How should I cool my 5970 VRMs?

[BOARBEAR]

Performance does not scale linearly with clock speed at all.  (try 500mhz and 800mhz and you will see)

Of course it does. Unless perhaps if you are using SDK 2.6.

A simply test shows it does not.

870Mhz  397Mhash/s 870/397=2.191
800Mhz 359Mhash/s 800/359=2.222
500Mhz 202Mhash/s 500/202=2.475

In theory it can't be linear, same reason why cpu does not scale linearly with core speed.
Don't spread false information

[1714066879]

1714066879

[1714066879]

1714066879

[1714066879]

1714066879

[cuz0882]

870 works great for me, 5970's stay under 70c. I've only had one pc that could run 910 and stay cool but it was only a 4 gpu rig. If they don't go over 80c I would not worry. Nice to shoot for 70c though. I think some people forget, not everyone is concerned with power consumption ratios.

Did you lose your 5970's stock fan cooler? Those Accelero Extreme coolers are terrible for mining. There not made for steady high temps.

[DeathAndTaxes]

Well, no. Voltage is only one part of power usage, higher clocked cards use more current than lower clocked cards at the same voltage. That's why overclocked cards get hotter even when not overvolted.

Yes, power consumption scales linearly with clockspeed. But performance also scales linearly with clockspeed.
Higher performance at equal MH/W for the cards, means higher efficiency for the overall rig since CPU, MB, RAM, etc remain the same.

Power consumption scales quadratically with voltage. Clockspeed -and  thus performance- does not. Therefore undervolting almost always gives better MH/W, but overclocking is completely sane from an efficiency POV.

Performance does not scale linearly with clock speed at all.  (try 500mhz and 800mhz and you will see)

I did and performance scaled linearly.  Not sure why you think it wouldn't.

A 5870/5970 gets roughly 0.46 MH (+/- .04) per Mhz.  That hold true from 500 Mhz to 1000 Mhz+.
Now for max efficiency you are going to need to find the sweet spot on memclock which varies based on coreclock.   If you see significantly less than that at any clock your settings are non-optimal or you sample is too small (should be at least 2000 shares, 5000 would be better). 

[DeathAndTaxes]

A simply test shows it does not.

870Mhz  397Mhash/s 870/397=2.191
800Mhz 359Mhash/s 800/359=2.222
500Mhz 202Mhash/s 500/202=2.475

In theory it can't be linear, same reason why cpu does not scale linearly with core speed.
Don't spread false information

I agree YOU should not spread false information.

Your numbers are garbage so likely you have suboptimal settings and the most likely culprit is memclock.

Do those results seem plausible to you. 


870Mhz  397Mhash/s 397/870= 0.41 MH per Mhz
800Mhz 359Mhash/s 359/800 = 0.44 MH per Mhz
500Mhz 202Mhash/s 202/500 = 0.46 MH per Mhz

So the card is getting more efficient at higher clock (and likely higher temp)?  Does that seem plausible to you?

Likely you normally run @ 870Mhz and have found a more optimizes memclock.  You sloppily moved core clock without modifying memclock introducing timing delays making the card less effective the further you move it from 870 Mhz.

For the record @ 500 MHz I get 225 MH (0.450 MH/MHz) and @ 820 I get 375 MH/s (0.457 MH/MHz).

[BOARBEAR]

A simply test shows it does not.

870Mhz  397Mhash/s 870/397=2.191
800Mhz 359Mhash/s 800/359=2.222
500Mhz 202Mhash/s 500/202=2.475

In theory it can't be linear, same reason why cpu does not scale linearly with core speed.
Don't spread false information

I agree YOU should not spread false information.

Your numbers are garbage so likely you have suboptimal settings and the most likely culprit is memclock.

Do those results seem plausible to you. 


870Mhz  397Mhash/s 397/870= 0.41 MH per Mhz
800Mhz 359Mhash/s 359/800 = 0.44 MH per Mhz
500Mhz 202Mhash/s 202/500 = 0.46 MH per Mhz

So the card is getting more efficient at higher clock (and likely higher temp)?  Does that seem plausible to you?

Likely you normally run @ 870Mhz and have found a more optimizes memclock.  You sloppily moved core clock without modifying memclock introducing timing delays making the card less effective the further you move it from 870 Mhz.

For the record @ 500 MHz I get 225 MH (0.450 MH/MHz) and @ 820 I get 375 MH/s (0.457 MH/MHz).

Your calculation is garbage.
202/500=0.404 not 0.46
397/870=0.45 not 0.41
These are just the multiplicative inverse of the number I gave above.
And I tested all above in identical situation except of the core frequency.

[DeathAndTaxes]

And I tested all above in identical situation except of the core frequency.

Oops on me reversing the order of the values.  Now that WAS foolish on my part.

I guess it still hasn't sunk in that the optimal memory clock at 500Mhz isn't the same as the optimal memory clock at 870 Mhz.  All you have proven if you are really bad at getting optimal performance from a 5870/5970 as your efficiency (in terms of MH/s per MHz) is about 10% lower than most miners get.

In case you ignored it from the prior post

"For the record @ 500 MHz I get 225 MH (0.450 MH/MHz) and @ 820 I get 375 MH/s (0.457 MH/MHz)."

and lastly (from prior post) this this stands:

A 5870/5970 with optimal settings gets roughly 0.46 MH (+/- ~.04) per Mhz.  That hold true from 500 Mhz to 1000 Mhz+.
Now for max efficiency you are going to need to find the sweet spot on memclock which varies based on coreclock.   If you see significantly less than that at any clock your settings are non-optimal or you sample is too small (should be at least 2000 shares, 5000 would be better).

[ssateneth]

And I tested all above in identical situation except of the core frequency.

Oops on me reversing the order of the values.  Now that was foolish.

Still I guess it still hasn't sunk in that the optimal memory clock at 500Mhz isn't the same as the optimal memory clock at 870 Mhz.  All you have proven if you are really bad and getting optimal performance from a 5870/5970 as you efficiency (MH per MHz) is about 10% lower than most miners get across the board.

Once again in case you ignored it from the prior post
"For the record @ 500 MHz I get 225 MH (0.450 MH/MHz) and @ 820 I get 375 MH/s (0.457 MH/MHz)."

and lastly

A 5870/5970 with optimal settings gets roughly 0.46 MH (+/- ~.04) per Mhz.  That hold true from 500 Mhz to 1000 Mhz+.
Now for max efficiency you are going to need to find the sweet spot on memclock which varies based on coreclock.   If you see significantly less than that at any clock your settings are non-optimal or you sample is too small (should be at least 2000 shares, 5000 would be better).

Thank you for laying the beatdown on this poor misguided soul. It was deserved.

[P4man]

Performance does not scale linearly with clock speed at all.  (try 500mhz and 800mhz and you will see)

Of course it does. Unless perhaps if you are using SDK 2.6.

A simply test shows it does not.

870Mhz  397Mhash/s 870/397=2.191
800Mhz 359Mhash/s 800/359=2.222
500Mhz 202Mhash/s 500/202=2.475

In theory it can't be linear, same reason why cpu does not scale linearly with core speed.

Of course CPU bound cache resident benchmarks do scale linearly with clockspeed. Try benchmarking Coremark, its linear from 0.1 Hz to 1 GHz.

As for your results, like I said, unless perhaps you are using SDK 2.6. In that case, you become memory bandwidth bound. Hashrate will still scale linearly with gpu clock but only if you can scale memory clock as well.

If you use SDK 2.1 or 2.4,  hashrate is not bandwidth bound, in fact you typically get a small increase as you lower memory clocks down to ~150-200 MHz. Yes negative scaling. Check the link and try yourself. But the result is perfectly linear performance scaling with GPU clock.

[johnyj]

Stay stock voltage without too much overclock are important IMO

With stock voltage and decent overclock, machine can run several weeks without interrupt, but if it crashes from time to time because of the stability issues caused by high temp or high frequency, then the time spent trouble shooting and fixing things will easily cost you many BTCs

[DeathAndTaxes]

Stay stock voltage without too much overclock are important IMO

With stock voltage and decent overclock, machine can run several weeks without interrupt, but if it crashes from time to time because of the stability issues caused by high temp or high frequency, then the time spent trouble shooting and fixing things will easily cost you many BTCs

Good point and as you get more cards and more rigs it makes more sense to become even more conservative.  That was something I learned the hard way.  You can't run 38 GPUs (19x 5970) at the same clocks you can run a single card that you can baby.  Well not without losing your sanity.  Not only are there stability issues but there are power and heat issues to which only get harder as the farm scales.  Dissipating 1KW of heat.  Meh.  Get a good fan.  Dissipating 5KW?  A little tougher.

[The-Real-Link]

Stay stock voltage without too much overclock are important IMO

With stock voltage and decent overclock, machine can run several weeks without interrupt, but if it crashes from time to time because of the stability issues caused by high temp or high frequency, then the time spent trouble shooting and fixing things will easily cost you many BTCs

Good point and as you get more cards and more rigs it makes more sense to become even more conservative.  That was something I learned the hard way.  You can't run 38 GPUs (19x 5970) at the same clocks you can run a single card that you can baby.  Well not without losing your sanity.  Not only are there stability issues but there are power and heat issues to which only get harder as the farm scales.  Dissipating 1KW of heat.  Meh.  Get a good fan.  Dissipating 5KW?  A little tougher.

*drools*.  Yeah same.  I've kept my clocks reasonable and I'm still having to troubleshoot a good amount here and there.  Cards get buggy, network crashes, etc.  I'd much rather be slightly conservative but have the systems run for a week or more without a hitch than reset things every day.  When you have that many machines you'd think it'd be a given.

[johnyj]

Unfortunately even at stock clock and voltage, I still get some crash recently when running cgminer (never had issues when running phoenix miner), so a stable miner is also critical

It's also time to look at VRM cooling. I have some very efficient heat pad came with water blocks for GTX295 years ago, they could reduce the VRM temp by at least 10 degrees. As I understand, the VRM location on 5970 card are very weak, maybe a heatsink on the back of the plate will help