GTX 590 a good choice ?


log in

Advanced search

Message boards : Number crunching : GTX 590 a good choice ?

Author Message
SupeRNovA
Volunteer tester
Avatar
Send message
Joined: 25 Oct 04
Posts: 131
Credit: 9,238,042
RAC: 0
Bulgaria
Message 1316036 - Posted: 16 Dec 2012, 15:10:33 UTC

Hello,

it is time to add one more card to my setup of 2 heavily overclocked EVGA GTX 295 water cooled cards. I'm thinking to buy Point of View / TGT GTX 590 with water block so i can add it on my setup. Is this a good choice or wait to see what is the performance of the new GTX 780 in Q1 2013 ? What do you think ?

Regards

____________

Team kizb
Send message
Joined: 8 Mar 01
Posts: 219
Credit: 3,709,162
RAC: 0
Germany
Message 1316042 - Posted: 16 Dec 2012, 15:23:00 UTC - in response to Message 1316036.
Last modified: 16 Dec 2012, 15:38:08 UTC

Putting a 590 in with 295's isn't a good idea as your 295s can only handle one WU per a GPU core (4 total at a time in your case) where the 590 could handle 2 or 3 WU's at a time per a GPU core. So you wouldn't be getting the optimal performance out of the 590.

I'd suggest putting the 590 in its own computer or pull the 295s and just run the single 590. According to FRED'S LIST a GTX 295 will do 1 WU in 130 sec and a GTX 590 will do 1 WU in 53 sec. So based on that you would get 2 WUs from the 295 every 260 sec and 2 WUs every 106 sec from the 590 if you where only running 1 WU at a time on each GPU core. Based on this the 590 would be able to do 4 WU's every 212 sec beating out your pair of 295's by 48 sec to complete the same number of WU's.

The thing I'm not sure about on Fred's list is if the numbers are based on only running 1 WU at a time on all the cards or are the results based off the optimal WU's per a card. Maybe someone else will be able to answer that question.
____________
My Computers:
Blue Offline
Green Offline
Red Offline

juan BFBProject donor
Volunteer tester
Avatar
Send message
Joined: 16 Mar 07
Posts: 5472
Credit: 313,437,458
RAC: 107,873
Brazil
Message 1316046 - Posted: 16 Dec 2012, 15:25:39 UTC
Last modified: 16 Dec 2012, 15:27:21 UTC

I have both 590/690, air cooled, my opinion 590 works to hot, to much trouble to keep it cool, and uses a lot of power, the 690 is better, produces almost the same RAC and works with less power and is cooler (but not to much, stills needs extra fans when used for crunching). In wather cooled you r only concers is your electric bill, the 590 uses about 25% more electric power to do the same job.

The 780 apears to be a good promise but you are comparing 2xGPU´s with single GPU (780) probabily they will launch a 790 with a 2xGPU version. The 15% increase is against the 680 not the 690.

The 2xGPu´s are great if you have few slots avaiable, or whant to buil an super-cruncher with few on them on a single MB, and knows how to handle the heat, otherwise a 2x670 for example is chaper and actualy produces more RAC and are easy to handle than a single 690.

But keep in mind, each case is a unique case, in my case that´s is my opinion, from a user point of view not a technical opinion.
____________

zoom314Project donor
Avatar
Send message
Joined: 30 Nov 03
Posts: 46801
Credit: 37,000,373
RAC: 2,765
United States
Message 1316069 - Posted: 16 Dec 2012, 15:48:58 UTC

Speaking of the devil, here's some tidbits on the 700 series, there may be 2 dual gpu cards, the GTX790 and the lower powered GTX785...
____________
My Facebook, War Commander, 2015

juan BFBProject donor
Volunteer tester
Avatar
Send message
Joined: 16 Mar 07
Posts: 5472
Credit: 313,437,458
RAC: 107,873
Brazil
Message 1316075 - Posted: 16 Dec 2012, 15:58:22 UTC - in response to Message 1316069.
Last modified: 16 Dec 2012, 16:00:30 UTC

Speaking of the devil, here's some tidbits on the 700 series, there may be 2 dual gpu cards, the GTX790 and the lower powered GTX785...

Oh noo... here we go again... 500 x 600 (past) ... 600 x 700 (next year) ... 700 x 800 int 2014? Thake a look...

http://videocardz.com/38693/nvidia-sticks-with-tsmc-for-20nm-maxwell-geforce-800-series

But the old question remains...WHy? If the server actualy can´t feed a 2x690 host (I hate the 100 limit!), imagine with 2x or more 790 ...
____________

SupeRNovA
Volunteer tester
Avatar
Send message
Joined: 25 Oct 04
Posts: 131
Credit: 9,238,042
RAC: 0
Bulgaria
Message 1316078 - Posted: 16 Dec 2012, 16:01:02 UTC - in response to Message 1316042.
Last modified: 16 Dec 2012, 16:21:49 UTC

Putting a 590 in with 295's isn't a good idea as your 295s can only handle one WU per a GPU core (4 total at a time in your case) where the 590 could handle 2 or 3 WU's at a time per a GPU core. So you wouldn't be getting the optimal performance out of the 590.

I'd suggest putting the 590 in its own computer or pull the 295s and just run the single 590. According to FRED'S LIST a GTX 295 will do 1 WU in 130 sec and a GTX 590 will do 1 WU in 53 sec. So based on that you would get 2 WUs from the 295 every 260 sec and 2 WUs every 106 sec from the 590 if you where only running 1 WU at a time on each GPU core. Based on this the 590 would be able to do 4 WU's every 212 sec beating out your pair of 295's by 48 sec to complete the same number of WU's.

The thing I'm not sure about on Fred's list is if the numbers are based on only running 1 WU at a time on all the cards or are the results based off the optimal WU's per a card. Maybe someone else will be able to answer that question.


I think there was some way i can set which gpu how many units to crunch or i'm wrong ?
I have 2 single pcb GTX 295 with voltage mode 1.12V all 4 gpu cores each one is overclocked to 800core and 1717shaders crunching the stock unit in the FRED application for 165sec i do not know how it is possible gtx 295 to crunch it for 130 sec. ... i have tried a lot of different drivers version and app version and this is the best i have achieved. Soon i will try 1.21 V on the GTX's 295 cards. These cards love to be clocked there are like on steroid when compare to the stock clocks and the time in the crunch is superb. I hope i can manage to pass 850 on core :D i do not know if this is safe or not but we will found out soon :)

I have both 590/690, air cooled, my opinion 590 works to hot, to much trouble to keep it cool, and uses a lot of power, the 690 is better, produces almost the same RAC and works with less power and is cooler (but not to much, stills needs extra fans when used for crunching). In wather cooled you r only concers is your electric bill, the 590 uses about 25% more electric power to do the same job.

The 780 apears to be a good promise but you are comparing 2xGPU´s with single GPU (780) probabily they will launch a 790 with a 2xGPU version. The 15% increase is against the 680 not the 690.

The 2xGPu´s are great if you have few slots avaiable, or whant to buil an super-cruncher with few on them on a single MB, and knows how to handle the heat, otherwise a 2x670 for example is chaper and actualy produces more RAC and are easy to handle than a single 690.

But keep in mind, each case is a unique case, in my case that´s is my opinion, from a user point of view not a technical opinion.

Thanks for the reply. Yes you are right i'm limited for PCI Express slots only 3
My custom water build keep the temp on my single pcb 2 x GTX 295 with voltage mode 1.12V all 4 gpu cores each one is overclocked to 800core and 1717shaders the temp on idle are 11 C and on when crunching SETI@HOME 30 C so basically i do not have any problems with the heat at all.
I want to play games on DX11 and 3D that's why i want to buy GTX 590 and the price in which i can buy it's the same as GTX 670, but GTX 670 is no near as fast for crunching as the GTX 590. i have 1500W PSU Thermaltake so it will be no problem with the W requirements too. My main concern is at the end of the month when i will pay for the electricity bill :D

So that's why i'm wondering if a wait for the GTX 780 which will be faster for chunching compare to GTX 680 or to buy gtx 590

GTX 690 is out of my reach because it is too expensive.
____________

Profile ivan
Volunteer tester
Avatar
Send message
Joined: 5 Mar 01
Posts: 639
Credit: 146,914,388
RAC: 65,286
United Kingdom
Message 1316086 - Posted: 16 Dec 2012, 16:21:28 UTC - in response to Message 1316075.
Last modified: 16 Dec 2012, 16:22:49 UTC

But the old question remains...WHy? If the server actualy can´t feed a 2x690 host (I hate the 100 limit!), imagine with 2x or more 790 ...

Heh! I'm trying to get my hands on a Xeon Phi. If the article I read is to be believed, this can show up as 61 cores each with 4 threads in Linux, so 244 cpus; in my dual-Xeon with Hyperthreading on, that's potentially 260 simultaneous jobs...
(Plus a couple on the GTX460 -- I'd have to pull the C1060 to fit the Phi.)
____________

SupeRNovA
Volunteer tester
Avatar
Send message
Joined: 25 Oct 04
Posts: 131
Credit: 9,238,042
RAC: 0
Bulgaria
Message 1316090 - Posted: 16 Dec 2012, 16:23:45 UTC - in response to Message 1316086.

But the old question remains...WHy? If the server actualy can´t feed a 2x690 host (I hate the 100 limit!), imagine with 2x or more 790 ...

Heh! I'm trying to get my hands on a Xeon Phi. If the article I read is to be believed, this can show up as 61 cores each with 4 threads in Linux, so 244 cpus; in my dual-Xeon with Hyperthreading on, that's potentially 260 simultaneous jobs...


This will be wonderful if somebody can use that power for crunching SETI@HOME and i will be curios what will be the power per watt ration of this card :)
____________

Horacio
Send message
Joined: 14 Jan 00
Posts: 536
Credit: 75,848,887
RAC: 18,731
Argentina
Message 1316108 - Posted: 16 Dec 2012, 17:22:19 UTC - in response to Message 1316078.

I think there was some way i can set which gpu how many units to crunch or i'm wrong ?

AFAIK, there is no way. All the GPUs in the host will be using the some number of concurrent WUs, unless, you do a multi BOINC installation and then configure it so each instance of BOINC ignores certain GPUs and then you make the app_info different for each version... But if you are luky enough to get all this working then your host will be seen as 2 different hosts with separate statistics and total credits (if it matters)...

Also, if you run different GPUs with very different speeds in the same host you may get troubles with WUs beeing aborted with the error that they are taking too long time to finish, the estimated times to completion will be always wrong and they will be always changing which can lead to a lot of headaches...
____________

juan BFBProject donor
Volunteer tester
Avatar
Send message
Joined: 16 Mar 07
Posts: 5472
Credit: 313,437,458
RAC: 107,873
Brazil
Message 1316129 - Posted: 16 Dec 2012, 17:52:23 UTC - in response to Message 1316108.
Last modified: 16 Dec 2012, 17:54:39 UTC

I think there was some way i can set which gpu how many units to crunch or i'm wrong ?

AFAIK, there is no way. All the GPUs in the host will be using the some number of concurrent WUs, unless, you do a multi BOINC installation and then configure it so each instance of BOINC ignores certain GPUs and then you make the app_info different for each version... But if you are luky enough to get all this working then your host will be seen as 2 different hosts with separate statistics and total credits (if it matters)...

Also, if you run different GPUs with very different speeds in the same host you may get troubles with WUs beeing aborted with the error that they are taking too long time to finish, the estimated times to completion will be always wrong and they will be always changing which can lead to a lot of headaches...


Actualy, there is a way, and works, but you need to know what you are doing.

Check this post at Vyper´s site, he teach you how to do that:

http://vyper.kafit.se/wp/index.php/2011/02/04/running-different-nvidia-architectures-most-optimal-at-setihome/

Just a warming: Do not try if you don´t know realy what you are doing.
____________

Horacio
Send message
Joined: 14 Jan 00
Posts: 536
Credit: 75,848,887
RAC: 18,731
Argentina
Message 1316139 - Posted: 16 Dec 2012, 18:12:12 UTC - in response to Message 1316129.

I think there was some way i can set which gpu how many units to crunch or i'm wrong ?

AFAIK, there is no way. All the GPUs in the host will be using the some number of concurrent WUs, unless, you do a multi BOINC installation and then configure it so each instance of BOINC ignores certain GPUs and then you make the app_info different for each version... But if you are luky enough to get all this working then your host will be seen as 2 different hosts with separate statistics and total credits (if it matters)...

Also, if you run different GPUs with very different speeds in the same host you may get troubles with WUs beeing aborted with the error that they are taking too long time to finish, the estimated times to completion will be always wrong and they will be always changing which can lead to a lot of headaches...


Actualy, there is a way, and works, but you need to know what you are doing.

Check this post at Vyper´s site, he teach you how to do that:

http://vyper.kafit.se/wp/index.php/2011/02/04/running-different-nvidia-architectures-most-optimal-at-setihome/

Just a warming: Do not try if you don´t know realy what you are doing.

That link is exactly which I meant with "unless you do a multi BOINC installation"...
____________

SupeRNovA
Volunteer tester
Avatar
Send message
Joined: 25 Oct 04
Posts: 131
Credit: 9,238,042
RAC: 0
Bulgaria
Message 1316209 - Posted: 16 Dec 2012, 20:44:09 UTC - in response to Message 1316129.

I think there was some way i can set which gpu how many units to crunch or i'm wrong ?

AFAIK, there is no way. All the GPUs in the host will be using the some number of concurrent WUs, unless, you do a multi BOINC installation and then configure it so each instance of BOINC ignores certain GPUs and then you make the app_info different for each version... But if you are luky enough to get all this working then your host will be seen as 2 different hosts with separate statistics and total credits (if it matters)...

Also, if you run different GPUs with very different speeds in the same host you may get troubles with WUs beeing aborted with the error that they are taking too long time to finish, the estimated times to completion will be always wrong and they will be always changing which can lead to a lot of headaches...


Actualy, there is a way, and works, but you need to know what you are doing.

Check this post at Vyper´s site, he teach you how to do that:

http://vyper.kafit.se/wp/index.php/2011/02/04/running-different-nvidia-architectures-most-optimal-at-setihome/

Just a warming: Do not try if you don´t know realy what you are doing.


Thank You very much that was the link i have been read in the past :)

____________

Message boards : Number crunching : GTX 590 a good choice ?

Copyright © 2014 University of California