Hi!

a) What is the cheapest possible graphics card for crunching PS3grid.net workunits?
b) What is the best/fastest possible graphics card for crunching PS3grid.net workunits?

Thanks!
____________

Hi!

a) What is the cheapest possible graphics card for crunching PS3grid.net workunits?
b) What is the best/fastest possible graphics card for crunching PS3grid.net workunits?

Thanks!

Hi!

a) What is the cheapest possible graphics card for crunching PS3grid.net workunits?
b) What is the best/fastest possible graphics card for crunching PS3grid.net workunits?

Thanks!

a) Geforce 8400GS - but better run it on a single core PC, as otherwise the other units won't finish before the deadline, because the card is sooo slow. Only few shaders at a low clock speed
b) Geforce GTX280 - if you can afford it

8800GT 512MB
gdf

8800GT 512MB
gdf

Thanks! :)

So, is this the correct model?

Does this card support CUDA 2? Will it work fast enough in PCI Express 1.1? I only have PCI Express 1.1 motherboard.

Where to get the latest drivers for that card? I have no experience on using NVIDIA cards (I have an ATI card at the moment).

Henri.

I run 3 8800GT, 1 in each computer. They work very well and I see not much slowdown with normal windows operation. There is some noticed depending on what I'm running, most mostly not too bad. Some people have reported they can't even use their computer when it runs here. My two at work I use all day, heavily with little noticed slowdown and they are both P4-HT, running full boinc using both CPU and GPU. These were well priced for me, not to expensive like the high end cards. Mine are XFX brand and are single slot wide, an important thing to consider as some computers, especially all mine, can not take double wide cards without sacrificing another PCI slot, which I could not do as I have other boards and no empty slots to move them to. These are PCIe x16 2.0 but that is backward compatible with PCIe x16 1.1 slots. They worked fine in my PCIe x 16 1.1 slots.

Mine came with a double life-time warranty. I guess that means when I die, I get to take it with me to the afterlife ;)

Number of stream processors (shown as cores) is important, less processors means it the application takes longer.

512MB memory would be good.

There is 1 device supporting CUDA

Device 0: "GeForce 8800 GT" (640MHz version)
Major revision number: 1
Minor revision number: 1
Total amount of global memory: 536543232 bytes
Number of multiprocessors: 14
Number of cores: 112
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 16384 bytes
Total number of registers available per block: 8192
Warp size: 32
Maximum number of threads per block: 512
Maximum sizes of each dimension of a block: 512 x 512 x 64
Maximum sizes of each dimension of a grid: 65535 x 65535 x 1
Maximum memory pitch: 262144 bytes
Texture alignment: 256 bytes
Clock rate: 1.62 GHz
Concurrent copy and execution: Yes

Test PASSED

There is 1 device supporting CUDA

Device 0: "GeForce 8800 GT" (600MHz Version x 2)
Major revision number: 1
Minor revision number: 1
Total amount of global memory: 536543232 bytes
Number of multiprocessors: 14
Number of cores: 112
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 16384 bytes
Total number of registers available per block: 8192
Warp size: 32
Maximum number of threads per block: 512
Maximum sizes of each dimension of a block: 512 x 512 x 64
Maximum sizes of each dimension of a grid: 65535 x 65535 x 1
Maximum memory pitch: 262144 bytes
Texture alignment: 256 bytes
Clock rate: 1.51 GHz
Concurrent copy and execution: Yes

Test PASSED

See some of the other reports by users in other threads.

Links to downloads are on the front page. NVIDIA supports their product well on their website. Visit the CUDA Zone section for the CUDA drivers. See the FAQ section for a list of cards that are supported here.

8800GT 512MB
gdf

I recomended 8800 GTS 512 - 128 procesors....

It is 2.0 PCI-E but it should work without problems on PCI-E 1.0

drivers: http://www.nvidia.com/object/cuda_get.html

8800GT 512MB
gdf

I recomended 8800 GTS 512 - 128 procesors....

It is 2.0 PCI-E but it should work without problems on PCI-E 1.0

drivers: http://www.nvidia.com/object/cuda_get.html

I think the FX9800GTX+ 512MB is a good (ot better) solution. You have to pay 10 EURO more but you will get more power. Isn't it? Or am I wrong?

Thanks again!

One more thing. I want the card to be as silent as possible. It would be very annoying to crunch ~24/7, if the GPU fan yells all the time. So, what exact make and model is quiet enough?

Henri.
____________

And let's get serious about the "most effective card" question.

MrS

Once upon a time... 2 weeks ago... we all happy owners of:

# Geforce 8800 GTS - 320/640M - 96 shader units
# GeForce 8800 GTX - 768M - 128 shader units
# GeForce 8800 Ultra - 768M - 128 shader units

cruched 6.25 application on happy linux OS....

Do U think is possible make us happy in the future? Now we can't help the project but we want !!!


sorry but... I was so happy 2 weeks ago :-))

...I'd like to know how fast a GTX260 is "in real world", because it's considerably more expensive than the 9800GTX+, the most expensive G92 card, but has about the same maximum GFlops.

...

And let's get serious about the "most effective card" question.

Buying anything smaller than a G92-based card is not the way to go - they are not that much cheaper but much slower. I'd like to know how fast a GTX260 is "in real world", because it's considerably more expensive than the 9800GTX+, the most expensive G92 card, but has about the same maximum GFlops.

Let's consider these cards: 8800GT, 8800GTS 512, 9800GTX+ and GTX280. I collect current pricing for Germany and the GFlops from the wiki (here, here and here).

Doing that, the
- 8800GT has 504 GFlops for 110€ -> 4.58 GFlops/€
- 8800GTS 512 has 624 GFlops for 130€ -> 4.80 GFlops/€
- 9800GTX+ has 705 GFlops for 155€ -> 4.54 GFlops/€
- GTX280 has 933 GFlops for 350€ -> 2.66 GFlops/€

So the 8800GTS 512 seems to be the most efficient card of these. However, you have to take into account that you also need a PC to run the card in and one CPU core. I'll use my main rig to give an example of what I mean by that.

On average my 9800GTX+ needs 44569s/WU with the 6.41 client. That means running 24/7 it earns 3850 credits/day or 6265 with the 6.43 client. Since I sacrifice one CPU core I loose about 1000 credits/day (Q6600@3GHz running QMC). Therefore the net gain by running GPU-Grid is 2580 or 5265 credits/day. Assuming linear performance scaling with the GFlops rating a 8800GTS 512 would earn 5545 credits/day which is a net win of 4545 credits/day.

Therefore the 8800GTS 512 gives you 35 credits/day/€ and the 9800GTX+ 34 credits/day/€ and the 8800GTS 512 is the efficiency winner. However, are 700 credits/day worth a one time investment of 25€ for you? Your choice.. I certainly made mine ;)

Of course you could always overclock either card.. but I don't think the software is already that stable. I'd rather have the additional speed guaranteed. And going with a 55 nm chip doesn't help much but doesn't hurt either.

MrS

Hello

I just got a 9800GT for 110€ (the same price as the 8800GT and the same G92Chip & Clockspeed)
Running time for one WU is ~21h on 6.43 on a Core2 Duo ~2.1Ghz (E6300)
I have it running together with a Seti WU on the other Core.
(Last night I stopped the Seti Projekt to see, if it has an impact on the GPUgrid WU time, but it doesn´t seem so.)


Regards, Thorsten "KyleFL"

I've one 8800GT and one GTX280 running. The 8800GT needs 11:39h for one WU and I got 1987.41 credits, that's ca. 170 cr/h. The card works on a AMD Penom 9850 BE. The GTX280 needs only 7:50h for one WU and I got 3232.06 for it, that's ca. 415 cr/h. Are this other WUs, or why the credits are higher?
____________

I've one 8800GT and one GTX280 running. The 8800GT needs 11:39h for one WU and I got 1987.41 credits, that's ca. 170 cr/h. The card works on a AMD Penom 9850 BE. The GTX280 needs only 7:50h for one WU and I got 3232.06 for it, that's ca. 415 cr/h. Are this other WUs, or why the credits are higher?

It's the new credit award with app v6.42, your 8800GT will also start to get 3232/WU when it runs v6.42 (or higher) ;-)

Kokomiko,

both of your cards are rather fast. Are they overclocked? What are the shader clocks on both? Are you running Win or Linux? Using my values as reference a stock GTX280 would need ~9:20h.

MrS

Which driver are you using?

MrS

The last WU finished by your 8800GT was still using 6.41, hence the lower credits.

So your 8800GT is not overclocked by you, but is clocked way higher than the stock 1500 MHz. Interesting though, it's clearly faster than my 9800GTX+ with fewer shaders and a lower shader clock. Which driver are you using?

MrS

So the 8800GTS 512 seems to be the most efficient card of these. However, you have to take into account that you also need a PC to run the card in and one CPU core. I'll use my main rig to give an example of what I mean by that.

So the 8800GTS 512 seems to be the most efficient card of these. However, you have to take into account that you also need a PC to run the card in and one CPU core. I'll use my main rig to give an example of what I mean by that.

I do not understand. Cannot I run two SETI@home WUs + gpugrid all at once with my dual core Pentium D 920 (and NVIDIA)?

Henri.

It is whell known that some 3D prodecents make 3D cards with higher clocks that references...

Sure. The point is that he's about 30 min faster than me with 112 shaders at 1.73 GHz, whereas I have 128 shaders at 1.83 GHz. That's a difference worth investigation. My prime candidate would be the Vista / Vista 64 driver.

"Flops" are misleading, I think the number of stream processors plays a bigger role, and frankly, I was never a big fan of SLIs.

Well.. no. Flops are calculated as "number of shaders" * "shader clock" * "instructions per clock per shader". The latter one could be 2 (one MADD) or 3 (one MADD + one MUL), but it's constant for all G80/90/GT200 chips. So Flop are a much better performance measure than "number of shaders", because they also take the frequency into account.

And SLI.. yeah, just forget it for games. And for folding you'd have to disable it anyway.

MrS

Thanks for the info once again, guys!

It's a bit sad that one CPU core is wasted even if the GPU is used. Can they change this someday?

Henri.
____________

It is not wasted. If I understood it right the CPU is needed to feed the GPU with data...
It's the same with Folding@home on the GPU, but they only need about 5% of one core and they are planning to distribute an application that only uses the GPU without the need to use the CPU in the future.

Don't know if this would also be possible with the application here on PS3GRID...
____________

pixelicious.at - my little photoblog

Remember the flops formula is the best the GPU can do (peak), but very few real world applications can issue the max instructions every cycle

Yes, we're only calculating theoretical maximum Flops here, the real performance is going to be lower. This "lower" is basically the same factor for all G8x / G9x chips, but GT200 received a tweaked shader core and could therefore show higher real world GPU-Grid-performance with the same Flops rating. That's why I asked for the GTX260 :)

Edit: and regarding CPU usage, F@H also needed 100% of one core in GPU1. The current GPU2 client seems tremendously improved. Maybe whatever F@H did could also help GPU-Grid?

MrS

In every of my finished workunits I find the following text:

<core_client_version>6.3.10</core_client_version>
<![CDATA[
<stderr_txt>
# Using CUDA device 0
# Device 0: "GeForce 9800 GT"
# Clock rate: 1512000 kilohertz
MDIO ERROR: cannot open file "restart.coor"
called boinc_finish
</stderr_txt>
]]>

What I´m wondering about is the MDIO Error Message. What exactly is meant by that?
The restults are always valid and the credits are granted. But it seems, that 21.5h computing time on a GF9800GT is a little bit to slow, isn´t it?

What I´m wondering about is the MDIO Error Message.

I have the same card, but mine needs ~58600 seconds for one work unit.

http://www.ps3grid.net/results.php?hostid=7528

The older ones at 61.000 seconds where done at a lower VRAM speed of only 500MHz, but finally I found the solution to force the highest Powermizer settings.

First think is to check if your card runs at designed speed while crunching, get yourself a copy GPUZ to find out =)

But it seems, that 21.5h computing time on a GF9800GT is a little bit to slow, isn´t it?

Back in the old days (hmm, 2 weeks ago?) I once managed to force the card to Powermizer level 3, but then the improvement was only from 48.000 to 47.000. Those were the days of the faster Cuda 1.x 6.25 app.

I can't guarantee that the recent decrease in crunching time is only the result of the VRAM now running full speed, as I also updated from 177.68 to 177.70 almost at the same time.
So there might or might not be an impact from the graphics driver, but I doubt that, as no one else reported any improvements.

Following the folding@home GPU2 thread in our teams forum, it seemed that the biggest speedup can be achieved by overclocking the shaders. Core and memory had only little impact.

Should be the same here...

edit:

Well, there are some small differences between 9800GT and GTX+...
The GT is basically a relabeled 8800GT with the same (shader +12MHz) clock speeds and only 112 shader units. The GTX+ has 128 shaders at 1836MHz versus 1500/1512MHz on x800 GT side.

But still, more that 20h is way to long and needs to be investigated ;-)

My 9800GT is running at stock speed:

600 / 900 / 1500 (looked it up in GPU-Z 0.2.7)
Driver is 177.83
PhysX is running OK

Shaders: 1512000 kilohertz is detected by the 6.43 Application (it´s in the text I posted - so the 1500Mhz Shaderclockspeed seem to be working)

running time is always around 78026.66 like that task.
http://www.gpugrid.net/result.php?resultid=49731


Very strange..


Thanks for your help, Kyle

My 9800GT is running at stock speed:

600 / 900 / 1500 (looked it up in GPU-Z 0.2.7)
Driver is 177.83
PhysX is running OK

Shaders: 1512000 kilohertz is detected by the 6.43 Application (it´s in the text I posted - so the 1500Mhz Shaderclockspeed seem to be working)

running time is always around 78026.66 like that task.
http://www.gpugrid.net/result.php?resultid=49731


Very strange..


Thanks for your help, Kyle

I also have a Gainward Bliss 9800GT 512MB, but mine is 20000 fas^ter than yours was. No idea why yours was so slow...

No, I have taken them by their word on the 14day returning right and sent the 9800GT back .... :)
I have a GTX260 now - just wanted to let you know.


110€ for the 9800GT
210€ for the GTX260

and the Leadtek GTX260 is MUCH more silent than the Gainward 9800GT (but gets ~20° hotter - 73° vs ~53°)
The 9800GT was a lot louder than the 7800GT I had before -- the GTX260 is better in that way, too, so I won´t complain about the upgrade. Should have gone with the GTX260 from the beginning :)


Regards, KyleFL

- GTX280 has 933 GFlops for 350€ -> 2.66 GFlops/€

Can I soon after the Christmas of 2008 buy passively cooled GTX280 with just 300 €? Is it completely impossible to have passively cooled (and stable) GTX280 with 100% GPU load 24/7?

A GTX280 has 150+W power consumption.. do I need to say more? And, btw., 55 nm isn't going to help much.

Henri,

The ASUS cards have fairly quiet fans in my experience. My 9600 GSO has one of the big "glaciator" fans, and the noise increase over the power supply fan is negligible at most (I really can't tell a difference to be honest).

The ASUS cards have fairly quiet fans in my experience. My 9600 GSO has one of the big "glaciator" fans, and the noise increase over the power supply fan is negligible at most (I really can't tell a difference to be honest).