mfaktc: a CUDA program for Mersenne prefactoring

[James Heinrich]

Originally Posted by ET_

What is the Street Price of this monster?

Seems to be hovering around US$1800 right now, which is getting close to double MSRP, due to high demand and low supply. Things should calm down in a little while when there's more supply.

[chalsall]

Originally Posted by TheJudger

New performance king but slightly behind Tesla V100 in terms of energy efficency.

Like, um, wow!!!

Despite the capex, the investment might make sense over the life of the kit based on the TDP.

[petrw1]

Originally Posted by James Heinrich

Seems to be hovering around US$1800 right now, which is getting close to double MSRP, due to high demand and low supply. Things should calm down in a little while when there's more supply.

I pre-ordered my 2080Ti right after the announcement for $1200US.

[ET_]

Originally Posted by petrw1

I pre-ordered my 2080Ti right after the announcement for $1200US.

Still expensive, having a GTX 980 idle at home...

[ATH]

Originally Posted by TheJudger

Hello!

1. Installed Visual Studio 2017.8 "Community"
2. Installed CUDA Toolkit 10 for Windows
3. Installed MinGW as on of many options for GNU Make on Windows. In MinGW folder I've copied bin/mingw32-make.exe to bin/make.exe because I'm lazy. Careful when updating mingw32-make.exe...
4. Configure Environment for "x64 Native Tools-Command Promt" - add MinGW/bin and CUDA/bin to PATH variable.

The just open "x64 Native Tools-Command Promt" and change into the directory with the mfaktc source files and run

make -f Makefile.win

I had to adjust some settings in Makefile.win:

CUDA_DIR = "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0"

CC = cl
CFLAGS = /Ox /Oy /GL /W2 /fp:fast /I$(CUDA_DIR)\include /I$(CUDA_DIR)\include\cudart /nologo

NVCCFLAGS = --ptxas-options=-v
CUFLAGS = -ccbin "C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Tools\MSVC\14.15.26726\bin\Hostx86\x64" -x cu -I$(CUDA_DIR)\/include --machine 64 --compile -Xcompiler "/wd 4819" -DWIN64 -Xcompiler "/EHsc /W3 /nologo /O2 /FS" $(NVCCFLAGS)

# generate code for various compute capabilities
# NVCCFLAGS += --generate-code arch=compute_11,code=sm_11 # CC 1.1, 1.2 and 1.3 GPUs will use this code (1.0 is not possible for mfaktc)
# NVCCFLAGS += --generate-code arch=compute_20,code=sm_20 # CC 2.x GPUs will use this code, one code fits all!
NVCCFLAGS += --generate-code arch=compute_30,code=sm_30 # all CC 3.x GPUs _COULD_ use this code 
NVCCFLAGS += --generate-code arch=compute_35,code=sm_35 # but CC 3.5 (3.2?) _CAN_ use funnel shift which is useful for mfaktc
NVCCFLAGS += --generate-code arch=compute_50,code=sm_50 # CC 5.x GPUs will use this code
NVCCFLAGS += --generate-code arch=compute_60,code=sm_60 # CC 6.x GPUs will use this code
NVCCFLAGS += --generate-code arch=compute_70,code=sm_70 # CC 7.x GPUs will use this code
# NVCCFLAGS += --generate-code arch=compute_75,code=sm_75 # CC 7.5 GPUs will use this code

Oliver

I compiled mfaktc this way without any errors but when running it, it does not work:
ERROR: cudaGetLastError() returned 8: invalid device function


Is CUDA10 too advanced for compute capability 3.5 (Titan Black) ? I did add 3.5 in Makefile and as I said no error:
NVCCFLAGS += --generate-code arch=compute_35,code=sm_35 # but CC 3.5 (3.2?) _CAN_ use funnel shift which is useful for mfaktc

