I already posted a similar question in the Newbies section but now I have 5 posts I would like to ask the 'Big Boys'

I put a 7950 into my own computer and I got it to mine scrypt at 630 Khash.
Then I decided to build a custom rig for it and ordered a cheap motherboard, the Asus F2A55-M LE (AMD socket FM2 with a trinity processor).
When I put the SAME 7950 card on this motherboard, I got the following error from cgminer: "Maximum buffer memory device 0 supports says 536870912, Your scrypt settings come to 1422917632"
I had to lower my thread-concurrency all the way down to 7168 to get it to work. Also I can not push the intensity past 13 on this board or I will get hardware errors. On the first motherboard (Intel H67 chipset) I was mining happily at intensity 20 and thread-concurrency 21504 without any errors.

I have tried all different BIOS settigs on the new board, disabling the onboard GPU and much more, but I can't get the performance out of the 7950 card that I got on the Intel board.
Does anybody have any idea why this is happening? Obviously I'm going to order another motherboard but I would like to make an educated decision about what I'm going to buy now I know there are big performance differences between boards.
Any insight would be much appreciated.

That's not an error, that's information. Did you leave out the part where it says "decrease TC or increase LG?"?

There is also the issue of "memory speed", and "buffers", and "controller chips", and "bus-speeds", and all sorts of other motherboard/bios/hardware related potential issues... "turbo-boosting", "shared-channels", "spread-spectrum", "voltage limits", "throttling from heat"... yadda, yadda, yadda... lol. Oh, and PSU power quality, under-loaded, and over-loaded.

Most higher priced MoBo's tend to have better/faster/tuned controls and functions that cheap motherboards skimp-on, to save a buck.

(Odd fact... Memory, although it says one speed... may default to the "real speed", as the "advertised speed" is an "overclock-stable" speed. You usually have to manually set that speed/clock and the times. Otherwise the BIOS simply uses the "safe" settings that the memory was "designed" for, internally. Also note, some memory speed/voltage is locked to the CPU speed/voltage clocks, since CPU's now have internal memory controllers available.)

Did you cleanly remove the drivers after changing them?

May also have to do with slots... Not all 16x slots are 16x... that is just the size of them.

Using 1 card, any of those MAY be 16x... but add another card, and chances are, they drop to 8x each.... or one 16x and one 8x, etc...

Thus, running "both the same", they are limited to 8x speeds/bus...

However, scrypt mining is temperamental, and highly dependent on reverse settings from bitcoin mining. (Must have higher memory clocks, lower core clocks, and they must be tuned... Just making one higher is not "tuned". They have a specific "balance" that has to be found... This is NOT the same for every card/mobo, and is unique for almost every setup.)

EG... core/mem
800/950 fail
800/1000 crap
800/1024 ok
800/1048 great
800/1100 perfect
800/1110 great
800/1120 crap
800.1125 fail

For bob, with the same setup... 800/1024 might be the best...

Also, workload I=13 G=1 is the "best" speed... less or more, and it degrades.

Of all the apps I've run in my entire life, I've never seen one like cgminer where it can run so differently on each system it installed in. There are so many variables. Even two video cards of the same make in the same PC don't always behave the same way .

OK guys,

It has been 2 days of exhaustive testing and I'm starting to understand why some people on this forum say that tuning a miner is like 'Black Magic'. Some crazy things happen with different settings and different hardware.

The 2 boards I have been testing are the ASRock B75M-GL R2.0 with a Intel Celeron G1610 on it, and a Gigabyte F2A85-XM HD3 with a AMD A4-5300 on it.
Ofcourse the first step was to see if the AMD F2A85 chipset board had the same problem as my previous F2A55 board. I set it up, installed windows, started cgminer and got the same message: "Maximum buffer memory device 0 supports says 536870912, Your scrypt settings come to 1422917632"
I went back into the BIOS, which was more elaborate as the one on the F2A55 board, and found a setting where I could choose for my default GPU: 'Onboard', 'NB PCI expess Bus' or 'SB PCI express Bus'
The setting was set to 'NB PCI express Bus' so I changed it to 'SB PCI express Bus' and rebooted. To my big surprise, cgminer started happily at Intensity 20 and TC 21504. So problem soved

I then also set up the ASRock board with the Intel Celeron and tested the same setup. No problems there.
I then added a 2nd 7950 card to each board, as I tested with one first. No problems either.
Then I switched both boards to use the 'onboard' video by default, so the 7950's could do the mining and I still had a useable desktop. Again no problems.
An added bonus is that the AMD A4-5300 has a small radeon GPU onboard that could be used for mining too It didn't do much but still 25Khash on scrypt so why not use it.

The Gigabyte board with the AMD A4-5300 was 15 Euros more expensive than the ASRock board with the Celeron, but the AMD did give me the extra 25Khash so that did level out evenly.

So which solution did I choose? And more important, WHY?
I'm going for the ASRock board with the Intel Celeron processor, for something that I noticed and I can't explain: The 7950 GPU's on the ASRock board run cooler than the pair on the Gigabyte board.
I know every GPU is different and produces different amounts of heat, so I swapped the GPU's over between boards. Same result. Now the pair that ran hot on the Gigabyte board ran cooler on the ASRock board and the pair that was cooler before now runs hotter on the Gigabyte board.
And it wasn't by a small margin either. The temp difference was 8 to 10 degrees Celcius !
I really can't explain why this is happening. I have spent this whole day trying configurations to find out but I don't have any explanation except that I can tell you for sure that it was happening every time. I read the temps as listed by cgminer, afterburner and GPU-Z. As far as I know, the temps are registered by the GPU card and just passed on to the driver, I can't see how the motherboard can influence this reading and I also can't explain why a card would physically run hotter. Maybe one day I'll figure it out, or maybe one of the members on this forum has some ideas

Anyway, I hope this info might help any of you in the future. Thanks for helping me out, I'm going mining