I notice evga has a superclocked version of the GT240 out. About the only difference seems to be the memory clocks and the price tag. Has anybody tried one of these?

I was looking at replacing a couple of old GTX260 (65nm) with say 3 of these. Not as fast I know with only 96 cuda cores compared to the 216 of the GTX260. The GTX260 is getting rather hard to find these days. Also the nvidia drivers seem to have issues with more than 3 gpus under Win7.
____________
BOINC blog

The only thing EVGA overclocked was the memory,
http://www.legitreviews.com/article/1144/1/

I would not call 1800MHz superclocked!
1700MHz is normal. All my GT240 cards with GDDR5 are clocked at 1800MHz or more. My GPUs are clocked to between 610MHz and 640MHz (depending on the card) and the shaders are linked (about 1490 to 1500MHz). At these speeds they operate slightly better than half the speed of a GTX260. So three would do about 55% more work that one GTX260 and use about the same power - might cost slightly more though.

The setup is also important. A good CPU backing the GPU makes some difference. On my i7 I only found a 3 or 4% difference in using GDDR5 rather than GDDR3 (when both are overclocked), but on other systems the difference was more. I think a good CPU hides some of the shortcomings of GDDR3.
By the way, there is DDR3 as well as GDDR3! If anyone gets the DDR3 card it will be noticeably slower than a GDDR5 card.

Two GT240s in an i-7 (W7 x64) (GPU 612MHz, GDDR5 1800MHz, Shaders 1491MHz) RAC 28K.

GT240 (GPU 615, GDDR5 1824, Shaders 1498) on a Quad Opteron @ 2.1GHz

Tasks take around 11h, but there are a few that finish around 9h30min.

but I think it must much more slower than my old 8800 GT

Any GT240 is a good card for crunching here, but it is about their relative values.
If you can get a GT240 DDR3 for 20 or 30% less than a GT240 GDDR5 then it’s a good deal. If there is not much price difference then the GDDR5 card would be the one to go for. But some of the cards are better designed than other, so that has to be considered too.
The best designed card I have seen uses DDR3; it is a Gigabyte GBG016. It has a Large fan compared to many cards and the GPU is steady at 640MHz. It shipped with a Factory Overclocking of 600MHz (up from 550). I have not seen another GT240 reach that for GPU speed, but I am sure a few people will have managed it.

The Gigabyte's FOC offset much of the performance loss it would have by not using GDDR5, especially when compared to a native GDDR5 card. That said it is slightly slower than my overclocked GT240s with GDDR5.

It is presently very difficult to compare the cards when in different systems crunching one of many different tasks, and I think performance varies from WU to WU anyway.
All in I think we are only talking about 15% performance difference between native DDR3 and GDDR5 cards. Even a highly overclocked GDDR5 card might only be 30% faster than a natively clocked DDR3. In my opinion the top GDDR5 cards, that are 30% more expensive are not worth the money.

I picked up my last GT240 for £50 – a real bargain given that my first card was £65 and used DDR3.
It has GDDR5 and overclocks reasonably well (GPU 615MHz, GDDR5 1800MHz, Shaders 1498MHz).

Perhaps someday when I am really bored I will strip a GT240 and put silly heatsinks and fans on it (rather than the cheap tin that comes with most of them) just to see what can be got out of the card. I think that just using better adhesive and putting small heatsinks on the RAM could make a difference. In the past I found that positioning a 12cm case fan to blow directly onto a card made a lot of difference; temps dropped by over 10degC.
When I am overclocking in this way I try to get the temps down and then up the clocks until I come close to (but under) the previous temps, then tweak. For GPUGrid a good margin of error is required – There is no point overclocking a card by 15% only to find that 10% of tasks fail. You would be better off only overclocking to 10% and getting no failures.

http://www.gigabyte.com.tw/Products/VGA/Products_Overview.aspx?ProductID=3183

- a superclocked GTX260, 96MHz faster GPU core; performs as well as a GTX275!

I guess they have stopped manufacturing GTX275 and GTX285 GPU cores so there will be no such versions?

If it only uses the same power, then in terms of power usage vs performance it would be on a power with a GT240 (but crunch around 2.5 times as many tasks)!

Pity none of the manufacturers managed to use GDDR5 (to reduce the heat & power usage) - the GDDR5 prices are no longer too high. Such cards might have been useful over the last 6months, as would GTX275 and GTX285 versions!

GT200 doesn't know what GDDR5 is ;)

MrS
____________
Scanning for our furry friends since Jan 2002

I thought GDDR5 might have been usable if it could mimic GDDR3; with 25% higher clocks, so the card would benefit from reduced heat & power usage?

But on second thoughts, even if GT200 could be introduced to GDDR5 (perhaps by GT215), the limited heat and power benefits might not be worth the effort - higher GDDR5 latency at only 125% GDDR3 rates would actually reduce performance. The latency is only outweighed at much higher rates.

NVidia would have to redesign the memory controller to support GDDR5. That's not something GPU manufacturers generally do, since they normally update their chips quite frequntly anyway (refresh 6 months, new 1 year .. slowing down recently). If they update the silicon and card design, they can just as well update the architecture.. which, for nVidia, is Fermi. And remember: Fermi was not intended to be hot, slow and late ;)

MrS
____________
Scanning for our furry friends since Jan 2002

Well I ordered a few today. Not the overclocked version, but the normal (supposedly GDDR5) version. Hopefully the motherboard can cope with 3 in the one box.
____________
BOINC blog

What did you order and which board is it?
- I'm guessing you mean 3 GT240 cards.
Hope you read the Motherboard manual, and that you actually have three PCIE x16 slots.

What did you order and which board is it?
- I'm guessing you mean 3 GT240 cards.
Hope you read the Motherboard manual, and that you actually have three PCIE x16 slots.

You should be OK with that setup.
I tried a board with 3 but the third PCIE dropped to 1X and prevented any from working. Two ran on the same board, but it beeped on restart every time and occasionally restarted itself (but it was an MSI K9A12 Plat board, 790FX & SB600)! I have two on a different system working well.
Good Luck,

Beyond,
I noticed that the top card I have, that sits very close to a chipset heatsink, is about 10 degrees warmer than the lower card. Confirmed it is the heatsink/airflow by swapping the cards. Essentially the chipset heatsink is radiating heat onto the back of the card (GT240).

As the GT240 fan is on the front this cannot cool the back down, so the chipset is heating up the GPU from the rear.

So in one respect having a fan at the back of the GPU is therefore preferable, even if the fan is that of another GPU!

As I said in my earlier post I did try 3 but it did not work.
My K9A2 can in theory take 4 PCIE cards, but the system struggled with two GT240s, never mind four. At the time I did not have four cards to play with, but I might give it a go sometime.
I have however had 2 GPU cards and another non-GPU card just underneath the lower GPU, and a GPU sandwiched between 2 other cards in a tight midi case.
The problem with three or more is likely to be the exhaust air being brought back around and over the cards, especially the middle card(s). This would occur when the case fans are pushing too much air over the cards and the cards cannot expel all that air, so this warm air is brought back over the cards.
I alleviated this by reversing the airflow on a side case fan (turning it around, so that it blows air out from the GPUs), after installing a Big front case fan. I also found that leaving back plates off helps the air flow hugely. When I had my power greedy HD450 sandwiched between a sound card and a SATA RAID card in a midi case, I had to put a large fan blowing in from the front, and remove the back plates just to get it stable. The case had no side fan, and just a standard front fan was not enough to run the card stably. When I left the door off the temps plummeted, but it was too noisy so I took the back blanking plates off (a happy medium).

Tried to put 4 GT240 cards into my K9A12 Motherboard.
System boots, Boinc Loads but only sees 2 cards. GPU-Z can see all 4 cards as can EVGA Precision and the system, just not Boinc!

Tried latest drivers and Boinc Beta.
Edited cc_config to use all GPUs.

Still only sees 2 of the 4 cards!

Anyone?

