ATI Radeon 5770 on Core i5 750 @3.7GHz
 

[CODE]F:\Eigene Dateien\Prime95\ppsieve-cl>tpsieve-cl-x86-windows.exe -i 480000-484999
_09dec2010.txt -p 2601e12 -P 2602e12
tpsieve version cl-0.2.3c (testing)
Found K's from 3 to 9999999.
Found N's from 480000 to 485000.
nstart=480000, nstep=28
Read 16904938 terms from NewPGen format input file `480000-484999_09dec2010.txt'

tpsieve initialized: 3 <= k <= 9999999, 480000 <= n < 485000
Sieve started: 2601000000000000 <= p < 2602000000000000
Resuming from checkpoint p=2601016269443073 in tpcheck2601e12.txt
Thread 0 starting
Detected 160 multiprocessors (800 SPUs) on device 0.
nstep changed to 22
CL setup complete.
cthread_count = 10240
2601017751453061 | 40467*2^481227+1
p=2601017752129537, 24.71M p/sec, 0.68 CPU cores, 1.8% done. ETA 18 Dec 08:04
p=2601019389743105, 25.79M p/sec, 0.71 CPU cores, 1.9% done. ETA 18 Dec 07:47
2601020071189121 | 825711*2^481790-1
p=2601021025259521, 27.26M p/sec, 0.74 CPU cores, 2.1% done. ETA 18 Dec 07:32
2601022079467697 | 9536745*2^483550+1
p=2601022663659521, 27.31M p/sec, 0.75 CPU cores, 2.3% done. ETA 18 Dec 07:24
2601022757818439 | 8955831*2^483464+1
p=2601024314118145, 27.50M p/sec, 0.75 CPU cores, 2.4% done. ETA 18 Dec 07:18
p=2601025963528193, 27.49M p/sec, 0.75 CPU cores, 2.6% done. ETA 18 Dec 07:15[/CODE]

ATI Default Catalyst 10.11

Thanks Brain for the benchmark, i don't think its using all the processors available in the GPU? that number just seems really slow. I get about 223.5M p/sec on my GTX 465 and it is barely using 25% of the GPU power. Maybe it is starting 1 multi thread out of 160.

Any thoughts? can any one else confirm?

I'm getting 9.5M p/sec using one core of an Athlon X2 5000+ (2.6GHz). Running two threads gives me 1.75 CPU usage and a rate of 16.5M p/sec.

i5 M460 laptop, 4 threads

File 485-490k @ 991T
85m p/sec
3.99x2527MHz (CPU-Z says 2660MHz)

edit:

[QUOTE=Oddball;225417]...Switching from the 64 bit version to the SSE2 version boosts performance to an astonishing 134M p/sec :shock:[/QUOTE]

Oh, looks like I should try that next time.

edit: Nope. Just 58m p/sec

i7-2600k @ 4.3GHz, 8 threads

File 480-485k @ 4435T
250.2m p/sec
7.97 CPU
Blocksize 3650k
Chunksize 32k

Is it possible to tweak the code so it performs only a twin sieve? If yes, how?

I'm looking for a way to sieve for triples or quadruples without having NewPGen do the initial sieving. I tried fermfact but it does not remove enough candidates due to sieving only the +1 side so I end up with gigantic files.

Thanks
Peter

I thought the source for newpgen was unavailable but I just found it at [URL]http://jpenne.free.fr/NewPGen/[/URL].