Someone mentioned there were other drivers with computational errors like ver 295.73.

Is there an updated white or black list of drivers pinned somewhere? I had the misfortune to have these 340.52 drivers installed on 2 machines, wondering why I was receiving no GPU units, and when trying to correct the issue dropped down to another blacklisted driver.

IMO, the best Driver for the older cards in Win 7 and below is 266.58 WHQL. That driver doesn't use the Full CPU core to run AstroPulse tasks the way newer drivers do. No need for the -sleep switch or anything else, great for Dual core CPUs. It also works fine with CUDA versions up to 3.2. This host is using it; All tasks for computer 6813106 It also works fine with my GTS 250.

The XP Link is here, http://www.nvidia.com/object/winxp-266.58-whql-driver.html
You'll have to search for the other OS versions.

It wasn't released for the Quadro FX series of laptop workstation GPU's but I'll try hacking the NVHDC.inf to allow them.
If that fails, are 266.45 or 267.79 (both available for Quadro FX) equivalent in functionality? I can also get any in the list on this post: http://setiathome.berkeley.edu/forum_thread.php?id=75978&postid=1593087#1593087

---

IMO, the best Driver for the older cards is 266.58 WHQL. That driver doesn't use the Full CPU core to run AstroPulse tasks the way newer drivers do. No need for the -sleep switch or anything else, great for Dual core CPUs. It also works fine with CUDA versions up to 3.2. This host is using it; All tasks for computer 6813106 It also works fine with my GTS 250.

The XP Link is here, http://www.nvidia.com/object/winxp-266.58-whql-driver.html
You'll have to search for the other OS versions.

Or just use http://www.oldapps.com/nvidia.php

Someone linked that for me on another post but that's ForceWare and the installers exclude Quadro FX cards. I can attempt hacking them but would rather find a Quadro FX driver without computational errors.
Thanks.

---

It wasn't released for the Quadro FX series of laptop workstation GPU's but I'll try hacking the NVHDC.inf to allow them.
If that fails, are 266.45 or 267.79 (both available for Quadro FX) equivalent in functionality? I can also get any in the list on this post: http://setiathome.berkeley.edu/forum_thread.php?id=75978&postid=1593087#1593087

Use the 266.45 driver as after 266.58 the behaviour greatly changes. ;-)

Cheers.

It wasn't released for the Quadro FX series of laptop workstation GPU's but I'll try hacking the NVHDC.inf to allow them.
If that fails, are 266.45 or 267.79 (both available for Quadro FX) equivalent in functionality? I can also get any in the list on this post: http://setiathome.berkeley.edu/forum_thread.php?id=75978&postid=1593087#1593087

Use the 266.45 driver as after 266.58 the behaviour greatly changes. ;-)

Cheers.

Installing 266.45 now. Thanks!

---

Some positive news, I think - I just noticed in the BOINC commit logs that it is accounting for the combination of CUDA 6.5+ and NV GPUs with Compute Capability < 2.0 and preventing use of NV GPUs for OpenCL in that situation.

---

Soli Deo Gloria

Some positive news, I think - I just noticed in the BOINC commit logs that it is accounting for the combination of CUDA 6.5+ and NV GPUs with Compute Capability < 2.0 and preventing use of NV GPUs for OpenCL in that situation.

No, as far as I can see it prevents the Use of the Pre-Fermi GPUs full stop, Not just OpenCL, But Cuda too:

client: add some comments to clarify that the CUDA warning message and notice will appear only once for each relevant GPU.

cannot be used for CUDA or OpenCL computation with CUDA driver 6.5 or later

340.52 drivers display Cuda 6.5 support for Pre-Fermi GPUs:

06-Oct-2014 21:26:13 [---] CUDA: NVIDIA GPU 0: GeForce 9800 GT (driver version 340.52, CUDA version 6.5, compute capability 1.1, 512MB, 408MB available, 504 GFLOPS peak)

and supposedly crunches Cuda22, Cuda23, Cuda32, Cuda42 and Cuda50 apps O.K, (there hasn't been mass reports of validity problems), There'll just be no Cuda65 apps for pre-fermis,
I think Charlie has jumped the Gun a bit, at least for Windows (for single GPU use at least, mixed GPU use is another matter)

Claggy

Didn't consider the CUDA side, but my thinking was the preventative measure was a good step - means that future BOINC versions will cover this case for all projects instead of just trying to handle it within NV AstroPulse builds.

And it's all development stuff, anyway, it can be changed to allow CUDA again in future work.

---

Soli Deo Gloria

It wasn't released for the Quadro FX series of laptop workstation GPU's but I'll try hacking the NVHDC.inf to allow them.
If that fails, are 266.45 or 267.79 (both available for Quadro FX) equivalent in functionality? I can also get any in the list on this post: http://setiathome.berkeley.edu/forum_thread.php?id=75978&postid=1593087#1593087

Use the 266.45 driver as after 266.58 the behaviour greatly changes. ;-)

Cheers.

266.58 is available for Quadro FX mobile GPU's... I just had trouble finding a working d/l location but here they are still up at nVidia.


64 bit: http://www.nvidia.com/object/notebook-win7-winvista-64bit-266.58-whql-driver.html

32 bit: http://www.nvidia.com/object/notebook-win7-winvista-266.58-whql-driver.htm

Thanks for the help determining the best driver for these older workstations.

---

Here is some correspondence that I initiated on BOINC alpha, when I noticed that BOINC 7.4.26 disabled use of my FX3800M GPU.

I'm not trying to stir the pot, but instead, am just trying to relay information, and get information. Note that I was not aware of this thread, when I sent my original email. I'm still trying to comprehend the decisions being made here, but my gut tells me it might be the wrong call to blanket-blacklist GPUs in BOINC.

What I'd like to see is:
- A simple downloadable test case where I can prove to myself that the application returns valid data on some drivers, and invalid data on other drivers.
- Do we know if CUDA 6.5 results (which don't use OpenCL) are affected?
- Can we see the quoted NVIDIA responses regarding them indicating that they will not fix the issue?
- Can we get a clear understanding of why the project can't use validation logic in the validator, and server scheduling logic, to completely mitigate the problem?
- Are you aware that NVIDIA plans to support the R340 branch until April 1, 2016, per http://nvidia.custhelp.com/app/answers/detail/a_id/3473
- Does anyone else agree that the blanket blacklist approach, built into BOINC, may not be the right solution here?

Regards,
Jacob




==================================================================

> From: j...klein@msn.com
> To: charlie...@ssl.berkeley.edu
> Date: Sun, 2 Nov 2014 09:19:22 -0500
> CC: b..._alpha@ssl.berkeley.edu
> Subject: Re: [boinc_alpha] BOINC 7.4.26 - Drops support for older GPUs - Showstopper?
>
> Charlie:
> I've had applications cause problems in the past (WCG HCC GPU tasks cause severe lag on GPU connected to display), and I had to disable those applications or work around the problem (excluded that app on that GPU). I've had drivers cause problems in the past (NVIDIA 320x range introduced a problem that killed CUDA on monitor sleep), and I had to uninstall those drivers or work around the problem (told monitor to blank instead of sleep, and communicated with NVIDIA, until driver update fixed problem). I also read the articles, and understand that, supposedly, the drivers are silently returning invalid OpenCL results. I have not seen any NVIDIA correspondance, even quoted, indicating that they introduced OpenCL problems that they do not intent to fix.
>
> Handling this problem, in my opinion, should be the job of a) the validator to invalidate those results, and b) the project maintainer to actively exclude certain hardware/driver combos that cause problems. I don't think BOINC should blanket blacklist the GPUs. What if the CUDA results are currently valid? What happens if NVIDIA decides to fix the OpenCL issue? They are, as seen, still releasing driver updates, even for my laptop GPU here. Does the user even have a BOINC control to turn the GPU support back on if the issue gets fixed? Or would the workaround for that scenario involve downgrading BOINC?
>
> I know where to get older drivers from. In fact, I keep every driver version I've ever used, because I know they sometimes cause problems. And I know of the "Beta or newer" NVIDIA hyperlink.
>
> Are we really black-listing my GPU here? If so, I may have to downgrade BOINC on it, to even attempt to do possibly-valid-CUDA work while using the latest supported released NVIDIA drivers. Note: Did you see "The Release 340 drivers will continue to support these products until April 1, 2016" in my link http://nvidia.custhelp.com/app/answers/detail/a_id/3473 ? I have every reason to believe that the issue has a possibility of being fixed.
>
> Richard:
> No, I had not seen that SETI link. After reading it, I'd want to do my own testing. Is there a readily available test case where I can prove for myself that a certain driver version works or doesn't work, for OpenCL/Astropulse? You asked "why stay up to date on the driver?" Well, I always run the latest drivers on my systems, because they generally fix several things, and if they break something I want to be the first to know so I can report it. I don't know why NVIDIA released a CUDA 6.5 driver that happens to supposedly silenty break OpenCL results, but I'll probably keep it. It has nothing to do with CUDA 6.5 being important, or sacrificing
> OpenCL. It has everything to do with "using the latest, to find and the
> fix any problems."
>
> I don't feel like it's right to have BOINC disable the GPU entirely. Did we even test if a CUDA app still runs fine on it? Can't we have the projects make the decisions on how to block these, like they've done in the past? How will we get BOINC to work right, if NVIDIA does fix the issue -- require an older version of BOINC?
>
> A frustrated user who thinks this current blanked blacklist solution is the wrong approach,
> Jacob
>

==================================================================

>
>
> > Subject: Re: BOINC 7.4.26 - Drops support for older GPUs - Showstopper?
> > From: charlie...@ssl.berkeley.edu
> > Date: Sun, 2 Nov 2014 01:52:43 -0800
> > CC: b..._alpha@ssl.berkeley.edu
> > To: j...klein@msn.com
> >
> > Hi Jacob,
> >
> > Not only is that the intent, but it is the reason we withdrew 7.4.23 and 7.4.24. We have found that once CUDA 6.5 is installed on Macs, OpenCL applications (including Astropulse 7.05 and Collatz Conjecture) cause frequent kernel panics (i.e, they crash the OS!) on systems with older NVIDIA GPUs. In addition, the people working on SETI@home have found that OpenCL applications return incorrect results on Windows computers with older GPUs which have the latest NVIDIA drivers (those which have CUDA 6.5), even though it appears to the user that it works properly.
> >
> > Notice we are talking about OpenCL applications, not just CUDA apps.
> >
> > Despite NVIDIA's claims that CUDA 6.5 still supports GPUs with compute capabilities 1.0 through 1.3, we have found that it does not. Please see:
> > <http://setiathome.berkeley.edu/forum_thread.php?id=75633&postid=1571811#1571811> and
> > <https://devtalk.nvidia.com/default/topic/774899/cuda-programming-and-performance/340-52-driver-breaks-opencl-support-for-pre-fermi-cc-1-x-cards/>.
> >
> > The Mac installer for BOINC 7.4.26 includes a ReadMe, which now says:
> > > CUDA UPGRADE WARNING: Do not upgrade to CUDA 6.5 or later if you have an older NVIDIA GPU with Compute Capability 1.3 or less. You can check your GPU’s Compute Capability at https://developer.nvidia.com/cuda-gpus.
> > >
> > > You can find older CUDA drivers at http://www.nvidia.com/object/mac-driver-archive.html. Note: after mounting the downloaded disk image, you may need to control-click on the CUDA installer package to open it.
> >
> > To use BOINC 7.4.26 and beyond with older NVIDIA GPUs, you must install an older version of the drivers. You can find older NVIDIA drivers for Windows at <http://www.nvidia.com/Download/Find.aspx?lang=en-us> by selecting "All" under "Recommended/Beta."
> >
> > Cheers,
> > --Charlie
> >
> > --
> > Charlie Fenton charlie...@ssl.berkeley.edu
> > BOINC / SETI@home Macintosh & Windows Programmer
> > Space Sciences Laboratory
> > UC Berkeley
> >

