cudaError 13 & not validate on CPU for compute capability 5.0

[yi-ji]

Hi,

My OS is Linux CentOS 7 (x86_64), I installed CUDA 8.0 and a NVIDIA Quadro K620 with compute capability 5.0.

First I met the "Can't compile with compute_10" problem (as in #153 ) so I change all compute_10 flags in Makefile into compute_50, and changed all sm_30/sm_35 into sm_50, according to my understanding that compute_xy and sm_xy should follow the compute capability of physical GPU which is 5.0 in my case.

but when I run ./cudaminer --device 1 --benchmark I got:

[2017-10-18 23:51:30] NVML GPU monitoring is not available.
[2017-10-18 23:51:30] 1 miner threads started, using 'scrypt' algorithm.
[2017-10-18 23:51:31] GPU #1: Quadro K620 with compute capability 5.0
[2017-10-18 23:51:31] GPU #1: interactive: 1, tex-cache: 0 , single-alloc: 0
[2017-10-18 23:51:31] GPU #1: 32 hashes / 4.0 MB per warp.
[2017-10-18 23:51:31] GPU #1: Performing auto-tuning (Patience...)
[2017-10-18 23:51:31] GPU #1: maximum total warps (BxW): 463
[2017-10-18 23:51:58] GPU #1: 94693.53 hash/s with configuration T6x16
[2017-10-18 23:51:58] GPU #1: using launch configuration T6x16
[2017-10-18 23:51:58] GPU #1: cudaError 13 (invalid device symbol) calling 'cudaMemcpyToSymbol(sha256_h, host_sha256_h, sizeof(host_sha256_h), 0, cudaMemcpyHostToDevice)' (sha256.cu line 415)

[2017-10-18 23:51:58] GPU #1: cudaError 13 (invalid device symbol) calling 'cudaMemcpyToSymbol(sha256_k, host_sha256_k, sizeof(host_sha256_k), 0, cudaMemcpyHostToDevice)' (sha256.cu line 416)

[2017-10-18 23:51:58] GPU #1: cudaError 13 (invalid device symbol) calling 'cudaMemcpyToSymbol(keypad, host_keypad, sizeof(host_keypad), 0, cudaMemcpyHostToDevice)' (sha256.cu line 417)

[2017-10-18 23:51:58] GPU #1: cudaError 13 (invalid device symbol) calling 'cudaMemcpyToSymbol(innerpad, host_innerpad, sizeof(host_innerpad), 0, cudaMemcpyHostToDevice)' (sha256.cu line 418)

[2017-10-18 23:51:58] GPU #1: cudaError 13 (invalid device symbol) calling 'cudaMemcpyToSymbol(outerpad, host_outerpad, sizeof(host_outerpad), 0, cudaMemcpyHostToDevice)' (sha256.cu line 419)

[2017-10-18 23:51:58] GPU #1: cudaError 13 (invalid device symbol) calling 'cudaMemcpyToSymbol(finalblk, host_finalblk, sizeof(host_finalblk), 0, cudaMemcpyHostToDevice)' (sha256.cu line 420)

[2017-10-18 23:51:58] GPU #1: cudaError 13 (invalid device symbol) calling 'cudaMemcpyToSymbol(pdata, host_pdata, 20*sizeof(uint32_t), 0, cudaMemcpyHostToDevice)' (sha256.cu line 423)

[2017-10-18 23:51:58] GPU #1: cudaError 13 (invalid device symbol) calling 'cudaMemcpyToSymbol(midstate, host_midstate, 8*sizeof(uint32_t), 0, cudaMemcpyHostToDevice)' (sha256.cu line 424)

[2017-10-18 23:51:58] GPU #1: Quadro K620 result does not validate on CPU (i=0, s=0)!
[2017-10-18 23:51:58] GPU #1: Quadro K620 result does not validate on CPU (i=1, s=0)!
[2017-10-18 23:51:58] GPU #1: Quadro K620 result does not validate on CPU (i=2, s=0)!
[2017-10-18 23:51:58] GPU #1: Quadro K620 result does not validate on CPU (i=3, s=0)!
repeating...

I am not sure about the reason. Is CudaMiner supporting CUDA 8.0 and compute capability 5.0 well enough? Or how should I modify the Makefile?

I tried several combinations with compute/sm_21/30/35/50, etc. but all I got were basically the same or just:

[2017-10-19 00:13:37] NVML GPU monitoring is not available.
[2017-10-19 00:13:37] 1 miner threads started, using 'scrypt' algorithm.
[2017-10-19 00:13:39] GPU #1: Quadro K620 with compute capability 5.0
[2017-10-19 00:13:39] GPU #1: interactive: 1, tex-cache: 0 , single-alloc: 0
[2017-10-19 00:13:39] GPU #1: 32 hashes / 4.0 MB per warp.
[2017-10-19 00:13:39] GPU #1: Performing auto-tuning (Patience...)
[2017-10-19 00:13:39] GPU #1: maximum total warps (BxW): 464
[2017-10-19 00:15:06] GPU #1:    0.00 hash/s with configuration T0x0
[2017-10-19 00:15:06] GPU #1: using launch configuration T0x0
[2017-10-19 00:15:07] workio thread dead, waiting for workers...
[2017-10-19 00:15:07] worker threads all shut down, exiting.

Many thanks for kind help. m(___)m