Maximising MH/J

[teukon]

Since you need network for mining you could ping the machine to see if it's running (not sure if the additional wattage used by the machine you're pinging from from time to time will average out more than what is consumed by the LEDs though)

Lol

[1713420400]

1713420400

[1713420400]

1713420400

[1713420400]

1713420400

[1713420400]

1713420400

[1713420400]

1713420400

[teukon]

Make sure your power supply is properly sized, peak efficiency is about 50% load. You probably should be running a 600W supply with an 80+ gold rating. It's tough to pick the proper supply without plugging it in and measuring it.

Switch your power supply from 120V to 240V. Most power supplies are ~2% more efficient at the higher voltage. If memory serves, power companies charge based on whichever phase is drawing the most power. If you have two lightbulbs connected to the same phase you will pay twice what you would pay if the lightbulbs were on separate phases. I can't find a reference for this though. I did find a page discounting this. I guess I'll have to run the experiment.

Make sure you are plugged in as close as possible to the breaker panel to avoid losses in your wiring. My project for the weekend is to drop a new 240V outlet into my basement (so I don't pop a breaker when I run the microwave.) Power strips, though useful, are energy wasters also.

http://www.anandtech.com/show/2624/3

My power supply situation is not optimal for minimising power consumption.

I'm using a Scythe Chouriki 2 850W power supply (80+ silver certified) and it claims to be about 89% efficient at a load of 300W, see http://www.techpowerup.com/reviews/Scythe/SPCR2-850P/5.html.  I was originally using a 650W OCZ supply of some kind but it was quite a bit louder than any other part of my system.

For the purposes of this, the psu is plugged directly into the wall.  In reality it is plugged into a 4-gang which is plugged into a 10 meter long 2-gang but I've put the power meter into the 4-gang and am only interested in the power draw at this point.  When I move I'll probably manage to arrange to have it plugged into the wall directly.

The voltage 'at the wall' is 243.6V.

[teukon]

That is amazing what voltage is a good voltage to keep cards at to optimize energy.  Saw in an older post you were running at 1.01v/900/300 with 735 mH/s.

I'm still not very sure about minimising voltage but if you are trying to maximise MH/J then I'm almost certain that you need the card voltages below 1V for 5850s.  Of course, you may be willing to pay more than the cost of power for your Bitcoins and so it may make sense to increase the voltage.

When I posted before I was not aware that the VRMs on the 5850s handle voltage in multiples of 0.0125V (I'm not certain of this but fairly confident).  Thus, I was really running at 1.0125V/900/300.

My cards are currently running at 0.9875V/850/300 (352.7 MH/s +/- 0.1) and 0.9875/895/300 (371.6 MH/s +/- 0.2) for a total of (724.3 +/- 0.2).  My power consumption at the wall with the cards running is 304W +/- 2W.  When I close the mining software the cards fall back to 157MHz@0.95V and the power consumption drops to 111W +/- 1W.  I'm still running from my hard-drive and preventing it from going into any kind of power saving mode with bitcoind so there is a saving I'm not applying here too.

I've not had time to verify the stability of this setup but the cards did run all night at 850 and 890 respectively.  895 is a new venture.

Running a card at 0.9875V at 850 MHz yields similar MH/J as running it at 1.0125V at 900 MHz (taking system load into account, if you just count the cards then going up to 1.0125V and clocking up to 900 MHz actually lowers my MH/J).

Honestly, finding the correct voltage and clock rate for the best stable MH/J requires testing as every card is different.  I'll point out that my cards both seem to be pretty good at overclocking so you may find it difficult to even get above 800 MHz at 1V.  I'm using Linux, the cards are configured to give no signal, and I'm using Zalman ZF3000A coolers, and the ambiant temperature where I am is currently peaking at 21*C so I have some natural advantages when it comes to stability too.  My good card clocks to 1025 MHz stably at stock voltage and went for 3 hours at 1110 MHz at 1.25V (461.1 Mh/s) so it's not bad even compared to 5870s!  I'm using everything I learnt in competing for high hash rates to maximise my efficiency at low voltages.  PM me if you'd like more details.

[mikeo]

Very informative thread -- thanks.

[vapourminer]

making sure the PSU is drawing fresh (ie not heated from the cards) air may help the PSU efficiency. Im unsure but I do know the hotter they get the less power they can deliver. whether cooler = slightly better efficiency is true or not I dunno but even 1 or 2 % = 3-6 watts on your setup. 

and man, 300 watts that whole system on those two cards? sweet! I pull 300 watts from the wall on a P4 prescott/single 5830 rig. no DVD, other cards, one HD.  Ill have to try undervolting the 5830. cant mess with CPU/mem stuff as that rig is a HP/Compaq business computer work gave me when they retired it. no BIOS options for the good stuff. that thing sat for ages in the basement, never thought Id would actually use it lol.

[steelhouse]

I stopped my cards and the power went down to 98 watts with the stock fans still on.  I did the test earlier in year before turning cards on and the base power was 78-86 watts.  However, the cpu-z ghz won't show 3.3 ghz but something less.

stock bios, stock settings
2500K intel sandy bridge, stock clockings
Gigabyte Z68MA-D2H-B3
2 dummy plugs and using intel graphics
8 gb ram (2 sticks of 4)
no hd, SSD
thermaltake 750W 80 plus bronze

6870 1.075/900/300 277 mh/s
6950 1.075/870/300 357 mh/s

total 376 watts  = 1.68 Mh/J

[teukon]

making sure the PSU is drawing fresh (ie not heated from the cards) air may help the PSU efficiency. Im unsure but I do know the hotter they get the less power they can deliver. whether cooler = slightly better efficiency is true or not I dunno but even 1 or 2 % = 3-6 watts on your setup. 

and man, 300 watts that whole system on those two cards? sweet! I pull 300 watts from the wall on a P4 prescott/single 5830 rig. no DVD, other cards, one HD.  Ill have to try undervolting the 5830. cant mess with CPU/mem stuff as that rig is a HP/Compaq business computer work gave me when they retired it. no BIOS options for the good stuff. that thing sat for ages in the basement, never thought Id would actually use it lol.

My PSU is certainly getting fresh air (not a problem with a rig mounted on a board).

I just discovered that increasing all the fans to maximum increases the load by 7W total right away but if I leave the system running for 15 minutes then the power consumption slowly drops until eventually I'm actually saving 6W (so down to 298W total)!  It seems the temperature of the gpus has a pretty serious effect on power consumption.  Unfortunately I cannot make much use of this because the miner is too loud to run like this 24-7.  It's a shame, 2.43 MH/J would have been awesome.

Undervolting my cards and maxing the fans really did a number on my temps: 44*C and 36*C respectively (the 44*C is sucking in warm air from the other card).  At 36*C and churning out 371.6 MH/s you might guess my 5850 was under water .

[teukon]

I stopped my cards and the power went down to 98 watts with the stock fans still on.  I did the test earlier in year before turning cards on and the base power was 78-86 watts.  However, the cpu-z ghz won't show 3.3 ghz but something less.

stock bios, stock settings
2500K intel sandy bridge, stock clockings
Gigabyte Z68MA-D2H-B3
2 dummy plugs and using intel graphics
8 gb ram (2 sticks of 4)
no hd, SSD
thermaltake 750W 80 plus bronze

6870 1.075/900/300 277 mh/s
6950 1.075/870/300 357 mh/s

total 376 watts  = 1.68 Mh/J

You are getting much lower base-system power consumption than I am (I'm at 111W with both cards idle).  I'm sure the HDD is one culprit but it's not enough on it's own to explain this gap.  I'm also only running 1GB of RAM.  I can't even pretent to compete with this, well played.  My high MH/J is probably partly due to the efficiency of the 5850s and partly because my cards are running at a much lower voltage (0.9875V).

[steelhouse]

It must be intel.  There are even lower watt intels out there.  I will try to go 1 volt, but with prices a $13, any mh/s at about 0.3 Mh/J is making money.

[teukon]

It must be intel.  There are even lower watt intels out there.  I will try to go 1 volt, but with prices a $13, any mh/s at about 0.3 Mh/J is making money.

Could be.  My main computer consists of an intel chip and board and consumes about 55W total at 100% system load.  I'd expect to lose about 70-80 MH/s by taking both cards from 1.075V to 1V so and for your Mh/J to increase significantly but if you are profitable at 0.3 Mh/J then surely the 70-80 MH/s is worth quite a lot more than the power you will save.

My personal motivation for undervolting is to decrease noise levels and I know I would be more profitable at stock voltage.

You are doing well to be profitable at 0.3 Mh/J.  I'd be at a loss below 0.5 MH/J, but I guess that's just power costs.  Ah well, I can't have all the advantages I guess.

[steelhouse]

I kept getting crashes looking for sweet spot, but

1000/800/400 both, too many crashes so I stopped here

6950 2.59 mh/J 328 mh/s
6870 2.85 mh/J 246 mh/s

311 watts wall  1.84 mh/J
98 watts base

[trentzb]

6950 2.59 mh/J 328 mh/s
6870 2.85 mh/J 246 mh/s

How did you get these numbers? Did you measure cards individually?

[Pipesnake]

Dual gpu cards are more efficient than having two single gpu cards.  Trading in your two 5850s for a single 5970 or 6990 will increase your mh/s and lower your energy use.

Mobile gpus are more efficient still.  My dream rig is a desktop mobo and power supply with 10 6990m gpus.  Obviously it is not cost efficient to build one (yet).

[teukon]

I kept getting crashes looking for sweet spot, but

1000/800/400 both, too many crashes so I stopped here

6950 2.59 mh/J 328 mh/s
6870 2.85 mh/J 246 mh/s

311 watts wall  1.84 mh/J
98 watts base

Nice.  How have you accounted for the proportion of the base wattage which is going to the cards in idle mode?

I found that the maximum overclock is pretty much a hard line at lower voltages.  My cards are at 850MHz and 895MHz.  If I clock either of them up by 5 MHz they crash instantly (tried twice with both).  They've been running at 850MHz and 895MHz with no problems for 36 hours now.  I experienced the same thing at 1.0125V on my slower card where 900MHz was very stable and 905MHz always crashed instantly.  I may have to start moving by 1 MHz intervals.

At higher voltages I get all sorts of nonsense.  Sometimes my good card will run fine at 1035MHz for 36 hours.  Sometimes it will crash when I try to run it at 1025MHz.  Sometimes I'll get kernel errors.  All at about 55*C!  I just don't understand it.

Note: My cards are very much dedicated to mining and I don't think any other program is trying to use the cards.  In particular, when I plug a monitor into a card I get no signal.  I've seen some people trying to run a GUI at the same time which can seriously affect stability.

[Dargo]

teukon,

Wow, you've got me beat big-time. I'm hashing at a similar rate with two 5850s (not the extreme version), but pulling a little under 400 watts from the wall (measured with Kill A Watt). Since you aren't booting from usb, I assume you have tried disabling legacy usb support? I've read that this is a bit power hungry (but of course don't do this if you use a usb keyboard for accessing bios). Also, if you happen to be using a usb wireless card, I discovered by accident that switching to an ethernet connection to my modem dropped power consumption by about 5 watts (and the wireless card is still plugged in). Did you see the biggest drops from undervolting the CPU and GPUs? I have an Athlon II 250 Regor CPU and my PSU is about 5% less efficient than yours - that is probably around +40 watts right there. Still, given what you have achieved, I ought to be able to get down to around 350 from the wall while staying above 700 Mh/s, which would be good enough for me.

[steelhouse]

6950 2.59 mh/J 328 mh/s
6870 2.85 mh/J 246 mh/s

How did you get these numbers? Did you measure cards individually?

I have the base of 98 watts turn one card on write watts, turn card off and turn other card on write watts.  subtract 98 from them and (mh/s)/watts = mh/J.  so I don't measure individually and I am not 100% of my baseline because it was once 86 watts.  I am amazed the slight lower voltage really cuts the watts.  I know I can do better.

[trentzb]

That would not be the total draw per card then. To find per card total consumption you would have to remove all cards, take measurement, insert a card, take idle measurement, run miner take final measurement. I saw ~20-~30 W per card at idle doing this which would drop your mh/j number down to where it sounds right. Your ~98W includes power consumed by GPUs at idle...right? They don't idle for free.

[xurious]

Reducing the RAM clock and worksize lost me far too much hash rate to be viable.

Reduce ram clock to about 1/3 your core clock, the 300mhz is a reference everyone throws out. I've done this on all my cards, it doesn't effect the performance. Even if you have to run it a bit higher, say 400mhz, that should be more than plenty and it's saving you 5-700mh/s of memory clock that isn't getting used.

The 6xxx series work better with 256 work size.

Haven't read through the entire posting yet, but have you tried underclocking/volting your hypertransport? Slowing down the pci/e lanes? I mean... if your card is using 16 lanes, and you only "need" one lane, then couldn't you reduce the lane to 1/16 the original speed and still maintain overall bandwidth?

Use services.msc in windows to disable stuff your pc does not need to be running. Could shave cpu cycles allowing you to slow your cpu down even further.

Edit: I get the idea you are not running windows. So scratch that last line.

[teukon]

teukon,

Wow, you've got me beat big-time. I'm hashing at a similar rate with two 5850s (not the extreme version), but pulling a little under 400 watts from the wall (measured with Kill A Watt). Since you aren't booting from usb, I assume you have tried disabling legacy usb support? I've read that this is a bit power hungry (but of course don't do this if you use a usb keyboard for accessing bios). Also, if you happen to be using a usb wireless card, I discovered by accident that switching to an ethernet connection to my modem dropped power consumption by about 5 watts (and the wireless card is still plugged in). Did you see the biggest drops from undervolting the CPU and GPUs? I have an Athlon II 250 Regor CPU and my PSU is about 5% less efficient than yours - that is probably around +40 watts right there. Still, given what you have achieved, I ought to be able to get down to around 350 from the wall while staying above 700 Mh/s, which would be good enough for me.

I have no idea but I've been told that my chip is very efficient so that must help a lot.

I think the bulk of my power saving may come from running the cards at low voltages.  At stock voltages I was using quite a bit more than 400W but with a 0.1V undervolt (1.0875V -> 0.9875V) on both cards 304W is my average draw.  Unfortunately, I had to lower the clocks on the cores to get stability back; 975MHz -> 845MHz and 1020MHz -> 895MHz respectively.  I was running at 850 and 895 but my 850 card crashed after 36 hours.  What voltages are your cards at?

I haven't disabled USB in the BIOS despite the fact that I have no USB in use.  I have no SATA, sound, firewire in use either but I left them on in the BIOS because I didn't see a difference in the power draw between enabling them and disabling them (I only checked power consumption in the BIOS though).  All I changed in the BIOS was the CPU voltage, CPU and RAM clocks, and I set the motherboard to be always switched on.  Given all the people talking about tweaking the BIOS I think I'm going to have to go back and have a proper look at this.

I also keep thinking about the possibility of doing away with the hard-drive but there's a fair bit of work involved in unhooking this for what amounts to 83p (1.34 USD) per month.

[teukon]

Reducing the RAM clock and worksize lost me far too much hash rate to be viable.

Reduce ram clock to about 1/3 your core clock, the 300mhz is a reference everyone throws out. I've done this on all my cards, it doesn't effect the performance. Even if you have to run it a bit higher, say 400mhz, that should be more than plenty and it's saving you 5-700mh/s of memory clock that isn't getting used.

The 6xxx series work better with 256 work size.

I'm at 300MHz.  I meant that decreasing to 180MHz and dropping the worksize from 256 to 128 seems to do me no good no matter what my other settings are (Sapphire HD5850 Xtreme).

Haven't read through the entire posting yet, but have you tried underclocking/volting your hypertransport? Slowing down the pci/e lanes? I mean... if your card is using 16 lanes, and you only "need" one lane, then couldn't you reduce the lane to 1/16 the original speed and still maintain overall bandwidth?

No, I originally tried to underclock/undervolt everything as sensibly as I could but ended up but couldn't boot at all and ended up having to reset the BIOS/CMOS/whatever.  Since then I've just touched the CPU and RAM setting and have been afraid to touch any of the dozens of other settings for breaking my board (I have no experience with this kind of thing).  If you can give me a little more detail about this I'll try it and see if it helps.

Use services.msc in windows to disable stuff your pc does not need to be running. Could shave cpu cycles allowing you to slow your cpu down even further.

Edit: I get the idea you are not running windows. So scratch that last line.

Yep.  You can safely assume that I'm not running any 'services' at all.  Got a daemon or two running but powertop thinks they're all benign.  Linux is much easier to configure with respect to power consumption than the BIOS because of all the helper tools and documentation so assume that the OS related stuff is largely done.