the url about it?

I have a G210 on order to replace a ATI that wasn't a high enough version. So I am very interested in the question.

Thanks,
Tom

We sort of can.

Here is what I did. I used optimized application developed by Lunatics group. When you install the application, it lets you select which CudaXX to use.

Like Claggy noted I have GTX 460 card, it uses Fermi chip and supposed to work best with Cuda42 (CUDA 4.2). When I installed the Lunatics optimized application, I selected to use Cuda42. The application creates app_info.xml file. This file tells boinc manager how to run the project.

---

Ok, now I'm confused. Can we control which cudaXX we get? If so would someone point me to the url about it?

I have a G210 on order to replace a ATI that wasn't a high enough version. So I am very interested in the question.

Thanks,
Tom

We sort of can.

Here is what I did. I used optimized application developed by Lunatics group. When you install the application, it lets you select which CudaXX to use.

Like Claggy noted I have GTX 460 card, it uses Fermi chip and supposed to work best with Cuda42 (CUDA 4.2). When I installed the Lunatics optimized application, I selected to use Cuda42. The application creates app_info.xml file. This file tells boinc manager how to run the project.

The Lunatics installer allows you to select the version, right or wrong.
There is some good readme guidance on what version and settings should work best for most cards.

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

I didn't see a great deal of difference between 2.3 and 3.2 on my GT 8500. So I just went with 3.2. I also haven't bothered to update the drive since 306.23

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

I am running a couple of GeForce 210's. With stock Seti.
And it is not clear if this card is "pre-Fermi" or "post-Fermi". I looked at the link into the GeForce website but there was no specific displayed on that page.

I want to set my *.cfg files for each level of cuda22, 23, etc. It looks as if I need to change things depending on the card. So is it pre-Fermi etc/

Thanks,

Tom

---

A proud member of the OFA (Old Farts Association).

I am running a couple of GeForce 210's. With stock Seti.
And it is not clear if this card is "pre-Fermi" or "post-Fermi".

It's Pre-Fermi, Fermi's are compute capability 2.0 and 2.1, your G210 is compute capability 1.2:

<stderr_txt>
setiathome_CUDA: Found 1 CUDA device(s):
Device 1: GeForce 210, 1024 MiB, regsPerBlock 16384
computeCap 1.2, multiProcs 2
pciBusID = 1, pciSlotID = 0
clockRate = 1402 MHz
In cudaAcc_initializeDevice(): Boinc passed DevPref 1
setiathome_CUDA: CUDA Device 1 specified, checking...
Device 1: GeForce 210 is okay

Claggy

Thank you Claggy. Once again you are a font of information!

As you know in a stock Seti you a likely to get any of the various CUDAxx's (at least I have been).

And I have a *.cfg file for each one of the CUDAxx apps.

What should I be putting in them?

a) Leave blank?
b) Set the priority to "abovenormal" but otherwise leave it at the defaults (eg. blank)?
c) All the sample files have higher numbers, that based on "Pre-Fermi" are probably not useful.

Here is an example of the config file.
-------------------------------------------------------------------
;;; This configuration file is for optional control of Cuda Multibeam x41zc
;;; Currently, the available options are for
;;; application process priority control (without external tools), and
;;; per gpu priority control (useful for multiple Cuda GPU systems)
[mbcuda]
;;;;; Global applications settings, to apply to all Cuda devices
;;; You can uncomment the processpriority line below, by removing the ';', to engage machine global priority control of x41x
;;; possible options are 'belownormal' (which is the default), 'normal', 'abovenormal', or 'high'
;;; For dedicated crunching machines, 'abovenormal' is recommended
;;; raising global application priorities above the default
;;; may have system dependant usability effects, and can have positive or negative effects on overall throughput
processpriority = abovenormal
;;; Pulsefinding: Advanced options for long pulsefinds (affect display usability & long kernel runs)
;;; defaults are conservative.
;;; WARNING: Excessive values may induce display lag, driver timeout & recovery, or errors.
;;; pulsefinding blocks per multiprocessor (1-16), default is 1 for Pre-Fermi, 4 for Fermi or newer GPUs
pfblockspersm = 1
;;; pulsefinding maximum periods per kernel launch (1-1000), default is 100, as per 6.09
pfperiodsperlaunch = 200

;[bus1slot0]
;;; Optional GPU specifc control (requires Cuda 3.2 or newer app), example
;processpriority = abovenormal
;pfblockspersm = 8
;pfperiodsperlaunch = 200

The last 3 listed are the "defaults" for the sample file. eg. abovenormal, 8, 200.
-----------------------------------------------------------------------
This is one I have for CUDA22. What should it be?

I am running this board on an Intel Duo 2 cpu (without hyper threading) that processes an average baseline seti wu every 4 odd hours on each cpu. The gpu unit runs on a nominal .24 cpu in the app_config.xml file because anything higher takesover a cpu. So what is my "best choice" for optimal "thruput"? 2 baseline wu and a slow (maybe 5 hours) CUDAxx unit? Or 1 baseline wu and devote the other cpu to trying to run the CUDAxx unit(s) faster?

I haven't been able to get the Lunatics benchmark to finish running. It errors out when trying to copy .dll files.

