the problem on this one is that you only can actually manage only one GPU. in my case with a 5970 card that has 2 GPU's, i only can do the settings on the first gpu DEVICE=0.
AMDOverdriveCtrl --help
AMDOverdriveCtrl --adapter-index=
Topic: Underclock mem speed on HD 5970 linux? (Read 11840 times)
AMDOverdriveCtrl --help
AMDOverdriveCtrl --adapter-index=
the problem on this one is that you only can actually manage only one GPU. in my case with a 5970 card that has 2 GPU's, i only can do the settings on the first gpu DEVICE=0.
i've allready wrote on the open discussion board here https://sourceforge.net/projects/amdovdrvctrl/forums/forum/1335752/topic/4026230/index/page/1
The official clock boundaries are set in the bios. I don't know much about Linux, on the other hand, I know from experience it's not too complicated to modify your bios with RBE (which is a windows tool if I recall. Yet once it's done, you're all good).
I found a solution using "ATI Overclocking Utility" referenced here: http://www.overclock.net/software-news/514993-phoronix-ati-overclocking-utility-linux-2.html#post6338431
There should be instructions on that page for installing it. On ubuntu I also had to "sudo apt-get install libqt4-network" in order to get the util to run.
This utility appears to change whatever values aticonfig is using for bounds so once you lower the memory speeds in the utility it will expand the range available to aticonfig. I could clock my 5870 as low as 160Mhz, however anything under 900Mhz or so seemed to affect hash speed (~340MH/s @ 900Mhz, ~290MH/s @ 300Mhz using poclbm).
There should be instructions on that page for installing it. On ubuntu I also had to "sudo apt-get install libqt4-network" in order to get the util to run.
This utility appears to change whatever values aticonfig is using for bounds so once you lower the memory speeds in the utility it will expand the range available to aticonfig. I could clock my 5870 as low as 160Mhz, however anything under 900Mhz or so seemed to affect hash speed (~340MH/s @ 900Mhz, ~290MH/s @ 300Mhz using poclbm).
I have an 5870 as well, but when I use this utility on my Ubuntu machine, it seems to result in more "invalid or stale " shares.
I'm thinking about getting my windows machine working so I can see if there is any difference.
I would love a low mem version of poclbm. apparently diablo isn't effected by the low mem, but I couldn't get more than 318mh on diablo compared to the 346mh I get on poclbm. I tried nearly every sdk and ati driver combo I could find, none of diablo's came close after waiting for it to ramp up as well.
When running in linux I noticed that poclbm showed a reduced hash rate when I lowered the memory clocks under 900Mhz, however DiabloMiner has stayed steady all the way down to 160Mhz. IIRC poclbm wasn't having that problem when running under windows.
Hmmm, might have to put a special order request into mOmchill for a linux super-cool low-mem version ... unless it is Xorg sticking its nose in again.
When running in linux I noticed that poclbm showed a reduced hash rate when I lowered the memory clocks under 900Mhz, however DiabloMiner has stayed steady all the way down to 160Mhz. IIRC poclbm wasn't having that problem when running under windows.
You can underclock the memory by modifying the bios using Radeon Bios Editor for the profile with the highest voltage. unfortunately, I saw no gain by lowering my memory clock on poclbm on my gentoo installation, and I actually lost mhash. I tried it at 925 core and 600 mem and 925 core / 900 mem beat it by 12 mhash on a asus 5870.
Did you try running it at 300? Thats what works for me in Windows with the 5870 at least. I get similar results as you at 600 but for some reason 300 works better. Lets me get up to 950 core as well.
yes, results were just as poor. for some reason the low memory clock just wasn't helping in linux, but my friend has seen great results of a 600 memory clock in windows (lower temps and same speed). for linux though, I've gotten nothing but lower temps AND lower speed.
I found a solution using "ATI Overclocking Utility" referenced here: http://www.overclock.net/software-news/514993-phoronix-ati-overclocking-utility-linux-2.html#post6338431
There should be instructions on that page for installing it. On ubuntu I also had to "sudo apt-get install libqt4-network" in order to get the util to run.
This utility appears to change whatever values aticonfig is using for bounds so once you lower the memory speeds in the utility it will expand the range available to aticonfig. I could clock my 5870 as low as 160Mhz, however anything under 900Mhz or so seemed to affect hash speed (~340MH/s @ 900Mhz, ~290MH/s @ 300Mhz using poclbm).
There should be instructions on that page for installing it. On ubuntu I also had to "sudo apt-get install libqt4-network" in order to get the util to run.
This utility appears to change whatever values aticonfig is using for bounds so once you lower the memory speeds in the utility it will expand the range available to aticonfig. I could clock my 5870 as low as 160Mhz, however anything under 900Mhz or so seemed to affect hash speed (~340MH/s @ 900Mhz, ~290MH/s @ 300Mhz using poclbm).
Thanks Ian. I'd played with utility before but hadn't noticed that it was frigging with the peak range back in aticonfig .. wth, how does that work? Anyway happy to have access to bigger clock range without flashing the BIOS.
You can underclock the memory by modifying the bios using Radeon Bios Editor for the profile with the highest voltage. unfortunately, I saw no gain by lowering my memory clock on poclbm on my gentoo installation, and I actually lost mhash. I tried it at 925 core and 600 mem and 925 core / 900 mem beat it by 12 mhash on a asus 5870.
Did you try running it at 300? Thats what works for me in Windows with the 5870 at least. I get similar results as you at 600 but for some reason 300 works better. Lets me get up to 950 core as well.
You can underclock the memory by modifying the bios using Radeon Bios Editor for the profile with the highest voltage. unfortunately, I saw no gain by lowering my memory clock on poclbm on my gentoo installation, and I actually lost mhash. I tried it at 925 core and 600 mem and 925 core / 900 mem beat it by 12 mhash on a asus 5870.
note: when you modify the bios using RBE, you can't mess with the clock speeds once you are booted into X, because it will only let you set the default low value you were stuck with before.
note: when you modify the bios using RBE, you can't mess with the clock speeds once you are booted into X, because it will only let you set the default low value you were stuck with before.
Unfortunately not as far as I know, it appears to apply the settings to all GPUs (at least in the case of my multi-5870 setup). If you're just looking to have different memory clock settings through what you can do is use the utility to set all of the clocks down to 160Mhz (or whatever your GPU minimum is), then if you go back to aticonfig you should have the full 160Mhz-1300Mhz range available to set as you like.
I'll also note that I've found the overclocking utility to be a bit flaky, sometimes it will fail to set the clocks (if you run it from the terminal you'll see it print errors). I've gotten it to set everything eventually by turning down the clocks in increments (set to 900Mhz, then 600Mhz, then 300Mhz, then 160Mhz).
I'll also note that I've found the overclocking utility to be a bit flaky, sometimes it will fail to set the clocks (if you run it from the terminal you'll see it print errors). I've gotten it to set everything eventually by turning down the clocks in increments (set to 900Mhz, then 600Mhz, then 300Mhz, then 160Mhz).
I found a solution using "ATI Overclocking Utility" referenced here: http://www.overclock.net/software-news/514993-phoronix-ati-overclocking-utility-linux-2.html#post6338431
Hey thanks, this is nice. Do you know how to specify which GPU the utility will access, in the case of multi-GPU setups?
It looks like the 5870 is using ~30-35 fewer watts @ 160Mhz mem vs. 1200Mhz mem, pretty huge savings. DiabloMiner doesn't seem to care about the memory speeds either.
And you'd be surprised but many games are still playable at 300 mem clocks + miner running LOL Core clock is what makes the big difference even in-games. I do not however recommend playing games with those clocks :p
While I did notice a, well, noticeable amount of Mh/s rise, the biggest plus was the MUCH lower Power consumption and heat output. MUCH lower as in like 10C and 30+W
While I did notice a, well, noticeable amount of Mh/s rise, the biggest plus was the MUCH lower Power consumption and heat output. MUCH lower as in like 10C and 30+W
I found a solution using "ATI Overclocking Utility" referenced here: http://www.overclock.net/software-news/514993-phoronix-ati-overclocking-utility-linux-2.html#post6338431
There should be instructions on that page for installing it. On ubuntu I also had to "sudo apt-get install libqt4-network" in order to get the util to run.
This utility appears to change whatever values aticonfig is using for bounds so once you lower the memory speeds in the utility it will expand the range available to aticonfig. I could clock my 5870 as low as 160Mhz, however anything under 900Mhz or so seemed to affect hash speed (~340MH/s @ 900Mhz, ~290MH/s @ 300Mhz using poclbm).
There should be instructions on that page for installing it. On ubuntu I also had to "sudo apt-get install libqt4-network" in order to get the util to run.
This utility appears to change whatever values aticonfig is using for bounds so once you lower the memory speeds in the utility it will expand the range available to aticonfig. I could clock my 5870 as low as 160Mhz, however anything under 900Mhz or so seemed to affect hash speed (~340MH/s @ 900Mhz, ~290MH/s @ 300Mhz using poclbm).
I'd be interested in this as well. With my 5870 under Windows the CCC would only let me bring the memory speeds down to 900Mhz, however using MSI Afterburner I could bring them down to 600Mhz so it should be possible without flashing the bios or otherwise modding the card.
With Afterburner you should be able to drop down to 300Mhz. Dropping from 600 to 300 actually gives me +2Mhash/sec.
After you enable unofficial overclocking, Afterburner will show 600 as the min. You need to set it to 600, and then restart Afterburner. After the restart 300 should be available.
I'd be interested in this as well. With my 5870 under Windows the CCC would only let me bring the memory speeds down to 900Mhz, however using MSI Afterburner I could bring them down to 600Mhz so it should be possible without flashing the bios or otherwise modding the card.
As for reasoning, the hash performance always stayed consistent regardless of the memory speed (which makes sense since mining barely requires any reads or writes), but dropping the memory speeds from the default 1200Mhz to 900Mhz results in lower temperatures and saves ~10-12W.
As for reasoning, the hash performance always stayed consistent regardless of the memory speed (which makes sense since mining barely requires any reads or writes), but dropping the memory speeds from the default 1200Mhz to 900Mhz results in lower temperatures and saves ~10-12W.
I've read about Windows guys having ability to under-clock memory down to 300Mhz on the HD 5XXX cards is there a tool to do this in linux?
Aticonfig seems to be getting its minimum mem clock speeds from somewhere, GPU BIOS, how do the windows clock tools get around that?
E.g:
$ aticonfig --odgc --adapter=all
Adapter 0 - ATI Radeon HD 5900 Series
Core (MHz) Memory (MHz)
Current Clocks : 725 1000
Current Peak : 725 1000
Configurable Peak Range : [550-1000] [1000-1500]
GPU load : 99%
Adapter 1 - ATI Radeon HD 5900 Series
Core (MHz) Memory (MHz)
Current Clocks : 725 1000
Current Peak : 725 1000
Configurable Peak Range : [550-1000] [1000-1500]
GPU load : 98%
indicates the minimum mem. clock speed is 1000 MHz, configurable range [1000-1500].
Is there a --pplib-cmd that will do this perhaps? (related does someone have a list of the --pplib-cmd options?)
Now that you mention it: what are the benefits of this? Anything else besides slightly lower wattage and temps?
I was only considering wattage and temperatures ... but if hashes can be increased all the better. Spinning the mem. clock more than it needs to is just inefficient. I'm purely hashing with these cards right now and want to see how much can be squeezed out for a feasibility of a bigger GPU cluster application also.
A wider question is whether it is really necessary to use Xserver and display adapter config to make the OS GPU aware for OpenCL clustering. Are there other routes? The whole Xorg, display adapter integration seems a bit kludgy when what we want is to get computational access to the GPUs.
Jump to: