At peak MH/j Linux has a higher MH/s rate than Windows from the testing we did.  This probably has a lot to do with the fact Linux in general has a lot less 'overhead' than Windows.

I'll give you an example, card A is an ATI 5770 we found on this specific card the optimal MH/j was core 947mhz, memory 321mhz...  Under Windows this card averaged 197.9 MH/s over a 15 minute period, with those exact same settings we are getting 212.3 MH/s under Linux.  15 MH/s isn't a lot, but you factor in we were able to then remove the HD and the CPU load (and thus power usage) is lower, the Linux system has a much better profit margin than the Windows.

Now, our measurements aren't exact...  we took a network power switch/battery backup, and pulled snmp power usage every 15 seconds for a 15 minute period.  Then we wrote a script that would move the core and memory settings up and down until it found a sweet spot in MH/j for a specific card.  We ran this test on 8 different cards from 2 different vendors in the 5770 line, the results are similar for all the cards in that test.  We ran the same test on a smaller batch of 5850's and 6950's again with similar results.

If all you know is Windows, and you're setting up a single mining rig, then it probably doesn't make sense to worry about those numbers as much...  if you're trying to setup a bunch of rigs and plan to operate long term in a 20-30% profit margin...  every little bit of cost savings helps.

Not to mention, that writing a script to remotely move the core and memory mhz settings remotely in windows would not have been an easy undertaking.  IMHO (and it's just an opinion) Linux wins hands down as the better dedicated mining rig OS.

Did you read my post?  I'm assuming when you write MHZ, you really mean MH/s.  First, we were trying to find the most profitable settings, surprisingly as you overclock, while you're MH/s increases, often your MH/j tends to decrease, meaning while you get more BTC, it costs you more per BTC to generate them.

So, while AMDOverdriveCtrl doesn't allow you to overclock as much as certain Windows tools, the sweet spot was *always* within the range of a linux system without having to jump through hoops.Now your point probably is that makes them even...  but it doesn't...

In theory you're having to buy a Windows license to run that dedicated mining rig...  that costs money...  you're most likely putting Windows on a hard drive, which costs a lot more than network booting, or thumb drives, that costs extra money...  Windows requires dummy plugs for each card, that costs money, and makes management that much harder...  and remote management takes a little more setup time, costing you time, and as we all know time is money...  More money spent means less profit

So yeah, if you know what you're doing in Linux, it's more profitable than Windows for dedicated mining rigs.

AMDOverdriveCtrl works with 6950's and 6970's just fine.

Try the x64 version and see if it works.  Probably not a graphical problem.

Linux hands down is going to be more profitable if you know what you're doing.  We ran extensive tests on 58xx and 69xx systems running both Windows XP and Ubuntu 11 even with the "limited" overclocking controls we found Linux to be faster by far than Windows.  Keep in mind that we first searched for the sweet spot of profitability, sometimes overclocking reduces your profit margin.

Long term this is a game about maximizing profit, and while maxing out MH/s seems like the thing to do, you should be much more concerned with MH/J, we found that over clocking outside of the 'normal' settings, resulted in a lower MH/J in all cases.

No that would just be in the search warrant they're required to give you a copy of (at least here in the US) when they perform the search.

I don't think it is true...  easy enough to prove me wrong.  Where did it happen?

I'm sorry but I call BS...  It's easy enough to prove us wrong...  post the town and state he lives in...  or if it's rural, the county and state...  That search warrant is public record, so it will be easy enough for one of us to call and confirm there was a DEA raid there.

Most modern chips use less power when they're idling than when they're running under load...  so you're paying more for power to mine even if you normally left your computer on 24/7.

You won't get anywhere near 420MH/s out of that card...  More like 325-335...  It's not a reference card so you can't unlock the shaders on it...  Otherwise it's a good card.

Something is wrong with your airflow then...  I'm getting 212 Mh/s out of a 5770 in a dual card config, one is 58C and the other 67C.

Have you underclocked the memory down to 300mhz?  What software are you using?  SDK version?

You should be getting 200+ Mh/s out of a 5770...  You should spend sometime tweaking the settings and learning how to get the most out of your card(s) before spending money on new cards.

Send me a PM, I've already got the gpu temps into snmpd on linux and getting that into opennms or nagios would be pretty straight forward.

You're focusing on just the cards, not the infrastructure, power costs, or labor...  If you're paying $.25 per kwh and I'm paying $.08 kwh...  things become unprofitable for you, way before they do for me.

Also scale as well...  if you have 1 machine making $1 USD a day profit... it might not be worth the headache....  but if I have 200 machines making $1 a day profit it might very well be worth the headache.

My experience with the 5770's is that at some point they slow down due to the heat...  That is somewhere north of 80C when that happens, my Mh/s rate dropped like a stone, it was reproducible on my cards...  once I drop the temp down no more problems.  Might just be my model of the 5770's... but I wouldn't want to run a card that hot 90+C for very long.

More importantly, there are pools out there with 0% or .5% fees...  It's worth the effort to use a pool even 8Gh/s.

Try this:

-k phatk DEVICE=0 VECTORS BFI_INT FASTLOOP=false AGGRESSION=9

I'm running 5770's at 960 with that command and getting about 212-213 Mh/s

Also, what is the temp on the card?  It will slow down a lot if the temp gets too high.

I have 6950's running and did see a slight bump in using phatz (4 or 5 Mh/s), and underclocking the memory doesn't hurt performance in my experience and does seem to help reduce heat.  You notice I didn't state that he would see a Mh/s bump from reducing the memory clock, only that he should be able to find a sweet spot where it doesn't hurt Mh/s and does reduce heat.

Where to begin...  first Linux, I'd install it, phoenix, and use 2.4 SDK and phatz kernel.  I'd experiment with underclocking the memory on the card, you should be able to find a sweet spot, and at the same time reduce heat production (you don't need the memory speed for mining, by reducing it, you don't hurt anything on the card).

I would expect you should be able to get 8.4+ Gh/s without having to overclock the core.

Just out of curiosity, how do you tell what worksize you need for a specific card?