nowy stress test dla GPU

[matszpk]

napisałem program clgpustress do stresowania GPU. Program obciąża GPU i jego pamięć. Potrafi szybko wykryć błędy w pamięci i może być pomocny do testowania pamięci GPU. Ponieważ jest to groźne narzędzie radzę przeczytać plik README.md. Program znajduje się w repozytorium github: https://github.com/matszpk/clgpustress.
Program na razie działa pod Linuksem i był testowany na moim Radeon HD 7850. Na razie może mieć poważne błędy i jest nadal rozwijany.

EDIT: Program nadal jest w intensywnej fazie rozwoju i może mieć błędy. Akurat, zrobiłem błąd, który powodował wyświetlanie FAILED COMPUTATIONS  :deadman: (bez żadnego powodu). Jakiekolwiek znajdowanie błędów będzie przez ze mnie docenione. Przepraszam za błędy i zachęcam do ostrożnego testowania programu.

[matszpk]

Dobra. wrzuciłem binaria mojego programu na mój serwer. są w http://files.nativeboinc.org/offtopic/clgpustress/.
Są tam binaria dla linux32, linux64 no i oczywiście dla win64. Program jak zwykle nie ma GUI (uruchamia się z linii poleceń).
Już poprawiłem kilka błędów i uzupełniłem README.md

[PDH]

Hej, binarka dla WIN z nvidią OpenCL1.1 wywala błąd

Kod [Zaznacz] Rozwiń

The procedure entry point clReleaseDevice could not be located in the dynamic link library

[matszpk]

pewnie spowodowane tym, że NVIDIA ma OpenCL 1.1 zamiast 1.2. spróbuję to jakoś rozwiązać.

[matszpk]

poprawiłem już wszystkie binarki. dziwię dlaczego nvidia nie chciała do tej pory przygotować swojej implementacji OpenCL 1.2  :bad:
PS: są już binarki dla win32.

[PDH]

Okey uruchamia się, lecz podczas wykonywania testu wywala z kolei

Kod [Zaznacz] Rozwiń

C:\Users\mystique\Downloads\clgpustress-0.0.4.2-windows32\clgpustress-0.0.4.2-wi
ndows32>gpustress -G -W128 -S32 -B2 -T0
CLGPUStress 0.0.4.2 by Mateusz Szpakowski. Program is distributed under terms of
the GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD CTRL-C.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=1966080, memory=480 MB, workFactor=128, blocksNum=2,
      computeUnits=15, groupSize=1024, passIters=32, builtinKernel=0,
      inputAndOutput=no
Calibrating Kernel for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black...
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------
Error in processing command line: Don't understand command line argument "-O3"!
OpenCL error happened: clBuildProgram, Code: -11

C:\Users\mystique\Downloads\clgpustress-0.0.4.2-windows32\clgpustress-0.0.4.2-wi
ndows32>gpustress -G
CLGPUStress 0.0.4.2 by Mateusz Szpakowski. Program is distributed under terms of
the GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD CTRL-C.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=3932160, memory=960 MB, workFactor=256, blocksNum=2,
      computeUnits=15, groupSize=1024, passIters=32, builtinKernel=0,
      inputAndOutput=no
Calibrating Kernel for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black...
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------
Error in processing command line: Don't understand command line argument "-O3"!
OpenCL error happened: clBuildProgram, Code: -11

C:\Users\mystique\Downloads\clgpustress-0.0.4.2-windows32\clgpustress-0.0.4.2-wi
ndows32>
C:\Users\mystique\Downloads\clgpustress-0.0.4.2-windows32\clgpustress-0.0.4.2-wi
ndows32>


[matszpk]

Cholerna NVIDIA  :wth:. gdyby ich OpenCL zwracał właściwe kody błędów.. to by było ok. Myslałem że problem jest rozwiązany .... a tu klops.
Dobra. zrobię poprawkę dla NVIDII.

[matszpk]

Poprawione. możecie próbować.

[necavi]

Teraz działa :boing:   Poprzednia clgpustress-0.0.4.2-windows64 nie chciała pracować.
Na sprzęcie 650Ti/Q6600 na gpu miałem 64*C, folding którym ją katuję grzeje kartę na 71*C, więc jeżeli ma ten program obciążać na max to jeszcze mu trochę zostało do zagospodarowania mocy w mojej grafie. Przy okazji, obciążenie procesora 25% ładnie się rozłożyło na wszystkie rdzenie :p_arr:

[matszpk]

ok. możesz poeksperymentować z opcją -S, która ustawia liczbę wywołań kerneli na jeden przebieg (jeśli obciązenie karty jest za niskie zwiększ liczbę); opcją -j, która ustawia ile operacji ma wykonać kernel przed zapisem do pamięci (jeśli bandwidth za niski, zmniejsz liczbę). Program mocno może obciążać radeony hd 7000, ze względu na ich konstrukcję. Wyprobujcie też dwa pozostałe testy (opcja -T1 lub -T2).
PS: Sprawdzcie też osiągi przez ten program (no i porównajcie je z maks. osiągami karty). Program je pokazuje: przepustowość i GFLOPS.

[krzyszp]

Na 2x7770 działa ok, obciążając obie karty naraz.

[matszpk]

Co do kart nVidii, to np.: kernele nie zostały zoptymalizowane, i może to źle działać (np.: kompilator nie rozwija pętli, ale powinien bo domyslny poziom opt to 3) lub też program źle konfiguruje pracę (za duży groupSize), albo po prostu GPU głównie ładuje/zapisuje dane i nic poza tym nie robi. przyczyn może być kilka w tym arch GPU. Program w zamierzeniu był robiony do sprawdzania pamięci GPU pod obciązeniem. Podajcie też osiągi czyli mem bandwidth i ilość GFLOPS.

[krzyszp]

Kod [Zaznacz] Rozwiń

CLGPUStress 0.0.4.3 by Mateusz Szpakowski. Program is distributed under terms of the GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETTER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD CTRL-C.

Preparing StressTester for
  #0 AMD Accelerated Parallel Processing:Capeverde
    SetUp: workSize=655360, memory=160 MB, workFactor=256, blocksNum=2,
      computeUnits=10, groupSize=256, passIters=32, builtinKernel=0,
      inputAndOutput=no
Calibrating Kernel for
  #0 AMD Accelerated Parallel Processing:Capeverde...
Kernel Calibrated for
  #0 AMD Accelerated Parallel Processing:Capeverde
  BestKitersNum: 15, Bandwidth: 47.759 GB/s, Performance: 537.289 GFLOPS
Program build log:
  AMD Accelerated Parallel Processing:Capeverde
:--------------------

Preparing StressTester for
  #1 AMD Accelerated Parallel Processing:Capeverde
    SetUp: workSize=655360, memory=160 MB, workFactor=256, blocksNum=2,
      computeUnits=10, groupSize=256, passIters=32, builtinKernel=0,
      inputAndOutput=no
Calibrating Kernel for
  #1 AMD Accelerated Parallel Processing:Capeverde...
Kernel Calibrated for
  #1 AMD Accelerated Parallel Processing:Capeverde
  BestKitersNum: 15, Bandwidth: 47.6289 GB/s, Performance: 535.825 GFLOPS
Program build log:
  AMD Accelerated Parallel Processing:Capeverde
:--------------------

#1 AMD Accelerated Parallel Processing:Capeverde passed PASS #10
Approx. bandwidth: 37.2806 GB/s, Approx. perf: 419.406 GFLOPS, elapsed: 00:00:01.440
#0 AMD Accelerated Parallel Processing:Capeverde passed PASS #10
Approx. bandwidth: 32.6943 GB/s, Approx. perf: 367.811 GFLOPS, elapsed: 00:00:01.642
#1 AMD Accelerated Parallel Processing:Capeverde passed PASS #20
Approx. bandwidth: 39.0145 GB/s, Approx. perf: 438.914 GFLOPS, elapsed: 00:00:02.816
#0 AMD Accelerated Parallel Processing:Capeverde passed PASS #20

[matszpk]

dzięki krzyszp :). myślę w twoim przypadku program działa całkiem dobrze. mam spore wątpliwości co do działania na kartach nVidii.

[PDH]

OK, gra i buczy na NVIDII:)

Kod [Zaznacz] Rozwiń

#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 119.066 GB/s, Approx. perf: 1339.49 GFLOPS, elapsed: 00:00:27
.550
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 117.175 GB/s, Approx. perf: 1318.22 GFLOPS, elapsed: 00:00:34
.423
^C
C:\Users\mystique\Downloads\clgpustress-0.0.4.3-windows64>gpustress -G -W640 -S3
20 -B2 -j19 -T2
CLGPUStress 0.0.4.3 by Mateusz Szpakowski. Program is distributed under terms of
the GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD CTRL-C.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=9830400, memory=2400 MB, workFactor=640, blocksNum=2,
      computeUnits=15, groupSize=1024, passIters=320, builtinKernel=2,
      inputAndOutput=no
Kernel KitersNum: 19
Trying to compile without optimizing compiler flags...
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 126.396 GB/s, Approx. perf: 2401.53 GFLOPS, elapsed: 00:01:03
.712
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 137.425 GB/s, Approx. perf: 2611.07 GFLOPS, elapsed: 00:02:02
.312
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 138.687 GB/s, Approx. perf: 2635.06 GFLOPS, elapsed: 00:03:00
.378
^C
C:\Users\mystique\Downloads\clgpustress-0.0.4.3-windows64>gpustress -G -W64 -S32
0 -B2 -j29 -T2
CLGPUStress 0.0.4.3 by Mateusz Szpakowski. Program is distributed under terms of
the GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD CTRL-C.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=983040, memory=240 MB, workFactor=64, blocksNum=2,
      computeUnits=15, groupSize=1024, passIters=320, builtinKernel=2,
      inputAndOutput=no
Kernel KitersNum: 29
Trying to compile without optimizing compiler flags...
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 98.982 GB/s, Approx. perf: 2870.48 GFLOPS, elapsed: 00:00:08.
135
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 104.159 GB/s, Approx. perf: 3020.62 GFLOPS, elapsed: 00:00:15
.867
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 103.118 GB/s, Approx. perf: 2990.42 GFLOPS, elapsed: 00:00:23
.676
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 101.62 GB/s, Approx. perf: 2946.98 GFLOPS, elapsed: 00:00:31.
601
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 101.415 GB/s, Approx. perf: 2941.04 GFLOPS, elapsed: 00:00:39
.542
^C
C:\Users\mystique\Downloads\clgpustress-0.0.4.3-windows64>gpustress -G -W64 -S32
0 -B2 -j7 -T2
CLGPUStress 0.0.4.3 by Mateusz Szpakowski. Program is distributed under terms of
the GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD CTRL-C.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=983040, memory=240 MB, workFactor=64, blocksNum=2,
      computeUnits=15, groupSize=1024, passIters=320, builtinKernel=2,
      inputAndOutput=no
Kernel KitersNum: 7
Trying to compile without optimizing compiler flags...
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 168.876 GB/s, Approx. perf: 1182.13 GFLOPS, elapsed: 00:00:04
.768
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 175.125 GB/s, Approx. perf: 1225.88 GFLOPS, elapsed: 00:00:09
.367
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 176.354 GB/s, Approx. perf: 1234.48 GFLOPS, elapsed: 00:00:13
.933
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 175.507 GB/s, Approx. perf: 1228.55 GFLOPS, elapsed: 00:00:18
.521
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 172.571 GB/s, Approx. perf: 1208 GFLOPS, elapsed: 00:00:23.18
8
^C
C:\Users\mystique\Downloads\clgpustress-0.0.4.3-windows64>

[matszpk]

Dzięki  :p_arr:. eksperymentuje z groupSize. okazuje się za ustawienie maks ilości workitemów w grupie nie jest najlepsze. wystarczy taka ilość żeby zapchać CU. jutro opublikuje nową wersję z nowym testem :).

[matszpk]

Już jest wersja 0.0.5 z nowym testem i nową opcją '-g' (--groupSize) do ustawiania workGroupSize.

[sknd]

Kod [Zaznacz] Rozwiń


[ja@komp clgpustress-0.0.5-linux64]$ ./gpustress -N
./gpustress: /usr/lib/libOpenCL.so.1: no version information available (required by ./gpustress)
CLGPUStress 0.0.5 by Mateusz Szpakowski. Program is distributed under terms of the GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETTER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD 'CTRL-C' KEY COMBINATION.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX 650
    SetUp: workSize=524288, memory=128 MB, workFactor=256, blocksNum=2,
    computeUnits=2, groupSize=1024, passIters=32, testType=0,
    inputAndOutput=no
Calibrating Kernel for
  #0 NVIDIA CUDA:GeForce GTX 650...
Trying to compile without optimizing compiler flags...
Kernel Calibrated for
  #0 NVIDIA CUDA:GeForce GTX 650
  BestKitersNum: 11, Bandwidth: 13.9759 GB/s, Performance: 115.301 GFLOPS
Program build log:
  NVIDIA CUDA:GeForce GTX 650
:--------------------


#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #10
Approx. bandwidth: 13.7472 GB/s, Approx. perf: 113.414 GFLOPS, elapsed: 00:00:03.124
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #20
Approx. bandwidth: 13.8864 GB/s, Approx. perf: 114.563 GFLOPS, elapsed: 00:00:06.217
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #30
Approx. bandwidth: 13.8958 GB/s, Approx. perf: 114.641 GFLOPS, elapsed: 00:00:09.308
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #40
Approx. bandwidth: 13.8653 GB/s, Approx. perf: 114.389 GFLOPS, elapsed: 00:00:12.405
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #50
Approx. bandwidth: 13.8381 GB/s, Approx. perf: 114.164 GFLOPS, elapsed: 00:00:15.509
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #60
Approx. bandwidth: 13.8785 GB/s, Approx. perf: 114.498 GFLOPS, elapsed: 00:00:18.604
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #70
Approx. bandwidth: 13.8848 GB/s, Approx. perf: 114.55 GFLOPS, elapsed: 00:00:21.697
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #80
Approx. bandwidth: 13.8896 GB/s, Approx. perf: 114.589 GFLOPS, elapsed: 00:00:24.789
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #90
Approx. bandwidth: 13.8898 GB/s, Approx. perf: 114.591 GFLOPS, elapsed: 00:00:27.881
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #100
Approx. bandwidth: 13.8905 GB/s, Approx. perf: 114.596 GFLOPS, elapsed: 00:00:30.973
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #110
Approx. bandwidth: 13.8874 GB/s, Approx. perf: 114.571 GFLOPS, elapsed: 00:00:34.066
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #120
Approx. bandwidth: 13.86 GB/s, Approx. perf: 114.345 GFLOPS, elapsed: 00:00:37.165
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #130
Approx. bandwidth: 13.8851 GB/s, Approx. perf: 114.552 GFLOPS, elapsed: 00:00:40.258
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #140
Approx. bandwidth: 13.8804 GB/s, Approx. perf: 114.513 GFLOPS, elapsed: 00:00:43.352
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #150
Approx. bandwidth: 13.8637 GB/s, Approx. perf: 114.375 GFLOPS, elapsed: 00:00:46.450
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #160
Approx. bandwidth: 13.882 GB/s, Approx. perf: 114.526 GFLOPS, elapsed: 00:00:49.544
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #170
Approx. bandwidth: 13.8821 GB/s, Approx. perf: 114.527 GFLOPS, elapsed: 00:00:52.638
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #180
Approx. bandwidth: 13.888 GB/s, Approx. perf: 114.576 GFLOPS, elapsed: 00:00:55.731
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #190
Approx. bandwidth: 13.8032 GB/s, Approx. perf: 113.876 GFLOPS, elapsed: 00:00:58.842
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #200
Approx. bandwidth: 13.7979 GB/s, Approx. perf: 113.832 GFLOPS, elapsed: 00:01:01.955
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #210
Approx. bandwidth: 13.7708 GB/s, Approx. perf: 113.609 GFLOPS, elapsed: 00:01:05.074
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #220
Approx. bandwidth: 13.7805 GB/s, Approx. perf: 113.689 GFLOPS, elapsed: 00:01:08.191
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #230
Approx. bandwidth: 13.7687 GB/s, Approx. perf: 113.592 GFLOPS, elapsed: 00:01:11.310
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #240
Approx. bandwidth: 13.7497 GB/s, Approx. perf: 113.435 GFLOPS, elapsed: 00:01:14.434
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #250
Approx. bandwidth: 13.7637 GB/s, Approx. perf: 113.55 GFLOPS, elapsed: 00:01:17.554
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #260
Approx. bandwidth: 13.767 GB/s, Approx. perf: 113.577 GFLOPS, elapsed: 00:01:20.674
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #270
Approx. bandwidth: 13.816 GB/s, Approx. perf: 113.982 GFLOPS, elapsed: 00:01:23.783
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #280
Approx. bandwidth: 13.8757 GB/s, Approx. perf: 114.474 GFLOPS, elapsed: 00:01:26.878
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #290
Approx. bandwidth: 13.8439 GB/s, Approx. perf: 114.212 GFLOPS, elapsed: 00:01:29.980
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #300
Approx. bandwidth: 13.8819 GB/s, Approx. perf: 114.526 GFLOPS, elapsed: 00:01:33.074
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #310
Approx. bandwidth: 13.8319 GB/s, Approx. perf: 114.113 GFLOPS, elapsed: 00:01:36.179
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #320
Approx. bandwidth: 13.8609 GB/s, Approx. perf: 114.352 GFLOPS, elapsed: 00:01:39.278
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #330
Approx. bandwidth: 13.8837 GB/s, Approx. perf: 114.54 GFLOPS, elapsed: 00:01:42.372
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #340
Approx. bandwidth: 13.8686 GB/s, Approx. perf: 114.416 GFLOPS, elapsed: 00:01:45.469
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #350
Approx. bandwidth: 13.8827 GB/s, Approx. perf: 114.532 GFLOPS, elapsed: 00:01:48.562
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #360
Approx. bandwidth: 13.8362 GB/s, Approx. perf: 114.149 GFLOPS, elapsed: 00:01:51.666
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #370
Approx. bandwidth: 13.8192 GB/s, Approx. perf: 114.008 GFLOPS, elapsed: 00:01:54.774
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #380
Approx. bandwidth: 13.7744 GB/s, Approx. perf: 113.639 GFLOPS, elapsed: 00:01:57.892
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #390
Approx. bandwidth: 13.7736 GB/s, Approx. perf: 113.632 GFLOPS, elapsed: 00:02:01.011
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #400
Approx. bandwidth: 13.8688 GB/s, Approx. perf: 114.418 GFLOPS, elapsed: 00:02:04.108
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #410
Approx. bandwidth: 13.8836 GB/s, Approx. perf: 114.54 GFLOPS, elapsed: 00:02:07.201
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #420
Approx. bandwidth: 13.8664 GB/s, Approx. perf: 114.397 GFLOPS, elapsed: 00:02:10.299
#0 NVIDIA CUDA:GeForce GTX 650 passed PASS #430
Approx. bandwidth: 13.8852 GB/s, Approx. perf: 114.552 GFLOPS, elapsed: 00:02:13.392
^C


po dwóch minutach się skaplem ze chyba samo się nie skończy i wyłaczyłem, rozgrzało kartę do 48 stopni. aha - nie wiem czy to ważne, ale procek był zajęty boincem cały czas  ;)

[PDH]

Testy pomyślne, dla s>256 karta u mnie obciążona na full

Kod [Zaznacz] Rozwiń

.777
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 156.523 GB/s, Approx. perf: 1408.71 GFLOPS, elapsed: 00:00:01
.035
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 155.917 GB/s, Approx. perf: 1403.25 GFLOPS, elapsed: 00:00:01
.293
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #60
Approx. bandwidth: 155.313 GB/s, Approx. perf: 1397.81 GFLOPS, elapsed: 00:00:01
.552
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #70
Approx. bandwidth: 155.916 GB/s, Approx. perf: 1403.24 GFLOPS, elapsed: 00:00:01
.810
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #80
Approx. bandwidth: 155.314 GB/s, Approx. perf: 1397.82 GFLOPS, elapsed: 00:00:02
.070
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #90
Approx. bandwidth: 155.916 GB/s, Approx. perf: 1403.24 GFLOPS, elapsed: 00:00:02
.328
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #100
Approx. bandwidth: 155.314 GB/s, Approx. perf: 1397.82 GFLOPS, elapsed: 00:00:02
.587
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #110
Approx. bandwidth: 154.122 GB/s, Approx. perf: 1387.1 GFLOPS, elapsed: 00:00:02.
848
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #120
Approx. bandwidth: 156.523 GB/s, Approx. perf: 1408.7 GFLOPS, elapsed: 00:00:03.
106
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #130
Approx. bandwidth: 155.917 GB/s, Approx. perf: 1403.25 GFLOPS, elapsed: 00:00:03
.364
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #140
Approx. bandwidth: 155.313 GB/s, Approx. perf: 1397.82 GFLOPS, elapsed: 00:00:03
.623
^C
C:\Users\mystique\Downloads\clgpustress-0.0.5-windows64>gpustress -G -W64 -S256
-B2 -T2 -j3 -g512
CLGPUStress 0.0.5 by Mateusz Szpakowski. Program is distributed under terms of t
he GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD 'CTRL-C' KEY COMBINATION.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=491520, memory=120 MB, workFactor=64, blocksNum=2,
    computeUnits=15, groupSize=512, passIters=256, testType=2,
    inputAndOutput=no
Kernel KitersNum: 3
Trying to compile without optimizing compiler flags...
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 172.831 GB/s, Approx. perf: 518.492 GFLOPS, elapsed: 00:00:01
.863
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 182.329 GB/s, Approx. perf: 546.986 GFLOPS, elapsed: 00:00:03
.630
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 182.329 GB/s, Approx. perf: 546.986 GFLOPS, elapsed: 00:00:05
.397
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 182.329 GB/s, Approx. perf: 546.986 GFLOPS, elapsed: 00:00:07
.163
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 182.432 GB/s, Approx. perf: 547.296 GFLOPS, elapsed: 00:00:08
.929
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #60
Approx. bandwidth: 182.019 GB/s, Approx. perf: 546.058 GFLOPS, elapsed: 00:00:10
.699
^C
C:\Users\mystique\Downloads\clgpustress-0.0.5-windows64>gpustress -G -W64 -S256
-B2 -T3 -j3 -g512
CLGPUStress 0.0.5 by Mateusz Szpakowski. Program is distributed under terms of t
he GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD 'CTRL-C' KEY COMBINATION.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=491520, memory=120 MB, workFactor=64, blocksNum=2,
    computeUnits=15, groupSize=512, passIters=256, testType=3,
    inputAndOutput=no
Kernel KitersNum: 3
Trying to compile without optimizing compiler flags...
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 154.865 GB/s, Approx. perf: 464.596 GFLOPS, elapsed: 00:00:02
.080
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 161.551 GB/s, Approx. perf: 484.653 GFLOPS, elapsed: 00:00:04
.073
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 165.116 GB/s, Approx. perf: 495.347 GFLOPS, elapsed: 00:00:06
.024
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 163.107 GB/s, Approx. perf: 489.321 GFLOPS, elapsed: 00:00:07
.999
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 163.107 GB/s, Approx. perf: 489.321 GFLOPS, elapsed: 00:00:09
.974
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #60
Approx. bandwidth: 163.107 GB/s, Approx. perf: 489.321 GFLOPS, elapsed: 00:00:11
.949
^C
C:\Users\mystique\Downloads\clgpustress-0.0.5-windows64>gpustress -G -W128 -S64
-B2 -T0 -g256
CLGPUStress 0.0.5 by Mateusz Szpakowski. Program is distributed under terms of t
he GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD 'CTRL-C' KEY COMBINATION.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=491520, memory=120 MB, workFactor=128, blocksNum=2,
    computeUnits=15, groupSize=256, passIters=64, testType=0,
    inputAndOutput=no
Calibrating Kernel for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black...
Trying to compile without optimizing compiler flags...
Kernel Calibrated for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
  BestKitersNum: 11, Bandwidth: 120.93 GB/s, Performance: 997.673 GFLOPS
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 87.1641 GB/s, Approx. perf: 719.104 GFLOPS, elapsed: 00:00:00
.923
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 88.5067 GB/s, Approx. perf: 730.18 GFLOPS, elapsed: 00:00:01.
833
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 100.19 GB/s, Approx. perf: 826.568 GFLOPS, elapsed: 00:00:02.
637
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 96.3503 GB/s, Approx. perf: 794.89 GFLOPS, elapsed: 00:00:03.
473
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 100.19 GB/s, Approx. perf: 826.568 GFLOPS, elapsed: 00:00:04.
277
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #60
Approx. bandwidth: 92.2621 GB/s, Approx. perf: 761.162 GFLOPS, elapsed: 00:00:05
.149
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #70
Approx. bandwidth: 91.4233 GB/s, Approx. perf: 754.242 GFLOPS, elapsed: 00:00:06
.030
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #80
