I have tested with BOINC 6.3.8, driver 177.67, Gefore 8800 GT and fedora 8.

gdf

I have tested with BOINC 6.3.8, driver 177.67, Gefore 8800 GT and fedora 8.

gdf

It is likely that people with older drivers will have to update.

gdf

It's still constantly crashing on the system with the 8800GTX

All crashed with the same error:

<core_client_version>6.3.8</core_client_version>
<![CDATA[
<message>
process exited with code 1 (0x1, -255)
</message>
<stderr_txt>
# Using CUDA device 0
# Device 0: "GeForce 8800 GTX"
# Clock rate: 1350000 kilohertz
Cuda error: Kernel [copy_mul] failed in file 'com.cu' in line 43 : invalid device function .

</stderr_txt>
]]>


Using Ubuntu x64 8.04, Boinc v6.3.8 and the Nvidia 177.67 drivers.

---

Down with the Kredit Kops!!!

It's still constantly crashing on the system with the 8800GTX

All crashed with the same error:

6.3.8

process exited with code 1 (0x1, -255)


# Using CUDA device 0
# Device 0: "GeForce 8800 GTX"
# Clock rate: 1350000 kilohertz
Cuda error: Kernel [copy_mul] failed in file 'com.cu' in line 43 : invalid device function .


]]>


Using Ubuntu x64 8.04, Boinc v6.3.8 and the Nvidia 177.67 drivers.

I am sorry but your card has compute capability 1.0 instead of 1.1. So, it is not supported by the current application. Note that 1.1 is the the CUDA toolkit, but the version of your GPU (in practice it does not have atomic operations).

I will add a better error handling into next version. The previous code to handle 1.0 does not work for us under CUDA2.

gdf

Which means if I have crunched all 6.37 WUs, I can throw away my 8800 GT which was working pretty fine until now and which I bought for the inital GPU tests, because CUDA 2.0 doesn't support my card?
Great...

---

pixelicious.at - my little photoblog

The previous code to handle 1.0 does not work for us under CUDA2.

Do you mean code to handle the errors or alternative code to do the calculations? I'd suspect the latter and thus G80 based cards would still be able to contribute after the fix is in place.

MrS

---

Scanning for our furry friends since Jan 2002

8800GT/8800GTSv2 (512mo) support compute capability 1.1, so it's compatible with CUDA2.0 as GF9 and GTX260/280

However, 8400/8600, 8800GTS320/640 , 8800GTX/ultra are not compatible, I think

Where can I find such information? I tried on google, but maybe with the wrong key words...

Now with the 6.38 using driver 177.68 on a 9800GT it seems the unit will take 60400 seconds, compared to the other recent apps that is an improvement after all:

6.25 -> 48.000s
6.34 -> 74.500s
6.37 -> 64.500s


But unfortunatelly not back to the 6.25 level...

It seems an upgrade to the Linux drivers are necessary for the 6.38 application.

On this host all WU run with 6.38 app and 173.14 driver failed in the first two seconds with the following error...

<core_client_version>6.3.8</core_client_version>
<![CDATA[
<message>
process exited with code 1 (0x1, -255)
</message>
<stderr_txt>
# Using CUDA device 0
# Device 0: "GeForce 8800 GT"
# Clock rate: 1458000 kilohertz
Cuda error in file '../force/vdw_texture.cu' in line 127 : feature is not yet implemented.

</stderr_txt>
]]>

After a manual upgrade to the 177.67 driver and the WU runs OK - Note Envy does not load the 177.xx driver, a manual installation is necessary.

System Spec:
Dual core E4300 overclocked @2.4GHz
8800GT (I am unable to change clocks on the card using nvclock when running under 177.67 driver)
Ubuntu 8.04 x64
Boinc 6.3.8
Application switch 6.37->6.38



/edit
the WU failed after 18minutes with the following....

<core_client_version>6.3.8</core_client_version>
<![CDATA[
<message>
process exited with code 1 (0x1, -255)
</message>
<stderr_txt>
# Using CUDA device 0
# Device 0: "GeForce 8800 GT"
# Clock rate: 1728000 kilohertz
MDIO ERROR: cannot open file "restart.coor"
Cuda error: Kernel [frc_sum_kernel_dihed] failed in file 'force.cu' in line 516 : unspecified launch failure.

</stderr_txt>
]]>

Seems unrelated to the earlier error, maybe the increase in clocks on the card makes it unstable? Does anyone know how to reduce the clocks under 177.67 drivers?

---

8800GT/8800GTSv2 (512mo) support compute capability 1.1, so it's compatible with CUDA2.0 as GF9 and GTX260/280

However, 8400/8600, 8800GTS320/640 , 8800GTX/ultra are not compatible, I think

Phew, seems I was lucky...
I upgraded to the latest 177.67 drivers on the computer with the 8800GT, and it is crunching the first WU with 6.38 - seems to work fine as far as I can tell now...

