Will 4x pci slot effect cuda crunching


log in

Advanced search

Message boards : Number crunching : Will 4x pci slot effect cuda crunching

1 · 2 · Next
Author Message
Profile Paul D Harris
Volunteer tester
Send message
Joined: 1 Dec 99
Posts: 1123
Credit: 33,598,472
RAC: 0
United States
Message 1136454 - Posted: 5 Aug 2011, 21:34:31 UTC

I have asus p6t motherboard and the video slots are the 1st slot is 16x and the 2nd slot is 4x.I have 2 460 cards slot 1 runs at 16x and slot 2 runs at 4x and I can not change this in the bios. So my question is what lga 1366 boards have 2 16x slots and will the second slot running at 4x affect any cuda wu's slower than the 16x slot doing cuda wu's?
____________

Profile dnolanProject donor
Avatar
Send message
Joined: 30 Aug 01
Posts: 1255
Credit: 46,413,545
RAC: 27,253
United States
Message 1136460 - Posted: 5 Aug 2011, 21:47:49 UTC
Last modified: 5 Aug 2011, 21:51:42 UTC

I can't definitively say if it will affect Nvidia Seti crunching, but I have a board running X16 X4 that I do Mikyway on ATI HD 5870 cards in both slots, and it does not seem to affect the speed of the second card for that project, making me think it shouldn't have a huge effect here either, but can't say for sure. Right now I am doing Seti on the X4 card in this system here, so if you feel like it, you could try to find another system with a 5870 in it and compare times to mine. As for a 1366 board with 2 X16 slots, the other system I have uses an Asus Sabertooth X58 board that has 2 X16 and one X4, it's a pretty good board and not that expensive. I'm sure others will chime in with other choices.

-Dave

[edit] added link to the system with an x4 card in it, it still has 2 cards in it, but I have it excluding device 0 via cc_config to keep heat and power down. My x58 system has not been crunching lately, but that one is here if you want to look at any of its info...
____________

OzzFan
Volunteer tester
Avatar
Send message
Joined: 9 Apr 02
Posts: 13625
Credit: 31,071,022
RAC: 20,353
United States
Message 1136468 - Posted: 5 Aug 2011, 21:55:00 UTC - in response to Message 1136454.

When it comes to crunching, there is little difference between an x4 and an x16 slot for a card. Simply make sure that the x4 slot is x16 size in length, since an x16 card will not physically fit in an x4 slot.




I don't mean to split hairs, but to enlighten and education, I feel it is worth pointing out the difference in using the 'x' before or after the numerical designation.

For instance, when talking about AGP, the 'x' after the number meant it was that much faster than the first bus speed; 8x meant it was 8 times faster than the first speed (32bit x 66MHz). Or when talking about CD/DVD/HD/Blu-Ray discs, the 'x' after the number means it is that many times faster than the original speed, for CD-ROMs 1x was 150KiB/sec, so 2x would be 300KiB/sec and 4x would be 600KiB/sec. For DVD-ROMs 1x is roughly 1.32MiB/sec, 2x is 2.64MiB/sec, etc.


When talking about PCI Express, the 'x' goes before the number to signify how many PCIe "lanes" there are, as each lane has a fixed speed and bitrate. Therefore it should be written as x1, x2, x4, x8, x16 and so on.

Profile Lint trapProject donor
Send message
Joined: 30 May 03
Posts: 867
Credit: 27,405,902
RAC: 24,694
United States
Message 1136478 - Posted: 5 Aug 2011, 22:11:06 UTC - in response to Message 1136454.


...will the second slot running at 4x affect any cuda wu's slower than the 16x slot doing cuda wu's?


It will some.

My ancient P43 board has 1 x16 (ver 2) and 1 x4 (ver 1) slot and I had my 8800 GTS in the x4 slot for a while. The slow slot had maybe a 10% impact on MB crunching.

The faster your card is or the more demanding (data wise) the application, the bigger impact you could possibly see when using a slower slot.

Crunching the same Nvidia AP with a 460 using a 16x(v2) or 4x(v1) could make quite a difference, I think.

Lt

Profile Wiggo
Avatar
Send message
Joined: 24 Jan 00
Posts: 7358
Credit: 96,965,126
RAC: 66,615
Australia
Message 1136479 - Posted: 5 Aug 2011, 22:22:08 UTC - in response to Message 1136478.

I run 2 setups here with 2x 9800GT's in them and they both have a 16x/4x slot combination but there is absolutely no difference in crunching between them no matter which slot they are in. ;)

Cheers.
____________

Profile Paul D Harris
Volunteer tester
Send message
Joined: 1 Dec 99
Posts: 1123
Credit: 33,598,472
RAC: 0
United States
Message 1136481 - Posted: 5 Aug 2011, 22:23:44 UTC

The only problems I am having is that the the newer drivers crashes my system unless I run the supplied driver 258.96 when I run the newer drivers 26x.xx drivers I have blue screens and dump and or a black screens. The second card when doing cuda wu runs around 93F though is hot asus web site says 105F it can run.
____________

Profile Lint trapProject donor
Send message
Joined: 30 May 03
Posts: 867
Credit: 27,405,902
RAC: 24,694
United States
Message 1136486 - Posted: 5 Aug 2011, 22:41:02 UTC - in response to Message 1136479.
Last modified: 5 Aug 2011, 22:48:21 UTC

I run 2 setups here with 2x 9800GT's in them and they both have a 16x/4x slot combination but there is absolutely no difference in crunching between them no matter which slot they are in. ;)

Cheers.



Of course impact will also depend some on the characteristics of each particular system, cpu, ram, etc.
Lt
edited...

Profile Paul D Harris
Volunteer tester
Send message
Joined: 1 Dec 99
Posts: 1123
Credit: 33,598,472
RAC: 0
United States
Message 1136495 - Posted: 5 Aug 2011, 23:42:05 UTC - in response to Message 1136481.

The only problems I am having is that the the newer drivers crashes my system unless I run the supplied driver 258.96 when I run the newer drivers 26x.xx drivers I have blue screens and dump and or a black screens. The second card when doing cuda wu runs around 93F though is hot asus web site says 105F it can run.

Edit of my post
I got the F and C confused the temps are in Celsus and not Fahrenheit and Nvidia instead asus and it is 104C max temp..
____________

Profile dnolanProject donor
Avatar
Send message
Joined: 30 Aug 01
Posts: 1255
Credit: 46,413,545
RAC: 27,253
United States
Message 1136507 - Posted: 6 Aug 2011, 0:24:36 UTC

Don't know what video card you're using, but generally speaking, 93C is pretty darn toasty to run a video card at continuously. I wouldn't do it unless I was just waiting for a card to die so I could get something new, especially if the max is only about 10C above that. I also assume that's the GPU temp, which means the VRM temps are probably somewhat higher than that. I don't like to run my cards 24/7 over mid-70s, but I also would like them to last a good long while, given that I paid a fair amount for them. If it was my setup, I'd look at what I could do to bring the temp down some.

-Dave
____________

Profile BMaytum
Volunteer tester
Avatar
Send message
Joined: 3 Apr 99
Posts: 100
Credit: 3,866,873
RAC: 3,983
United States
Message 1136596 - Posted: 6 Aug 2011, 5:00:21 UTC - in response to Message 1136507.

Don't know what video card you're using, but generally speaking, 93C is pretty darn toasty to run a video card at continuously. I wouldn't do it unless I was just waiting for a card to die so I could get something new, especially if the max is only about 10C above that. .... If it was my setup, I'd look at what I could do to bring the temp down some.

-Dave


I fully agree with Dave on this, you REALLY should explore ways to reduce GPU temperature!

I use Asus' SmartDoctor II to control (& monitor) my Asus ENGTX580 videocard (I use it to crunch 4 Main Beam WUs simultaneously). I've configured Fan Control to step up vidcard fan speed as GPU temperatures increase, and my GPU temperature stays below 60C (140F) when crunching full load. Note: SmartDoctor requires you have an appropriate Asus vidcard - I've read there are other 3rd party free utilities to control Fan speed and monitor GPU temperatures - maybe others here can point you to some.
____________
Sabertooth Z77, i7-3770K@4.2GHz, GTX680, W8.1Pro x64
P5N32-E SLI, C2D E8400@3Ghz, GTX580, Win7SP1Pro x64 & PCLinuxOS2014 x86

Profile Paul D Harris
Volunteer tester
Send message
Joined: 1 Dec 99
Posts: 1123
Credit: 33,598,472
RAC: 0
United States
Message 1136601 - Posted: 6 Aug 2011, 5:12:47 UTC - in response to Message 1136597.

I have some 120mm fans positioned in my rigs in front of the GPU cards to force feed their cooling fans some outside air.

How these things can run this hot and still survive still astounds me.

I run side off to try to get them the coolest ambient air possible.

I don't understand how you folks can cool them within a closed case.

But.....I am the local Seti nutcase, and will do whatever it takes to get the most out of my rigs that I can.

Meow.

