GPUPI Superpi

[flanker]

Zajímavý a docela nový benchmark GPUPI Superpi. Kdy vlastně výpočet LUdolfova čísla na XX miliónů destrinných míst počítá dle volby GPU a nebo CPU.
Benchmark stáhnete zde:
https://www.overclockers.at/downloads/G ... %201.2.zip

Eventuálně i potřebný Visual C++ Redistributable Package:
http://www.microsoft.com/en-us/download ... x?id=40784

Někdo může na procesorech Intel mít problém spustit program přes CPU výpočet, potom může pomoci instalace intel OpenCL driver
https://software.intel.com/en-us/articl ... cl-drivers
eventuálně dll knihovna/knihovny, tak nainstalovat MPSS 3.4 release for Microsoft* Windows
https://software.intel.com/en-us/articl ... hive#34n-1

Uvádějte případně jak výkon GPU, tak i výsledek s CPU. Výpočet bude probíhat na 20M míst (přednastaveno).
--------------------------------------------------------------------------------------------------------------

Zde moje výsledky:
R9-270X Toxic



FX-9590@4715 MHz

[Vlastik80]

Device: NVIDIA GeForce GTX 770 (8 CUs, 1137 MHz) 2GB VRAM bez OC
OpenCL 1.1 CUDA 6.5.29 is ready. (driver 344.65 WHQL - r343_00-295)

Compiling OpenCL kernels ... done.

Calculating 1.000.000.000th digit of PI. 20 iterations.

Allocated device memory : 335546368 Bytes
Batch Size : 20M
Reduction Size : 64

00h 00m 00.298s Batch 1 finished.
00h 00m 01.831s Batch 2 finished.
00h 00m 03.346s Batch 3 finished.
00h 00m 06.022s Batch 4 finished.
00h 00m 12.019s Batch 5 finished.
00h 00m 17.475s Batch 6 finished.
00h 00m 19.008s Batch 7 finished.
00h 00m 20.523s Batch 8 finished.
00h 00m 23.198s Batch 9 finished.
00h 00m 29.191s Batch 10 finished.
00h 00m 34.646s Batch 11 finished.
00h 00m 36.179s Batch 12 finished.
00h 00m 37.694s Batch 13 finished.
00h 00m 40.369s Batch 14 finished.
00h 00m 46.366s Batch 15 finished.
00h 00m 51.824s Batch 16 finished.
00h 00m 53.356s Batch 17 finished.
00h 00m 54.871s Batch 18 finished.
00h 00m 57.546s Batch 19 finished.
00h 01m 03.540s Batch 20 finished.
00h 01m 08.836s PI value output -> 5895585A0

Device time for pi calculation: 67.601 s
Device time for memory reduction: 1.236 s

... s CPU (i5-2500K bez OC) to ani zkoušet nebudu, to je tak na hodinu.

[dexterav]

to by musela byť 2500k asik 10x pomalšia ako FX



i7 5960X 4,4GHz
3x4GB DDR4 2200 cl 15 15 15 37 CR1

Strix
1516/8GHz
344.65

W8.1Pro

[flanker]

wow...5960x na poli CPU tady nic nedá v tomhle multithread superpičku...To už vidím i dle zahraničí.

Ta grafika takdy drsné skóre

[dexterav]

Mna by este zaujímali vysledky gk110 a tej 2500k

[flanker]

2500k odhaduju tak o 2minutky horší jak FX s 8 vlákny. Zajímavý myslím bude i Radeon 290/290x díky výpočetnímu výkonu. To bude zajímavej souboj s 980 (která zas jde excelentně taktovat, viz ta tvá )

[HunterGtNet]

r9-290
Device time for pi calculation: 22.903 s
Device time for memory reduction: 0.963 s

[flanker]

Hunter, můžeš nahodit screen? Wow... Až neuvěřitelný výsledek..

[HunterGtNet]

[Vlastik80]

Device: Intel(R) Core(TM) i5-2500K (4 CUs, 3300 MHz) (CPU bez OC)
OpenCL 1.2 is ready. Windows 8.1 x64, 12GB DDR3 1600 MHz

Compiling OpenCL kernels ... done.
Calculating 1.000.000.000th digit of PI. 20 iterations.

Allocated device memory : 335546368 Bytes
Batch Size : 20M
Reduction Size : 64

00h 00m 04.028s Batch 1 finished.
00h 00m 43.946s Batch 2 finished.
00h 01m 23.616s Batch 3 finished.
00h 02m 14.626s Batch 4 finished.
00h 03m 42.911s Batch 5 finished.
00h 05m 11.876s Batch 6 finished.
00h 05m 51.462s Batch 7 finished.
00h 06m 30.942s Batch 8 finished.
00h 07m 21.896s Batch 9 finished.
00h 08m 47.334s Batch 10 finished.
00h 10m 15.031s Batch 11 finished.
00h 10m 54.666s Batch 12 finished.
00h 11m 34.025s Batch 13 finished.
00h 12m 24.727s Batch 14 finished.
00h 13m 51.948s Batch 15 finished.
00h 15m 20.843s Batch 16 finished.
00h 16m 00.637s Batch 17 finished.
00h 16m 39.753s Batch 18 finished.
00h 17m 30.943s Batch 19 finished.
00h 18m 56.275s Batch 20 finished.
00h 20m 19.545s PI value output -> 5895585A0

Device time for pi calculation: 1210.556 s
Device time for memory reduction: 8.988 s

> pravěk nezklamal, skoro půl hodiny

[dexterav]

tak je to default 3,3GHz
a napr to FX 4,7GHz
daj ho na tých 4,7 a bude to lepšie, tie vlákna tam určite chýbajú
nesklamala ani 290

[Vlastik80]

AllCPU meter mi ukazoval 3600 MHz, takže tipuji že 3300 je základ, při zatížení CPU taktne samo na 3600 (3700)... Na 4700 to nedám, nemám na to chlazení. Měl by to spíš zkusit HEAD, má stejné CPU (snad zatím) a tuším že dal max takt kolem 5200MHz, což by bylo zajímavé sledovat výsledek.

Pokud je toto OT, tak se omlouvám a klidně to smažte...

[flanker]

výhodu budou mít určitě vlákna, takže i7 s HT budou s FX poblíž. Haswell bude myslím i lepší (prý tneprogram má i support FMA, což právě FX či Haswellu pomůže)

[HunterGtNet]

Device: Intel Intel(R) Core(TM) i5-3570K (4 CUs, 3400 MHz)
OpenCL 1.2 AMD-APP (1573.4) is ready.

Compiling OpenCL kernels ... done.

Calculating 1.000.000.000th digit of PI. 20 iterations.

Allocated device memory : 335546368 Bytes
Batch Size : 20M
Reduction Size : 64

00h 00m 03.166s Batch 1 finished.
00h 00m 34.337s Batch 2 finished.
00h 01m 05.590s Batch 3 finished.
00h 01m 46.836s Batch 4 finished.
00h 02m 51.469s Batch 5 finished.
00h 03m 58.687s Batch 6 finished.
00h 04m 29.906s Batch 7 finished.
00h 05m 01.077s Batch 8 finished.
00h 05m 40.147s Batch 9 finished.
00h 06m 40.841s Batch 10 finished.
00h 07m 45.597s Batch 11 finished.
00h 08m 16.722s Batch 12 finished.
00h 08m 47.816s Batch 13 finished.
00h 09m 28.979s Batch 14 finished.
00h 10m 33.594s Batch 15 finished.
00h 11m 40.631s Batch 16 finished.
00h 12m 11.773s Batch 17 finished.
00h 12m 42.946s Batch 18 finished.
00h 13m 22.674s Batch 19 finished.
00h 14m 23.540s Batch 20 finished.
00h 15m 25.208s PI value output -> 5895585A0

Device time for pi calculation: 900.951 s
Device time for memory reduction: 24.257 s

[r34ktor]

HunterGtNet: Tvůj výsledek mi připadá dost slabý. Mě to spočetlo o více než třetinu rychleji. Procesor mám mírně přetaktován - turbo pro všechna jádra 4200 MHz.

Device: Intel Intel(R) Core(TM) i5-3570K (4 CUs, 3392 MHz)
OpenCL 1.2 AMD-APP (1573.4) is ready.

Compiling OpenCL kernels ... done.

Calculating 1.000.000.000th digit of PI. 20 iterations.

Allocated device memory : 335546368 Bytes
Batch Size : 20M
Reduction Size : 64

00h 00m 02.481s Batch 1 finished.
00h 00m 26.713s Batch 2 finished.
00h 00m 50.846s Batch 3 finished.
00h 01m 17.584s Batch 4 finished.
00h 01m 55.612s Batch 5 finished.
00h 02m 37.430s Batch 6 finished.
00h 03m 01.728s Batch 7 finished.
00h 03m 25.862s Batch 8 finished.
00h 03m 52.636s Batch 9 finished.
00h 04m 29.187s Batch 10 finished.
00h 05m 09.468s Batch 11 finished.
00h 05m 33.668s Batch 12 finished.
00h 05m 57.696s Batch 13 finished.
00h 06m 24.374s Batch 14 finished.
00h 07m 02.361s Batch 15 finished.
00h 07m 44.101s Batch 16 finished.
00h 08m 08.325s Batch 17 finished.
00h 08m 32.370s Batch 18 finished.
00h 08m 58.949s Batch 19 finished.
00h 09m 35.445s Batch 20 finished.
00h 10m 13.119s PI value output -> 5895585A0

Device time for pi calculation: 589.433 s
Device time for memory reduction: 23.686 s

[flanker]

tak máš o 700 MHz více než on, to není zanedbatelný rozdíl taktu

[r34ktor]

Je to o 600 MHz (standardní turbo násobič pro plné vytížení je 36). Ale že by o 17% vyšší frekvence přinesla zlepšení 35% se mi nezdá. Tam se musí promítat i jiné vlivy.

[HunterGtNet]

sory jsem mel 32bit verzi spustenou

Device: Intel Intel(R) Core(TM) i5-3570K (4 CUs, 3400 MHz)
OpenCL 1.2 AMD-APP (1573.4) is ready.

Compiling OpenCL kernels ... done.

Calculating 1.000.000.000th digit of PI. 20 iterations.

Allocated device memory : 335546368 Bytes
Batch Size : 20M
Reduction Size : 64

00h 00m 02.181s Batch 1 finished.
00h 00m 23.732s Batch 2 finished.
00h 00m 45.322s Batch 3 finished.
00h 01m 08.993s Batch 4 finished.
00h 01m 42.651s Batch 5 finished.
00h 02m 19.624s Batch 6 finished.
00h 02m 41.158s Batch 7 finished.
00h 03m 02.539s Batch 8 finished.
00h 03m 26.188s Batch 9 finished.
00h 04m 04.362s Batch 10 finished.
00h 04m 39.899s Batch 11 finished.
00h 05m 01.385s Batch 12 finished.
00h 05m 22.859s Batch 13 finished.
00h 05m 46.469s Batch 14 finished.
00h 06m 20.043s Batch 15 finished.
00h 06m 56.899s Batch 16 finished.
00h 07m 18.638s Batch 17 finished.
00h 07m 40.055s Batch 18 finished.
00h 08m 03.536s Batch 19 finished.
00h 08m 35.854s Batch 20 finished.
00h 09m 09.184s PI value output -> 5895585A0

Device time for pi calculation: 524.526 s
Device time for memory reduction: 24.658 s

[blaho]

i7 2600K @4.0Ghz + GTX 780 @1110 /6416
http://i.imgur.com/5s0eiFe.jpg

[dexterav]

Cpu si mal potestovat tiez