Is the GTX 295 still the best card for CUDA? (Better than a GTX 470?)


log in

Advanced search

Message boards : Number crunching : Is the GTX 295 still the best card for CUDA? (Better than a GTX 470?)

1 · 2 · Next
Author Message
Profile lupo
Send message
Joined: 29 Aug 10
Posts: 91
Credit: 4,736,407
RAC: 0
United States
Message 1050079 - Posted: 21 Nov 2010, 0:28:37 UTC

I just picked up 3 GTX 470's for my CUDA system but I have the opportunity to pick up a few 295's. Are the 295's still better than a 470 for CUDA work or are there any programming changes coming down in the future that will make the GTX 470 superior?

Thanks,
Adam

Profile arkaynProject donor
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3688
Credit: 48,714,398
RAC: 6,995
United States
Message 1050090 - Posted: 21 Nov 2010, 1:07:00 UTC

Jason over at Lunatics is starting to work on the next generation of the CUDA app now that the 3.2 developer tools are out.

Still a ways to go though.
____________

Brkovip
Avatar
Send message
Joined: 18 May 99
Posts: 274
Credit: 144,162,576
RAC: 7,367
United States
Message 1050149 - Posted: 21 Nov 2010, 4:25:54 UTC - in response to Message 1050107.

Yes, the 295 is still the card to beat. Although on some projects, ATI is a killer.

Not here, though.

Two 295's in one rig are a powerful couple of crunchers.

I have 2 465's in another rig, and they are rather lame.

Very dissapointed in them, I be.


Overclock them!
____________

Profile lupo
Send message
Joined: 29 Aug 10
Posts: 91
Credit: 4,736,407
RAC: 0
United States
Message 1050173 - Posted: 21 Nov 2010, 8:31:23 UTC - in response to Message 1050090.

So, the question is this. If it were your money, would you keep the GTX 470 or spend the cash on a GTX 295? Do you think the single 470 will surpass the processing power of a 295 once the new version of the CUDA app is out?

Also, if you know, what would the new CUDA app do to more fully utilize the power of the 470 making it a better performer that the 295? (If you follow my question.)

is there some part of the hardware in the 470 that is not currently utilized in the current CUDA software that will be more fully utilized with the future version of the CUDA app?

I have a chance to get 2 295's for about $200 each, but I need to make the decision in the next 36-48 hours. :(

Adam

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 5051
Credit: 73,816,534
RAC: 12,983
Australia
Message 1050174 - Posted: 21 Nov 2010, 8:44:52 UTC - in response to Message 1050173.

... Do you think the single 470 will surpass the processing power of a 295 once the new version of the CUDA app is out?
470, maybe, 480 definitely.

Also, if you know, what would the new CUDA app do to more fully utilize the power of the 470 making it a better performer that the 295? (If you follow my question.)
Use cuda kernels targeted to the wider memory bus, caching & enhanced concurrency features.

is there some part of the hardware in the 470 that is not currently utilized in the current CUDA software that will be more fully utilized with the future version of the CUDA app?
Yes

I have a chance to get 2 295's for about $200 each, but I need to make the decision in the next 36-48 hours. :(


Awesome cards, if you took them you won't be disappointed. They are at the end of their lifecycle though (won't be getting a lot faster), whereas the Fermis are at the beginning of theirs( Will be getting a lot faster). The best choice for you would depend on how much patience you have.

Jason



____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Profile gizbar
Avatar
Send message
Joined: 7 Jan 01
Posts: 586
Credit: 21,087,774
RAC: 0
United Kingdom
Message 1050200 - Posted: 21 Nov 2010, 10:06:06 UTC

It's not really much of a comparison though, is it?

A 295 is a dual GPU card, and a 470 is a single. You'd need 4 470's to even get close to 2 295's. But as has been said, there won't be any more 295's coming, unless you can pick them up second hand, and they are already updating the 470's and 480's with the 580, which runs cooler and overlocks better, and commands the price premium of being the new kid on the block.

Plus as has also been said, there may not be much more that can be done to optimise the code for the 295's, whereas there is a lot to be learned about the Fermi code and optimising that.

The killer question is:- Immediate results from some 295's, or possibly better results from a newer spec card but over a longer period of time.

Giz.

____________


A proud GPU User Server Donor!

Profile HelliProject donor
Volunteer tester
Avatar
Send message
Joined: 15 Dec 99
Posts: 704
Credit: 90,002,239
RAC: 78,538
Germany
Message 1050204 - Posted: 21 Nov 2010, 10:27:07 UTC

The real Question is - How long do we have to wait for a Fermi optimized version!

If Nvidia can follow his own Roadmap - next Year we will have the first Kepler GPUs...

Helli
____________
A loooong time ago: My first Credits

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 5051
Credit: 73,816,534
RAC: 12,983
Australia
Message 1050212 - Posted: 21 Nov 2010, 12:12:10 UTC - in response to Message 1050204.
Last modified: 21 Nov 2010, 12:13:52 UTC

The real Question is - How long do we have to wait for a Fermi optimized version!

If Nvidia can follow his own Roadmap - next Year we will have the first Kepler GPUs...

Helli


You saw the first fermi optimisations introduced with x32f ... unfortunately the particular optimisations related mainly to VLAR, for about a 20% improvement, which the project now doesn't send out to GPUs, so the benefit there is obliterated.

Now you just have to wait until we work out how to program the stinking things ;)

Jason
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Profile HelliProject donor
Volunteer tester
Avatar
Send message
Joined: 15 Dec 99
Posts: 704
Credit: 90,002,239
RAC: 78,538
Germany
Message 1050213 - Posted: 21 Nov 2010, 12:41:26 UTC - in response to Message 1050212.

