Note that Martin is using a third party application at SETI. We don't yet know for certain whether the stock app still works, although I have a volunteer tester working on it offline and the preliminary signs are good.
User JohnSheridan at GPUGrid (who seems to share an account with MartinBen :P) seems to be having memory size problems:
# Total amount of global memory: -2147483648 bytes
OK - any jobs called Gravitational Wave S6 LineVE to Search 1.10 (SSE2) crash out with "computation error" every time.
Not sure why but the error log points to einstein_S6LV1_1.10_windows_intelx6_SSE2.exe causing the issue so might just be a faulty group of work units?
This is a CPU only app, so definitely not directly related to Kepler GPUs. Anyway, I brought this to the attention of the devs. Did you change anything else (apart from the GFX card)??
This is a CPU only app, so definitely not directly related to Kepler GPUs. Anyway, I brought this to the attention of the devs. Did you change anything else (apart from the GFX card)??
CU
HBE
HBE
No - I first uninstalled the old nVidia drivers, turned off the PC, removed the 580 card, installed the new 680 card, switched pc back on then installed the new nVidia drivers etc (version 301.10)
Reviews like this makes me wonder if it's not better to go with cheaper cards for compute.
Looks a little disappointing, but considering the price tags and the power consumtion GTX680 is the winner from my point of view.
@Bikeman
Quote:
As Oliver has pointed out in another thread, NVIDIA is working hard on the driver problem that causes the issues with E@H [...] A fix should be available in the next release in a couple of weeks. By that time, you will have many more reviews of the card's performance and maybe even lower prices, so if I had to do this decision, I would wait a bit.
These are good arguments against buying now. Unfortunately I have to set up a new computer now.
@MartinBen
Quote:
No - I first uninstalled the old nVidia drivers, turned off the PC, removed the 580 card, installed the new 680 card, switched pc back on then installed the new nVidia drivers etc (version 301.10)
You will not get any tasks here (at least not via the "normal" way). Driver versions above 290 are not allowed.
OK, at least there is SETI. As far as I can see already some WUs validated.
Michael
Yes. We're seeing high-quality validation in offline bench testing. No alarm signals regarding use of these cards at SETI, using either stock or Lunatics third-party applications.
The speeds, however, seem disappointing, and may remain so until fresh apps specifically written to take advantage of the Kepler processors and memory architecture become available.
I ordered a EVGA GTX 680 this afternoon. It should be here on Tuesday. I plan to run it with my 3930K in Linux and the latest driver available when the time comes.
I brought my EVGA GTX 680 online this evening and am happy to report that 9 BRP4 tasks completely successfully with one validated so far. The others are pending. I did see one task fail however. I am running Linux with the latest 295.33 NVIDIA driver and kernel 3.3 both compiled with the latest GCC 4.7. Like with my GTX 580, I am running three tasks at once which is an optimal number of tasks for the 1.5 GB 580. The GTX 680 can perhaps scale to a higher number of tasks but I have not tested that yet.
At the moment I am running the card via a board with a 3.0 GHz Q6600. BRP4 CUDA performance is not quite as good with this CPU as with the newer overclocked i7 processors. My plan is to run this new card via my 3930K system currently set at 4.48 GHz but that system is finshing up a very large task from another project. It will be another day or so before I can move the card. I am hopeful that once I move the card to the other board that the completion time per task will be even better.
Note that Martin is using a
)
Note that Martin is using a third party application at SETI. We don't yet know for certain whether the stock app still works, although I have a volunteer tester working on it offline and the preliminary signs are good.
User JohnSheridan at GPUGrid (who seems to share an account with MartinBen :P) seems to be having memory size problems:
# Total amount of global memory: -2147483648 bytes
RE: OK - any jobs called
)
This is a CPU only app, so definitely not directly related to Kepler GPUs. Anyway, I brought this to the attention of the devs. Did you change anything else (apart from the GFX card)??
CU
HBE
HBE
RE: This is a CPU only app,
)
No - I first uninstalled the old nVidia drivers, turned off the PC, removed the 580 card, installed the new 680 card, switched pc back on then installed the new nVidia drivers etc (version 301.10)
@Logforme RE: Reviews
)
@Logforme
Looks a little disappointing, but considering the price tags and the power consumtion GTX680 is the winner from my point of view.
@Bikeman
These are good arguments against buying now. Unfortunately I have to set up a new computer now.
@MartinBen
You will not get any tasks here (at least not via the "normal" way). Driver versions above 290 are not allowed.
Michael
Team Linux Users Everywhere
RE: MartinBen over at
)
OK, at least there is SETI. As far as I can see already some WUs validated.
Michael
Team Linux Users Everywhere
RE: OK, at least there is
)
Yes. We're seeing high-quality validation in offline bench testing. No alarm signals regarding use of these cards at SETI, using either stock or Lunatics third-party applications.
The speeds, however, seem disappointing, and may remain so until fresh apps specifically written to take advantage of the Kepler processors and memory architecture become available.
I ordered a EVGA GTX 680 this
)
I ordered a EVGA GTX 680 this afternoon. It should be here on Tuesday. I plan to run it with my 3930K in Linux and the latest driver available when the time comes.
Does E@H use Single Precision
)
Does E@H use Single Precision or Double Precision math ?
RE: Does E@H use Single
)
SP!
I brought my EVGA GTX 680
)
I brought my EVGA GTX 680 online this evening and am happy to report that 9 BRP4 tasks completely successfully with one validated so far. The others are pending. I did see one task fail however. I am running Linux with the latest 295.33 NVIDIA driver and kernel 3.3 both compiled with the latest GCC 4.7. Like with my GTX 580, I am running three tasks at once which is an optimal number of tasks for the 1.5 GB 580. The GTX 680 can perhaps scale to a higher number of tasks but I have not tested that yet.
At the moment I am running the card via a board with a 3.0 GHz Q6600. BRP4 CUDA performance is not quite as good with this CPU as with the newer overclocked i7 processors. My plan is to run this new card via my 3930K system currently set at 4.48 GHz but that system is finshing up a very large task from another project. It will be another day or so before I can move the card. I am hopeful that once I move the card to the other board that the completion time per task will be even better.
Results (3 tasks per GPU):
GTX 680: 3669 Sec Avg (Stock)
GTX 580: 3836 Sec Avg (850 MHz)