==================================================================

>
> Date: Sun, 2 Nov 2014 09:50:13 +0000
> From: r.hasel...@btopenworld.com
> Subject: Re: [boinc_alpha] BOINC 7.4.26 - Drops support for older GPUs - Showstopper?
> To: j...klein@msn.com; charlie...@ssl.berkeley.edu
> CC: b..._alpha@ssl.berkeley.edu
>
> There are concerns that other changes introduced by NVidia in the 340.xx driver range, apart from cuda 6.5, introduce incompatibilities with existing applications. You mention Seti@Home - I assume you have read http://setiathome.berkeley.edu/forum_thread.php?id=75633&postid=1571811#1571811? The OpenCL-based Astropulse application appears to be incompatible with this hardware/driver combination.
> NVidia have reproduced the problem in internal testing, but have declared the issue "won't fix". It doesn't seem worth tracking down a solution at the application level for one last driver increment before these legacy cards reach end-of-life.
> May I ask why you *need* driver 340.84 on your FX 3800M? Is cuda 6.5 so important to you that you are prepared to sacrifice OpenCL?
>

==================================================================

>
> >
> > On Nov 1, 2014, at 10:48 PM, Jacob Klein <j...klein@msn.com> wrote:
> >
> > > Charlie,
> > >
> > > I noticed your recent checkins, that involve "If CUDA driver 6.5 or later is installed, prevent use of NVIDIA GPUs with Compute Capability < 2.0 and show explanation in Event Log and Notices." But I'm having a problem with that logic.
> > >
> > > I happen to have an NVIDIA Quadro FX3800M that has been working great on projects that support it, including Albert@Home, Einstein@Home, Seti@Home, and Seti Beta. Tonight, I even upgraded its drivers to the latest from NVIDIA, released this month, version 340.84, which supports CUDA 6.5.
> > >
> > > However, when I upgraded BOINC from 7.4.22 to 7.4.26, I noticed that it stopped using my GPU, with a message:
> > > NVIDIA GPU 0: Quadro FX 3800M cannot be used for CUDA or OpenCL computation with CUDA driver 6.5 or later
> > >
> > > Surely, a mistake was made, right? Just because CUDA 6.5 is installed, does NOT mean that GPUs of older compute capabilities should be disabled, does it? I'll refer you to this article, which NVIDIA has released in March, which basically says that the new 343+ versions of the drivers will not be supporting older GPUs:http://nvidia.custhelp.com/app/answers/detail/a_id/3473
> > >
> > > What was the intent of your BOINC checkin changes? Can you please fix this, so that I can do computational work again on my GPU that is running the latest (CUDA 6.5 supported) NVIDIA drivers? I'll copy/paste Event Logs from both versions, at the bottom of this email.
> > >
> > > Regards,
> > > Jacob
> > >
> > >
> > >
> > > 7.4.26:
> > > 11/2/2014 12:30:07 AM | | Starting BOINC client version 7.4.26 for windows_x86_64
> > > 11/2/2014 12:30:07 AM | | log flags: file_xfer, sched_ops, task, checkpoint_debug, scrsave_debug, unparsed_xml
> > > 11/2/2014 12:30:07 AM | | log flags: work_fetch_debug
> > > 11/2/2014 12:30:07 AM | | Libraries: libcurl/7.33.0 OpenSSL/1.0.1h zlib/1.2.8
> > > 11/2/2014 12:30:07 AM | | Data directory: C:\ProgramData\BOINC
> > > 11/2/2014 12:30:07 AM | | Running under account jacob_000
> > > 11/2/2014 12:30:07 AM | | NVIDIA GPU 0: Quadro FX 3800M cannot be used for CUDA or OpenCL computation with CUDA driver 6.5 or later
> > > 11/2/2014 12:30:07 AM | | CUDA: NVIDIA GPU 0 (not used): Quadro FX 3800M (driver version 340.84, CUDA version 6.5, compute capability 1.1, 1024MB, 862MB available, 634 GFLOPS peak)
> > > 11/2/2014 12:30:07 AM | | OpenCL: NVIDIA GPU 0 (not used): Quadro FX 3800M (driver version 340.84, device version OpenCL 1.0 CUDA, 1024MB, 862MB available, 634 GFLOPS peak)
> > > 11/2/2014 12:30:07 AM | | OpenCL CPU: Intel(R) Core(TM) i7 CPU Q 740 @ 1.73GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 3.0.1.15216, device version OpenCL 1.2 (Build 80752))
> > > 11/2/2014 12:30:07 AM | | No usable GPUs found
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | Albert@Home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | Albert@Home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | Albert@Home | Missing coprocessor for task PB0049_004A1_292_1
> > > 11/2/2014 12:30:07 AM | Albert@Home | Missing coprocessor for task p2030.20131211.G181.84-02.68.S.b0s0g0.00000_2064_1
> > > 11/2/2014 12:30:07 AM | Albert@Home | Missing coprocessor for task p2030.20131211.G181.84-02.68.S.b0s0g0.00000_2112_1
> > > 11/2/2014 12:30:07 AM | Albert@Home | Missing coprocessor for task PB0049_004A1_276_1
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | Poem@Home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task p2030.20131211.G177.56+00.31.C.b5s0g0.00000_3408_0
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task p2030.20131211.G177.56+00.31.C.b1s0g0.00000_2640_0
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task p2030.20131211.G177.56+00.31.C.b1s0g0.00000_2624_1
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task PB0055_00191_246_0
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task p2030.20131211.G177.56+00.31.C.b1s0g0.00000_1872_0
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task p2030.20131211.G177.56+00.31.C.b1s0g0.00000_0_0
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task p2030.20131211.G177.56+00.31.C.b1s0g0.00000_2560_0
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task PB0055_00191_336_1
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task p2030.20131211.G177.56+00.31.C.b1s0g0.00000_144_1