---

pixelicious.at - my little photoblog

# Device 0: "GeForce 8800 GT"
# Clock rate: 1728000 kilohertz

1.73 GHz is way above the stock clock speed of any 8800 GT, so that might cause instability. Did you clock it that high or did it happen automatically after the software update?

MrS

---

Scanning for our furry friends since Jan 2002

My 8800GT runs at 1.7Ghz (shader), but it is a ZOTAC AMP! edition, so it's overclocked by default.
Even though the gpugrid app reports it as ""GeForce 8800 GT"; Clock rate: 1674000 kilohertz"
Using GPUz it reports it as 1700Mhz shader....

And it runs gpugrid fine, and it's about 90 minutes faster compared to the stock speed 8800GT that I also have.

---

Down with the Kredit Kops!!!

# Device 0: "GeForce 8800 GT"
# Clock rate: 1728000 kilohertz

1.73 GHz is way above the stock clock speed of any 8800 GT, so that might cause instability. Did you clock it that high or did it happen automatically after the software update?

MrS

It's the default setting on that card. It's a BFG 8800GT. Previously, with 173.xx drivers I was able to use nvclock to reduce the GPU and memory clocks: see the first set of results: # Clock rate: 1458000 kilohertz. However, with 177.67 nvclock crashes the machine if I try to adjust the GPU clocks. The Nvidia-Settings panel has adjusters for the clocks, but no "apply" button. I think someone had some xorg.conf settings that may allow me to adjust, but I can't recall the details just now.

That box trashed a further 4 WU earlier this evening at reduced CPU clocks, so it may well be the GPU clock.

---

Note Envy does not load the 177.xx driver, a manual installation is necessary.

Then I will wait until Envy makes them available.

You need to edit the config file and in the section that lists the videocard, should be like:

Section "Device"
Identifier "Videocard0"
Driver "nvidia"

just above the EndSection put in
Option "Coolbits" "1"

Save and either re-start X or reboot system.

---

Down with the Kredit Kops!!!

Is anyone else encountering these errors.....

Cuda error: Kernel [frc_sum_kernel_angle] failed in file 'force.cu' in line 487 : unspecified launch failure.

Since upgrade to 177.xx drivers, every WU on this box has failed after 10 to 20 minutes with the above error.

It has been suggested that the card overclock is to blame, but I can't reduce the clock under 177.xx drivers.

With 173.xx drivers I had occasional errors of this type, but my card was clocked at 650/950 as opposed to 700/1000 under 177.xx driver.

Another machine is running 6.38 under 177.13 driver without fault at 700/1000.

So, does anyone have this error? How did you fix it? Has anyone got a suggestion to reduce the clock under 177.xx without crashing the machine?

Thanks!

/edit

You need to edit the config file and in the section that lists the videocard, should be like:

Sorry I should have explained more. I have the coolbits options set and using either the sliders, manually entering the clock speeds, or using nvclock results in a rapid white flashing screen and unresponsive system :(

---

Sorry I should have explained more. I have the coolbits options set and using either the sliders, manually entering the clock speeds, or using nvclock results in a rapid white flashing screen and unresponsive system :(

I get that also, on both the Zotac AMP! and Stock 8800GT :(
Flashing white screen (not good when yer Epileptic!), then it would reboot...

That's with just trying to move the clock speed 1MHz up or down.....

---

Down with the Kredit Kops!!!

I get that also, on both the Zotac AMP! and Stock 8800GT :(
Flashing white screen (not good when yer Epileptic!), then it would reboot...

That's with just trying to move the clock speed 1MHz up or down.....

Well, I'm glad it's not just me then....(you know what I mean!).

I'm not sure that the factory overclock is the cause of the Cuda error: Kernel [frc_sum_kernel_angle] failed in file 'force.cu' in line 487 : unspecified launch failure, but it is a good candidate. When I managed to lower the clocks under 173.xx I had only a few like this, but now it's every WU.

I notice in this thread Mr S reports some WU similar (Windows I think).

It's a bit odd that another box here with a Zotac 8800GT is ok at 700/1000.

....that's another WU just gone, this time it got to over an hour I think. Oh well, I'm not going to fix it tonight and I have work in the morning. Off to get some sleep.

---

Have you tried just increasing the fan speed?
I have the fans running at 100% on both 8800GT's, As I noticed that the cards do tend to warm up a bit.

Mind you that also depends on the noise of the fan, and what you can tolerate noise wise.
I'm one of those peeps that could quite easily fall asleep with a Dyson running :)

---

Down with the Kredit Kops!!!

The fact that the application is slower depends on some driver issue related with latest versions 177

http://forums.nvidia.com/index.php?showtopic=75466

It will be sorted alone with new drivers in the future.

gdf