CUDA Cards

Message boards : Number crunching : CUDA Cards
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3

AuthorMessage
Profile Dirk Sadowski
Volunteer tester

Send message
Joined: 6 Apr 07
Posts: 7105
Credit: 147,663,825
RAC: 5
Germany
Message 969234 - Posted: 8 Feb 2010, 13:01:45 UTC - in response to Message 969202.  
Last modified: 8 Feb 2010, 13:02:28 UTC


nVIDIA can CUDA, so you could let run only MB.
You could let run on ATI a CPU/ATI GPU hybrid app of AP.

nVIDIA -> MB
ATI -> AP

'NVIDIA or ATI - Which is giving better performance'


Not all new is better: 'Nvidia 195.62? You can keep it!'


Open your eyes if you visit a forum! ;-)


____________
[Optimized project applications, for to increase your PC performance (double RAC)!][Overview of abbreviations, which are used often in forum and their meaning.]
ID: 969234 · Report as offensive
Profile Dirk Sadowski
Volunteer tester

Send message
Joined: 6 Apr 07
Posts: 7105
Credit: 147,663,825
RAC: 5
Germany
Message 969235 - Posted: 8 Feb 2010, 13:10:54 UTC - in response to Message 969102.  
Last modified: 8 Feb 2010, 13:12:00 UTC

[...]
I got the GTS 250, and am very happy with it, except for one small nit ---> it's the last of the line that WON'T run the GPU app for the Milkyway project. Something to do with CUDA 3.x. Of course, there was no way for me to know this at the time of purchase (5 months ago).
[...]


AFAIK, CUDA V3.0 is still BETA, or?
And MilkyWay send this out?

Why you couldn't let run MilkyWay on an 'old' GTS250? I couldn't imagine why.
But, I have no experiences with this project..

So MilkyWay accept only new GTx2xx series GPUs?


____________
[Optimized project applications, for to increase your PC performance (double RAC)!][Overview of abbreviations, which are used often in forum and their meaning.]
ID: 969235 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14679
Credit: 200,643,578
RAC: 874
United Kingdom
Message 969238 - Posted: 8 Feb 2010, 13:55:45 UTC - in response to Message 969235.  

So MilkyWay accept only new GTx2xx series GPUs?

I have no experience with MilkyWay either.

But it's been written often enough on various BOINC boards that they require double-precision hardware, so boards based on G200-series chips (GTX 2xx), not the G92b chip in the GTS 250.

Nothing to do with the CUDA 3.0 software at all.
ID: 969238 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 970672 - Posted: 15 Feb 2010, 7:49:06 UTC - in response to Message 968992.  

[quote]

[http://www.nvidia.com/object/cuda_learn_products.html] ->
[http://www.nvidia.com/object/product_geforce_210_us.html]

16 CUDA cores (new).. nVIDIA name now the shader cores CUDA cores.. ;-)

So it have 16 shader cores or 2 CUDA cores (old)..
I wouldn't expect too much from this.

One of my 5 OCed GTX260-216 have a ~ 15,000 RAC, but have 216 shader/27 CUDA cores (old).

The 210 use max. 30.5 W, if CUDA then ~ 2/3 so ~ 20 W.

It's not easy to guess which RAC you could reach.. the specs are like a 8500 GT. Maybe you find in the top_host_list a PC with 210 or 8500 GT.
[http://setiathome.berkeley.edu/top_hosts.php]
I would guess ~ 1,000 RAC.

If you follow the first URL, there are all GPUs which can CUDA.
If a very well performance/wattage ratio, I would go with the new 40nm GPU chip, 210/GT220/240 GPUs.

[http://www.nvidia.com/object/product_geforce_gt_240_us.html]
69 W max. - ~ 46 W CUDA - 96 shader cores - 12 CUDA cores (old) - maybe 5,000 RAC.


I've read this thread, and am trying to decide which video card to start putting in my farm. I purchased a couple low end cards last year (8000 series I believe) and though they worked, the RAC was very low, below 1000 for sure. I tried them as they were the best PCI based CUDA cards I could find, but as their RAC is so low, I've decided that I will just skip doing multi-cards in these boards (only one PCI-E slot, the rest are PCI) and go with one "winner" card, it's much more cost effective, even though those cards were well under $50.