15/04/2010 21:05:28 Starting BOINC client version 6.10.45 for windows_x86_64
15/04/2010 21:05:28 Config: report completed tasks immediately
15/04/2010 21:05:28 Config: use all coprocessors
15/04/2010 21:05:28 log flags: file_xfer, sched_ops, task
15/04/2010 21:05:28 Libraries: libcurl/7.19.7 OpenSSL/0.9.8l zlib/1.2.3
15/04/2010 21:05:28 Data directory: C:\ProgramData\BOINC
15/04/2010 21:05:28 Running under account X
15/04/2010 21:05:28 Processor: 4 AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2]
15/04/2010 21:05:28 Processor: 512.00 KB cache
15/04/2010 21:05:28 Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 htt pni cx16 syscall nx lm svm sse4a osvw ibs skinit wdt page1gb rdtscp 3dnowext 3dnow
15/04/2010 21:05:28 OS: Microsoft Windows Vista: Ultimate x64 Edition, Service Pack 1, (06.00.6001.00)
15/04/2010 21:05:28 Memory: 4.00 GB physical, 8.21 GB virtual
15/04/2010 21:05:28 Disk: 931.51 GB total, 700.11 GB free
15/04/2010 21:05:28 Local time is UTC +1 hours
15/04/2010 21:05:28 NVIDIA GPU 0: GeForce GT 240 (driver version 19745, CUDA version 3000, compute capability 1.2, 475MB, 257 GFLOPS peak)
15/04/2010 21:05:28 NVIDIA GPU 1: GeForce GT 240 (driver version 19745, CUDA version 3000, compute capability 1.2, 475MB, 257 GFLOPS peak)

15/04/2010 21:12:16 Re-reading cc_config.xml
15/04/2010 21:12:16 Re-read config file
15/04/2010 21:12:16 Config: report completed tasks immediately
15/04/2010 21:12:16 Config: use all coprocessors
15/04/2010 21:12:16 log flags: file_xfer, sched_ops, task

<cc_config>
<options>
<report_results_immediately>1</report_results_immediately>
<ncpus>0</ncpus>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>

Thanks,

Removed 1 card, leaving 3 but Boinc only sees 2. The system sees 3 as does GPUZ and EVGA Precision and NVidia control panel.

As it was doing nothing I removed the 3rd card as well.

Pity, 4 GT240 cards would do the same work as a GTX295.

Any ideas and I can put them back in.

Anyone?

I don't really know what you mean, dummy load?
Two cards presently work, but when I add a third or fourth they are not seen by Boinc.
Thanks,

I will have another play about with it. Fortunately I have 2 KVM's to try. I dont have a dummy VGA/terminator, though I might be able to make one.
Reminds me of having to set up a dummy network port or bus networking.
Odd that the 2nd card works without any dummy connector.

Tried the KVM switches (both), but no joy.
The computer (device manager) can see all 4 cards and that they have the same driver. GPUZ can see them as can NVidia Control Panel, but Boinc is blind.

Thanks Paul, I forgot about extending the monitors after I wired up the KVM.
Boinc is now able to see 3 of the 4 cards; an improvement.

I use Vista, so for anyone else with this problem that has Vista/Win7,
Right Click on the desktop, Personalize, Display Settings, then extend your monitors to the other cards.

I have 7 cards listed here! (Device Manager correctly shows 4 GT240's.
In Display Settings there are,
Four, (Default Monitor) on NVIDIA GeForce GT 240,
and three, Generic non plug and play Monitor on NVIDIA GeForce GT 240

So, to get the fourth card working...

I uninstalled the drivers, restarted and reinstalled the drivers, restarted again, and then looked at the monitor extentions. Initially I could see all four cards in the list but when I extended the monitor one of the listed devices vanished. I think this suggests that one of my connections is not working, so I probably need to get a different video cable or get a dummy VGA plug.

I did restart Boinc again.

