Undervolting a 5870 at 0.95v

[jamesg]

Hey gigavps,   what kind of savings do you get from the rig going from same clocks at stock to the new .95V?  Is there a substantial wattage savings?

I am thinking of starting to undervolt my rigs as electricity is killer but stability is real concern. 

Not sure to be honest. I don't have a kilawatt that works on 208v. My main concern with undervolting is better stability and longevity of the cards. Also seeing a decrease in my power bill is nice.

[1714124624]

1714124624

[rograz]

Power draw increases exponentially with voltage so even a seemingly small decrease of 0.1 can be quite a big efficiency increase. The reduced frequency the cards can handle with the new voltage often follow a more linear scale.

[Littleshop]

Yeah that sounds like a pretty good cruising speed.  I wonder if these can be volted down any further?  On 5970s you can go down to .90v (610/200) for 550 MH/s, and possibly even lower on the voltage (in the .86-.88v range).  Has anybody had any luck with sub-.95v on 5870s?

I run my 5870s at 650/150/.899v currently for around 295Mh/s.

I tried the 750/150/.95v on a 5x5870 rig and things crashed before they even got started.

About 1/2 my cards can take 700/150/.95v

I use this as my 'summer' setting for some of my miners.

[johnyj]

I have 2 under-volted 5870's.

One at 840/155 and 1v = 384 mh/s
The other at 755/155 and .95v = 347 mh/s

If I try to set the first card to .95v I have to set the engine clock too low to around 400, otherwise it crashes.
The 2nd card is happy as can be at .95v

That's a 5% drop in power usage and 10% drop in hash rate, not very attractive IMO

But of course if you have plenty of hash power and want to reduce the noise/heat, it will help

[TheHarbinger]

I have 2 under-volted 5870's.

One at 840/155 and 1v = 384 mh/s
The other at 755/155 and .95v = 347 mh/s

If I try to set the first card to .95v I have to set the engine clock too low to around 400, otherwise it crashes.
The 2nd card is happy as can be at .95v

That's a 5% drop in power usage and 10% drop in hash rate, not very attractive IMO

But of course if you have plenty of hash power and want to reduce the noise/heat, it will help

All my 5870's are set at 1.000v and 825/150 clocks.  So each rig (5 cards) gets about 1850Mh/s @ 650W.

[jamesg]

I have 2 under-volted 5870's.

One at 840/155 and 1v = 384 mh/s
The other at 755/155 and .95v = 347 mh/s

If I try to set the first card to .95v I have to set the engine clock too low to around 400, otherwise it crashes.
The 2nd card is happy as can be at .95v

That's a 5% drop in power usage and 10% drop in hash rate, not very attractive IMO

But of course if you have plenty of hash power and want to reduce the noise/heat, it will help

All my 5870's are set at 1.000v and 825/150 clocks.  So each rig (5 cards) gets about 1850Mh/s @ 650W.

Do you mind running a rig at 800/150/.959v as seeing what kind of power draw your are getting?

[TheHarbinger]

I have 2 under-volted 5870's.

One at 840/155 and 1v = 384 mh/s
The other at 755/155 and .95v = 347 mh/s

If I try to set the first card to .95v I have to set the engine clock too low to around 400, otherwise it crashes.
The 2nd card is happy as can be at .95v

That's a 5% drop in power usage and 10% drop in hash rate, not very attractive IMO

But of course if you have plenty of hash power and want to reduce the noise/heat, it will help

All my 5870's are set at 1.000v and 825/150 clocks.  So each rig (5 cards) gets about 1850Mh/s @ 650W.

Do you mind running a rig at 800/150/.959v as seeing what kind of power draw your are getting?

Can't really do it, since I have a mix of different vendor cards on each rig and they are running headless.  Well, I mean technically, I could do it, but the thought of taking a rig off line to intentionally make it unstable for testing, isn't very appealing to me.  Every card I have seems to be happy at 1.000v and 825/150 clocks.  I'm sure some could go to higher clocks and/or lower voltage, but I'm not going to muck with them just to squeeze out another 20Mh/s or save another 10 watts. 

[johnyj]

On 5870, there are only 3 effective voltage, distributed in range:

0.95v
0.96v-1.06v
1.07v-1.17v

0.96v and 1.06v actually gave same readout on power meter, so you may run 825Mhz on 0.96v, but that will not improve efficiency

[Gomeler]

Anyone take a multimeter to their cards to see how fine-tuned the voltage steps are? I know Afterburner is accurate on 5870s(I measured it back in the day, it is good up to 1.55vgpu on the 5870). But, I'm not sure how cgminer adjusts voltages and if it is as accurate.

[italeffect]

On 5870, there are only 3 effective voltage, distributed in range:

0.95v
0.96v-1.06v
1.07v-1.17v

0.96v and 1.06v actually gave same readout on power meter, so you may run 825Mhz on 0.96v, but that will not improve efficiency

I have found the same thing measuring wattage at the wall. (except I thought there was a step in there at 1v as well) You always want to be at the highest voltage in the bracket and whatever the highest clocks that will support.

That's a 5% drop in power usage and 10% drop in hash rate, not very attractive IMO

I didn't include the wattage so you can't draw that conclusion. I don't have the per-card wattages at the wall (cause I don't feel like interrupting mining to measure) but based on extensive earlier testing dropping to 1v and .95v on some cards I'm up to about 2.85 mh/s per watt, vs. earlier at 2.3, 2.5 per watt etc.

(1 rig = 1x 6950 unlocked, 1x 5970, 2x 5870 = 1767 mh/s @ 619 watts) And the "efficiency" could actually be much higher except for that 6950 is at stock voltage. As soon as I flash the bios and undervolt it the total should be even better.

If you are concerned about power usage like I am, it certainly makes sense to undervolt.
If the rig ever hangs and I have to reboot it maybe I'll measure the wattage on the individual cards at stock and undervolted to provide the hard numbers.

[jamesg]

Well this is interesting. I have two 5870s on miner100. I set them to run at 800/150/.959v but for some reason they are running at 850 core!

http://gigamining.com/mgpumon/

Anyone know why that would happen? Or better yet, I'm wondering if I can run all my 5870s at 850/150/.959v.   

[rograz]

That's a 5% drop in power usage and 10% drop in hash rate, not very attractive IMO

1v to 0,95v is not a 5% reduction in power usage, it's not linear it's exponential.

[zvs]

On 5870, there are only 3 effective voltage, distributed in range:

0.95v
0.96v-1.06v
1.07v-1.17v

0.96v and 1.06v actually gave same readout on power meter, so you may run 825Mhz on 0.96v, but that will not improve efficiency

it'd make the card run hotter, so the fans would spin faster?

so it would improve efficiency by reducing wear & tear on the card, fans, and whatever watts the fans take from spinning faster

[rograz]

2 ASUS EAH5870 V2 1gb cards i've got doesn't seem to be able to set anything below 1,06V for me, anyone else who got experience with undervolting these cards? Else I I'll just have to add them to the rigs I plan to take down when profitability goes down I guess :/

[Grinder]

Anyone know why that would happen? Or better yet, I'm wondering if I can run all my 5870s at 850/150/.959v.   

I doubt those read outs are correct. I have not been able to run any 5850, 5870 or 5970 GPUs reliably at more than 750 MHz at 0.95v, and 780 MHz at 0.97v.

[rograz]

Agreed Grinder, not a single one of 15 my cards running at 0,95V can do above 740 24/7 stable so it would either have to be one crazy bin or the card simply isn't running at that voltage.

What happens with the 5xxx card's is that if you set a voltage the card can't handle it will default to the next step above it or the stock voltage in some cases. At least this is how all of my cards seems to behave (hard to tell since some non ref boards doesn't support voltage monitoring in gpu-z)

Like the Asus boards I just mentioned a few posts earlier, they are both flashed to 0,95 but draw exactly as much and clock as high as when I undervolted them in windows to 1,06V.

[jamesg]

Agreed Grinder, not a single one of 15 my cards running at 0,95V can do above 740 24/7 stable so it would either have to be one crazy bin or the card simply isn't running at that voltage.

I now have 10 cards running 850/150/.959v! Two are running 825/150/.959v.

http://gigamining.com/mgpumon/

See rigs 100, 107 and 110. The 110 rigs has 5k shares on each card.

I think the .959v over .95v makes a big difference.

[this time]

It is possible to run a 5870 at 850/150/.95 but unusual. I have one that will do it. I've also had them that won't do 850 on anything less than 1.100. And if you got the 5870 cards from bens bargains they won't underclock well in my experience.

You have to always check with a kill a watt to make sure that the readings you are getting are actually dropping the power consumption. The software readouts are not always accurate in any of the programs, including gpuz. And if you have a serious farm it may not be worth going through every card and figuring out what it will do.

[DeathAndTaxes]

VRM are only capable of certain discrete voltage values.

So if 0.959v isn't a valid voltage the card will actually be running at voltage > 0.959V whatever the next higher step is.  Which according to johnyj (not verified by me) is 1.06V

So 0.959V, 0.96V 1.0V, 1.0123456789V it doesn't matter the card is running at the next higher valid step which may be 1.06V.

The only way to verify for sure is to pick a static clock (lower is better) and connect rig to watt meter (like kill-a-watt).  Change voltage and look for a change in wattage.  No change in wattage = no actual change in voltage.  Since wattage is going to have some variance anyways it may require measuring power instead (kWh) and time to get average wattage.

So something like

Set clocks to a static 700 Mhz.
Set voltage to stock.
Measure power for ~10 minutes. 
Divide power by exact time to get avg wattage.
Lower voltage and try again.

You will notice something like this

Voltage: 1.05V, 1.04V, 1.03V, 1.02V 1.01V = same wattage.  Then at some point the wattage will drop.  THAT IS THE DISCRETE STEP.

[rograz]

Was gonna post but D&T pretty much got it all and then some :p