I have a 120 fan blowing on mine with the side panel off and it still runs this hot. I bought this just to crunch with and it matches my first card. It's an old case with the psu at the top. I think I need a new case with the psu at the bottom to give some air around the card as it is now it sits at the bottom of the case and not much air flows around it because it sits at the bottom of the case.
____________

Profile Paul D Harris
Volunteer tester
Send message
Joined: 1 Dec 99
Posts: 1123
Credit: 33,598,472
RAC: 0
United States
Message 1136609 - Posted: 6 Aug 2011, 5:22:35 UTC - in response to Message 1136607.

Mine is the lower one that gets hot from the top card if I had some space at the bottom of the case I think it would run cooler.
____________

Profile BMH
Avatar
Send message
Joined: 27 May 99
Posts: 329
Credit: 93,465,912
RAC: 97,219
United Kingdom
Message 1136647 - Posted: 6 Aug 2011, 8:15:32 UTC

Also, regular dust clearing from the heatsink of that bottom card in particular may be in order with it being sat in the bottom of your case. Once heasinks become clogged temperatures quickly rise.
____________
Brian.

Profile Paul D Harris
Volunteer tester
Send message
Joined: 1 Dec 99
Posts: 1123
Credit: 33,598,472
RAC: 0
United States
Message 1136944 - Posted: 6 Aug 2011, 19:39:01 UTC - in response to Message 1136647.

Also, regular dust clearing from the heatsink of that bottom card in particular may be in order with it being sat in the bottom of your case. Once heasinks become clogged temperatures quickly rise.


That's very good thinking there.
I have since updated to the new beta driver 280.19 and my temps have dropped to upper card is 60C and the bottom is at 71C and I am doing 2 cuda ap wu's per card also so the new driver is cooler on my cards.

____________

Terror Australis
Volunteer tester
Send message
Joined: 14 Feb 04
Posts: 1717
Credit: 205,856,298
RAC: 28,071
Australia
Message 1137163 - Posted: 7 Aug 2011, 10:54:06 UTC

PCIE numbers seem to make little difference in my opinion. I run a number of multi card computers of interest here, one has 2xPCIE 16 sockets running at 16x and 4x. Two other boards have 3 slots running at 8x, 8x and 1x.

I have not been able to detect any real difference in the crunching times between those cards on the faster slots and those in the slower slots. If there is any difference, it is less than the varieations in crunching times that naturally occurs between units.

T.A.

zoom314Project donor
Avatar
Send message
Joined: 30 Nov 03
Posts: 46521
Credit: 36,866,624
RAC: 4,974
United States
Message 1137273 - Posted: 7 Aug 2011, 16:43:00 UTC

I have an x16 and an x4 pcie slot in this old pc, with a pair a of GTX295 single pcb cards in Her that I run 6 hours a night, they do pretty well really.
____________
My Facebook, War Commander, 2015

Profile John Clark
Volunteer tester
Avatar
Send message
Joined: 29 Sep 99
Posts: 16515
Credit: 4,418,829
RAC: 0
United Kingdom
Message 1137280 - Posted: 7 Aug 2011, 16:54:57 UTC - in response to Message 1137163.

PCIE numbers seem to make little difference in my opinion.


I tend to agree with T.A. on the basis that all the crunching work is done on the NV card RAM. The only time the PCI-E bandwidth is tested is when the data is transferred to the GPU for crunching, and any small support the CPU gives.

The only times PCI-E bandwidth is, therefore, tested is when data is uploaded to crunch and completed results is downloaded.

____________
It's good to be back amongst friends and colleagues



Profile [seti.international] Dirk SadowskiProject donor
Volunteer tester
Avatar
Send message
Joined: 6 Apr 07
Posts: 7102
Credit: 60,948,168
RAC: 17,194
Germany
Message 1137311 - Posted: 7 Aug 2011, 17:55:39 UTC
Last modified: 7 Aug 2011, 17:58:34 UTC

For 2 1/2 years, after S@h had the first CUDA app, someone made a test with his IIRC 9800 GPU and different PCIe slot speeds.
The forum search is limited one year, so not possible to find again this thread.

He had the GPU at PCIe 1.0 x16 speed.. also at
x8 and saw 3 % loss..
x4 and saw 10 % loss..

The only way for to stop the discussion and find a conclusion about PCIe speed and S@h CUDA.. would be if someone have a mobo with different PCIe slot speeds and the time for to make a offline test with bench WUs.. (for to repeat this test).

Or, someone which have same GPUs at different slot speeds in one machine (if this would be the same result)..


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC. - SETI@home needs your help. -
____________
BR

SETI@home Needs your Help ... $10 & U get a Star!

Team seti.international

Das Deutsche Cafe. The German Cafe.

1 · 2 · Next

Message boards : Number crunching : Will 4x pci slot effect cuda crunching

Copyright © 2014 University of California