Well, we wait...as always. ;-)

But keep in mind that the lifetime of the Fermi GPUs is limited. :-)

Helli
____________
A loooong time ago: My first Credits

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 5051
Credit: 73,816,534
RAC: 12,983
Australia
Message 1050215 - Posted: 21 Nov 2010, 12:58:50 UTC - in response to Message 1050213.
Last modified: 21 Nov 2010, 12:59:47 UTC

But keep in mind that the lifetime of the Fermi GPUs is limited. :-)
Hahaha, I know what you mean. Thankfully Fermi draws on '80's-90's supercomputer architecture, which isn't something that's going to date quickly. Apps come and go, but the skills involved remain.

Jason
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Profile SciManStevProject donor
Volunteer tester
Avatar
Send message
Joined: 20 Jun 99
Posts: 4876
Credit: 82,954,814
RAC: 39,670
United States
Message 1050216 - Posted: 21 Nov 2010, 13:03:02 UTC

The 480's are very powerful indeed. I have two of them water cooled, and overclocked, and each will deliver a RAC of well over 15,000, actually around 18,000. With only 2 GPU's and a fast processor, that easily puts my rig in the top 20 computers. That is much harder to do with just 1 295, which as mentioned is actually two cards in one. The rigs in the top 20 have several GPU's. When Jason has finished optimizing them, the Fermi's will be formidable indeed. I have all the patience in the world.

Steve
____________
Warning, addicted to SETI crunching!
Crunching as a member of GPU Users Group.
GPUUG Website

Profile Westsail and *Pyxey*
Volunteer tester
Avatar
Send message
Joined: 26 Jul 99
Posts: 338
Credit: 20,538,216
RAC: 0
United States
Message 1050242 - Posted: 21 Nov 2010, 17:17:52 UTC

Absolutely thrilled with Fermi here...Over in Collatz the 460 is exactly 3x as fast as my 260 and 10 degrees cooler. :)
<10 minutes for 460 vs. 30+ minutes on the old 260
Haven't had a chance to compare on Seti yet..Sounds like when Jason gets done they will rock'n roll!

Jason, Will there be multiple apps for different versions of GF1xx or a single app for all Fermi? I ask cause of the architecture differences between 460/580 and 465/470/480. Thanks in advance!
____________
"The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 5051
Credit: 73,816,534
RAC: 12,983
Australia
Message 1050244 - Posted: 21 Nov 2010, 17:22:13 UTC - in response to Message 1050242.
Last modified: 21 Nov 2010, 17:23:02 UTC

Jason, Will there be multiple apps for different versions of GF1xx or a single app for all Fermi? I ask cause of the architecture differences between 460/580 and 465/470/480. Thanks in advance!


Likely a single 'flexible' build, since nVidia's drivers now contain sophisticated dispatch mechanisms to select amongst suitable kernels based on compute capability. In addition, device parameters allow scaling the kernel launch characteristics at run time (which I'm actually experimenting with now).

Jason
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

zoom314Project donor
Avatar
Send message
Joined: 30 Nov 03
Posts: 46492
Credit: 36,843,177
RAC: 5,103
United States
Message 1050286 - Posted: 21 Nov 2010, 22:59:20 UTC

I'm still planning on getting a pair of GTX295 v2 cards to go with My EVGA CO-OP GTX295 card.
____________
My Facebook, War Commander, 2015

Profile Joshua
Avatar
Send message
Joined: 19 Mar 06
Posts: 13
Credit: 385,654
RAC: 0
United States
Message 1050331 - Posted: 22 Nov 2010, 17:04:56 UTC

I had asked myself this exact same question.. do I get a second 295 or do I go with a 470...

At the moment you can easily get a 295 for 225-300 and for the same price you can get 470. I decided to get two 470's instead, because FERMI is where is at! Fermi is a new technology, and I've seen good results on a non-optimized app, so would you imagine what the performance will be on a optimized app?

I currently have 2 PNY 470's that I will be water colling soon. They run a bit hot, but that's typical for a Fermi card. I like the performance so far.

I love my 295, but I know its a dying gen video card, once the optimized Fermi app comes out, we will see more and more FERMI cards on top.

Just want to give Jason, and BIG BIG thank you for continuing his work on optimizing the Lunatics apps for all the Fermi owners! Thanks Jason!
____________
Any program is only as good as it is useful.
-Linus Torvalds

Profile HelliProject donor
Volunteer tester
Avatar
Send message
Joined: 15 Dec 99
Posts: 704
Credit: 90,002,239
RAC: 78,538
Germany
Message 1050336 - Posted: 22 Nov 2010, 17:38:56 UTC - in response to Message 1050331.
Last modified: 22 Nov 2010, 17:39:24 UTC

I had asked myself this exact same question.. do I get a second 295 or do I go with a 470...
...


My 295er do 9.6 mid range (AR 0.426) WUs in one Hour.

Can the GTX470 do the same? If your Answer is Yes - buy a FERMI, is the
Answer No then it's a question of Money. ;-)

Helli
____________
A loooong time ago: My first Credits

Profile Frizz
Volunteer tester
Avatar
Send message
Joined: 17 May 99
Posts: 271
Credit: 5,852,934
RAC: 0
New Zealand
Message 1050352 - Posted: 22 Nov 2010, 19:49:50 UTC - in response to Message 1050107.


Although on some projects, ATI is a killer.

Not here, though.


The optimized app for Astropulse (in alpha stadium) looks very promising. A mainstream HD6850 crunches a unit in under 1 hour!

1 · 2 · Next

Message boards : Number crunching : Is the GTX 295 still the best card for CUDA? (Better than a GTX 470?)

Copyright © 2014 University of California