Until thursday cuda-sieving run without any failure. Since then every task abort with following stderr:

Update:
done an additional test (from sieve-subforum):

/var/lib/boinc-client/projects/www.primegrid.com $ ./primegrid_tpsieve_1.35_x86_64-pc-linux-gnu__cuda23 -p42070e9 -P42070030e6 -k 1201 -K 9999 -N 2000000 -c 60 -M2 --device 0
tpsieve version cuda-0.2.2d (testing)
Compiled Nov 8 2010 with GCC 4.3.3
nstart=76, nstep=32
Didn't change nstep from 31
tpsieve initialized: 1201 <= k <= 9999, 76 <= n < 2000000
42070000070587 | 9475*2^197534+1
42070000154219 | 6023*2^934790-1
42070000198537 | 3373*2^1046686+1
42070001803331 | 5237*2^486598-1
42070003062431 | 7465*2^1994555-1
42070003101727 | 4207*2^1054290+1
42070003511309 | 6057*2^1043547+1
42070005645821 | 3633*2^119620-1
42070006307657 | 1513*2^1771812+1
42070006388603 | 2059*2^1816098+1
42070007177519 | 5437*2^1121592+1
42070007396759 | 7339*2^1803518+1
42070007733361 | 7007*2^1691614-1
42070008458437 | 7095*2^1422761-1
42070008823897 | 4639*2^952018+1
42070008858187 | 2893*2^317690+1
42070010190569 | 5625*2^1903125+1
42070011430123 | 3821*2^1406279+1
42070012209011 | 9405*2^360411-1
42070012301263 | 1957*2^1185814+1
42070013521999 | 1965*2^404493+1
42070013970587 | 7143*2^1462422+1
42070013989247 | 5037*2^838603+1
42070016416499 | 4571*2^466510-1
42070017332953 | 6237*2^1916994+1
42070018235321 | 1941*2^363948+1
42070019117111 | 2523*2^999263-1
42070019542387 | 8587*2^1703626+1
42070021901227 | 6589*2^1149693-1
42070023987581 | 9811*2^318944+1
42070024242289 | 8319*2^1792800-1
42070024339237 | 9257*2^1170495+1
42070024532551 | 4311*2^1690093+1
42070024936837 | 5679*2^1726142+1
42070024995961 | 9111*2^1707153+1
42070026021997 | 4039*2^1819590+1
42070026719239 | 9981*2^629165-1
42070027452199 | 1323*2^854008+1
42070028029061 | 8205*2^1394191-1
42070029006583 | 5943*2^663870+1
Found 40 factors

I don't know if it's related, but my CUDA tasks are now taking about 21-22 minutes, vs. the 18-20 minutes they were taking before. It appears as if something changed.

I think John mentioned something about transitioning to a new phase of work.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

I don't know if it's related, but my CUDA tasks are now taking about 21-22 minutes, vs. the 18-20 minutes they were taking before. It appears as if something changed.

I think John mentioned something about transitioning to a new phase of work.

Update:
done an additional test (from sieve-subforum):

/var/lib/boinc-client/projects/www.primegrid.com $ ./primegrid_tpsieve_1.35_x86_64-pc-linux-gnu__cuda23 -p42070e9 -P42070030e6 -k 1201 -K 9999 -N 2000000 -c 60 -M2 --device 0
tpsieve version cuda-0.2.2d (testing)
Compiled Nov 8 2010 with GCC 4.3.3
nstart=76, nstep=32
Didn't change nstep from 31
tpsieve initialized: 1201 <= k <= 9999, 76 <= n < 2000000
42070000070587 | 9475*2^197534+1
42070000154219 | 6023*2^934790-1
.
.
.
42070027452199 | 1323*2^854008+1
42070028029061 | 8205*2^1394191-1
42070029006583 | 5943*2^663870+1
Found 40 factors

stderr looks good:

Can't open init data file - running in standalone mode Sieve started: 42070000000000 <= p < 42070030000000 Thread 0 starting Detected GPU 0: GeForce 9800 GTX+ Detected compute capability: 1.1 Detected 16 multiprocessors. Thread 0 completed Sieve complete: 42070000000000 <= p < 42070030000000 count=955289,sum=0x2dbc17167afb6a8d Elapsed time: 15.38 sec. (0.02 init + 15.36 sieve) at 1962751 p/sec. Processor time: 0.41 sec. (0.03 init + 0.38 sieve) at 79333053 p/sec. Average processor utilization: 1.57 (init), 0.02 (sieve) called boinc_finish

