[solved] nVidia 295.51 beta driver problems - Failed tasks - Cannot create a symbolic link in a registry key that already has subkeys or values. (0x3fc) - exit code 1020 (0x3fc)
Language
Copyright © 2024 Einstein@Home. All rights reserved.
RE: Is not finding a gpu
)
In this case, no. That's what error 100 means, regardless of when it occurs.
Oliver
Einstein@Home Project
Good news everyone, NVIDIA
)
Good news everyone, NVIDIA released a new stable version 301.34 that should fix the problem. Could those with affected systems please test and report back:
- Does this driver fix the problem for GeForce 600 series GPUs?
- Can this driver be used on older GPUs as well (doesn't seem to be the case)?
- If so, does it fix the problem for those GPUs as well?
Thanks very much,
Oliver
Einstein@Home Project
301.34 is a 670 and 690
)
301.34 is a 670 and 690 driver.
I will have a Gigabyte 670 in on Monday, and can test then. Going to be running some benchmarks on the same rig my 680 is in, so I will also report if the higher clock rate "beats" a 680 on this project as well.
I have my gigabyte 670
)
I have my gigabyte 670 installed using the latest unofficial driver, 301.40. I know you wished for 301.34, but these drivers fix several bugs that were in 301.34. These drivers are listed on guru3d, and an nvidia rep has posted a sticky on nvidias forums in regards to how these drivers are working for those who wish to use them in order to gain feedback
He says a unified driver will be coming out shortly, so it is my understanding that the 301.40 will be the "base" for either the next beta or whql official drivers which they will post on nvidia.com
Will turn monitor sleep function on and leave it running over night and let them run. Will report results by 9 am central us time tomorrow.
Thanks for the info and
)
Thanks for the info and looking forward to your update!
Oliver
Einstein@Home Project
No issues at all. Put
)
No issues at all.
Put monitor settings to fall asleep after 1 minute, and let it run. No tasks failed, or failed to start.
For those interested, the 670 appears to scale with multiple tasks with about the same rate as the 680. This was expected, but my personal reason for using Kepler, besides faster performance at GPUgrid whenever they release CUDA 4.2, is the VAST amount of power reduction and heat that is dispersed from these cards.
$500 EVGA 680 has to run at 65% fan speed in order to hover around 54-55C, while my $400 Gigabyte 670 (triple fans), only needs 50% to hover around 52C. For those interested.
I have posted a question in regards to the CUDA monitor bug at NVIDIA forums to the admin in regards to seeing if we can get official word from them if this is indeed solved FOR A FACT. I realize that I have not had any issues, but would still like to see some word from NVIDIA, so that I may pass this around to the other projects as well.
Will let you know if/when he responds.
Currently using a dev driver
)
Currently using a dev driver 302.59 FYI, just in case you needed to know. Still no issues.
New OFFICIAL NVIDIA drivers
)
New OFFICIAL NVIDIA drivers are coming out next week
http://forums.nvidia.com/index.php?showtopic=218060&st=220
Devs/admins Is there any
)
Devs/admins
Is there any chance that next week you could allow 300.83 drivers? Would like to run a test and see if having PCIe 3.0 running on my x79 makes any difference. Currently this has been disabled in later drivers as they certify mobos. Mine is an ASUS sabertooth, so I'm not worrired. Would like to see if it impacts task duration at all
Thanks
RE: Is there any chance
)
No, since this is a production/stable project and not meant for testing purposes. You may attach your client to our test project where the specific NVIDIA driver version restrictions discussed here are not in place.
Best,
Oliver
Einstein@Home Project