Approx. bandwidth: 96.3503 GB/s, Approx. perf: 794.89 GFLOPS, elapsed: 00:00:06.
866
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #90
Approx. bandwidth: 100.44 GB/s, Approx. perf: 828.631 GFLOPS, elapsed: 00:00:07.
668
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #100
Approx. bandwidth: 89.7909 GB/s, Approx. perf: 740.775 GFLOPS, elapsed: 00:00:08
.565
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #110
Approx. bandwidth: 95.777 GB/s, Approx. perf: 790.16 GFLOPS, elapsed: 00:00:09.4
06
^C
C:\Users\mystique\Downloads\clgpustress-0.0.5-windows64>gpustress -G -W128 -S64
-B2 -T1 -g128
CLGPUStress 0.0.5 by Mateusz Szpakowski. Program is distributed under terms of t
he GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD 'CTRL-C' KEY COMBINATION.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=245760, memory=60 MB, workFactor=128, blocksNum=2,
    computeUnits=15, groupSize=128, passIters=64, testType=1,
    inputAndOutput=no
Calibrating Kernel for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black...
Trying to compile without optimizing compiler flags...
Kernel Calibrated for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
  BestKitersNum: 13, Bandwidth: 172.095 GB/s, Performance: 1677.93 GFLOPS
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 98.5984 GB/s, Approx. perf: 961.335 GFLOPS, elapsed: 00:00:00
.408
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 105.029 GB/s, Approx. perf: 1024.03 GFLOPS, elapsed: 00:00:00
.791
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 109.014 GB/s, Approx. perf: 1062.89 GFLOPS, elapsed: 00:00:01
.161
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 104.756 GB/s, Approx. perf: 1021.37 GFLOPS, elapsed: 00:00:01
.545
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 110.816 GB/s, Approx. perf: 1080.46 GFLOPS, elapsed: 00:00:01
.908
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #60
Approx. bandwidth: 106.138 GB/s, Approx. perf: 1034.85 GFLOPS, elapsed: 00:00:02
.288
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #70
Approx. bandwidth: 106.984 GB/s, Approx. perf: 1043.09 GFLOPS, elapsed: 00:00:02
.664
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #80
Approx. bandwidth: 103.145 GB/s, Approx. perf: 1005.66 GFLOPS, elapsed: 00:00:03
.054
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #90
Approx. bandwidth: 104.213 GB/s, Approx. perf: 1016.08 GFLOPS, elapsed: 00:00:03
.441
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #100
Approx. bandwidth: 101.839 GB/s, Approx. perf: 992.93 GFLOPS, elapsed: 00:00:03.
836
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #110
Approx. bandwidth: 104.484 GB/s, Approx. perf: 1018.72 GFLOPS, elapsed: 00:00:04
.222
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #120
Approx. bandwidth: 105.305 GB/s, Approx. perf: 1026.73 GFLOPS, elapsed: 00:00:04
.604
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #130
Approx. bandwidth: 101.325 GB/s, Approx. perf: 987.92 GFLOPS, elapsed: 00:00:05.
001
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #140
Approx. bandwidth: 105.859 GB/s, Approx. perf: 1032.12 GFLOPS, elapsed: 00:00:05
.382
^C
C:\Users\mystique\Downloads\clgpustress-0.0.5-windows64>gpustress -G -W128 -S64
-B2 -T3 -g384
CLGPUStress 0.0.5 by Mateusz Szpakowski. Program is distributed under terms of t
he GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD 'CTRL-C' KEY COMBINATION.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=737280, memory=180 MB, workFactor=128, blocksNum=2,
    computeUnits=15, groupSize=384, passIters=64, testType=3,
    inputAndOutput=no
Calibrating Kernel for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black...
Trying to compile without optimizing compiler flags...
Kernel Calibrated for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
  BestKitersNum: 15, Bandwidth: 115.788 GB/s, Performance: 1736.82 GFLOPS
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 85.6485 GB/s, Approx. perf: 1284.73 GFLOPS, elapsed: 00:00:01
.410
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 93.4771 GB/s, Approx. perf: 1402.16 GFLOPS, elapsed: 00:00:02
.702
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 95.0976 GB/s, Approx. perf: 1426.46 GFLOPS, elapsed: 00:00:03
.972
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 91.9108 GB/s, Approx. perf: 1378.66 GFLOPS, elapsed: 00:00:05
.287
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 95.8528 GB/s, Approx. perf: 1437.79 GFLOPS, elapsed: 00:00:06
.547
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #60
Approx. bandwidth: 94.8733 GB/s, Approx. perf: 1423.1 GFLOPS, elapsed: 00:00:07.
820
^C
C:\Users\mystique\Downloads\clgpustress-0.0.5-windows64>gpustress -G -W512 -S128
-B2 -T3 -g512
CLGPUStress 0.0.5 by Mateusz Szpakowski. Program is distributed under terms of t
he GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD 'CTRL-C' KEY COMBINATION.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=3932160, memory=960 MB, workFactor=512, blocksNum=2,
    computeUnits=15, groupSize=512, passIters=128, testType=3,
    inputAndOutput=no
Calibrating Kernel for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black...
Trying to compile without optimizing compiler flags...
Kernel Calibrated for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
  BestKitersNum: 17, Bandwidth: 104.749 GB/s, Performance: 1780.73 GFLOPS
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 83.5599 GB/s, Approx. perf: 1420.52 GFLOPS, elapsed: 00:00:15
.419
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 89.8974 GB/s, Approx. perf: 1528.26 GFLOPS, elapsed: 00:00:29
.752
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 88.4945 GB/s, Approx. perf: 1504.41 GFLOPS, elapsed: 00:00:44
.312
^C
C:\Users\mystique\Downloads\clgpustress-0.0.5-windows64>gpustress -G -W128 -S128
-B2 -T0 -g512
CLGPUStress 0.0.5 by Mateusz Szpakowski. Program is distributed under terms of t
he GPLv2.

