Number of WU thread block time (minutes-seconds)
10 2 2000 8:12
10 5 4000 8:30
10 10 10000 9:23

So if you look at this table 10/10000 is slowest ,and 2/2000 is fastest.

BUT

if you test it then results is opposite but value p/sec show right direction

C:\Users\IntelC\Desktop\p>wwwwcl64.exe -p1334568e10 -P1334569e10 -TWallSunSun -t2 -b2000
wwwwcl v2.1.9, a GPU program to search for Wieferich and WallSunSun primes
Sieve started: 13345680000000000 <= p < 13345690000000000

Sieve complete: 13345680000000001 <= p < 13345690000000000 269325538 primes tested
Elapsed time: 49.36 sec. (1.15 init + 47.27 sieve) at 5498225 p/sec.
Processor time: 16.27 sec. (1.34 init + 14.93 sieve).
Seconds spent in CPU and GPU: 0.37 (cpu), 82.66 (gpu)


C:\Users\IntelC\Desktop\p>wwwwcl64.exe -p1334568e10 -P1334569e10 -TWallSunSun -t5 -b4000
wwwwcl v2.1.9, a GPU program to search for Wieferich and WallSunSun primes
Sieve started: 13345680000000000 <= p < 13345690000000000

Sieve complete: 13345680000000001 <= p < 13345690000000000 269325538 primes tested
Elapsed time: 49.05 sec. (2.39 init + 45.41 sieve) at 5326883 p/sec.
Processor time: 20.19 sec. (2.57 init + 17.61 sieve).
Seconds spent in CPU and GPU: 18446744073708.04 (cpu), 212.99 (gpu)


C:\Users\IntelC\Desktop\p>wwwwcl64.exe -p1334568e10 -P1334569e10 -TWallSunSun -t10 -b10000
wwwwcl v2.1.9, a GPU program to search for Wieferich and WallSunSun primes
Sieve started: 13345680000000000 <= p < 13345690000000000

Sieve complete: 13345680000000001 <= p < 13345690000000000 269325538 primes tested
Elapsed time: 40.56 sec. (4.45 init + 34.91 sieve) at 4857745 p/sec.
Processor time: 22.92 sec. (4.60 init + 18.31 sieve).
Seconds spent in CPU and GPU: 18446744073694.67 (cpu), 407.05 (gpu)
____________
92*10^^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^⁴⁹⁸¹⁰¹-1 CRUS PRIME
2022202116^¹³¹⁰⁷²+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie!

For my 6870 those setting are very slow
Drivers 12.6
2 WU in 2:45
____________
92*10^^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^⁴⁹⁸¹⁰¹-1 CRUS PRIME
2022202116^¹³¹⁰⁷²+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie!

1 CPU thread (HT off) can't keep up with a 570 on WSS. In fact on my new machine (2x570) it takes basically a whole i7 to keep it fed. I have 2 prpclient instances that I use there, and both are set to threads=2, blocksize=8000, and I have HT off. Just one more data point. Your mileage will vary.

I don't think there's much point in setting more threads than you have "cores" available. It is probably counterproductive.

--Gary

I start this thread to show how benchmark doesn't show anything. If I tune up result with benchmark time, then in real life that will be slowest option.
In my test -TWallSunSun -t10 -b10000 give me faster elapsed time: but when I run those settings in real life got slowest result ( 10 WU for 9:23)

Contra-productive IS, that is no doubt in that, but benchmark timer show best result :)
____________
92*10^^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^⁴⁹⁸¹⁰¹-1 CRUS PRIME
2022202116^¹³¹⁰⁷²+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie!