The latest new entries:

Watts   Mhps       Watts/Mhps   Mhps/Watt
1135          563       0.496          2.016
800           1760       0.455          2.2

This makes my claim of 2.41 MH/J more plausible and I still get to be an anomylous result!  Good work guys .

Agreed.  My two 5850s are identical on the back (all codes are the same) with the sole exception of a black circular stamp for which the first reads "TEST PASS 61" and the second "TEST PASS 63".  I'd guess they were near to each other from their creation.  At stock voltage one card hits max stable clocks at around 970 MHz and will often crash instantly at 990 MHz, the other goes to 1020, crashing at 1040.  After a 0.1V undervolt I did all the testing again and found max stable clocks of 845 MHz and 895 MHz.

I tried swapping them round on the board thinking it was a heat issues, and later tweaking the fan speeds to keep the cards at the same temperature, but still the 'lucky' card was quite a bit faster.

My lucky card lived up to its name last night when I decided to start mining namecoin.  It found a block within 90 minutes of me starting .

No GUI required.  AMDOverdriveCtrl can work in batch mode (hence the '-b' flag I gave above).  I don't remember exactly but if you run the command once it will generate some default profiles based on your cards and put them in "$HOME/.AMDOverdriveCtrl".  You can then edit the profile and apply it with the line I gave above.

Here's a profile I'm currently using for my 5850 if you can't get it to generate one for some reason (the voltage may be a little low for a 5830).

I'm getting 723 MH/s from 300W (2.41 MH/J) but I spent extra on a good PSU and on aftermarket coolers for my two cards.

530W for 800 Mh/s seems a little high but not grossly so.  I have a friend with a 6990 managing 800 MH/s at 520W and he's perfectly happy with this.  I'd guess you are overvolting your cards and if so and you can stomach coming down to 750 MH/s you might save 50+ watts by lowering the clocks and voltages on the cards.

Agreed.

SSH will not affect your temps.  RAM clock speed will.  Core clock speed will.

Card voltage makes a very big difference to temperature.  I'm able to get my temps down to 48*C and 35*C for 351 MH/s and 372 MH/s respectively (Sapphire HD5850 Xtreme cards) and a good 10*C of this was due to lowering the voltage below stock.

Good air flow is the most important factor (unless your cards are water cooled then focus on water flow ).  Think about the air currents in your case and experiment with fan locations and rig orientations (heat rises and has a small effect).  For each test give you rig at least 10 minutes to reach a stable temperature and try to keep the ambient temperature the same for all tests.

The most obvious reason why some of your cards are hotter than the others is because they are all lined up in the case and they are each sucking in air which is being heated by the card in front of it, a Graphics card centipede if you will.

The most effective remedy is to use PCIe risers and position your cards so that they all get good airflow.  If you can't do this then position fans to blow air towards the motherboard inbetween the cards (most cards including yours suck in air from the front and blow it out the sides).

After market cooling is another option but is not as cheap as PCIe risers.  Beware of the coolers which are 3 slots wide because you will then have practically no breathing room between the cards at all (something I learned the hard way).

Wow!  I'm impressed with the large savings you are making by tweaking your BIOS.  switching off firewire for 5W, man!  When I switched off firewire the only detectable difference was that the firewire port wasn't lit up, the power drain didn't change at all for me.

In this case I stand corrected and do absolutely set about tweaking your BIOS as much as you can.  But even so, the efficiency of the PSU could save/cost you 50W so is worth considering at least.

I'm afraid I can't advise on PSUs.  I only have one miner and I bought my PSU with noise as the primary consideration.  For me the PSU is critical in building a mining rig (along with the cards and motherboard).  The chip, drives, RAM, OS, et cetera are good to get right but of secondary importance.  I personally never skimp on the power supply but for maximum profit you may want to go with something cheapish which can manage 86%+ at your mining load.  A typical positive attribute and one many people pay a lot for is for the PSU to be at least x% efficient at any load - give that you really don't need this shop around and look at the efficiency vs. load graphs for cheap PSUs.

As for voltages on the cards it really depends what you want out of your system.

• To maximise MH/J you will probably want to drop the voltage quite a bit (0.1V or more) and a little more still if you have many cards in one rig.
• To maximise profit then you may want to overvolt your cards by a small amount (0.05V is my guess) but this is assuming you have cooling sorted.
• To maximise MH/s then I find 1.25V is best (higher voltages hurt my stability due to the increased heat).

Of course, there are many factors: The cost of power; The market value of bitcoins; Your cooling; Your noise tolerance; How good at overclocking your particular cards are; even the OS!  I would get a PSU which is efficient at the power you expect to draw and for +/- 30%, use a power meter to measure power consumption at the wall, find your maximum stable clocks for a range of different card voltages and note power consumption/temperature/noise each time, and the calculate what works best for you.