There were some errors:

c:\msys64\home\ath\mfaktc-0.21\src\tf_common.cu(242): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'
c:\msys64\home\ath\mfaktc-0.21\src\tf_common.cu(242): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'
c:\msys64\home\ath\mfaktc-0.21\src\tf_common_gs.cu(169): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'
c:\msys64\home\ath\mfaktc-0.21\src\tf_common.cu(242): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'
c:\msys64\home\ath\mfaktc-0.21\src\tf_common_gs.cu(169): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'
gpusieve.cu(1371): warning C4244: '=': conversion from '__int64' to 'uint32', possible loss of data
gpusieve.cu(1385): warning C4244: '=': conversion from '__int64' to 'uint32', possible loss of data
gpusieve.cu(1400): warning C4244: '=': conversion from '__int64' to 'uint32', possible loss of data
gpusieve.cu(1416): warning C4244: '=': conversion from '__int64' to 'uint32', possible loss of data
gpusieve.cu(1450): warning C4244: '=': conversion from '__int64' to 'uint32', possible loss of data
gpusieve.cu(1466): warning C4244: '=': conversion from '__int64' to 'uint32', possible loss of data
gpusieve.cu(1506): warning C4244: '=': conversion from '__int64' to 'uint32', possible loss of data
gpusieve.cu(1522): warning C4244: '=': conversion from '__int64' to 'uint32', possible loss of data
gpusieve.cu(1558): warning C4244: '=': conversion from '__int64' to 'uint32', possible loss of data
gpusieve.cu(1273): warning C4996: 'cudaThreadSetCacheConfig': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(1112): note: see declaration of 'cudaThreadSetCacheConfig'
gpusieve.cu(1599): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'
gpusieve.cu(1621): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'
gpusieve.cu(1645): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'
c:\msys64\home\ath\mfaktc-0.21\src\tf_common.cu(242): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'
c:\msys64\home\ath\mfaktc-0.21\src\tf_common_gs.cu(169): warning C4996: 'cudaThreadSynchronize': was declared deprecated
c:\cuda10\include\cuda_runtime_api.h(947): note: see declaration of 'cudaThreadSynchronize'

[kriesel]

Originally Posted by ATH

I compiled mfaktc this way without any errors but when running it, it does not work:
ERROR: cudaGetLastError() returned 8: invalid device function
...
Is CUDA10 too advanced for compute capability 3.5 (Titan Black) ?

OK, you compiled for CUDA10. Have you confirmed the installed NVIDIA driver supports CUDA10? Confirmed the CUDArt....dll is CUDA10 also? (Via file names, or separate tools?) My recollection is invalid device function error 8 shows up when there's a mismatch.

Have you tried compiling a smaller simpler sample project? How did that go? One that prints out driver supported CUDA version, runtime dll supported version, and gpu model & CUDA CC level would be good. Such might be quickly created from (a copy of) mfaktc by deleting most of mfaktc. Keep just the part necessary for producing the following output, and accepting a device number as input, or spin through integers starting from zero until there's no gpu there:

CUDA version info
  binary compiled for CUDA  6.50
  CUDA runtime version      6.50
  CUDA driver version       9.10

CUDA device info
  name                      GeForce GTX 480
  compute capability        2.0
  maximum threads per block 1024
  number of multiprocessors 15 (480 shader cores)
  clock rate                1451MHz

I take https://docs.nvidia.com/cuda/pdf/CUD...river_NVCC.pdf page 23 to mean CUDA10 toolkit supports CC 3.0 and up.

Also https://en.wikipedia.org/wiki/CUDA#GPUs_supported.

[ATH]

Yes, I installed the drivers that came with CUDA10 and added the CUDA 10 dll files:

CUDA version info
  binary compiled for CUDA  10.0
  CUDA runtime version      10.0
  CUDA driver version       10.0

I also tried compiling with CUDA 9.2 which I still have installed and copy the CUDA 9.2 dll files to the folder. Again it compiled without error, but same error message when running it:

CUDA version info
  binary compiled for CUDA  9.20
  CUDA runtime version      9.20
  CUDA driver version       10.0

I'll just wait for Oliver's CUDA10 Windows binaries and test if those works on my card. Otherwise the old binaries I have are good enough, I just wanted to compile them myself if I could.

[bayanne]

Originally Posted by TheJudger

Hello,

finally I was able to put my hands on a Turing (RTX 20x0 series) card. Because of this I was excited and I was right, Turing is a beast for mfaktc.

