Reducing electric power ?

[drkirkby]

Originally Posted by bhelmes

A peaceful and pleasant day for you,

I thought of adding a nvidia geforce 1030 card (passiv, 30 Watt) and use cudalucas instead of mprime.
The cpu (Ryzan 3600) use ca. 110 Watt, and the computer in idle time 30 Watt.

Is cudalucas still useful ? I installed Cuda 8.0 version and the nvidia-driver 430, but the binary cudalucas goes to Cuda 6.0

Is there a binary update for cudalucas ?

I have not used cudalucas, but GpuOwl
https://mersenneforum.org/forumdisplay.php?f=171
will do LL and PRP tests with an Nvidia card. But I found on my mid-range card, an Nvidia Quadro P2200 75 W GPU, that double-precession maths is very slow, so the card was effectively useless for PRP testing, being painfully slow. I would expect you will find the same with your 30 W card.

You may well find your card would be better used for trial factoring, which was certainly the case with my 75 W card, but of course you will never find a prime number with that.

[kriesel]

Originally Posted by bhelmes

I thought of adding a nvidia geforce 1030 card (passiv, 30 Watt) and use cudalucas instead of mprime.
The cpu (Ryzan 3600) use ca. 110 Watt, and the computer in idle time 30 Watt.

Is cudalucas still useful ? I installed Cuda 8.0 version and the nvidia-driver 430, but the binary cudalucas goes to Cuda 6.0

Is there a binary update for cudalucas ?

If interested in efficiency, use GTX10xx on TF, and Gpuowl on GPUs with much better DP performance relative to their SP or int32 performance. CUDALucas is LL-only (no PRP or GEC or proof capabilities), lacks even the Jacobi symbol check that can be applied to LL, and is slower on the same GPU and algorithm and exponent than a recent version of Gpuowl. CUDALucas was a good program, and still may be used in verifying new Mersenne prime discoveries in the future, but should not be used for ordinary first time tests or DC when there are alternatives, due to lower performance compounded by lower reliability. (Although, I run CUDALucas occasionally on old slow NVIDIA Quadro GPUs that are not capable of running Gpuowl, doing LLDC. Their SP/DP ratios tend to be better DP performance than most GTX. And some incorporate ECC ram on the GPU.)

Some underclocking & perhaps undervolting can help efficiency and power consumption, and usually can be performed with vendor-supplied performance tuning software that's often used for overclocking. Test carefully for reliability. Gpuowl PRP with GEC is excellent in detecting errors.

CUDALucas is available, including on the mersenne.ca software mirror, for a variety of CUDA levels, but most precompiled are for Windows.
There's lots more info available in the reference info, including CUDALucas and Gpuowl-specific threads.