If MH/J is important to you then let us know how efficient you manage to make your rig.  Mine is at 2.41 Mh/J.  My PSU is 89% efficient at my load, I'm running off of a HDD, and I'm running two Sapphire HD5850 Xtreme cards.

I don't know about 5830s but with my 5850s I can underclock the RAM using AMDOverdriveCtrl.  Create a profile and set the RAM clock speed to the speed that you want and then apply it with.

('-i 0' is the interface, run 'AMDOverdriveCtrl --help' to get a list of your interfaces.)

This worked for me with early versions of Catalyst (11.5 and earlier).  I was not able to overclock above the BIOS maximum but could go below the minimum with this trick.  It is important that you set the same RAM clock setting for all three performance levels.

I meant that I configure my cards to not drive a display at the same time as mining.  If I plug a monitor into one of my cards I get "no-signal", rather than some console or GUI mode.  I drive and monitor the cards by ssh.  This is the logical conclusion to tricks such as 'disabling desktop effects' or 'disabling Flash hardware acceleration'.  The tricky bit is still being able to modify voltage and clocks (aticonfig is not very sophisticated, makes bad assumptions, and is closed source) but it's at least possible with my cards.  I'd go into more detail but it's very much off topic for this thread.  I have no idea how to do this in Windows I'm afraid.

I honestly think that there is a big difference between 80% and 90% efficiency, enough to dwarf all other considerations apart from the voltages of the cards.  Ideas such as:
- Using 1 stick of RAM rather than 2;
- Underclocking the CPU;
- Booting from a USB drive or over ethernet to save on a HDD;
- Switching off USB/SATA/Firewire/Sound in the BIOS if you are not using them,
are all near-enough trivial compared to getting the voltages of the cards right and a high efficiency PSU, particularly if your system uses more than 1 card.


This is just wrong.  An efficient PSU is more important if the rest of your system is inefficient.  You will be saving more watts with the upgrade.

Beware: I'm about to start doing calculations and I have a reputation for making bad assumptions.  Take this with a pinch of salt.

Suppose you have a system consuming 400W with a PSU which is 80% efficient at that power draw (so it's pulling 500W from the wall and is only able to provide 80% of that to the system).  If you swap the PSU out for one that can provide 400W at 90% efficiency then you power draw at the wall will fall to 444W, saving 56W.

Of course, depending how much power costs you and how long you expect to be running the miner for it may not be cost effective to go with a more efficient PSU.

Awesome!

I wonder if anyone's built a rig with K'Nex?

Wouldn't the draw at the wall be 25% greater?  If a PSU draws 1250W from the wall and is 80% efficient then it is supplying 1250W * 80% = 1000W to the miner.

Honestly I don't know what the efficiency rating on the PSU means but if by 80% they mean that only 80% of the power it receives is passed on then the draw at wall would have to be 1250W.

Yes, I was getting around 402 Mh/s (970 MHz) on one card and 423 Mh/s (1020 MHz) on the other completely stable.  Undervolting by 0.1V took me down to 351 Mh/s (845 MHz) and 372 Mh/s (895 MHz) respectively and I'm still testing for stability so they may drop a little more (currently they've been running like this for over 48 hours).  The bulk of this hashing power comes from high clock rates and although I believe I have lucky cards I'm fairly sure this also has something to do with removing the GUI.  If you are using Linux then feel free to PM me for details.

Because your clocks are lower to start with you will probably not lose as much hashrate as I did with a 0.1V undervolt.  Also, because your clocks are lower, your cards will be drawing less power anyway so you might manage around 1.0125V on all three cards at under 450W.  Just as with overvolting, undervolting will give you diminishing returns; as you drop the voltage in 0.0125V increments you'll find that you'll lose larger and larger chunks of hashing power and save smaller and smaller chunks of power.  To maximise MH/J you'd probably have to go lower than 0.9875V but to maximise profit you'd be better off with something near stock voltage.  If you can handle the heat then I'd recommend going with the highest voltage that your PSU can handle comfortably and efficiently (try to find a graph of the power efficiency of your PSU).  Depending on the efficiency of your base system I'm guessing 1V or 1.0125V.

Yes, it's good to remind people every so often on this thread that maximising Mh/J is not the same as maximising profit.  I've seen many people complain about this on the few threads which talk about maximising Mh for fixed hardware (or hardware+clock rate).

This is a very efficient setup.  Might I enquire as to what your voltages and clock rates are for this silent mode?  Noise is the most important factor for me.

Same here.  The 'Bitcoin hits version 0.3' story.  I had some trouble getting started too (TOR/privoxy/blockcontrol related) but stuck with it and generated 5 blocks in a few months with one core of my Intel Atom D510 (the other was F@H).  As with many of the people here in June I 'bought into the hype' and bought some BTC at bitcoinmarket while the value was at an all-time high of 0.012 USD per BTC! (up from 0.008 USD the previous day).  I continued to buy as the price surged upwards to but no longer had the stomach for it when it hit 0.05 USD per BTC.

I'd forgotten about BTC for a good number of months and was reminded of it in April (I think by a friend that I'd given 150 BTC to to get him started) and was totally shocked by the market value.

My favourite bit of my story is when I accidentally entered the total USD I was willing to pay for a 200 BTC sell order but was supposed to enter the price per BTC and my order ended up being filled by the system at a ridiculous 4.5 USD per BTC!  Fortunately, Dustin was able to reverse this 200-times market value bid for me.

You're welcome.  450W for 3 cards looks perfectly possible.  Just be aware that your maximum stable clocks will reduce quite a lot. (I lost 130 MHz from both cores with a 0.1V undervolt).

I finally took the time to give the BIOS a good stripping.  I took out everything non-essential (USB, SATA, onboard graphics, firewire, sound, and various motherboard bits) and lowered the voltages and clocks of everything I could as far as they would go (a 0.3V undervolt for CPU, RAM, and NB).  I didn't touch the special DRAM pages as they really looked quite hard.  My CPU is now at 6% usage but I don't notice any sluggishness as one might expect.  The total power saving of this was 4W so I'm now down to 300W.

Unfortunately, my card at 850 MHz crashed this morning (it lasted for 36 hours or so) so I've had to pull the clock rate back to 845 MHz and this has cost me 2.1 MH/s.

My system is mining at 722.8 MH/s (+/- 0.2) and draws 300W (+/- 3W depending on temperature) giving me 2.41 MH/J.  I've decided that the HDD stays so this is probably as far as I'm going to go with this.

In summary it seems that the most important factors for determining MH/J are:

1) Low voltages on the cards.
2) Efficient PSU (just increasing mine just from 89% to 92% would save me 10W).

Thank you all, I've sent my tips to those whose comments have helped me.

I'm on Linux and I'm still able to get these error messages, even with earlier versions of the phatk patch from this thread.  For me they start happening if my clocks are too high and become more frequenty as I increase the clocks.

For my 5850 at stock volts, 1015MHz doesn't generate any errors and with 1020 MHz they are very occasional but become more frequent as I ramp up to 1035MHz (my card always freezes at 1040MHz).

When I overvolted my 5850 to 1.25V and took it to 1110 MHz for 3 hours I got a few but noticed that I was generating good shares with the same OpenCL work request as was throwing errors so it seems the error doesn't invalidate the whole second and likely just a very small portion of it.  I have a screenshot here.

At lower voltages I never see these errors.  Either the card runs error free or crashes and the MHz line dividing these two states is pretty fine in my experience.

It's interesting that the kernel version might affect the frequency of such errors but they don't bother me.

Thanks for this latest kernel patch Diapolo.  My clock rates increased by 0.3 MH/s each.  2x5850: 722.2 MH/s -> 722.8 MH/s.  I'll send another small tip.

I did save changes and reboot to read the new power drain but didn't go as far as booting the system in full and running the mining software (would have taken hours to test all of the different settings).  I did notice that the power savings were negligible so I didn't bother but I have to admit the 4W in my 304W read-out bugs me.

I'm not up on AMD or Intel stuff.  I've not touched PCIe buses, northbridge, southbridge, et cetera.  Do you think that more than a 1-2W saving could be made here?

I'd love to know more about this.  For me I've found I do better if I start from cold and dive straight into a high clock rate.  It may be a good idea to make my idle voltage the same as my performance voltage to ease the transition.  I've found that changing clocks while mining can cause the drivers to freeze if the clocks are high enough.

I once left my good card running at 1025 MHz at stock volts for 12 hours and it was fine.  Then I change the clock rate to 1020 while mining and it crashed instantly.  I reset and started it at 1020 MHz and it was fine.  Indeed, I can often get 1030 MHz like this and it runs fine once it's there.  I can sometimes get 1035 MHz like this but even if I do the card will crash after a few hours.  I get more and more kernel warnings the higher I take the clock but even at 1030 MHz the extra 5 MHz is a greater boost to profit than the hardware errors are a drain.

Perhaps this has something to do with temperature.  When the cards are at idle they are at much lower temperatures and can probably take the jump more easily.  I've found that if I turn my fans up I can reach higher clock rates.

I wrote a script to change my cards clock rates on the fly if the temperature left a certain range but all too often it would crash the cards so I stopped using it (I wish I could control my fans in software).

Anyway, I'm certainly going to increase my idle voltage from 0.95V to 0.9875V so that at least the voltage isn't changed when I start/stop mining.  This could be a good stability tip for people who do pool mining and occasionally have no work to do due to server issues.