I am more and more confused ...

done as mentioned by ralf

Computation Error: no candidates found for p=1507576375167301.

done as mentioned by ralf

$ ./primegrid_tpsieve_1.35_x86_64-pc-linux-gnu__cuda23 -p1507576000000000 -P1507579000000000 -k 5 -K 9999 -N 3M -c 60 -M2 --device 0

sieving stops after reporting some factors
stderr says

Computation Error: no candidates found for p=1507576375167301.

second test:

Computation Error: no candidates found for p=1507576223882821

With the comment that the Fermi GPU Code is somewhat different from the non-Fermi Code here is the result of my test run:

C:\Users\Ralf\Desktop\tpsieve-cuda>tpsieve-cuda-x86-windows.exe -p 1507576G -P 1507579G -k 5 -K 9999
-N 3M -c 60 -M2 -q
tpsieve version cuda-0.2.2d (testing)
nstart=86, nstep=38
nstep changed to 32
tpsieve initialized: 5 <= k <= 9999, 86 <= n < 3000000
Sieve started: 1507576000000000 <= p < 1507579000000000
Thread 0 starting
Detected GPU 0: GeForce GTX 460
Detected compute capability: 2.1
Detected 7 multiprocessors.
p=1507578852913153, 5.266M p/sec, 0.03 CPU cores, 95.1% done. ETA 26 Nov 18:57
Thread 0 completed
Waiting for threads to exit
Sieve complete: 1507576000000000 <= p < 1507579000000000
Found 162 factors
count=85841579,sum=0x7bd6428c20144c75
Elapsed time: 570.27 sec. (0.14 init + 570.13 sieve) at 5262348 p/sec.
Processor time: 18.33 sec. (0.16 init + 18.17 sieve) at 165083027 p/sec.
Average processor utilization: 1.14 (init), 0.03 (sieve)
____________

Watch out you are using Windows, I am using Linux
changed now NVIDIA
old (with failure) NVIDIA-Linux-x86_64-260.19.12.run
new (to come) NVIDIA-Linux-x86_64-256.53.run

____________
Sysadm@Nbg
my current lucky number: 3749*2^1555697+1
PSA-PRPNet-Stats-URL: http://u-g-f.de/PRPNet/

The BOINC version delivers the same results:

C:\Users\Ralf\Desktop\tpsieve-cuda>tpsieve-cuda-boinc-x86-windows.exe -p 1507576G -P 1507579G -k 5 -K 9999 -N 3M -c 60 -M2 -q
tpsieve version cuda-0.2.2d (testing)
nstart=86, nstep=38
nstep changed to 32
tpsieve initialized: 5 <= k <= 9999, 86 <= n < 3000000
Found 162 factors

stderr.txt:

18:59:25 (5108): Can't open init data file - running in standalone mode
Sieve started: 1507576000000000 <= p < 1507579000000000
Thread 0 starting
Detected GPU 0: GeForce GTX 460
Detected compute capability: 2.1
Detected 7 multiprocessors.

Thread 0 completed
Sieve complete: 1507576000000000 <= p < 1507579000000000
count=85841579,sum=0x7bd6428c20144c75
Elapsed time: 569.63 sec. (0.14 init + 569.49 sieve) at 5268336 p/sec.
Processor time: 16.47 sec. (0.14 init + 16.33 sieve) at 183688364 p/sec.
Average processor utilization: 1.00 (init), 0.03 (sieve)
19:08:54 (5108): called boinc_finish

____________

Watch out you are using Windows, I am using Linux
changed now NVIDIA
old (with failure) NVIDIA-Linux-x86_64-260.19.12.run
new (to come) NVIDIA-Linux-x86_64-256.53.run

result to come round about 19:30 CET / 18:30 UTC
it is still running, so I hope ...
____________
Sysadm@Nbg
my current lucky number: 3749*2^1555697+1
PSA-PRPNet-Stats-URL: http://u-g-f.de/PRPNet/

I've started a test run with Maverick (10.10) amd64 and the 260.19.21 drivers and it looks like a driver or linux problem:

ralf@quadriga:~/Desktop$ ./tpsieve-cuda-boinc-x86_64-linux -p 1507576G -P 1507579G -k 5 -K 9999 -N 3M -c 60 -M2 -q
tpsieve version cuda-0.2.2d (testing)
Compiled Nov 8 2010 with GCC 4.3.3
nstart=86, nstep=38
nstep changed to 32
tpsieve initialized: 5 <= k <= 9999, 86 <= n < 3000000
ralf@quadriga:~/Desktop$ cat stderr.txt
Can't open init data file - running in standalone mode
Sieve started: 1507576000000000 <= p < 1507579000000000
Thread 0 starting
Detected GPU 0: GeForce GTX 460
Detected compute capability: 2.1
Detected 7 multiprocessors.
Computation Error: no candidates found for p=1507576091573129.
called boinc_finish

____________

take my driver as mentioned before
or my teammates driver (he is still running successfully)

Yup. Repeated crashes here too. I've read about problems with the 260.19.xy driver series and the reports seem to be true.
____________

I think this is a driver problem. I have no problems with my 260GTX with driver 256.44. After downgrading the driver on my other computer (2x460GTX) from 260.19.xx to 256.53 i have no errors. I test both driverversions of the 260.19, i get errors with both.
Both systems run with Debian GNU Linux testing.
____________

done, success, Found 162 factors!

stderr

Will run in standalone mode.
Sieve started: 1507576000000000 <= p < 1507579000000000
Thread 0 starting
Detected GPU 0: GeForce 9800 GTX+
Detected compute capability: 1.1
Detected 16 multiprocessors.

Thread 0 completed
Sieve complete: 1507576000000000 <= p < 1507579000000000
count=85841579,sum=0x7bd6428c20144c75
Elapsed time: 1672.97 sec. (0.11 init + 1672.86 sieve) at 1793476 p/sec.
Processor time: 30.25 sec. (0.12 init + 30.13 sieve) at 99576438 p/sec.
Average processor utilization: 1.08 (init), 0.02 (sieve)
called boinc_finish

Interesting. I sieved several thousand G for PSA (PPR3M) with tpsieve and Linux without problems. I'll send Ken a PM.
____________

I suspect the bug with 260 drivers is a timing issue. Manual sieving usually uses the non-BOINC version, which has a separate thread for sieving.

I'm using the 260.19.14 driver, and I haven't had too many problems yet; but I haven't tried the newest work.
____________

I suspect the bug with 260 drivers is a timing issue. Manual sieving usually uses the non-BOINC version, which has a separate thread for sieving.

I'm using the 260.19.14 driver, and I haven't had too many problems yet; but I haven't tried the newest work.

With the old Workunits (1800 credits) i had also no problems with driverversion 260.19.21, but after PG send out new work (2314 credits) all wus ended with error.

tested driver on 460GTX:
260.19.21: works fine with old workunits, error with new workunits
260.19.12: erros with new workunits, not tested with old
256.53: works good with new workunits, not tested with old
256.44: works fine with old and new, but is 40 seconds slower on old wus in comparison with 260.19.21
____________

256.53: works good with new workunits, not tested with old

For those sticking with 260 drivers (or unable to leave them for some reason), it looks like adding "-m 8" to the command line may make the results better for Fermi users. I believe you can add "<cmdline>-m 8</cmdline>" to your app_info.xml file after "</coproc>", but I wouldn't swear that's right.
____________

I've already asked to get a new client installed with -m 8 default for Fermi. As for -m 2 and a GTS 250, though, that's just way smaller than it should have to be.

By the way, I got a mod at nVIDIA interested but not yet convinced. Can anyone verify that a particular command line works with earlier drivers but not with the current ones? Try one of these programs and a command line like this, for example:

./tpsieve-cuda-(version) -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64
____________

I've already asked to get a new client installed with -m 8 default for Fermi. As for -m 2 and a GTS 250, though, that's just way smaller than it should have to be.

By the way, I got a mod at nVIDIA interested but not yet convinced. Can anyone verify that a particular command line works with earlier drivers but not with the current ones? Try one of these programs and a command line like this, for example:

./tpsieve-cuda-(version) -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64

After work (22:00 UTC) i can make same tests with the 260 driver. I can also make same tests with 260GTX if needed.

2xGTX460 driver 256.53:

<cmdline> -m 8</cmdline> works (same runtime, same cpu-time; but i had only time for some fast tests)

./tpsieve-cuda-x86_64-linux -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64 --device 0
......
Thread 0 completed
Waiting for threads to exit
Sieve complete: 420700000000000 <= p < 420701000000000
Found 208 factors
count=29703006,sum=0x69ccf6011cb93a06
Elapsed time: 250.82 sec. (0.04 init + 250.78 sieve) at 3987928 p/sec.
Processor time: 4.78 sec. (0.04 init + 4.74 sieve) at 210974032 p/sec.
Average processor utilization: 1.04 (init), 0.02 (sieve)

____________

GT240(GT215) driver 195.36.31
<cmdline>-m 6</cmdline> seems to be the fastest

boinc@vmware2k-3:~/Cuda/tpsieve$ ./tpsieve-cuda-x86_64-linux -p42070e9 -P42070030e6 -k 1201 -K 9999 -N 3000000 -c 60 -M2 -m 2 -T -q --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=76, nstep=32
Changed nstep to 31
tpsieve initialized: 1201 <= k <= 9999, 76 <= n < 3000000
Sieve started: 42070000000000 <= p < 42070030000000
Thread 0 starting
Detected GPU 0: GeForce GT 240
Detected compute capability: 1.2
Detected 12 multiprocessors.
...
Elapsed time: 40.24 sec. (0.02 init + 40.22 sieve) at 749565 p/sec.
Processor time: 0.55 sec. (0.02 init + 0.53 sieve) at 56663058 p/sec.
Average processor utilization: 1.20 (init), 0.01 (sieve)

boinc@vmware2k-3:~/Cuda/tpsieve$ ./tpsieve-cuda-x86_64-linux -p42070e9 -P42070030e6 -k 1201 -K 9999 -N 3000000 -c 60 -M2 -m 4 -T -q --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=76, nstep=32
Changed nstep to 31
tpsieve initialized: 1201 <= k <= 9999, 76 <= n < 3000000
Sieve started: 42070000000000 <= p < 42070030000000
Thread 0 starting
Detected GPU 0: GeForce GT 240
Detected compute capability: 1.2
Detected 12 multiprocessors.
...
Elapsed time: 39.78 sec. (0.02 init + 39.76 sieve) at 758152 p/sec.
Processor time: 0.53 sec. (0.02 init + 0.51 sieve) at 59340001 p/sec.
Average processor utilization: 1.20 (init), 0.01 (sieve)

boinc@vmware2k-3:~/Cuda/tpsieve$ ./tpsieve-cuda-x86_64-linux -p42070e9 -P42070030e6 -k 1201 -K 9999 -N 3000000 -c 60 -M2 -m 6 -T -q --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=76, nstep=32
Changed nstep to 31
tpsieve initialized: 1201 <= k <= 9999, 76 <= n < 3000000
Sieve started: 42070000000000 <= p < 42070030000000
Thread 0 starting
Detected GPU 0: GeForce GT 240
Detected compute capability: 1.2
Detected 12 multiprocessors.
...
Elapsed time: 39.77 sec. (0.02 init + 39.76 sieve) at 758261 p/sec.
Processor time: 0.53 sec. (0.02 init + 0.51 sieve) at 58876435 p/sec.
Average processor utilization: 1.20 (init), 0.01 (sieve)

boinc@vmware2k-3:~/Cuda/tpsieve$ ./tpsieve-cuda-x86_64-linux -p42070e9 -P42070030e6 -k 1201 -K 9999 -N 3000000 -c 60 -M2 -m 8 -T -q --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=76, nstep=32
Changed nstep to 31
tpsieve initialized: 1201 <= k <= 9999, 76 <= n < 3000000
Sieve started: 42070000000000 <= p < 42070030000000
Thread 0 starting
Detected GPU 0: GeForce GT 240
Detected compute capability: 1.2
Detected 12 multiprocessors.
...
Elapsed time: 56.30 sec. (0.02 init + 56.29 sieve) at 535576 p/sec.
Processor time: 0.52 sec. (0.02 init + 0.50 sieve) at 60289382 p/sec.
Average processor utilization: 1.20 (init), 0.01 (sieve)

I'm runing linux64 with the older kernel 2.6.27-17 and driver 195.36.31 (cuda3.0). Your kernel is 2.6.37-3 and we don't know your driver-version. IIRC, cuda3.2 needs a driver higher than 262.x

I've already asked to get a new client installed with -m 8 default for Fermi. As for -m 2 and a GTS 250, though, that's just way smaller than it should have to be.