20/04/2010 11:48:51 Starting BOINC client version 6.10.45 for windows_x86_64
20/04/2010 11:48:51 Config: report completed tasks immediately
20/04/2010 11:48:51 Config: use all coprocessors
20/04/2010 11:48:51 log flags: file_xfer, sched_ops, task
20/04/2010 11:48:51 Libraries: libcurl/7.19.7 OpenSSL/0.9.8l zlib/1.2.3
20/04/2010 11:48:51 Data directory: C:\ProgramData\BOINC
20/04/2010 11:48:51 Running under account X
20/04/2010 11:48:51 Processor: 4 AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2]
20/04/2010 11:48:51 Processor: 512.00 KB cache
20/04/2010 11:48:51 Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 htt pni cx16 syscall nx lm svm sse4a osvw ibs skinit wdt page1gb rdtscp 3dnowext 3dnow
20/04/2010 11:48:51 OS: Microsoft Windows Vista: Ultimate x64 Edition, Service Pack 1, (06.00.6001.00)
20/04/2010 11:48:51 Memory: 4.00 GB physical, 8.17 GB virtual
20/04/2010 11:48:51 Disk: 931.51 GB total, 677.85 GB free
20/04/2010 11:48:51 Local time is UTC +1 hours
20/04/2010 11:48:51 NVIDIA GPU 0: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 257 GFLOPS peak)
20/04/2010 11:48:51 NVIDIA GPU 1: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 257 GFLOPS peak)
20/04/2010 11:48:51 NVIDIA GPU 2: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 257 GFLOPS peak)


20/04/2010 13:27:12 Re-reading cc_config.xml
20/04/2010 13:27:12 Re-read config file
20/04/2010 13:27:12 Config: report completed tasks immediately
20/04/2010 13:27:12 Config: use all coprocessors


Thanks,

Finally, another cable brought Success!

20/04/2010 16:45:05 Starting BOINC client version 6.10.45 for windows_x86_64
20/04/2010 16:45:05 Config: report completed tasks immediately
20/04/2010 16:45:05 Config: use all coprocessors
20/04/2010 16:45:05 log flags: file_xfer, sched_ops, task
20/04/2010 16:45:05 Libraries: libcurl/7.19.7 OpenSSL/0.9.8l zlib/1.2.3
20/04/2010 16:45:05 Data directory: C:\ProgramData\BOINC
20/04/2010 16:45:05 Running under account X
20/04/2010 16:45:05 Processor: 4 AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2]
20/04/2010 16:45:05 Processor: 512.00 KB cache
20/04/2010 16:45:05 Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 htt pni cx16 syscall nx lm svm sse4a osvw ibs skinit wdt page1gb rdtscp 3dnowext 3dnow
20/04/2010 16:45:05 OS: Microsoft Windows Vista: Ultimate x64 Edition, Service Pack 1, (06.00.6001.00)
20/04/2010 16:45:05 Memory: 4.00 GB physical, 8.17 GB virtual
20/04/2010 16:45:05 Disk: 931.51 GB total, 677.68 GB free
20/04/2010 16:45:05 Local time is UTC +1 hours
20/04/2010 16:45:05 NVIDIA GPU 0: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 281 GFLOPS peak)
20/04/2010 16:45:05 NVIDIA GPU 1: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 281 GFLOPS peak)
20/04/2010 16:45:05 NVIDIA GPU 2: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 281 GFLOPS peak)
20/04/2010 16:45:05 NVIDIA GPU 3: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 281 GFLOPS peak)

Hardly superclocked (from 550MHz to 600MHz, with linked shaders and modest GDDR increases, 1800MHz), but I think it's fair to say the GT 240 has now been tried ;)

Thanks guys,

I suppose you can crank up the shaders much more than the core and GPU-Grid reacts well to shader clock :)

MrS
____________
Scanning for our furry friends since Jan 2002

Thanks for the tip.
I will leave them as is for today, and have a go at upping the shaders tomorrow, after I pass my milestone :)
Although the case door is off, temps are between 54 and 64 deg C, so I should have a bit of leeway yet. The door also has a fan, in a very nice place and I can add another fan to the front of the Antec, nice and low, just opposite the GPUs.
On single card systems and dual card systems I can usually get slightly over 600MHz for these cores with the shaders tied; about 615MHz or 620MHz (though some can reach 640MHz) so if the shaders clock well then perhaps they will go past 1550MHz with the core kept at 600MHz?

I was actually considering changing that system (not the case), no chance of that now!

I wouldn't be surprised if your shaders hit 1.7 GHz :)

MrS
____________
Scanning for our furry friends since Jan 2002

For now I am keeping the shaders at a slightly more modest 1.6GHz on the system with the four GTX240 cards.

I have other systems to test on, but when I tried a DDR3 Gigabyte GT240 I could not hit 1.7 without an almost immediate failure. Mind you the GPU on that card is at 640MHz and the Shaders now seem stable at 1.65. Not bad considering the NVidia ref. rates are 550MHz and 1340MHz.

The real test is to run the cards without any task failures.

- Typo, should read GT 240 (no X).

The 4 cards ran overnight without error :)
GPU 600MHz, Shaders 1600MHz.

I've got a 8600GT which runs at 540 / 1180 MHz default. It's a 65 nm chip and easily reaches about 650 / 1700 MHz :p

MrS
____________
Scanning for our furry friends since Jan 2002

Alas the Gigabyte GT 240 failed a task when the shaders were at 1650MHz, and after 6h! The GPU was only 51 deg C, so I think the overall card cooling (capacitors and smaller chips) might have been the issue. The system was running 4 CPU tasks, so I'm sure that would not have helped; the case temps would have been a fair bit warmer as a result. I think I will leave all the cards at 1625MHz for now, to see how they get on for a few days. I may have a go at a different card in another system when I get the time.

On your 8600GT comparison,
You are comparing a seasoned GPU with 32 shaders to that of one with 96 shaders.
Although the 8600GT has a default clock of 540MHz with shaders at 1180, its big brother, the 8600GTS, sports a 675MHz GPU and 1450MHz shaders; so there was lots of potential there!
Was it stable crunching over long periods of time at 1.7GHz?
Did you use non-standard cooling?

If my 1625MHz shader rate stands the test of time, that is still a 21% increase :)

Well machine is finally back home. Its got 3 x GT240's in it from evga. Only two recognised by Win 7 initially until I shoved a spare KVM cable into the back of number 3.

I asked the computer shop if they have any dummy vga plugs, but they have never heard of them. A quick google and while there are plans on how to make them on the net, I can't find anyone selling them in Australia.

Anyway its going for now after reinstalling Windows and setting it up the way I have all the others. I'll post some pics to the blog later.
____________
BOINC blog

Temperature is fine, as is yours. It's actually happily running Collatz.
And, sure, it's a smaller chip.. but that doesn't really matter. What's important is the design (pretty similar), the process node (65 vs 40 nm) and the chip voltage. NVidia appears to be very generous here on the 860GT, it could have saved ~5W under load (with lower voltage) and probably still reached similar yield. On GT240 they probably didn't "overshot" that much with voltage, because here the power consumption matters: they need to stay below the limit of the PCIe slot.

MrS
____________
Scanning for our furry friends since Jan 2002

For now I'm reasonably happy with what I am getting out of these cards.
As you suggest, I suspect I am being limited by the voltage, but measuring GPU heat does not tell me anything about board temperatures so perhaps if I can reduce the heat the cards can be tweeked slightly better.

I do have one open system that I will play with further. First would like to put heat spreaders onto the RAM, but this may not be physically possible, due to the GPU heatsink. I still have to add a system fan and possibly another fan towards the back of the card, or onto the motherboards chipset heatsink, as this would be radiating heat onto the card. I will keep an eye on the temperatures for improvement. If they drop I will try to up the clocks again and test for stability.

After that I may turn the voltage up ever so slightly, and see if I can get some more from the cards shaders, but there is no way I am going to up the voltage on the 4 cards in the same system - that's just asking for trouble!

Don't expect much from lowering the temps - it does not affect maximum clock speed much and your temperature is already low, i.e. there's not much room for improvement anyway. Similar for RAM heat sinks - it's seldom they lead to any measureable improvements (logic: if they needed cooling, they'd already have it). Improved case airflow never hurts, though!

And strictly speaking: if 100% GPU fan speed keeps the chip temperature below 90°C any GPU OC is voltage limited ;)
But that doesn't mean I'd suggest increasing voltages, as it also reduces lifetime. Not sure what I'd do if I could increase it via software on my cards..

but there is no way I am going to up the voltage on the 4 cards in the same system - that's just asking for trouble!