Bitcoin Forum

I am using CGminer on a dual 7970 mining rig. 

I have the config file adjusted to load these settings:

Engine:  1150
Mem:  300
Intensity: 9
fan: 70


Using Windows 7

Everything works great, and they hash away at 690 MH/s each, but the damn memory clocks refuse to go to anything lower that 1,000.  And they will only go that low after I manually set them within CGminer.  Am I missing something? 

2 BTC bounty to the first person who can help me out and successfully get my mem speeds to where I have seen them on other systems. 

Have you tried clocktweak?  I don't have a 7000 series card, but it lets me lower my mem to 300mhz easily on my various 5830s, 5850s, and 6970.

Why not just use Afterburner?

Edit the MSIAfterburner.cfg

[ATIADLHAL]
EnableUnofficialOverclocking   = 1
UnofficialOverclockingEULA   = I confirm that I am aware of unofficial overclocking limitations and fully understand that MSI will not provide me any support on it
UnofficialOverclockingMode   = 2
AccessibilityCheckingPeriod   = 0

And your good to go.

I would prefer to do it within CGminer

If it's anything like my non-reference 6970s, then the best you can do is a 150 clock mem-diff unless you flash the BIOS.  Afterburner let me get my memory down to 685 on my 6970s and that was the best I could do without flashing the BIOS (which I won't do because I don't trust myself to do it).

I have a 6850 and run at 925/250 (so 250 Mhz mem clock) with Afterburner and have not flashed my bios

Yeah it would be interesting if you could test 300 memclock on the 7970 and if that is as effective as on the 5970s etc.

Try flashing BIOS and report back !

Without killing your card.

Reference 7xxx cards are like the reference 6xxx cards.  Without a change in the card firmware, they are locked to a minimum memclock related to the GPU clock.  I believe you can only go 150 down without it resetting to stock speeds.  And mucking about with the firmware this early in their release cycle could be risky.

I can lower my mem clocks down to 300mhz or lower on my 6970 no problem.  I've never touched the firmware.  It is a reference design.

So why can I not get it to even go to 600 if 150 is the limit? 

I think you need to do this to the MSIAfterburner.cfg file.

On my 6970s, I wasn't able to go to 685 until I modified this file.

I am going to try the afterburner beta tonight and see if that works. 

Yes, make sure it's the Beta.  It didn't work for me with the other versions.

The current version (Beta 12) of afterburner does NOT work correctly with the 7970.
I'm fighting that battle right now.
I can set whatever I want in AB, but the card just mostly ignores it when I start mining.
The best I've been able to get so far is 1000 core/800 mem @ .893v
I'm hoping when Beta 14 is released (possibly today) it might fix this bug.
The author of AB says the issues are due to AMD jacking around with the drivers and causing stuff to break. :)

Make sure you're using 2.2.0 Beta 10.  It's one of the only versions currently working for setting voltages and memclocks on a 7970.  The other option is manually hex editing the BIOS and flashing.  Apparently the BIOS signature check is disabled in Windows in the current driver, but I haven't tested it (works on Linux though).

Beta 10 didn't work at all on my XFX 7970. I had to get Beta 11 to make it at least partially work.
Unwinder said that the latest AMD drivers (12.3) removed some dlls and broke the unofficial overclocking feature.
Hopefully AB Beta 14 will find a fix...

Oh...   interesting.  I haven't tried updating the drivers on the rig since I started mining.  I started using Linux shortly after.  Looks like BIOS flashing may end up being the solution for newer drivers unless ATI decides to leave the old RivaTuner stuff in the driver.

Using the beta allowed me to lower the clock. 

So who deserves the bounty?  I will honor the wishes of the forum! 

The latest beta?  If so, it looks like you would have got it done regardless (you suggested "the beta" before anyone else did)  :)

I thought so too, but I did not want to be seen as suggesting that!

I seem to be missing something here.
The several posts above mine appear to suggest that the issue has been resolved.
Has it? Not for me.
I still can't drop my mem clocks below the AMD default of Core Clock - 150mV.
Has this been resolved and I missed it?

So how does it impact performance? AFAIK, you need SDK 2.6 for radeon 7 cards, and with SDK 2.6, hashrate scales pretty good with mem speed (IOW, if you lower your mem speed, hashrates take a nosedive).
Am I wrong?

Very simple question and only you can decide: Did any of the specific info posted help you reach the answer (did you already know it or not). If it didn't help no bounty, if it did bounty.

Not with the latest AMD drivers it doesn't.
Version 8.96 (Feb 14) pretty much breaks AB completely.
In fact, i had to uninstall it to get my card to even run a stock clocks.
Hopefully Unwinder will fix it soon, but it's in AMD's lap now.

That's the old driver version I was talking about.
With the new driver (8.96), it breaks AB, even Beta 14.

No problem, this whole thing confuses the heck out of me! :)
I'm debating rolling back to an earlier driver, but I'll probably wait at least until I get my water block installed later this week.

That's good info, thanks.
AMD is doing everything they can to drag users through mud...

Try using ati tray tools as well. I dont know if it supports the 79xx series yet but, I used it to lower the memclock on my 6950 to 150mhz (confirmed via gpuz) where all other tools failed to set it this far down.