I'm looking at buying a new computer after having this one since 2005 or 2006, so it's well deserved.

I'm not quite sure what I will be allowed to spend on it, BUT, throwing money totally aside, I am curious.

Lets say you're building a workstation. I know the graphics card(s) seem to be at least equally important, or probably more important to the cpus. What is the best one:

NVIDA GeForce GTX 295 vs. NVIDA Quadro FX 5800

Now I realize you can get 2 or 3 of either.....and I mean, fully loaded, lets say the most of either you can get hooked up. Or, just 1 of each. Whatever, or both scernios.

5800 has more memory
295's clock is faster I think?
5800 has the full 4.0 shader
295 has more memory bandwidth/throughput

Using the exact same computer, which does better on non-specific boinc projects? Do we have any data on this? I can't believe nobody has used the 5800 yet to crunch data.

Keep in mind that the higher end cards take up enough space for two, so make sure that if you want to fill all of those slots, make sure they are spaced properly, ideally, you want it to go

E = Slot that cards can go in
P = Other kinds of PCI slots that will not be filled with CUDA cards

E P E P E P

or something like that. You dont want "E E"

---

30+ Computers heading our way! Currently at the "Zomg we need to talk to our tech expert at the co-op about this first!!!" stage. 16 Lab machines and 14+ Staff machines each with 2.2Ghz CPUs and 256MB ram. Think they balance? The RAM certainly is bad

The GTX 295 gives you the most bang for the buck.

Both in terms of purchasing and powerdraw vs. crunching power.

In an earlier thread someone discussed efficiency and my machine was highest in terms of science done vs. power draw.

If you only wan't a regular machine to work on and then let it use seti in it's spare time then it's another story and there are tons of options there.

Kind regards Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

The GTX 295 gives you the most bang for the buck.

Both in terms of purchasing and powerdraw vs. crunching power.

If by 'buck' you mean dollars, greenbacks ... money in general, then I don't agree with this oft-repeated statement.

At my local store, a GTX 295 running at 1242MHz costs £349.93 (that's GBP).

You can get two GTX 275 cards (same manufacturer) running at 1404MHz for £158.61 each, or £317.22 - 10% cheaper.

Faster, cheaper - sounds good value to me. But they don't list the power draw, unfortunately.

What the GTX 295 undoubtably does give you is most bang for the slot, and hence most bang for the box.

What the GTX 295 undoubtably does give you is most bang for the slot, and hence most bang for the box.

the NVIDIA Tesla C1060 is expensive approx. $ 1600 to 1800 in Canadian dollars approx £ 900 in British Pounds
would there be any advantage in using the NVIDIA Tesla C1060 in cruching the SETI@home CUDA apps ?

---

The GTX 295 gives you the most bang for the buck.

Both in terms of purchasing and powerdraw vs. crunching power.

If by 'buck' you mean dollars, greenbacks ... money in general, then I don't agree with this oft-repeated statement.

At my local store, a GTX 295 running at 1242MHz costs £349.93 (that's GBP).

You can get two GTX 275 cards (same manufacturer) running at 1404MHz for £158.61 each, or £317.22 - 10% cheaper.

Faster, cheaper - sounds good value to me. But they don't list the power draw, unfortunately.

What the GTX 295 undoubtably does give you is most bang for the slot, and hence most bang for the box.

Just doing a quick poke around the internet the 295 used about 10% more power then the 275. I do not know about the seti performance of the two, but if the 295 is more then 10% faster. It would then have a greater "performance per watt". Which in the long term would still make it the best "bang for the buck".

Sometimes it is better to spend a little more up front to same money in the long run. You just have to weigh in the time to recoup the extra cost.

---

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

You also need to include number of CUDA processors on the board for comparison.

A GTX216 has 27 Cuda cores vs. a GTX285 has 30 cuda cores and a GTX295 has 60 CUDA cores even if it is a dual GTX260 in a single shell and thus should have got 54 cores instead of those 30 that is actually included.

If you compare all those facts and the frequency of ram and shader core then we have a fair comparison against all cards even if it is Tesla, Quadro or whatever flavour you want..

It's not a easy one put if you have the time and energy you could calculate the "bang for the buck" ratio . :)

Kind regards Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

I am not concerned about bank for your buck, bang for your watt, or any of that. I'm asking if you built one computer cost not being an issue.....your goal being to crunch non-specific boinc project WUs the fastest - What would your GPU be?

I know if you factor in the money thing, the 295 almost assuredly has the win vs. the 5800.

Im pretty happy with my 1GB NVIDIA GeForce GTS 250. Has 128 processor cores. and cost 180 US dollars. Combine that with an i7 cpu and you have a good cruncher.

---

[/quote]

Old James

Nobody has attempted to answer my question. Hmm.

Nobody has attempted to answer my question. Hmm.

i'll answer it for you.

me, no cost,

BFG H2OC 295+C1060

url to 1060 card: this card is purely for cuda applications.....


http://www.nvidia.com/object/product_tesla_c1060_us.html

this setup will give you 3 cuda cores to work with. note that the 1060 has no outlets, it's purely a crunching card.

5800 is just overkill, the 1060 can be bought on tigerdirect, and the 295 h20oc with it's liquid block can be bought on newegg.

---

I recommend Secunia PSI: http://secunia.com/vulnerability_scanning/personal/
Go Georgia Tech.

"5800 is just overkill, the 1060 can be bought on tigerdirect, and the 295 h20oc with it's liquid block can be bought on newegg."

Yes, I have also looked at the 1060....but I heard some of it's advantages were mitigated by......something in the newest cards.

But what do you mean when you said the "5800 is just overkill". Does anybody know of any examples, or actual benchmarks of the 5800 vs the 295 or the 295 + the 1060 to see which is better and if so, how much better it is?

Since one is traditionally for the work station and one traditionally for a gaming desktop they are never really compared against each other - I just see specs.

But since graphics are so HUGE in crunching, this is very important for us. I do 8 boinc programs...so I'd like a non-specific project answer, not just a seti answer.

If people aren't sure, I like guesses better than nothing. Thanks zpm for giving his opinion on the matter. I am not sure why if money wasn't a concern why you'd say something was overkill though. If there was another reason not to go with it.

I don't think it's clear from the data I have, which would be faster.

(3)295 vs. 295 + 1060 vs. (3)5800 vs. other configurations you guys can think of

5800 quadro is designed for CAD. and cost a ton, you get the same thing with the 1060 - some memory and a gui via card. and nvidia page doesn't show the gflops.

you can look at the specs of all cards at:
http://www.nvidia.com/object/cuda_learn_products.html

and tbh, Multi-threaded apps can compete(as shown with AQUA@HOME) with single core gpu's.

hands down, the 295's with 2 cores are the best bet, until the g300 chipset's are released.

the 1060's would be my second choice as you're paying for the cores, not the hdmi, dvi. and you get the same number of cores. also, the chances of trouble with the tesla's is less b/c it doesn't rely on an extended desktop. and people with 4 cuda cards or more have trouble with getting them to run in boinc the last time i checked.

1. 295
2. C1060
3. 5800

thats how i would rate the cards in my honest opinion.

1. 295
2. C1060
3. 5800

thats how i would rate the cards in my honest opinion.

Agreed...

Dedicated host /w single C1060
*note this one is pre-production, only 192core/1gig

---

"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

try not to be fooled by that host, he's running that computer in 32-bit, a tesla in 64-bit will process a bit faster.

try not to be fooled by that host, he's running that computer in 32-bit, a tesla in 64-bit will process a bit faster.

???
Really? Sounds like that would be worth looking into if it's the case. :) I had been under the impression the difference would be nil to negligible...Is this true for all cuda hosts? Must say, I have been thrilled with the performance of 2.3rt+xp32.

Here is the only other c1060 host I know of. Maybe someone has some data on Tesla crunch times. Anyone know if there are any on GPU grid etc?
host 5050341

---

"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

so you guys think ideal would be running 3 graphics cores with:

1(295) nvida
and
1 Telsa C1060

that that's the best you could do right now?

with vista ultimate 64 bit or something.........

yea, that would be my setup if i had the money:

another thing i would do is get a dual lga1366 server mb. 16 cpu cores.

Ananas at Simap has a nehalem server running with 2x L5520. 2.27GHz are not as high as possible but 60W TDP per socket sounds great.

---