Open ghost opened 7 years ago
Using -c
this error does not occur, so it is definitely related to the GPU functionality. Passing -g
works for a while, leads to tripcodes being found, but then moments later the application exits with the above error message. The --disable-gcn-assembler
flag does not affect the result.
For your information, mty_cl
does work, finds about 120Mtrip/s
on this instance type. merikens-tripcode-engine
finds about 40Mtrip/s until it crashes.
Interesting to note that merikens-tripcode-engine
seems to work fine on the g2 instance types, so perhaps related to the NVIDIA K80s?
Same here with p2.xlarge
instance and Deep Learning Base AMI (Ubuntu)
image
build cmd
./BuildAll.sh --enable-cuda --install
run log
Enabled Features: OpenCL/GCN CUDA SSE2/AVX/AVX2(x86_64)
CUDA DEVICE
===========
Device No.: 0
Device Name: Tesla K80
Multiprocessor Count: 13
Clock Rate: 824MHz
Compute Capability: 3.7
Compute Mode: cudaComputeModeDefault
PATTERN(S)
==========
0: "(hidden)" (regex)
1: "(hidden)" (regex)
2: "(hidden)" (regex)
TRIPCODES
=========
STATUS
======
Performing a forward-matching search on GPU(s)
for 2 patterns (2 chunks) with 6 to 8 characters:
CUDA0-0: 23.5M TPS, 128 blocks/SM
CUDA0-1: 23.7M TPS, 128 blocks/SM
0.001T tripcodes were generated in 0d 0h 0m 20s at:
38.17M tripcode/s (current)
38.17M tripcode/s (average)
On average, it takes 20.8 minutes to find one match at this speed.
No matches were found yet.
ERROR
=====
A corrupt tripcode was generated.
The hardware or device driver may be malfunctioning.
Please check the temperatures of CPU(s) and GPU(s).
Just compiled this on an p2.xlarge instance with the official NVIDIA image, and the Ubuntu 16.04 image.
Both give me the following result:
I've tried building both the regular and NVIDIA optimized version.
Any thoughts?