The 8800GTS is a G80 type and only supports compute capability of 1.0, whereas GPUDGrid currently requires CUDA 2.0 w/ compute capability 1.1 based on required hardware features (see atomic functions above). The G80 cards will never be supported because they can't do what GPUGrid requires in the calculations. Unfortunate, but that's just the way it is. You just happen to have one of the few cards that will not function based on its compute capability.

If you still have the card you might want to look into running it with Folding@home. I think it is capable, but I dont know for sure what their requirements are, you will need to read up on them yourself first of all, but it might be an option.

http://folding.stanford.edu/English/WinGPUGuide

Good Luck
____________

CUDA 2.2 has now been released.

What, if any, impact will that have on the availability of cards to participate?

Obviously Compute Capable 1.3 cards will still be able to participate, but what about all the 1.1 cards?
____________

Make that Cuda 2.3

along with NVIDIA Driver 190.38

So does my GTS 250, for now.

I would just like a heads up if GPUGRID decides to only support 1.3 and above some time in the future, especially with new cards expected in the Autumn!

In view of the new cards coming out in December. We are currently recommending 100 stream processors.

gdf

Just a small correction. Only compute capability 1.1 is required, so your card is supported by gpugrid.

gdf

Uninstall Boinc, clean the registry (CCleaner), reinstall latest Boinc and make sure you dont install it as a system process!

GTX275, GTX280, GTX285, GTX295 and Tesla 10 cards work.

GTX260 216sp uses a 55nm GT200 GPU core. Most should work with the project.

The Geforce GT 240 and Geforce GTS 240 cards should also work.

Many of the older GTX 216 192sp cards have problems running GPUGRID as do most G92 compute capable 1.1 cards, including the GTS250.

G90 cards no longer work.

Lesser cards such as the NVidia ION with only 16 shaders and the NVidia GT 220 would be unlikely to finish in time.

A G210 might finish in 3 or 4 days, or might time out.
The lowest card depends on the user and system. If the system is on and running GPUGrid tasks and has a good CPU it is more likely to finish for any given card.
The minimum reccomended would probably be a GT240, but if you have a GT220, or GT315 (basically the same thing) you could finish tasks if the system was running most of the time. A couple of months ago I got a task to finish on an ION (40 Boinc GFlops, the new version, 7 GFlops on older Boinc versions), but it did take about 5 days.

Exactly what specs does the GTX260 have?

G210 - No Use, Avoid at ALL costs!

GT230M is not as good as the GT220, which is not good.

http://www.gpugrid.net/forum_thread.php?id=1150

Hi scrap, welcome to the forums.
The GT 240 is a good card, very reliable, and energy efficient.
A good choice.

robertmiles, it is likely that if you removed one GT9800 card from your system and replaced it with 2 GT240 cards, it would use about the same power.
A GT9800 uses about 130W under load and the GT240 uses almost exactly half that!
The GT 240 does not require any additional power connectors; the PCIE slot is sufficient to power the card. If your PSU was able to support the GT9800 is should be able to support 2 GT240's.
That said, if you intend to overclock you should calculate how much power you actually use, give yourself some head room, and determine if your existing PSU is up to the task.

Koschi made the table some time ago.
All the Cards you list are good here.

Two cards to avoid, the GTS 240 and the OEM GT 330; they use G92 cores!
NVidia now have a G92 in the GeForce 8800, 9800, 100, 200 and for the set the GeForce 300 series.
Of course NVidia does not list such information on their site,
http://www.nvidia.com/object/product_geforce_gt_330_us.html

Hi Sid,
There have been a number of application updates. These have led to an increase in performance in most cards, but some have seen problems; notably CC1.1 cards (this includes the GTS 250). I experienced the same problem with my 8800GT and my GTS 250. I replaced them with GT 240 cards to good effect.

Until recently 2 types of task were available (6.03 and 6.71). The 6.71 tasks have now run out. These used the older application, and seemed to work better with CC1.1 cards but did not see the 60% increase in speed that the new 6.03 tasks (which use the newer application) see.

There have also been several Betas out over the last month or so. The latest also saw a 30-45% improvement, but again depends on the new application.

Normally motherboards have the onboard card set as default, but as soon as you plug in a PCIE card it becomes the card which has to be used with the monitor!

I tried what you are talking about recently, but I was not able to use my onboard card for the monitor after I plugged in a PCIE card – same on two different systems. Mind you one was a Dell and the other an HP system! So, perhaps on a better motherboard you could do this?
Check out the online motherboard manual before you buy one, it might say something about this (if it does not, expect that you cant).

I think any performance gain would depend on your use of the system; for browsing the web and working on documents it won’t be much, but if you play games or watch media 10hours a day it could be considerable.

In the past non-similar cards did not work together well, if at all. Although this was reported to have been fixed, I have my doubts that this is the case for all cards and systems. I tried this in the past with a GTX260 and an 8800GTS (Boinc reported two 8800GTS cards). After trying it recently with a GTX 260 and a GT240 I can say that it did not work for me at all, but again this is likely to be very motherboard dependant. I also had trouble with two GT240s on that Micro-Star (K9A2 Plat) board, but no issues on my Gigabyte board.

It would be useful if someone put together and maintained a failsafe (or best chance) description of how to do this.

The 9800GT might be quite a bit better than it is getting credit for in this thread.

First of all there are new "Green" or "low power" versions of the 9800GT that lack the extra power cord and draw all their power directly from the motherboard. The power draw for these "green" cards is under 70 watts.

Secondly, near as I can tell the 9800GT produces nearly as good as the GT240 on GPUGRID (depending on the exact version of each card). I'm not at all sure why it is listed as "might work"??

Anyhow for running GPUGRID the GT240 is most likely still a better card for the money but the 9800GT cards are a lot closer in performance than you might think.

The GT240 is better because it has Compute Capability 1.2, so it is better with CUDA. This is despite the fact that it only has 96 shaders. The 9800GT has 112 shaders, is often clocked higher and is a better card for playing games - it is not as good for crunching here. So the advice for people wanting to get a basic card to crunch with is to get a GT240. They are fairly inexpensive, simple to install and fit most systems. This does not mean people cant use a 9800GT, they may well be able to. Some of these cards work fine, others not so well.
On your Vista system you completed your last work unit in good time and got full bonus credits for early return. If that card constantly performed as well as that you could expect a return of 11,166 credits per day. A typical GT240 will get about 13K per day on Vista/W7, or 15K on XP.
Your Linux system is much faster at crunching than Vista and would get you about 12.6K per day.
Of course returns depends on a number of factors:

Your overall system spec, and in particular CPU speed (an OC can help)
CPU usage for other tasks (especially crunching CPU WUs for other projects)
Your operating system (Linux and XP are generally faster than Vista and Win7)
The optimizations in place
- free core/thread; to help speed up GPU tasks, use of swan_sync=0
- tasks you run, presently some betas (6.72) are faster (this changes)
- overclocked GPU (especially the shaders)
- NVidia configuration

I think in general, it is also fair to say that the GT240 is more reliable and returns more tasks in time for full credit. While as you point out, there are Green versions of the 9800GT, which do not require additional power connectors, these are not quite as fast as the cards that do have the extra power. I'm not sure why the GT240 is not called a green card?
There are many GPU manufacturers and each employ their own designs and specifications. This in part explains why some cards struggle while others work perfectly. It also explains some of the performance differences (with clock differences and firmware versions). So it’s a cloudy picture, and the simplest suggestion for a new cruncher, wishing to buy a cheap card to contribute is to get a GT240; it’s almost full-proof.
In the future it is possible that GPUGrid will move away from CC1.1 cards, and only use CC1.2 or above. So while some 9800 cards may be almost as good as a GT240 now, there is no guarantee that will be the case in the future. So if you have a good 9800 for crunching use it, but dont rush out to buy a new one to crunch with.

Thanks for your answer, you clearly know what you are doing.

I've currently got 2 different 9800GT cards in 2 different computers. Both are "green" and both seem to work just fine so I was wondering why they had the reputation that they do. Thanks again for the informative reply.

I'm not sure why the GT240 is not called a green card?

My dual "Nvidia 8800 GTS 512" are doing pretty well (12884.58) at the moment. Should they maybe be green? Or are they not cheap enough?

Yes!
Fermi cards can only run Fermi tasks:
Full-atom molecular dynamics v6.73 (cuda30)

They cannot run 6.03 Work Units.

To receive 6.73 tasks,
- create a profile for your Fermi,
- configure it to only accept 6.73 tasks:

http://www.gpugrid.net/prefs.php?subset=project


Run test applications? YES

Run only the selected applications ACEMD: YES
ACEMD ver 2.0: no
ACEMD beta: YES

This might be a bit closer to a translation:

Ah, there are no Chinese-speakers. I did not pass English.
I have a GT260 graphics card. What about XFX's use of Sha acceleration?

怎么样讯景的沙加速用？
我们不都讯景图形处理器。

The GTX460 cards are still only at the Beta stage, but they do work. At the present time the GTX460 is about half the speed of a GTX480. It is likely that application developments will improve this over time.

This post might help you decide what card to get and when,
Boinc GPU Card comparisons in GFLOPS

[edit] There has been a working application for the GTX460s and full tasks for a few days now. [/edit]

- GTS 450, 440, 430 are not out yet, or did I miss something?

- the GF 8800M GTS (G92, 64 shaders, 240 GFlops) is marked as green, whereas GF 320 (GT215, 72 shaders, 281 GFlops) is marked as "to slow". Considering the newer architecture of GF 320 it's equivalent to >400 GFlops of the old CUDA Hardware 1.1 chips.

- should be "too slow" instead of "to slow"

- GT240 is green, but GF 340 is yellow.. they're both the same cards

- G92 cards are marked as green for mobile and quadros, but yellow for desktops

Don't want to sound harsh.. keep up the good work guys :)

MrS
____________
Scanning for our furry friends since Jan 2002

The GF106 and GF108 Fermi cards have not been released yet. Going by recent reports, which say manufacturers have not received samples yet, they now look more likely to turn up in September than in August. When they do arrive there is no guarantee they will work straight out of the box. They are likely to be significantly different from GF100 and GF104 designs. So the scientists will probably take a few days to get working code. Perhaps coming soon rather than recommended.

NVidia said the entire range of Fermi cards will support CUDA, so it’s a reasonable bet that all but the slimmest Fermi’s can be made to work here.

By the way, there is only one GT240 and it uses a GT215 design, not a GT240 design; it’s listed twice, once as GT240 and once as GT215.
I think that GeForce 230 may be some sort of bespoke OEM product or more likely a GT 230M (158 GFlops), and therefore in the wrong chart. It would probably be too slow anyway; less than a GeForce GT 220. GT 230M's are also for laptops, which are generally not recommended for GPU crunching.

Although slower than a GT 240, both the GeForce GT 320 (72 shaders) and the GT 315 should work, both are CC1.2. The GT 325 is essentially a GT 220, so although it is slow it would get through 1 normal length task in about 24h. As listed the GT 210 would be too slow.