. Manually locked in the fan speed because I realized GPU utilization is couple % lower with automatic fan speed. 
Topic: Linux vs Windows GPU hash rate (Read 5927 times)
I hope XFX's Double Lifetime Warranty works when it breaks! Well the good thing about your pool is that sometimes there is some delay sometimes between receiving/sending and it gives my GPU's VRMs time to cool down. I was hitting 120C earlier on the VRMs when I was mining by myself and the card kept throttling every 2-3 minutes . Manually locked in the fan speed because I realized GPU utilization is couple % lower with automatic fan speed.
. Manually locked in the fan speed because I realized GPU utilization is couple % lower with automatic fan speed. 
VRMs below 125C
Too bad there is not prediction market yet. I'll bet to lifetime of your card
.
Seeing a 65 Mhash/s jump to 715 Mhash/s at the same core/mem with the newest poclbm_py2exe_20110127.
Thermally limited, as my stock cooler is @ 60% fan speed to keep it under 85C core and the VRMs below 125C to keep my ears from going deaf
. I believe 950 core @ 1.1625v is not bad at all.
Thermally limited, as my stock cooler is @ 60% fan speed to keep it under 85C core and the VRMs below 125C to keep my ears from going deaf
. I believe 950 core @ 1.1625v is not bad at all.
Personally, I think 650 Mhash/s is a pretty respectable number for a 5970. Who is getting more? The SDK 2.2/2.3 issues are mainly of concern to Linux users. You can't run both GPUs independently with SDK 2.2 and possibly 2.3 on Linux.
Put SDK 2.1 on my box with 10.12 Cats. Didn't notice a change in performance at all. Used Revo uninstaller to get rid of SDK 23 and Guru3d's driver sweeper to to cleanse it before I reinstalled. Weird...
What should I be expecting here?
What should I be expecting here?
Follow pyopencl instructions for manual installation.
http://wiki.tiker.net/PyOpenCL/Installation/Linux/Ubuntu
python-pyopencl depends on nvidia drivers
http://wiki.tiker.net/PyOpenCL/Installation/Linux/Ubuntu
python-pyopencl depends on nvidia drivers
It should work fine on Ubuntu. Use latest fglrx (ATI's proprietary driver) and stream 2.1
I failed on installing package python-pyopencl, because it provide some files as stream 2.1 does. How to fix it?
It should work fine on Ubuntu. Use latest fglrx (ATI's proprietary driver) and stream 2.1
Does m0mchil's miner work in Ubuntu? Or will I have to use another miner? Reading this, https://bitcointalksearch.org/topic/easy-ubuntu-python-opencl-mining-setup-2636 (Easy Ubuntu python OpenCL mining setup), it seems that it won't.
I'll get SDK 2.1 installed tomorrow and see how much it has improved and report back. Then I'll move to linux and see how that works out. Both GPU-Z and MSI Afterburner are reporting 99% usage, so something smells a little fishy.
Mem scales to 1200, but doesn't seem to influence hash rate so I've left it at 1000.
Memory speed is quite irrelevant for hashing. I had memory also on 1200, but moved back to 1000 without any significant drop in speed (ArtForz said the difference is something around 0.3%).
I can confirm that. SDK 2.1 on both systems, but I can reach 600Mhash/s on Linux boxes much easier (on lower clocks). From my investigation, miners on Windows are using GPUs with 92-95% load, but on Linux load is on 99%.
Use stream 2.1 or 2.2. Not sure why, but 2.3 is slower.
I can't speak to the GPU generation thing, but in my experience switching to Linux from Windoze gets you more of everything worth having on the exact same hardware.
I'm seeing a lot higher hash/s on other people's 5970 than what I'm getting and most of them are using Linux.
Would I see a higher hash rate if I moved from my current Win7 64-bit install to a Linux OS?
Currently getting 650 Mhash/s @ 950core/1000mem. Using 10.12 drivers with SDK 2.3 and using most current m0mchil client. Mem scales to 1200, but doesn't seem to influence hash rate so I've left it at 1000. Getting my second 5970 soon!
Would I see a higher hash rate if I moved from my current Win7 64-bit install to a Linux OS?
Currently getting 650 Mhash/s @ 950core/1000mem. Using 10.12 drivers with SDK 2.3 and using most current m0mchil client. Mem scales to 1200, but doesn't seem to influence hash rate so I've left it at 1000. Getting my second 5970 soon!
Jump to: