The GT240 is designed to fit into smaller cases, so that is its place.
Sticking two together would make it perform slightly better than a GTX260, but only just. There would be a place for such a card as there is a gap between the GT240 and the GTX260, in terms of cards using a G200 core, that work well here, and it would be more economic to run than a GTX260. That said it is only CC1.2 and there is likely to be lesser Fermi cards that would fill the role better (GTX440 or something), and Fermi is CC2.0.

I thought that.
Actually my quad GT240 system almost does the same amount of work as two GTX275's, slightly more than a GTX295 :)
It does not need any additional power connectors, but a dual card would be better with one, as it would give you more room to clock.

Never heard of such a card. It probably wouldn't make much sense for the manufacturer, as these things are made for games. Here a dual GT240 would have to face the less-than-ideal scaling via SLI (contrary to GPU-Grid) and the memory overhead due to replicating data for both chips (one larger chip woudln't have to do this). So unless a GT200 chip is considerably more expensive than 2 GT215 chips (for GT240) I can't see manufacturers even considering such an option.

Performance-wise it should be quite good at GPU-Grid, though: it brings along 2 x 96 = 192 shaders at 1.35 GHz, whereas the old GTX260 ran at 1.25 GHz. To tie a GTX275 (240 shaders) it would have to run at ~1.75 GHz, though. 4 GT240 should need approximately 1.56 GHz to tie a stock GTX295 while consuming an estimated 4 x 80 W = 320 W versus 289 W for the GTX295; going by TDP numbers and increasing the GT240 base value of 69 W at 1.34 GHz linearly for an OC of 1.56 GHz. Actual GPU-Grid power consumption will be lower for both cards. And since the architectures are quite similar we can assume a similar reduction factor for both.

MrS
____________
Scanning for our furry friends since Jan 2002

Yesterday I was told by a hardware salesman that there was a GT240x2. He wasn't sure about the manufacturer & said that it "might" be POV. I checked out their site & googled, but couldn't find anything that wasn't Chinese. I don't speak Chinese.

Does anyone know who makes the GT240x2 & how much it costs? I'm curious to know if it's cheaper to get 2 of these instead of 1 GTX295, if it'll get as much done, & maybe use less electricity.

As usual my opinions are somewhat tunnel versioned; A dual GT240 would not make for an especially good gaming card, but the G200 cores are on their way out. Last I heard manufacturing was all but over for most of the cards.
So that just leaves the gap between the GT240 and GTX260 to be filled with older 55nm or 65nm cards such as the GTS250.
I can’t see NVidia wanting to keep building with such old manufacturing techniques, when at the same time phasing out the G200 GPUs.

The TDP of a GT 240 is 69W, so that is 276W for 4 cards.
When natively clocked, I found that my Two GT 240s (DDR5) did slightly more work than my slightly overclocked GTX260sp216, which in turn, did slightly more than the reported work of a natively clocked GTX295 (on similar operating systems).
Running 4 GPU tasks, as opposed to zero GPU tasks and zero GPUs in the system, I measured a difference of about 243W. So as you said MrS, the actual running of the card “as a system” works out less, about 61W per card.

Steve, you are quite right to point out that I was talking about my overclocked GT240s and a natively clocked GTX295. However if you look at my GT240 system, and take the RAC, it is misleading because it has not yet levelled out since I moved the cards about (still rising). Another thing to take into consideration is the operating system (your streamlined XP vs my heavily used Vista system).
Although my GT240s are now overclocked, I am using stock voltages and the shaders are at 1.625GHz, so my OC is limited, and there is not an absolute need to rise to 80W per card. Of course a dual card using about 2 X 80W would be nice and would surely bring the reference clocks close to 1.7GHz as well as raising the GPU and RAM.

My 1.625GHz shaders brought the Boinc GFlops up from 281 to 312 peak (note that DDR3 cards are 257, 9% less than the native DDR5 version and 21% less than my OC’d GT240’s):
01/05/2010 14:35:14 NVIDIA GPU 0: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 312 GFLOPS peak)
01/05/2010 14:35:14 NVIDIA GPU 1: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 312 GFLOPS peak)
01/05/2010 14:35:14 NVIDIA GPU 2: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 312 GFLOPS peak)
01/05/2010 14:35:14 NVIDIA GPU 3: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 312 GFLOPS peak)

I would guess 4 GT240s (GDDR5) could do about 5% more than a GTX295 (at reference clocks), or at least come very close to matching a GTX295, and if you can run the cards with 1.625MHz shaders then that could rise to anywhere between 10 and 25% (if compared on the same operating system and to a reference GTX295, which of course might very well clock better). Four GT240 DDR3 @ 257 Boinc GFlops peak would not fair so well; about 5% less than the native GTX295, and substantially less than a highly overclocked GTX295. So it is an apples and oranges comparison.

Anyway, GT240’s are easier to find and cheaper to buy than GTX295’s, or Fermi’s for that matter!
Also, if a GT240 breaks you still have 3 running and it would only be £50 to £70 to replace one, whereas a GTX295 is not so readily replaceable.
On the other hand you could have 4 highly overclocked GTX295’s in the one box, and blow everything else away, right up until there is a highly optimized Fermi app.

At the minute there are many different WUs. Some long, some short and some Betas.
Linux is clearly faster at the minute.

My four OC'd GT240s are already bringing back about 43K RAC per day, still rising and should even out at over 50K; One of my other GT240s brings back over 14K per day, so I could reach 56K.
From what I can see, your 1 OC'd GTX285 + OC'd GTX295 system brings back 75K daily, and a GTX285 does more work than 50% of a GTX295.

Snow Crash, all you had to do was actually say what your daily return was; but you chose not to - you kept your cards to your chest. Why?
If you have a point just spit it out and say it. You are not helping!

By the way it was you that argued that XP was much better than Vista and W7; selective memory perhaps?

Oh, and per chance, are you now comparing your over clocked card, and not a natively clocked card? You know, just to keep everyone informed!
Voltages?

As I clearly stated, 2 GT240's (GDDR5) do more work than a GTX260sp216 - measured on the Exact same system and setup!

Your argument that CPU, Operating System, and x86/x64 does not matter is just wrong - so who is misleading who now?

I have a GTX470. It's TDP is 215W and its idle power consumption is 33W.
The GTX 480 has a TDP of 250W, and an idle power consumption of 47W. also.
For comparison,
- GTX275 TDP 219W
- GTX285 TDP 236W
- GTX295 TDP 289W

As for real world usage (on GPUGrid):
When the Fermi in my system is crunching, the system uses about 375W.
When the Fermi is not crunching it uses about 275W.
So when crunching it uses an extra 100W (in its present form), compared to being idle, which suggests the card only uses 137W to crunch on GPUGrid (mind you I'm having issues with the performance of this card, so no guarantees until I test it in another system).

I could work out what the idle power usage is if I swap it with a GT240:
The GT240 card only uses about 3.5 or 6W in itself, when idle, but the system uses an additional 10W in total; what we really need to know.

Initial figures suggest that the GTX480 is about 35% faster than a GTX285.
The GTX470 is about 21% slower than a GTX480, which would make it about 14% faster than a GTX285, at the minute!

So, a GTX470 can already do more work than a GTX285 (should tasks turn up) and uses less power :)
However, the present application has not been refined and developed for Fermi yet.
I expect they will do some basic refinements and release a rudimentary Fermi app, and then work on Fermi refinements over several months.
It might take CUDA refresh releases before we see a very efficient Fermi app.


Going back to compairing a GTX295 with 4 GT240's (much less fun):
Paul’s GTX295 (at stock & reference) on a Q9300 and W7 64bit:
Run Time 42,886.53
Credit 11,931.63
http://www.gpugrid.net/workunit.php?wuid=1377976
http://www.gpugrid.net/show_host_detail.php?hostid=52239
24X60X60= 86400seconds in a day.
86400 / 11931.63 = 2.0146 tasks per day.
Since the GTX295 has 2 cores, that is 4.029 tasks per day.
Or, 11931.63 X 4.029 = 48K credits per day.

My OverClocked GT240 with GDDR3 (now one of the 4 cards in the same system) Phenom II 940, W7 64bit:
82,240.24 sec for 11,931.63 points on a GT240
1.0505 tasks per day
Or 12535 points per day for that one card. Or 50,140 points per day.

Another one of my OverClocked GT240s, this time with GDDR5 (also one of the 4 in the same system):
30,686.46 sec for 5,144 credits
...
14484 credits per day for that one card.
Or 57939 points per day for 4 of these GDDR5 cards.


Overclocked GTX295 running on a faster operating system (XP) and supported by a faster CPU (i7-920)
12,861.44 sec for 4,247.95 credits:
57073 points per day.
http://www.gpugrid.net/workunit.php?wuid=1422132
http://www.gpugrid.net/result.php?resultid=2254916

I think the RAC is more informative because some tasks perform better than others. There is likely to be variation between card types and relative task crunching performance; a GT240 might crunch one task more quickly than another, but get the same points, for example.

Guys.. come on, don't let this discussion over hard facts turn into personal insults. And let's get a few things straight:

- OS, driver, CPU and whatever do influence the real world performance in some way. But they do so in a similar way for all cards within the same system, so as long as we're not discussing RAC values this doesn't matter (I don't intend to).

- The GT200 and GT215 chips have a very similar architecture (except no double precision for GT215, which doesn't matter here) and thus their shaders perform similarly, on a per clock basis.

- Neither of these cards is particularly memory bandwidth limited (not talking about DDR3 GT240). GT240 with GDDR5 has 386 GFlops and 54.4 GB/s bandwidth, i.e. 7.09 GFlop/GB. GTX295 has 2*894 GFlops and 2*119.9 GB/s, i.e. 7.46 GFlop/GB. That means the GTX295 has a little more memory bandwidth per raw compute capability, but both are clearly in the same league. To put things into perspective: a GT240 with DDR3 would be at 12.06 GFlop/GB, which I think we established costs about 15% GPU-Grid performance compared to the GDDR5 version. Based on linear interpolation one can drive a correction factor of 1.1% for the GTX295, but I'll ommit this.

Summing this up we can clearly say that in order to judge the cards performance potential it's sufficient to look at maximum theoretical GFlops. Real world performance will drop by a factor x with 0 < x < 1, but it will be the same x for both chips / cards (it's different for compute capability 1.1 cards, though).

Now where does that get us?

We've got GTX295 at stock speeds bringing in 2*894.2 = 1788.4 GFlops. Four GT240 GDDR5 land at 4*385.9 = 1543.6 GFlops, i.e. a factor of 1.16 lower. Conversely the GT240 has to be clocked 16% higher to tie a stock GTX295 and to make up for the lower amount of shaders. Make it 15% if you want to account for the memory bandwidth differences. 16% would be the 1.56 GHz shader clock for the GT240 I mentioned in my previous post. OC the GTX295 and you quickly reach the clock speed region GT240 can not reach. So I think we can all agree on:

I'm not saying they [GT240] are not good cards, they are, I'm just trying to make sure we present accurate stats to the best of our ability to help people make their own decisions.

Measured the actual idle power consumption of my GTX470 to only be 28W (this includes the power used by the motherboard to support the card).

Measured actual power usage difference of 4 GT240s when natively clocked and OC'd to; GPU 610MHz, GDDR5 1800MHz, Shaders 1625MHz:
9W total (yes that is for All 4 cards combined)!

Measured actual power usage of the above 4 overclocked GT240s when running GPUGrid tasks:
System used 340W when crunching GPUGrid tasks
System used 190W when not crunching GPUGrid tasks

Did that while crunching 3 CPU tasks, and then without CPU tasks:
299W crunching only GPUGrid tasks
145W crunching no tasks of any kind

Each costs the system 10W when idle, so the total Watt usage of 4 OC'd GT240's crunching on GPUGrid is 192W.
Four natively clocked cards use about 183W.

These figures should transfer well into other systems. But people should be more concerned about overall system efficiency than the card. A GT240 in an i7-980X is a poorly balanced system.

I'm going to look at the actual performance difference from one of the cards when natively clocked to the other overclocked cards, to see what I get for that extra 9W.

Well,
It would appear that under Vista, with optimized conditions and using Four natively clocked GT240's (GDDR5), you could get 53K per day at GPUGrid, as is.

Four overclocked GT240's could bring 56K, as is (several restarts may have held these figures back),
and the Beta app (6.22) could see 66K with 4 overclocked GT240's (again on Vista).

If XP is 11% faster, as reported;
4 natively clocked GT240's could get 58K per day on XP
and 4 OC'd GT240's could get 62K per day (72K with the new app).

Welcome to the forum,

Like your numbers :)

You are bringing some good systems with you.

Your GTX480 should do well in that 17-930, especially if you leave a thread free for it.

SK, the additional ~45W under GPU-Grid load you measured for a GT240 match very well to the power consumption number posted in that article - very nice!

MrS
____________
Scanning for our furry friends since Jan 2002

My position on 4x GT240 vs 1x GTX295:

Looks like roughly a tie except for one thing...

The GTX295 is about $500 while you can pick up a GDDR5 GT240 for $45-$60 each (after rebate).

Just ordered an XFX GT240 for $85 with 2 free softwares and a $30 rebate. Total cost after rebate $55 (free shipping, no tax).

Unfortunately rebates tend to be limited per household so I only got one.

My Gigabyte GDDR5 GT240 was only $45 after rebate.

At $55 each, 4 GT240 cards would cost $220.

Then again the GTX295 will fit in a single slot motherboard which are cheaper than the 4 slot main boards it would take to run 4x GT240s. Still 4x GT240s would end up being quite a bit cheaper...

If anyone is interested in the XFX GT240 @ $55 (after rebate):

http://www.newegg.com/Product/Product.aspx?Item=N82E16814150452

Several GT240 being so much cheaper is probably the point why the comparison has been brought up in the first place. And I think real GPU-Grid power consumption also favors the newer cards, by quite a margin.

MrS
____________
Scanning for our furry friends since Jan 2002

My four GT240s are now crunching 6.72 tasks (had to use an older driver on Vista 64bit). With only the shaders clocked high (1600MHz, the GPU core and RAM is native), these 4 cards could could get over 70K credits per day, on Vista! 86400/33000*6756=70.7K

- 2336422 1476441 15 May 2010 17:56:50 UTC 16 May 2010 3:12:30 UTC Completed and validated 32,999.91 3,178.07 4,503.74 6,755.61 Full-atom molecular dynamics v6.72 (cuda)