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Missing coprocessor for task p2030.20131211.G177.56+00.31.C.b1s0g0.00000_176_0
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | app version refers to missing GPU type NVIDIA
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | Application uses missing NVIDIA GPU
> > > 11/2/2014 12:30:07 AM | | Host name: Nitro
> > > 11/2/2014 12:30:07 AM | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7 CPU Q 740 @ 1.73GHz [Family 6 Model 30 Stepping 5]
> > > 11/2/2014 12:30:07 AM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 popcnt syscall nx lm vmx smx tm2 pbe
> > > 11/2/2014 12:30:07 AM | | OS: Microsoft Windows 8.1: Professional x64 Edition, (06.03.9600.00)
> > > 11/2/2014 12:30:07 AM | | Memory: 19.93 GB physical, 35.93 GB virtual
> > > 11/2/2014 12:30:07 AM | | Disk: 448.67 GB total, 328.64 GB free
> > > 11/2/2014 12:30:07 AM | | Local time is UTC -4 hours
> > > 11/2/2014 12:30:07 AM | | VirtualBox version: 4.3.12
> > > 11/2/2014 12:30:07 AM | Albert@Home | Found app_config.xml
> > > 11/2/2014 12:30:07 AM | Albert@Home | Your app_config.xml file refers to an unknown application 'hsgamma_FGRP2'. Known applications: 'einsteinbinary_BRP4', 'einsteinbinary_BRP5', 'einsteinbinary_BRP4G'
> > > 11/2/2014 12:30:07 AM | Poem@Home | Found app_config.xml
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Found app_config.xml
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Your app_config.xml file refers to an unknown application 'einsteinbinary_BRP4'. Known applications: 'einstein_S6CasA', 'einsteinbinary_BRP5', 'einsteinbinary_BRP4G'
> > > 11/2/2014 12:30:07 AM | Einstein@Home | Your app_config.xml file refers to an unknown application 'hsgamma_FGRP2'. Known applications: 'einstein_S6CasA', 'einsteinbinary_BRP5', 'einsteinbinary_BRP4G'
> > > 11/2/2014 12:30:07 AM | MindModeling@Beta | Found app_config.xml
> > > 11/2/2014 12:30:07 AM | SETI@home | Found app_config.xml
> > > 11/2/2014 12:30:07 AM | SETI@home | Your app_config.xml file refers to an unknown application 'setiathome_enhanced'. Known applications: 'astropulse_v6', 'setiathome_v7'
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | Found app_config.xml
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | Your app_config.xml file refers to an unknown application 'setiathome_enhanced'. Known applications: 'astropulse_v6', 'setiathome_v7', 'astropulse_v7'
> > > 11/2/2014 12:30:07 AM | | Config: use all coprocessors
> > > 11/2/2014 12:30:07 AM | | Config: event log limit 4000 lines
> > > 11/2/2014 12:30:07 AM | | Version change (7.4.22 -> 7.4.26)
> > > 11/2/2014 12:30:07 AM | Albert@Home | URL http://albert.phys.uwm.edu/; Computer ID 10320; resource share 1
> > > 11/2/2014 12:30:07 AM | Asteroids@home | URL http://asteroidsathome.net/boinc/; Computer ID 118229; resource share 0
> > > 11/2/2014 12:30:07 AM | ATLAS@home | URL http://atlasathome.cern.ch/; Computer ID 197; resource share 100
> > > 11/2/2014 12:30:07 AM | rosetta@home | URL http://boinc.bakerlab.org/rosetta/; Computer ID 1721823; resource share 100
> > > 11/2/2014 12:30:07 AM | drugdiscovery@home | URL http://boinc.drugdiscoveryathome.com/; Computer ID 16961; resource share 100
> > > 11/2/2014 12:30:07 AM | Poem@Home | URL http://boinc.fzk.de/poem/; Computer ID 211379; resource share 100
> > > 11/2/2014 12:30:07 AM | The Lattice Project | URL http://boinc.umiacs.umd.edu/; Computer ID 108139; resource share 0
> > > 11/2/2014 12:30:07 AM | boincsimap | URL http://boincsimap.org/boincsimap/; Computer ID 711285; resource share 100
> > > 11/2/2014 12:30:07 AM | climateathome | URL http://climateathome.com/climateathome/; Computer ID 1391; resource share 100
> > > 11/2/2014 12:30:07 AM | climateprediction.net | URL http://climateprediction.net/; Computer ID 1314488; resource share 100
> > > 11/2/2014 12:30:07 AM | Docking | URL http://docking.cis.udel.edu/; Computer ID 224149; resource share 100
> > > 11/2/2014 12:30:07 AM | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID 10383741; resource share 1
> > > 11/2/2014 12:30:07 AM | VirtualLHC@home | URL http://lhcathome2.cern.ch/vLHCathome/; Computer ID 69171; resource share 100
> > > 11/2/2014 12:30:07 AM | LHC@home 1.0 | URL http://lhcathomeclassic.cern.ch/sixtrack/; Computer ID 10316622; resource share 100
> > > 11/2/2014 12:30:07 AM | Beauty@LHC | URL http://lhcbathome.cern.ch/Beauty/; Computer ID 216; resource share 100
> > > 11/2/2014 12:30:07 AM | Milkyway@Home | URL http://milkyway.cs.rpi.edu/milkyway/; Computer ID 561351; resource share 0
> > > 11/2/2014 12:30:07 AM | MindModeling@Beta | URL http://mindmodeling.org/; Computer ID 47565; resource share 100
> > > 11/2/2014 12:30:07 AM | Quake-Catcher Network | URL http://qcn.stanford.edu/sensor/; Computer ID 32643; resource share 100
> > > 11/2/2014 12:30:07 AM | ralph@home | URL http://ralph.bakerlab.org/; Computer ID 31937; resource share 100
> > > 11/2/2014 12:30:07 AM | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 7211322; resource share 1
> > > 11/2/2014 12:30:07 AM | SETI@home Beta Test | URL http://setiweb.ssl.berkeley.edu/beta/; Computer ID 70486; resource share 1
> > > 11/2/2014 12:30:07 AM | correlizer | URL http://svahesrv2.bioquant.uni-heidelberg.de/correlizer/; Computer ID 15359; resource share 100
> > > 11/2/2014 12:30:07 AM | Citizen Science Grid | URL http://volunteer.cs.und.edu/csg/; Computer ID 4372; resource share 100
> > > 11/2/2014 12:30:07 AM | DNA@Home | URL http://volunteer.cs.und.edu/dna/; Computer ID 9698; resource share 100
> > > 11/2/2014 12:30:07 AM | WUProp@Home | URL http://wuprop.boinc-af.org/; Computer ID 65817; resource share 100
> > > 11/2/2014 12:30:07 AM | Bitcoin Utopia | URL http://www.bitcoinutopia.net/bitcoinutopia/; Computer ID 4351; resource share 0
> > > 11/2/2014 12:30:07 AM | Cosmology@Home | URL http://www.cosmologyathome.org/; Computer ID 207273; resource share 100
> > > 11/2/2014 12:30:07 AM | GPUGRID | URL http://www.gpugrid.net/; Computer ID 167515; resource share 100
> > > 11/2/2014 12:30:07 AM | RNA World | URL http://www.rnaworld.de/rnaworld/; Computer ID 30445; resource share 100
> > > 11/2/2014 12:30:07 AM | World Community Grid | URL http://www.worldcommunitygrid.org/; Computer ID 2809360; resource share 400
> > > 11/2/2014 12:30:07 AM | MindModeling@Beta | General prefs: from MindModeling@Beta (last modified 13-Nov-2013 15:43:52)
> > > 11/2/2014 12:30:07 AM | MindModeling@Beta | Host location: none
> > > 11/2/2014 12:30:07 AM | MindModeling@Beta | General prefs: using your defaults
> > > 11/2/2014 12:30:07 AM | | Reading preferences override file
> > > 11/2/2014 12:30:07 AM | | Preferences:
> > > 11/2/2014 12:30:07 AM | | max memory usage when active: 15305.53MB
> > > 11/2/2014 12:30:07 AM | | max memory usage when idle: 15305.53MB
> > > 11/2/2014 12:30:07 AM | | max disk usage: 341.51GB
> > > 11/2/2014 12:30:07 AM | | max upload rate: 30720 bytes/sec
> > > 11/2/2014 12:30:07 AM | | (to change preferences, visit a project web site or select Preferences in the Manager)
> > > 11/2/2014 12:30:07 AM | | [work_fetch] Request work fetch: Prefs update
> > > 11/2/2014 12:30:07 AM | | [work_fetch] Request work fetch: Startup
> > > 11/2/2014 12:30:07 AM | RNA World | Task cmsvm2_GA-p[e20-30MB_Lin64f]_1_Oryza-sativa-Japonica-Group_CM000147.lin.EMBL_RF00028_Intron_gpI_1349111823_64512_30 is 150.37 days overdue; you may not get credit for it. Consider aborting it.
> > > 11/2/2014 12:30:07 AM | | Not using a proxy
> > > 11/2/2014 12:30:08 AM | | Running CPU benchmarks
> > > 11/2/2014 12:30:08 AM | | Suspending computation - CPU benchmarks in progress
> > > 11/2/2014 12:30:08 AM | | Suspending network activity - time of day
> > > 11/2/2014 12:30:39 AM | | Benchmark results:
> > > 11/2/2014 12:30:39 AM | | Number of CPUs: 8
> > > 11/2/2014 12:30:39 AM | | 1774 floating point MIPS (Whetstone) per CPU
> > > 11/2/2014 12:30:39 AM | | 8603 integer MIPS (Dhrystone) per CPU
> > >
> > >
> > > 7.4.22:
> > > 11/2/2014 12:25:51 AM | | Starting BOINC client version 7.4.22 for windows_x86_64
> > > 11/2/2014 12:25:51 AM | | log flags: file_xfer, sched_ops, task, checkpoint_debug, scrsave_debug, unparsed_xml
> > > 11/2/2014 12:25:51 AM | | log flags: work_fetch_debug
> > > 11/2/2014 12:25:51 AM | | Libraries: libcurl/7.33.0 OpenSSL/1.0.1h zlib/1.2.8
> > > 11/2/2014 12:25:51 AM | | Data directory: C:\ProgramData\BOINC
> > > 11/2/2014 12:25:51 AM | | Running under account jacob_000
> > > 11/2/2014 12:25:51 AM | | CUDA: NVIDIA GPU 0: Quadro FX 3800M (driver version 340.84, CUDA version 6.5, compute capability 1.1, 1024MB, 862MB available, 634 GFLOPS peak)
> > > 11/2/2014 12:25:51 AM | | OpenCL: NVIDIA GPU 0: Quadro FX 3800M (driver version 340.84, device version OpenCL 1.0 CUDA, 1024MB, 862MB available, 634 GFLOPS peak)
> > > 11/2/2014 12:25:51 AM | | OpenCL CPU: Intel(R) Core(TM) i7 CPU Q 740 @ 1.73GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 3.0.1.15216, device version OpenCL 1.2 (Build 80752))
> > > 11/2/2014 12:25:51 AM | | Host name: Nitro
> > > 11/2/2014 12:25:51 AM | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7 CPU Q 740 @ 1.73GHz [Family 6 Model 30 Stepping 5]
> > > 11/2/2014 12:25:51 AM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 popcnt syscall nx lm vmx smx tm2 pbe
> > > 11/2/2014 12:25:51 AM | | OS: Microsoft Windows 8.1: Professional x64 Edition, (06.03.9600.00)
> > > 11/2/2014 12:25:51 AM | | Memory: 19.93 GB physical, 35.93 GB virtual
> > > 11/2/2014 12:25:51 AM | | Disk: 448.67 GB total, 328.65 GB free
> > > 11/2/2014 12:25:51 AM | | Local time is UTC -4 hours
> > > 11/2/2014 12:25:51 AM | | VirtualBox version: 4.3.12
> > > 11/2/2014 12:25:51 AM | Albert@Home | Found app_config.xml
> > > 11/2/2014 12:25:51 AM | Albert@Home | Your app_config.xml file refers to an unknown application 'hsgamma_FGRP2'. Known applications: 'einsteinbinary_BRP4', 'einsteinbinary_BRP5', 'einsteinbinary_BRP4G'
> > > 11/2/2014 12:25:51 AM | Poem@Home | Found app_config.xml
> > > 11/2/2014 12:25:51 AM | Einstein@Home | Found app_config.xml
> > > 11/2/2014 12:25:51 AM | Einstein@Home | Your app_config.xml file refers to an unknown application 'einsteinbinary_BRP4'. Known applications: 'einstein_S6CasA', 'einsteinbinary_BRP5', 'einsteinbinary_BRP4G'
> > > 11/2/2014 12:25:51 AM | Einstein@Home | Your app_config.xml file refers to an unknown application 'hsgamma_FGRP2'. Known applications: 'einstein_S6CasA', 'einsteinbinary_BRP5', 'einsteinbinary_BRP4G'
> > > 11/2/2014 12:25:51 AM | MindModeling@Beta | Found app_config.xml
> > > 11/2/2014 12:25:51 AM | SETI@home | Found app_config.xml
> > > 11/2/2014 12:25:51 AM | SETI@home | Your app_config.xml file refers to an unknown application 'setiathome_enhanced'. Known applications: 'astropulse_v6', 'setiathome_v7'
> > > 11/2/2014 12:25:51 AM | SETI@home Beta Test | Found app_config.xml
> > > 11/2/2014 12:25:51 AM | SETI@home Beta Test | Your app_config.xml file refers to an unknown application 'setiathome_enhanced'. Known applications: 'astropulse_v6', 'setiathome_v7', 'astropulse_v7'
> > > 11/2/2014 12:25:51 AM | | Config: use all coprocessors
> > > 11/2/2014 12:25:51 AM | | Config: event log limit 4000 lines
> > > 11/2/2014 12:25:51 AM | | Version change (7.4.26 -> 7.4.22)