While at Microcenter today, I grabbed a EVGA GT220 for around $75-80, but since I got home, I started reading up, found this thread, and hit Neweggs' site comparing cards. I like the thought of perf/watt, and while doing the research, found that the 240 Superclocked does idle/full load of 146/203 stock, a little more when OC'ed, and the GTX 260 Core 216 Superclocked is at 159/274 stock, at least at the review sites I found. I guess the main question now would be RAC/watt, because in the long haul, that has to come into play for utility budgetary reasons. I'm all for a higher RAC, all things being somewhat equal, but they can't be _too_ far out of whack.

Does anyone have a good idea as to the RAC/watt ratio of these 2 cards? I'm pretty sure at this point I'll be returning the GT220, it doesn't look like much of at folding card anymore, especially since the 240 Superclocked is under $100, and looks to be quite a bit better performer. One thing I noticed, there are 2 240 cards, one labeled Superclocked and the other not, Superclocked version has 512 meg, the other has a gig, both nearly the same specs (clock, DDR5, stream procs) though the Effective Memory Clock was 3588 MHz (Superclocked) and 850MHz (3.4Gbps effective) for the 1 gig model. Does more memory (after a certain point, whatever that is) necessairly equate to better performace in CUDA?

Any thoughts are appreciated.

ID: 970672 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 970817 - Posted: 16 Feb 2010, 2:14:14 UTC

Oh, also does anyone have a firm idea as to when the new 400 series is going to be released? I've heard maybe next month, maybe later, and the guesses of $ are anywhere from $3xx to $6xx+! Quite a price range, should be interesting...

ID: 970817 · Report as offensive
Profile Dirk Sadowski
Volunteer tester

Send message
Joined: 6 Apr 07
Posts: 7105
Credit: 147,663,825
RAC: 5
Germany
Message 970849 - Posted: 16 Feb 2010, 10:02:07 UTC - in response to Message 970672.  
Last modified: 16 Feb 2010, 10:03:25 UTC

[...]
While at Microcenter today, I grabbed a EVGA GT220 for around $75-80, but since I got home, I started reading up, found this thread, and hit Neweggs' site comparing cards. I like the thought of perf/watt, and while doing the research, found that the 240 Superclocked does idle/full load of 146/203 stock, a little more when OC'ed, and the GTX 260 Core 216 Superclocked is at 159/274 stock, at least at the review sites I found. I guess the main question now would be RAC/watt, because in the long haul, that has to come into play for utility budgetary reasons. I'm all for a higher RAC, all things being somewhat equal, but they can't be _too_ far out of whack.

Does anyone have a good idea as to the RAC/watt ratio of these 2 cards? I'm pretty sure at this point I'll be returning the GT220, it doesn't look like much of at folding card anymore, especially since the 240 Superclocked is under $100, and looks to be quite a bit better performer. One thing I noticed, there are 2 240 cards, one labeled Superclocked and the other not, Superclocked version has 512 meg, the other has a gig, both nearly the same specs (clock, DDR5, stream procs) though the Effective Memory Clock was 3588 MHz (Superclocked) and 850MHz (3.4Gbps effective) for the 1 gig model. Does more memory (after a certain point, whatever that is) necessairly equate to better performace in CUDA?

Any thoughts are appreciated.


A GT240 and 203 W at full load?
I couldn't believe.
From the nVIDIA site a GT240 use max. 69 W.

From my experiences if CUDA, my manufacturer OCed GTX260-216 use 3/4 of the max. W.
My 5 OCed GTX260-216 use ~ 140 W/GPU (in AMD 940 BE & 4 OCed EVGA), ~ 170 W/GPU (in Intel QX6700 & OCed GIGABYTE).

If a GT240 ~ 46 W with CUDA (~ 5,000 RAC) and a OCed GTX260-216 ~ 155 W if CUDA (~ 15,000 RAC) so you could decide very easy.. same performance/wattage ratio.
Depend of your money.. ;-)


In past nVIDIA said GTX3xx will come.. they said it over 1/2 year.
And now, they don't publish..

If a new GTXX*xx (*whatever) will come, I believe it earliest if you can buy it.

Maybe make a Inet search about.. and look if you could find info about.

If it's the same story like with GTX3xx GPUs, they should have the double performance like the current GTX2x series GPUs.

So, same wattage but double performance.


You speak about EVGA?
The SuperClocked is the manufacturer OCed 'SC'.

AFAIK..
GTS250 - SC
GTX260-216 - SC, SSC
GTX275/285 - SC, SSC, FTW
That's the max. level of the EVGA manufacturer OCed GPUs.
AFAIK, but the GIGABYTE SOC is the highest OCed GTX260-216.


____________
[Optimized project applications, for to increase your PC performance (double RAC)!][Overview of abbreviations, which are used often in forum and their meaning.]
ID: 970849 · Report as offensive
Profile hiamps
Volunteer tester
Avatar

Send message
Joined: 23 May 99
Posts: 4292
Credit: 72,971,319
RAC: 0
United States
Message 970878 - Posted: 16 Feb 2010, 14:29:56 UTC - in response to Message 970817.  

Oh, also does anyone have a firm idea as to when the new 400 series is going to be released? I've heard maybe next month, maybe later, and the guesses of $ are anywhere from $3xx to $6xx+! Quite a price range, should be interesting...

Not sure when they will realease new cards latest rumor I saw was March for 2 cards. They sure are helping ATI with their missed deadlines and total silence about it....
Official Abuser of Boinc Buttons...
And no good credit hound!
ID: 970878 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 66350
Credit: 55,293,173
RAC: 49
United States
Message 970886 - Posted: 16 Feb 2010, 15:03:47 UTC - in response to Message 970878.  

Oh, also does anyone have a firm idea as to when the new 400 series is going to be released? I've heard maybe next month, maybe later, and the guesses of $ are anywhere from $3xx to $6xx+! Quite a price range, should be interesting...

Not sure when they will realease new cards latest rumor I saw was March for 2 cards. They sure are helping ATI with their missed deadlines and total silence about it....

That and ATi is making their top card while Nvidia killed the 295 production, I wonder what the supply of 260, 275, 280 and 285 cards is like?
Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

ID: 970886 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 971059 - Posted: 18 Feb 2010, 2:40:40 UTC - in response to Message 970886.  

That and ATi is making their top card while Nvidia killed the 295 production, I wonder what the supply of 260, 275, 280 and 285 cards is like?
Drying up, unfortunately. I've been looking for them and the best I've found recently are the 260's, at least from reliable sources, as opposed to Fleabay...

Sutaru, you're right, those voltages were off, the comparisons I saw were between identical systems configured with the different cards, that being the only difference.

If a GT240 ~ 46 W with CUDA (~ 5,000 RAC) and a OCed GTX260-216 ~ 155 W if CUDA (~ 15,000 RAC) so you could decide very easy.. same performance/wattage ratio.
Depend of your money.. ;-)
You pretty much summed it up for me there. Thanks for the info, time to see what I can squeeze outa the piggy bank, that'll tell me which way I can go...

ID: 971059 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 66350
Credit: 55,293,173
RAC: 49
United States
Message 971069 - Posted: 18 Feb 2010, 3:33:58 UTC - in response to Message 971059.  

That and ATi is making their top card while Nvidia killed the 295 production, I wonder what the supply of 260, 275, 280 and 285 cards is like?
Drying up, unfortunately. I've been looking for them and the best I've found recently are the 260's, at least from reliable sources, as opposed to Fleabay...

Sutaru, you're right, those voltages were off, the comparisons I saw were between identical systems configured with the different cards, that being the only difference.

If a GT240 ~ 46 W with CUDA (~ 5,000 RAC) and a OCed GTX260-216 ~ 155 W if CUDA (~ 15,000 RAC) so you could decide very easy.. same performance/wattage ratio.
Depend of your money.. ;-)
You pretty much summed it up for me there. Thanks for the info, time to see what I can squeeze outa the piggy bank, that'll tell me which way I can go...

Well outside of ebay, I know of one or two resellers who might have a few of the 295's left, But of the one I know of, I'm not telling, Besides in June It may be only ebay, But I'll find out when I get there. You may not like ebay, But I've gotten some great deals from there and the QX6700 cpu that I'm using now came from there in October of 2006, Before Intel started selling the retail versions(Bleeding edge back then). :D
Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

ID: 971069 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 973505 - Posted: 23 Feb 2010, 12:24:05 UTC - in response to Message 971069.  