WARNING: THIS PROGRAM CAN OVERHEAT OR DAMAGE YOUR GRAPHICS CARD FASTER (AND BETT
ER)
THAN ANY FURMARK STRESS. PLEASE USE THIS PROGRAM VERY CAREFULLY!!!
RECOMMEND TO RUN THIS PROGRAM ON STOCK PARAMETERS (CLOCKS, VOLTAGES,
ESPECIALLY MEMORY CLOCK).
TO TERMINATE THIS PROGRAM PLEASE USE STANDARD 'CTRL-C' KEY COMBINATION.

Preparing StressTester for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
    SetUp: workSize=983040, memory=240 MB, workFactor=128, blocksNum=2,
    computeUnits=15, groupSize=512, passIters=128, testType=0,
    inputAndOutput=no
Calibrating Kernel for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black...
Trying to compile without optimizing compiler flags...
Kernel Calibrated for
  #0 NVIDIA CUDA:GeForce GTX TITAN Black
  BestKitersNum: 12, Bandwidth: 100.011 GB/s, Performance: 900.096 GFLOPS
Program build log:
  NVIDIA CUDA:GeForce GTX TITAN Black
:--------------------


#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #10
Approx. bandwidth: 87.3534 GB/s, Approx. perf: 786.18 GFLOPS, elapsed: 00:00:03.
687
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #20
Approx. bandwidth: 99.1713 GB/s, Approx. perf: 892.541 GFLOPS, elapsed: 00:00:06
.935
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #30
Approx. bandwidth: 97.077 GB/s, Approx. perf: 873.693 GFLOPS, elapsed: 00:00:10.
253
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #40
Approx. bandwidth: 91.7361 GB/s, Approx. perf: 825.625 GFLOPS, elapsed: 00:00:13
.765
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #50
Approx. bandwidth: 87.7345 GB/s, Approx. perf: 789.611 GFLOPS, elapsed: 00:00:17
.436
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #60
Approx. bandwidth: 90.0168 GB/s, Approx. perf: 810.152 GFLOPS, elapsed: 00:00:21
.015
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #70
Approx. bandwidth: 98.8664 GB/s, Approx. perf: 889.798 GFLOPS, elapsed: 00:00:24
.273
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #80
Approx. bandwidth: 83.1551 GB/s, Approx. perf: 748.396 GFLOPS, elapsed: 00:00:28
.147
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #90
Approx. bandwidth: 95.9196 GB/s, Approx. perf: 863.276 GFLOPS, elapsed: 00:00:31
.505
#0 NVIDIA CUDA:GeForce GTX TITAN Black passed PASS #100
Approx. bandwidth: 82.9836 GB/s, Approx. perf: 746.852 GFLOPS, elapsed: 00:00:35
.387
^C
C:\Users\mystique\Downloads\clgpustress-0.0.5-windows64>



[matszpk]

opublikowałem nową wersję 0.0.5.1 (tylko poprawki). jeśli chodzi o optymalizację kerneli dla NVIDII (nie mam sprzętu :dunno:) to jest to takie proste (np.: unroll loop może dać katastrofalne skutki na moim radeonie i innych też).
Dalsze plany: dorobienie GUI, nowe testy.

[matszpk]

Ktoś chciałby jeszcze przetestować program? testy na starszych urządzeniach (np.: radeonach hd 6000,5000) by się przydały. mógłbym coś poprawić

[matszpk]

dobra. już upublikowałem wersję z GUI (wersja 0.0.8 ). poprawiłem też kilka głupich błędów. jeśli to możliwe przetestujcie wersję z GUI.

[krzyszp]

Program w wersji GUI pokazuje krzaki w miejscu nazwy karty dla ATI (Devices list, na screenshocie widać), w Test Configs w dropdown'ie pokazuje tylko jedną kartę (ja mam dwie).
To jest w wxWidgets?

Kod [Zaznacz] Rozwiń


https://www.dropbox.com/s/ivvewll0wd6ntti/gpustress.png?dl=0

[matszpk]

Dzięki za zgłoszenie błędów. za chwilę opublikuje poprawioną wersję. zobaczymy czy będzie lepsza ;)

[matszpk]

juz jest wersja do testów pod windows dla ciebie:

Kod [Zaznacz] Rozwiń

http://files.nativeboinc.org/offtopic/clgpustress/testing/gpustress-gui.exe

EDIT: Upublikowałem nową wersję: 0.0.8c. sprawdźcie teraz.

[krzyszp]

Test configs (druga zakładka) pozwala już wybrać jedną z dwóch kart, niemniej w pierwszej zakładce, w oknie devices lists nic się nie zmieniło.

[matszpk]

był głupi błąd w obsłudzie stringów. aplikacja jest:

Kod [Zaznacz] Rozwiń

http://files.nativeboinc.org/offtopic/clgpustress/testing/gpustress-gui-2.exe

aha. nie powiedziałem. Apka jest zrobiona w FLTK 1.3. Testuje ją na WIndows 10  ;D

[krzyszp]

Już jest ok.

Jest możliwość, że w Devices List oprócz nazwy rdzenia będą dodatkowe informacje (typ karty, nazwa producenta, itd)? Obecnie, u mnie, jest tylko "Capevrde"
Wiem, że to wymaga zapewne zbudowania bazy, ale znacznie by polepszyło "look'n'feel" programu. Oczywiście doradzał bym też przejście na GTK lub QT... Ale to pierdoła. ;)

[matszpk]

gdy najedziesz na etykietę urządzenia to pokaże ci się tooltip z konfigiem.  myslałem, żeby dodać do listy informacje, ale wybrałem takie rozwiązanie.
FLTK jest lekkie. tylko, że nie wpiera stringów c++ (stąd te błędy). apka jest jeszcze wcześnej wersji rozwojowej. jeszcze będę ją testować.
póki co publikuje ulepszoną wersję.
Qt lub Gtk są cięzkie (kto by chciał binarkę ważącą kilkanaście megabajtów?).

[krzyszp]

Faktycznie, tooltip'a nie zauważyłem wcześniej.
Co do QT i rozmiarów binarki - zależy to tylko od tego, jak bardzo chcesz rozbudowywać apkę, w formie rozbudowanej te kilkanaście MB przestaną robić wrażenie. Jeszcze inaczej sytuacja wygląda w przypadku Linuksa, gdzie biblioteki są ściągane tylko w razie potrzeby. Ale to off-top, program działa dobrze i to jest najważniejsze :)

[matszpk]

nowa wersja 0.0.8.5 już dostępna. poprawiłem kilka poważnych błędów i dodałem kilka usprawnien. strona programu:

Kod [Zaznacz] Rozwiń

http://clgpustress.nativeboinc.org

[PDH]

Nie żebym się czepiał, ale binarki nie działają u mnie pod win xp a kompilowania źródła jakoś nie widzę %). System co prawda przeszedł już do historii, ale nadal ma rzeszę zwolenników. Wyskakuje komunikat o niepoprawnej aplikacji win32. Brakuje jakiejś biblioteki czy jak?

[krzyszp]

Nie żebym się czepiał, ale binarki nie działają u mnie pod win xp a kompilowania źródła jakoś nie widzę %). System co prawda przeszedł już do historii, ale nadal ma rzeszę zwolenników. Wyskakuje komunikat o niepoprawnej aplikacji win32. Brakuje jakiejś biblioteki czy jak?

Aż sprawdziłem nową kompilację u siebie. Nie mam XP, więc na 7-ce 64 bitowej działa dobrze, więc faktycznie chyba sprawa bibliotek.

[krzyszp]

A źródła są tutaj:

Kod [Zaznacz] Rozwiń

https://github.com/matszpk/clgpustress

[PDH]

Tak jak mówisz, na "ósemce" hula bo sprawdzałem tylko pod XP coś szwankuje.

[matszpk]

Do kompilacji używam VS2012. Tam piszą, że aplikacje kompilowane mogą działać pod Win7 lub nowszych. stąd problemy.
EDIT: Poza tym program jest pisany w C++11 i wymaga częściowej zgodności kompilatora z nim.

[matszpk]

opublikowałem wersję 0.0.8.7. (wersja 0.0.8.6 była głupim eksperymentem  XD ). Przetestujcie ją swoich maszynkach.

[matszpk]

Nowa wersja 0.0.8.8. Update zalecany poprawiłem poważny wyciek pamięci.

[matszpk]

jeśli ktoś jest chętny to zapraszam na testowanie wersji 0.0.9. Poza tym jest nowa strona:

Kod [Zaznacz] Rozwiń

http://clgpustress.nativeboinc.org
Wersje starszych niż 0.0.8.8 radzę porzucić (mają straszny błąd).
Program jest nadal w wersji alpha i może zawierać głupie błędy.

[Szopler]

Ten test ma jakiś limit czasu, czy działa aż do spalenia GPU ;) ?

PS.
Udało mi się 7770 nagrzać do ~71°C.