By the way, I got a mod at nVIDIA interested but not yet convinced. Can anyone verify that a particular command line works with earlier drivers but not with the current ones? Try one of these programs and a command line like this, for example:

./tpsieve-cuda-(version) -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64

I forgot to post my value for "-m 64":

boinc@vmware2k-3:~/Cuda/tpsieve$ ./tpsieve-cuda-x86_64-linux -p42070e9 -P42070030e6 -k 1201 -K 9999 -N 3000000 -c 60 -M2 -m 64 -T -q --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=76, nstep=32
Changed nstep to 31
tpsieve initialized: 1201 <= k <= 9999, 76 <= n < 3000000
Sieve started: 42070000000000 <= p < 42070030000000
Thread 0 starting
Detected GPU 0: GeForce GT 240
Detected compute capability: 1.2
Detected 12 multiprocessors.

Thread 0 completed
Waiting for threads to exit
Sieve complete: 42070000000000 <= p < 42070030000000
Found 69 factors
count=955289,sum=0x2dbc17167afb6a8d
Elapsed time: 42.30 sec. (0.02 init + 42.28 sieve) at 712943 p/sec.
Processor time: 21.24 sec. (0.02 init + 21.23 sieve) at 1420311 p/sec.
Average processor utilization: 0.96 (init), 0.50 (sieve)

boinc@vmware2k-3:~/Cuda/tpsieve$ ./tpsieve-cuda-x86_64-linux -p42070e9 -P42070030e6 -k 1201 -K 9999 -N 3000000 -c 60 -M2 -m 60 -T -q --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=76, nstep=32
Changed nstep to 31
tpsieve initialized: 1201 <= k <= 9999, 76 <= n < 3000000
Sieve started: 42070000000000 <= p < 42070030000000
Thread 0 starting
Detected GPU 0: GeForce GT 240
Detected compute capability: 1.2
Detected 12 multiprocessors.

Thread 0 completed
Waiting for threads to exit
Sieve complete: 42070000000000 <= p < 42070030000000
Found 69 factors
count=955289,sum=0x2dbc17167afb6a8d
Elapsed time: 42.37 sec. (0.02 init + 42.35 sieve) at 711830 p/sec.
Processor time: 19.92 sec. (0.02 init + 19.91 sieve) at 1514503 p/sec.
Average processor utilization: 0.96 (init), 0.47 (sieve)

test with driver 256.53 & 260.19.xx on GTX460

driver 256.53 (GTX460):
###################

./tpsieve-cuda-x86_64-linux -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64 --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=82, nstep=36
nstep changed to 32
tpsieve initialized: 1201 <= k <= 9999, 82 <= n < 3000000
Sieve started: 420700000000000 <= p < 420701000000000
Thread 0 starting
Detected GPU 0: GeForce GTX 460
Detected compute capability: 2.1
Detected 7 multiprocessors.
....
Thread 0 completed
Waiting for threads to exit
Sieve complete: 420700000000000 <= p < 420701000000000
Found 208 factors
count=29703006,sum=0x69ccf6011cb93a06
Elapsed time: 250.62 sec. (0.04 init + 250.58 sieve) at 3991101 p/sec.
Processor time: 4.68 sec. (0.04 init + 4.64 sieve) at 215520875 p/sec.
Average processor utilization: 0.98 (init), 0.02 (sieve)


driver 260.19.12 (GTX460):
#####################

./tpsieve-cuda-x86_64-linux -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64 --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=82, nstep=36
nstep changed to 32
tpsieve initialized: 1201 <= k <= 9999, 82 <= n < 3000000
Sieve started: 420700000000000 <= p < 420701000000000
Resuming from checkpoint p=420700689438721 in tpcheck420700e9.txt
Thread 0 starting
Detected GPU 0: GeForce GTX 460
Detected compute capability: 2.1
Detected 7 multiprocessors.
....
Thread 0 completed
Waiting for threads to exit
Sieve complete: 420700000000000 <= p < 420701000000000
Found 208 factors
count=29703006,sum=0x69ccf6011cb93a06
Elapsed time: 259.13 sec. (0.04 init + 259.09 sieve) at 3859950 p/sec.
Processor time: 4.66 sec. (0.04 init + 4.61 sieve) at 216829339 p/sec.
Average processor utilization: 1.02 (init), 0.02 (sieve)


driver 260.19.21 (GTX460):
#####################

./tpsieve-cuda-x86_64-linux -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64 --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=82, nstep=36
nstep changed to 32
tpsieve initialized: 1201 <= k <= 9999, 82 <= n < 3000000
Sieve started: 420700000000000 <= p < 420701000000000
Resuming from checkpoint p=420700090963969 in tpcheck420700e9.txt
Thread 0 starting
Detected GPU 0: GeForce GTX 460
Detected compute capability: 2.1
Detected 7 multiprocessors.
....
Thread 0 completed
Waiting for threads to exit
Sieve complete: 420700000000000 <= p < 420701000000000
Found 208 factors
count=29703006,sum=0x69ccf6011cb93a06
Elapsed time: 235.51 sec. (0.07 init + 235.43 sieve) at 3861472 p/sec.
Processor time: 4.42 sec. (0.04 init + 4.38 sieve) at 207737311 p/sec.
Average processor utilization: 0.59 (init), 0.02 (sieve)

____________

1. You have to use the BOINC version to get errors, it appears. Edit: Yes, I mean tpsieve-cuda-boinc-(version).
2. Please delete any tpcheck*.txt between runs.
____________

- make 10 test per driver (5 per card, 2x MSI N460GTX CYCLONE OC 1GD5)
- all tasks with driver 260.xx have errors, but the the number of factors before the task aborted is differing
- delete stderr.txt, tpfactors.txt and tpcheck*.txt (if available) between runs

driver 256.53 (GTX460):
#######################

./tpsieve-cuda-boinc-x86_64-linux -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64 --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=82, nstep=36
nstep changed to 32
tpsieve initialized: 1201 <= k <= 9999, 82 <= n < 3000000
...
Found 208 factors

more stderr.txt
shmget in attach_shmem: Invalid argument
Can't set up shared mem: -1
Will run in standalone mode.
Sieve started: 420700000000000 <= p < 420701000000000
Thread 0 starting
Detected GPU 0: GeForce GTX 460
Detected compute capability: 2.1
Detected 7 multiprocessors.

Thread 0 completed
Sieve complete: 420700000000000 <= p < 420701000000000
count=29703006,sum=0x69ccf6011cb93a06
Elapsed time: 248.63 sec. (0.04 init + 248.59 sieve) at 4022983 p/sec.
Processor time: 4.57 sec. (0.04 init + 4.52 sieve) at 221047039 p/sec.
Average processor utilization: 1.06 (init), 0.02 (sieve)
called boinc_finish


driver 260.19.12 (GTX460):
##########################

./tpsieve-cuda-boinc-x86_64-linux -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64 --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=82, nstep=36
nstep changed to 32
tpsieve initialized: 1201 <= k <= 9999, 82 <= n < 3000000

more stderr.txt
shmget in attach_shmem: Invalid argument
Can't set up shared mem: -1
Will run in standalone mode.
Sieve started: 420700000000000 <= p < 420701000000000
Thread 0 starting
Detected GPU 0: GeForce GTX 460
Detected compute capability: 2.1
Detected 7 multiprocessors.
Computation Error: no candidates found for p=420700026453163.
called boinc_finish


driver 260.19.21 (GTX460):
##########################

./tpsieve-cuda-boinc-x86_64-linux -p420700e9 -P420701000e6 -k 1201 -K 9999 -N 3000000 -c 60 -M 2 -T -m 64 --device 0
tpsieve version cuda-0.2.2e (testing)
Compiled Nov 27 2010 with GCC 4.3.3
nstart=82, nstep=36
nstep changed to 32
tpsieve initialized: 1201 <= k <= 9999, 82 <= n < 3000000

more stderr.txt
shmget in attach_shmem: Invalid argument
Can't set up shared mem: -1
Will run in standalone mode.
Sieve started: 420700000000000 <= p < 420701000000000
Thread 0 starting
Detected GPU 0: GeForce GTX 460
Detected compute capability: 2.1
Detected 7 multiprocessors.
Computation Error: no candidates found for p=420700154581337.
called boinc_finish

____________

That is the best option for now, yes. The second-best option is to include a -m argument with a low value in an app_info.xml file.
____________

Maybe a bug should be filed at Nvidia?
If this is a real driver-error it should be fixed by the developer of the driver if think.