Unmodified mfaktc 0.21 sources (just adjusted the Makefile) + CUDA 10.0.130 on Linux:

# ./mfaktc.exe -tf 66362159 73 74
mfaktc v0.21 (64bit built)
[...]
CUDA device info
  name                      GeForce RTX 2080 Ti
  compute capability        7.5
  max threads per block     1024
  max shared memory per MP  65536 byte
  number of multiprocessors 68
  clock rate (CUDA cores)   1635MHz
  memory clock rate:        7000MHz
  memory bus width:         352 bit
[...]
got assignment: exp=66362159 bit_min=73 bit_max=74 (28.83 GHz-days)
Starting trial factoring M66362159 from 2^73 to 2^74 (28.83 GHz-days)
 k_min =  71160531149400
 k_max =  142321062305090
Using GPU kernel "barrett76_mul32_gs"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
Oct 05 22:12 |    0   0.1% |  0.630  10m04s |   4118.14    82485    n.a.%
Oct 05 22:12 |    4   0.2% |  0.563   8m59s |   4608.22    82485    n.a.%
Oct 05 22:12 |    9   0.3% |  0.562   8m58s |   4616.42    82485    n.a.%
[...]
Oct 05 22:21 | 4612  99.9% |  0.599   0m01s |   4331.27    82485    n.a.%
Oct 05 22:21 | 4617 100.0% |  0.600   0m00s |   4324.05    82485    n.a.%
no factor for M66362159 from 2^73 to 2^74 [mfaktc 0.21 barrett76_mul32_gs]
tf(): total time spent:  9m 30.800s

This is a founders editions card, starting with a cold card. Power draw is ~260W on average so limited by power target. Temperature is a bit below 80°C and average clock is about 1680MHz once the card is "hot".

New performance king but slightly behind Tesla V100 in terms of energy efficency.

Oliver

Wow, look at the GHz-d/day figures!

[kriesel]

Originally Posted by James Heinrich

Seems to be hovering around US$1800 right now, which is getting close to double MSRP, due to high demand and low supply. Things should calm down in a little while when there's more supply.

Today's eBay listings have the RTX 2080 Ti ranging from $1200-1400 in ongoing auctions, and $1500 and up (way up) for buy it now.

[James Heinrich]

Originally Posted by kriesel

Today's eBay listings have the RTX 2080 Ti ranging from $1200-1400 in ongoing auctions, and $1500 and up (way up) for buy it now.

Buy It Now prices are useful, ongoing auction prices are mostly irrelevant. Most useful is the price of recently sold items, a version of which I use to update the price listing on my site.

[xx005fs]

Originally Posted by TheJudger

Hello,

finally I was able to put my hands on a Turing (RTX 20x0 series) card. Because of this I was excited and I was right, Turing is a beast for mfaktc.

Unmodified mfaktc 0.21 sources (just adjusted the Makefile) + CUDA 10.0.130 on Linux:

# ./mfaktc.exe -tf 66362159 73 74
mfaktc v0.21 (64bit built)
[...]
CUDA device info
  name                      GeForce RTX 2080 Ti
  compute capability        7.5
  max threads per block     1024
  max shared memory per MP  65536 byte
  number of multiprocessors 68
  clock rate (CUDA cores)   1635MHz
  memory clock rate:        7000MHz
  memory bus width:         352 bit
[...]
got assignment: exp=66362159 bit_min=73 bit_max=74 (28.83 GHz-days)
Starting trial factoring M66362159 from 2^73 to 2^74 (28.83 GHz-days)
 k_min =  71160531149400
 k_max =  142321062305090
Using GPU kernel "barrett76_mul32_gs"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
Oct 05 22:12 |    0   0.1% |  0.630  10m04s |   4118.14    82485    n.a.%
Oct 05 22:12 |    4   0.2% |  0.563   8m59s |   4608.22    82485    n.a.%
Oct 05 22:12 |    9   0.3% |  0.562   8m58s |   4616.42    82485    n.a.%
[...]
Oct 05 22:21 | 4612  99.9% |  0.599   0m01s |   4331.27    82485    n.a.%
Oct 05 22:21 | 4617 100.0% |  0.600   0m00s |   4324.05    82485    n.a.%
no factor for M66362159 from 2^73 to 2^74 [mfaktc 0.21 barrett76_mul32_gs]
tf(): total time spent:  9m 30.800s

This is a founders editions card, starting with a cold card. Power draw is ~260W on average so limited by power target. Temperature is a bit below 80°C and average clock is about 1680MHz once the card is "hot".

New performance king but slightly behind Tesla V100 in terms of energy efficency.

Oliver

This is honestly really impressive performance. For 1200$ you get about the same as Titan V would in trial factoring. Now I am just hoping an improvement like this for the next generation cards in LL that's as incredible as this speed bump from Pascal to Turing in trial factoring.