Well outside of ebay, I know of one or two resellers who might have a few of the 295's left, But of the one I know of, I'm not telling, Besides in June It may be only ebay, But I'll find out when I get there. You may not like ebay, But I've gotten some great deals from there and the QX6700 cpu that I'm using now came from there in October of 2006, Before Intel started selling the retail versions(Bleeding edge back then). :D

Actually, sadly, I quite like ebay, sometimes a bit too much. Many good deals to be had, sometimes too good to pass up, which can wreck havoc with the budget... :) I think I may be leaning towards the GTX 260 Core 216, seems to have similiar perf to the 280, but 50w less. Just have to get over that inital 110% bump in price over the 240, and I'll be ok... Then again, I know the 400 series cards are coming out _sometime_ soon, and if I save the $ now by getting the 240's, and spend it in 6-8 months on the lower end of new style series. Arrgh! This is a tough decision.

ID: 973505 · Report as offensive
hbomber
Volunteer tester

Send message
Joined: 2 May 01
Posts: 437
Credit: 50,852,854
RAC: 0
Bulgaria
Message 973710 - Posted: 25 Feb 2010, 11:07:42 UTC
Last modified: 25 Feb 2010, 11:16:25 UTC

My latest research revealed some interesting points about old grannie - 8800GT.
Some assumptions about these:
1.Most of them reach OC levels as high as 700/1728+/1000 (core/shaders/memory)
2.Max power consumption(if well cooled) do not exceed 90 W(in idle they suck anyway, but they are not meant to stay idle)
3. Computing power(when overclocked) of two 8800GTs exceeds fairly overclocked GTX 275, and its about same if no overclock at all. And they consume same amount 2x8800<=1x275
4. Computing power of two 8800GTs exceeds by far even well OCed 260, extra power consumption is about same like computing power, in percent. Keep in mind, 260s cannot reach that often OC levels as 275/285 can. Average OC for SETI crunching it around 650 core and < 1500 MHz for shaders. Memory is even worse. There are, of course, exceptions.

So far nothing interesting, except some more computing power, neglectable. Here comes biggest advantage - here u can find them for like 50-55 EURO, second hand. Which makes them far more better investment than still 150 EURO second hand GTX260s. Not to mentions price of new GT240s and their far weaker computing power. 8800GT is so much cheaper than new GT240 - u need to save power with 240s for over year to get same TCO(total cost of ownership) and 8800GT will still net u many more GFLOPs.(Electricity in Bulgaria is ~9 eurocents daytime and ~6.5 eurocents at night, per kW)
+8800GT are SLI capable(for GT240 I find controversary opinions, and if they get SLI, it will be slow, because of communication travelling thru PCI-e). Its kinda tricky to get SLI on X48/X38/P45/P35, but its not 7th level magic.

Still remains problem of having motherboard, capable of holding many of them. My old DFI X48-T2R will take 3 of these(8800GT) at x16, x16, x4. This MoBo can be still be found, new, with warranty for 75-80 EURO here.

U can also keep in mind, that being low level products laready, their price, when time to resell them, will drop insignificantly, while GT240s, second hand , will have huge drop as being new product atm.

P.S. Almost same apllies to 9800GTs, but they have a bit higher power consumptions(even 55 nm pieces) and are expensier than 8800Gts.
ID: 973710 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 973942 - Posted: 26 Feb 2010, 4:36:21 UTC - in response to Message 973710.  

Thanks for the info on the 8800's, was a good read. I ended up buying the 260 216's, mostly because the Intel boards in my farm only have one PCI-E 2 slot in them, and I wanted to get the most cost effective card in terms of perf/watt and cost. It was a little more than I wanted to spend, but since they are going to be going into stock Intel boards w/no OC'ing, I felt that this was the best way to go at this time.

Of course, when the 4xx series comes out, I will probably be sitting there thinking, Now if only I'd waited, I could have gotten x% more perf for only y% more money.. But I'll fight that thought off, as these are going to just be sitting and crunching for a long time, hopefully without much interference from me. :)

ID: 973942 · Report as offensive
Previous · 1 · 2 · 3

Message boards : Number crunching : CUDA Cards


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.