> > > 11/2/2014 12:25:51 AM | Albert@Home | URL http://albert.phys.uwm.edu/; Computer ID 10320; resource share 1
> > > 11/2/2014 12:25:51 AM | Asteroids@home | URL http://asteroidsathome.net/boinc/; Computer ID 118229; resource share 0
> > > 11/2/2014 12:25:51 AM | ATLAS@home | URL http://atlasathome.cern.ch/; Computer ID 197; resource share 100
> > > 11/2/2014 12:25:51 AM | rosetta@home | URL http://boinc.bakerlab.org/rosetta/; Computer ID 1721823; resource share 100
> > > 11/2/2014 12:25:51 AM | drugdiscovery@home | URL http://boinc.drugdiscoveryathome.com/; Computer ID 16961; resource share 100
> > > 11/2/2014 12:25:51 AM | Poem@Home | URL http://boinc.fzk.de/poem/; Computer ID 211379; resource share 100
> > > 11/2/2014 12:25:51 AM | The Lattice Project | URL http://boinc.umiacs.umd.edu/; Computer ID 108139; resource share 0
> > > 11/2/2014 12:25:51 AM | boincsimap | URL http://boincsimap.org/boincsimap/; Computer ID 711285; resource share 100
> > > 11/2/2014 12:25:51 AM | climateathome | URL http://climateathome.com/climateathome/; Computer ID 1391; resource share 100
> > > 11/2/2014 12:25:51 AM | climateprediction.net | URL http://climateprediction.net/; Computer ID 1314488; resource share 100
> > > 11/2/2014 12:25:51 AM | Docking | URL http://docking.cis.udel.edu/; Computer ID 224149; resource share 100
> > > 11/2/2014 12:25:51 AM | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID 10383741; resource share 1
> > > 11/2/2014 12:25:51 AM | VirtualLHC@home | URL http://lhcathome2.cern.ch/vLHCathome/; Computer ID 69171; resource share 100
> > > 11/2/2014 12:25:51 AM | LHC@home 1.0 | URL http://lhcathomeclassic.cern.ch/sixtrack/; Computer ID 10316622; resource share 100
> > > 11/2/2014 12:25:51 AM | Beauty@LHC | URL http://lhcbathome.cern.ch/Beauty/; Computer ID 216; resource share 100
> > > 11/2/2014 12:25:51 AM | Milkyway@Home | URL http://milkyway.cs.rpi.edu/milkyway/; Computer ID 561351; resource share 0
> > > 11/2/2014 12:25:51 AM | MindModeling@Beta | URL http://mindmodeling.org/; Computer ID 47565; resource share 100
> > > 11/2/2014 12:25:51 AM | Quake-Catcher Network | URL http://qcn.stanford.edu/sensor/; Computer ID 32643; resource share 100
> > > 11/2/2014 12:25:51 AM | ralph@home | URL http://ralph.bakerlab.org/; Computer ID 31937; resource share 100
> > > 11/2/2014 12:25:51 AM | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 7211322; resource share 1
> > > 11/2/2014 12:25:51 AM | SETI@home Beta Test | URL http://setiweb.ssl.berkeley.edu/beta/; Computer ID 70486; resource share 1
> > > 11/2/2014 12:25:51 AM | correlizer | URL http://svahesrv2.bioquant.uni-heidelberg.de/correlizer/; Computer ID 15359; resource share 100
> > > 11/2/2014 12:25:51 AM | Citizen Science Grid | URL http://volunteer.cs.und.edu/csg/; Computer ID 4372; resource share 100
> > > 11/2/2014 12:25:51 AM | DNA@Home | URL http://volunteer.cs.und.edu/dna/; Computer ID 9698; resource share 100
> > > 11/2/2014 12:25:51 AM | WUProp@Home | URL http://wuprop.boinc-af.org/; Computer ID 65817; resource share 100
> > > 11/2/2014 12:25:51 AM | Bitcoin Utopia | URL http://www.bitcoinutopia.net/bitcoinutopia/; Computer ID 4351; resource share 0
> > > 11/2/2014 12:25:51 AM | Cosmology@Home | URL http://www.cosmologyathome.org/; Computer ID 207273; resource share 100
> > > 11/2/2014 12:25:51 AM | GPUGRID | URL http://www.gpugrid.net/; Computer ID 167515; resource share 100
> > > 11/2/2014 12:25:51 AM | RNA World | URL http://www.rnaworld.de/rnaworld/; Computer ID 30445; resource share 100
> > > 11/2/2014 12:25:51 AM | World Community Grid | URL http://www.worldcommunitygrid.org/; Computer ID 2809360; resource share 400
> > > 11/2/2014 12:25:51 AM | MindModeling@Beta | General prefs: from MindModeling@Beta (last modified 13-Nov-2013 15:43:52)
> > > 11/2/2014 12:25:51 AM | MindModeling@Beta | Host location: none
> > > 11/2/2014 12:25:51 AM | MindModeling@Beta | General prefs: using your defaults
> > > 11/2/2014 12:25:51 AM | | Reading preferences override file
> > > 11/2/2014 12:25:51 AM | | Preferences:
> > > 11/2/2014 12:25:51 AM | | max memory usage when active: 15305.53MB
> > > 11/2/2014 12:25:51 AM | | max memory usage when idle: 15305.53MB
> > > 11/2/2014 12:25:51 AM | | max disk usage: 341.49GB
> > > 11/2/2014 12:25:51 AM | | max upload rate: 30720 bytes/sec
> > > 11/2/2014 12:25:51 AM | | (to change preferences, visit a project web site or select Preferences in the Manager)
> > > 11/2/2014 12:25:51 AM | | [work_fetch] Request work fetch: Prefs update
> > > 11/2/2014 12:25:51 AM | | [work_fetch] Request work fetch: Startup
> > > 11/2/2014 12:25:51 AM | RNA World | Task cmsvm2_GA-p[e20-30MB_Lin64f]_1_Oryza-sativa-Japonica-Group_CM000147.lin.EMBL_RF00028_Intron_gpI_1349111823_64512_30 is 150.37 days overdue; you may not get credit for it. Consider aborting it.
> > > 11/2/2014 12:25:51 AM | | Not using a proxy
> > > 11/2/2014 12:25:52 AM | | Running CPU benchmarks
> > > 11/2/2014 12:25:52 AM | | Suspending computation - CPU benchmarks in progress
> > > 11/2/2014 12:25:52 AM | | Suspending network activity - time of day
> > > 11/2/2014 12:26:23 AM | | Benchmark results:
> > > 11/2/2014 12:26:23 AM | | Number of CPUs: 8
> > > 11/2/2014 12:26:23 AM | | 1779 floating point MIPS (Whetstone) per CPU
> > > 11/2/2014 12:26:23 AM | | 8550 integer MIPS (Dhrystone) per CPU
> >
>
> _______________________________________________
> boinc_alpha mailing list
> b..._alpha@ssl.berkeley.edu
> http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_alpha
> To unsubscribe, visit the above URL and
> (near bottom of page) enter your email address.

Here is some correspondence that I initiated on BOINC alpha, when I noticed that BOINC 7.4.26 disabled use of my FX3800M GPU.

I'm not trying to stir the pot, but instead, am just trying to relay information, and get information. Note that I was not aware of this thread, when I sent my original email. I'm still trying to comprehend the decisions being made here, but my gut tells me it might be the wrong call to blanket-blacklist GPUs in BOINC.

What I'd like to see is:
- A simple downloadable test case where I can prove to myself that the application returns valid data on some drivers, and invalid data on other drivers.

There's Bench programs and test Wu's available at Lunatics for years

- Do we know if CUDA 6.5 results (which don't use OpenCL) are affected?

Pre-Fermi GPUs have been deprecated from the Cuda 6.5 SDK, So no Cuda 6.5 apps for Pre-Fermis,
There have been no mass reports of validations problems that we know of here using the Cuda22, Cuda23, Cuda42 or Cuda50 apps.

- Can we see the quoted NVIDIA responses regarding them indicating that they will not fix the issue?
- Can we get a clear understanding of why the project can't use validation logic in the validator, and server scheduling logic, to completely mitigate the problem?
- Are you aware that NVIDIA plans to support the R340 branch until April 1, 2016, per http://nvidia.custhelp.com/app/answers/detail/a_id/3473
- Does anyone else agree that the blanket blacklist approach, built into BOINC, may not be the right solution here?

Regards,
Jacob

It's reported on Nvidia Dev Forums here:

https://devtalk.nvidia.com/default/topic/774899/340-52-driver-breaks-opencl-support-for-pre-fermi-cc-1-x-cards/#4310165

and in Bug report here, Confirmed, and Won't fix:

https://developer.nvidia.com/nvbugs/cuda/edit/1554016

Please do not reply to this message
-------------------------------------------------------------------
The following items have been modified for this Bug:
- Status changed from "Open - in progress" to "Closed - Will not fix"

-------------------------------------------------------------------
Bug Information
-------------------------------------------------------------------
Requester: Raistmer _
NVIDIA Bug ID: 1554016
Date: 09/13/2014 2:13 AM
Company/Division: CUDA RegDev Program [DevZone]
Severity: High
Priority: 1 - High
Synopsis: OpenCL support broken in 340.52 driver for pre-FERMI cards
Description: With 340.52 driver OpenCL SETI@home AstroPulse app generates incorrect results on pre-FERMI hardware. With previous drivers like 337.50 beta app generates valid results on the same hardware
-------------------------------------------------------------------
Last comment from NVIDIA (09/16/2014 3:03 AM):
Thanks for the reporting and providing the repro case.
We have reproduced this issue and has assigned it to the appropriate developer team for investigation, we'll keep you posted once the further information becomes available. Thanks!

I believe the Blacklist approach is the right approach, at least for multi-GPU hosts, where the most Capable GPU is used as standard for whether work is sent.

Claggy

Claggy,

Thank you for responding. Regarding "There's Bench programs and test Wu's available at Lunatics for years", could you kindly explain how I could get started testing with that? I'm not familiar with "Bench" nor "Lunatics", and am really looking for a simple test program I can run to see the problem for myself.

Thanks in advance,
Jacob

I've sent Jacob an email with our APbench test suite, pre-populated with the test case application and results that we used to generate that NV bug report.

Thanks Richard.

Is it possible for us to diagnose which OpenCL functions are returning invalid results? Also, do we have an application outside of SETI, such as a standard OpenCL benchmark, that we can use to reproduce the problem? I'm trying to rule out an application problem here, somehow.

Note: I played with LuxMark last night, and it seemed to work okay on my FX3800M on R340 drivers on Windows 8.1 x64. Can someone else confirm that LuxMark does *not* demonstrate any problems?

Regards,
Jacob

Thanks Richard.

Is it possible for us to diagnose which OpenCL functions are returning invalid results? Also, do we have an application outside of SETI, such as a standard OpenCL benchmark, that we can use to reproduce the problem? I'm trying to rule out an application problem here, somehow.

Note: I played with LuxMark last night, and it seemed to work okay on my FX3800M on R340 drivers on Windows 8.1 x64. Can someone else confirm that LuxMark does *not* demonstrate any problems?

Regards,
Jacob

I did report back my confirmation results to Raistmer with the added question:

I then upgraded to driver 340.52, and re-ran the test. The repetitive pulse reports were byte-identical with my previous results, but the four single pulses were entirely missing. It seems to me that it might help NVidia track this down if you could identify which .cl kernels are used during single pulse processing, but not during repetitive pulse processing, and submit them for inspection. Or is that too simple?

His reply was

I added comments regarding possible affected kernels inside CL file too.

- so they should be accessible to anyone with credentials for accessing https://developer.nvidia.com/nvbugs/cuda/edit/1554016, but that doesn't include me.

Hello, I need your insight and help.
I am using this Video card, NVIDIA GeForce GTX 260 (896MB) driver: 311.06 OpenCL: 1.0

In the past we have never had a problem but now we are receiving
Computation error running seti@homev77.00 (cuda22)

We are not too familiar with much of the program but we support the efforts to run the data sets. Can you please tell me if we need to change something with our setup or will these errors clear themselves or just continue to build up in the task tab?

Hello, If we need to change video cards from NVidia GeForce. What video cards will work to crunch these numbers that you are aware of if any?

Thanks again

---

Thomas,

Look at this post here in this thread.

http://setiathome.berkeley.edu/forum_thread.php?id=75633&postid=1572217

Either go with older drivers for your graphic card or upgrade your GPU. Anything over and including the GTX 400s should be ok. Bang for the buck GTX 750 but not SLI compatible so if you are using it for gaming as well be aware of that.

I added comments regarding possible affected kernels inside CL file too.

- so they should be accessible to anyone with credentials for accessing https://developer.nvidia.com/nvbugs/cuda/edit/1554016, but that doesn't include me.

After status "closed, won't fix" I added few comments, status was changed to "open", but no human-based response was given. I'm afraid it's just bug in their bugtracking system that changes status every time new comment added.
I would not expect any improvements from NV in this area. They refuse to improve OpenCL support even for new cards, even when popular cross-platform benchmark developer requested such improvement (currently NV looks very bad in that OpenCL-based benchmark)...

If someone want to debug this issue more - code available at SETI repo, function under suspiction is SinglePulse find kernel. Cause rep pulses not affected I think only part of that single pulse find kernel behaves wrongly under new drivers.

Can you give more details about the suspected function? Can you paste which OpenCL function calls it makes?

I was able to determine that the following 4 OpenCL SDK applications (available here: https://developer.nvidia.com/opencl) will work for my [R337 337.88 Cuda6.0] driver, but will fail for my [R340 341.05 Cuda6.5] driver, on my FX 3800M on Windows 8.1 x64:
- oclFDTD3d
- oclDXTCompression
- oclQuasirandomGenerator
- oclConvolutionSeparable

Furthermore, Charlie found an interesting tidbit in the Cuda6.5 toolkit errata (http://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#ixzz3H3EPPdHB), that states:

Incorrect results from functions csr2csc() and bsr2bsc()
The csr2csc() and bsr2bsc() routines contain a bug in the CUDA 6.0 and 6.5 releases. As a consequence, csrsv(), csrsv2(), csrsm(), bsrsv2(), bsrsm2(), and csrgemm() may produce incorrect results when working with transpose (CUSPARSE_OPERATION_TRANSPOSE) or conjugate-transpose (CUSPARSE_OPERATION_CONJUGATE_TRANSPOSE) operations. These routines work correctly when the non-transpose (CUSPARSE_OPERATION_NON_TRANSPOSE) operation is selected. The bug has been fixed in the CUDA 7.0 release.


Does your application utilize any of those 6 functions, along with transpose or conjugate-transpose operations?

PC_single_pulse_kernel_FFA_update_reduce0 and other with similar name from here:

https://setisvn.ssl.berkeley.edu/svn/branches/sah_v7_opt/AP_BLANKIT/client/AstroPulse_Kernels.cl

I doesn't call any special function AFAIK.