Open evdcush opened 5 years ago
thanks a lot for your kindness!
Hi! I've tried to install different wheels (1.11, 1.12, 1.13) but I always have the same issue:
import tensorflow as tf Illegal instruction (core dumped)
I have an old processor but new RTX 2070 so I need cuda 10 vendor_id : GenuineIntel cpu family : 6 model : 30 model name : Intel(R) Core(TM) i5 CPU 750 @ 2.67GHz flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm sse4_1 sse4_2 popcnt lahf_lm pti ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid dtherm ida flush_l1d bugs : cpu_meltdown spectre_v1 spectre_v2 spec_store_bypass l1tf mds
Do you have an idea how to fix it?
Thank you
I'd love to see this wheel with cuda compute 3.7
:heavy_check_mark: MKL
AVX, AVX2
* FMA
Release link: https://github.com/evdcush/TensorFlow-wheels/releases/tag/tf-1.12.0-gpu-10.0
Wheel link: https://github.com/evdcush/TensorFlow-wheels/releases/download/tf-1.12.0-gpu-10.0/tensorflow-1.12.0-cp36-cp36m-linux_x86_64.whl
Install via:
pip install --no-cache-dir https://github.com/evdcush/TensorFlow-wheels/releases/download/tf-1.12.0-gpu-10.0/tensorflow-1.12.0-cp36-cp36m-linux_x86_64.whl
Requires:
glibc 2.27
(UNTESTED)-march=skylake
)Build Info:
4.15.0-38
,Xubuntu 18.04
,GCC-7
with-march=skylake
, python3.6.7
XLA JIT
(now supported by default in 1.12),nGraph
,mkl
,nccl 2.3
UPDATED: binary
The original release binary had issues with
ngraph
(core dumps :boom:), so I rebuilt the wheel withoutngraph
.Reading up on
ngraph
support, it's a bit surprising it's even allowed as a bazel config opt; it requires some very specific dependencies and only supports certain builds of TF. Future releases will not includengraph
.I only have my machine to test GPU builds on, so if you are having issues with the wheel despite matching/valid specs, please @ me here in a comment.