Thanks,
Tom Miller

---

A proud member of the OFA (Old Farts Association).

We sort of can.

Here is what I did. I used optimized application developed by Lunatics group. When you install the application, it lets you select which CudaXX to use.

Like Claggy noted I have GTX 460 card, it uses Fermi chip and supposed to work best with Cuda42 (CUDA 4.2). When I installed the Lunatics optimized application, I selected to use Cuda42. The application creates app_info.xml file. This file tells boinc manager how to run the project.

I wonder if creating an app_info.xml file for the stock CUDA apps would give me any more control? I could "ask" for cuda22/23?

Thanks,
Tom

---

A proud member of the OFA (Old Farts Association).

Just curious Tom, why don't you run the optimised apps?

---

I wonder if creating an app_info.xml file for the stock CUDA apps ...

Stock CUDA apps are identical these days to the optimized (that are inside Lunatics_Win??_v0.41_setup.exe)
(the .exe files differ in filenames and size - the stock is UPX compressed)

Right-Click your Stock CUDA apps .exe -> Properties ...

Screenshots from Lunatics_x41zc_win32_cuda32.exe and this stock file:
http://boinc2.ssl.berkeley.edu/sah/download_fanout/setiathome_7.00_windows_intelx86__cuda32.exe




 

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

BilBg I didnt know that they were exactly the same. That would explain a lot Of what Im seeing running my stock experiement. My I7 920 with a GTS250 had an avergae rack of 5400 running MB and AP with HT on running lunatics. I am nearly done with running 4 cores HT off with the GTS250 and my rack is nearing a peak of about 4250 running stock MB only no AP. My next test will be HT on. Im thinking I will get close to that 5400 Rac.

---

[/quote]

Old James

"Why aren't you running optimized apps?"

I have had varying results probably due to hardware issues with the optimized version(s). It also might be due to my lack of knowledge. So for now at least I am running just the stock stuff.

I have no pretensions of trying to become a "top" contributor. I am a LONG time Seti participant. I remember installing the windows screen saver and calling up with a modem every week or two.

So I am trying to keep this within the realm of being a hobby, not another part-time job. I think I have a machine that has been doing Seti crunching since 2006? I have just ordered a replacement cpu chip with "2!" cpu cores on it. Right now that machine (a HP dx2200) has far more total credits earned than anything else I own. But its average production rate is pretty low compared to the nearest next up machine. I don't want it to get too easily buried. I want my two faster machines to have to work for it :)

I might experiment with running the optimized stuff on a netbook that I have just started running Seti on. Currently it is tied up running 2 copies of Astropulse. So I can't even hardly process anything but gpu-based stuff right now. By about Tuesday they should both be done. I have turned off Astropulse for the time being. I want to run basic Seti on the netbook/netbook gpu for a while.

For your bemusement the Netbook has twice the Gigaflops on the GPU as the G210 card. But 1/4 the memory. So its not surprising to me that it is running at maybe half the speed. If I could figure out how to make the cmdline parms make it work faster that would be wonderful. So far I have shooting in the dark..

Thanks for asking.

Tom

---

A proud member of the OFA (Old Farts Association).

"Why aren't you running optimized apps?"

I have had varying results probably due to hardware issues with the optimized version(s). It also might be due to my lack of knowledge. So for now at least I am running just the stock stuff.

I have no pretensions of trying to become a "top" contributor. I am a LONG time Seti participant. I remember installing the windows screen saver and calling up with a modem every week or two.

So I am trying to keep this within the realm of being a hobby, not another part-time job. I think I have a machine that has been doing Seti crunching since 2006? I have just ordered a replacement cpu chip with "2!" cpu cores on it. Right now that machine (a HP dx2200) has far more total credits earned than anything else I own. But its average production rate is pretty low compared to the nearest next up machine. I don't want it to get too easily buried. I want my two faster machines to have to work for it :)

I might experiment with running the optimized stuff on a netbook that I have just started running Seti on. Currently it is tied up running 2 copies of Astropulse. So I can't even hardly process anything but gpu-based stuff right now. By about Tuesday they should both be done. I have turned off Astropulse for the time being. I want to run basic Seti on the netbook/netbook gpu for a while.

For your bemusement the Netbook has twice the Gigaflops on the GPU as the G210 card. But 1/4 the memory. So its not surprising to me that it is running at maybe half the speed. If I could figure out how to make the cmdline parms make it work faster that would be wonderful. So far I have shooting in the dark..

Thanks for asking.

Tom

Im not running Lunatics as a test of my machine running stock. A I7 920 with a GTS 250 nidia card in it. This machine has been crunching since Aug 2009.
I just wanted to test out the various configurerations to see what gives me the best rac. See my Running stock experiment thread down below. When I am done doing the test I will summarize what I found out about my machine.

The short summary is, I started with lunatics running MB and AP with ht on and the GPU. RAC was 5400. right now RAC is 4328 running HT off and the GPU and CPU doing stock MB only. Im still waiting for rac to flat line. Before I go to HT on.

At the moment Im not a fan of doing SETI on my phone, Wifes laptop, or my kindle. They all run way to hot for me in normal use. Why would I want to stress them further. But thats my choice. And I dont begrudge anyone else who wants to.

---

[/quote]

Old James