Your big@ss Hash Rate might actually be hurting you

[bitlane]

This really only applies to those who overclock their video cards.

I realize that every discussion around here has to do with how high people can tune their HASH RATES to.......

BUT, has anyone ever paid attention to their SHARES/HOUR ?

Recently I have been using my average SHARES/HOUR in Guiminer as a 'fine-tuning' tool of sorts and have achieved measurable results.
The math speaks for itself.

By simply turning down the overclock by 1 or 2 Mhz, I can increase my SHARES/HOUR on a single card by about 50 to 70 extra SHARES/HOUR...and able to maintain it.
(HD 6870 @ 302MH/s, 980 MHz GPU/Mem, 1280mV, 290+ SHARES/HOUR in Guiminer).

I have tried to counter this by adding more Voltage to the card, but only saw more heat generated and not more shares.
The 1-2 MHz drop in clock speeds was enough to increase output and not affect overall HASH RATES by too much. The trade-off is more than worth it.

Are you 'riding the ragged edge' with your overclock ? time to re-evaluate things. Save heat, card life and increase output.

Post#2 for me....
I'll shut up now..lol

Cheers,
Allan.

---------------------------------------
Did it help ?
Feel free to send something my way - 1QGjSCJ1tcz1kXnqB4KPqcY8kZz2Guwbef

[1715250003]

1715250003

[1715250003]

1715250003

[1715250003]

1715250003

[1715250003]

1715250003

[1715250003]

1715250003

[1715250003]

1715250003

[tysat]

How long of a period are you taking to get shares/hour?  It can vary.

[Graet]

i have noticed ppl that overclock "to extreme" often create a lot os "stale" or "invalid" shares. dropping a few Mhz can improve your accepted share and lower stale share rates thus giving a better return
hope thats a help

[bitlane]

How long of a period are you taking to get shares/hour?  It can vary.

I used 2 different pools with LONG rounds and monitored each for 3 hours on 2 seperate occasions/rounds.

**I should clarify though**

I run 3 cards (all 6870's) in the test machine in question, with the lowest card starting approx 260+ and the top card of the 3 doing 290+ Shares/hour during the testing/monitoring, using the most basic OpenCL setup in Guiminer with "-v -w128 -f1".
The same card was not always the highest producer, the 3 cards all maintained 260+ and each would peak at over 290/hour for periods of 30mins or more before leveling off to 280-ish.

Prior to this, clocking the cards as high as I could get them stable, producing nearly 315Mhash/s each.....they would never peak over 250 shares/hour.

[bitlane]

i have noticed ppl that overclock "to extreme" often create a lot os "stale" or "invalid" shares. dropping a few Mhz can improve your accepted share and lower stale share rates thus giving a better return
hope thats a help

My problem was never excessive Stales/Invalids.... it simply took a long time to produce anything (other than heat....lol).

Unfortunately, I have peaked as high as 5% Stales at times, but this is uncommon for the most part. I still see the odd stale share, same as always.
This 'tuning' has not affected Stales for the good or bad.

[sven]

that's why i test all my overclocked cards for graphical errors
if there is the slightest issue i clock it down to prevent invalid shares and possible freezes

[Pentium100]

That's why I do not overclock to the point where I need to increase the voltage. I would only get a few percent higher hashrate (assuming hashrate scales linearly with the GPU MHz) but more heat and power consumption. 1GHz (~310MH/s) IMO is enough for a 6870.

The time to calculate a share can vary, just like the time to calculate a block (assuming constant difficulty). It is interesting though, I'll try to calculate my shares/hour.

[Ryu.Hayabusa]

It can also randomly fail and generate nothing but bad hashes too.  I had mine running on the edge until I woke up the next morning and had -300 or so shares lol.

[Dargo]

I've noticed this too, and completely agree that you have to fine tune for shares/hour rather than hash rate. You have to be careful, though, about measuring the shares/hour due to variance. It would be nice if there were some standard and accurate way of doing this so people could have pissing contests in terms of shares/hour rather than Mh/s.

[btcking]

Thanks for the interesting info! I had no idea that large hash rates could actually result in less stable share counts if using overclocking. very enlightening.