phatk vs. poclbm effective hash rate

[padrino]

Like many others that have experimented I see an increase of about 5-10% in my reported hash rate when moving from poclbm to phatk.

On my 5770s running 935/1500

poclbm: 199 Mh/s 2011.beta4
phoenix with phatk: 207 Mh/s v1.48

I however noticed the reverse with my stats on deepbit:

poclbm: <400 Mh/s (varies lightly based on luck but always stays above 400 Mh/s)
phatk: ~375 Mh/s

To help account for variations I have deepbit average over 60 minutes and I've taken multiple samples to help account for the fact that deepbit will shift some depending on how fast a given set of shares are solved. The following questions come to mind:

1. Are the client values accurate with regard to Mh/s?
2. Is poclbm more efficient at managing the work queue and fetching/submitting shares?
3. Other factors?

I haven't done it yet but #1 should be able to be verified fairly quickly, in fact I assume if it were a problem it would have already been caught.

[SomeoneWeird]

Like many others that have experimented I see an increase of about 5-10% in my reported hash rate when moving from poclbm to phatk.

On my 5770s running 935/1500

poclbm: 199 Mh/s 2011.beta4
phoenix with phatk: 207 Mh/s v1.48

I however noticed the reverse with my stats on deepbit:

poclbm: <400 Mh/s (varies lightly based on luck but always stays above 400 Mh/s)
phatk: ~375 Mh/s

To help account for variations I have deepbit average over 60 minutes and I've taken multiple samples to help account for the fact that deepbit will shift some depending on how fast a given set of shares are solved. The following questions come to mind:

1. Are the client values accurate with regard to Mh/s?
2. Is poclbm more efficient at managing the work queue and fetching/submitting shares?
3. Other factors?

I haven't done it yet but #1 should be able to be verified fairly quickly, in fact I assume if it were a problem it would have already been caught.

Your stats on deepbit are the average shares you've submitted over the last 60 minutes. So if you have good luck, it will show you a higher mh/s, but luck, lower mh/s. I'd stick with phoenix.

[padrino]

Your stats on deepbit are the average shares you've submitted over the last 60 minutes. So if you have good luck, it will show you a higher mh/s, but luck, lower mh/s. I'd stick with phoenix.

I understand and that was why I discussed that in my post however maybe I was not clear enough, at first I thought the same thing however I can consistently over the course of a number of days repeat my findings, thus minimizing luck.

[padrino]

Ok... keep fooling yourself with that thought. If you think it long enough it becomes true, right?  
Only your mining program will show you accurate hash rates.

If my effective rate on deepbit is lower and I am making less BTC (luck aside as that variable changes from point to point, overtime converges) I should just continue to trust my local hash rate because it's obviously what matters more . Just throwing it out for consideration since the numbers indicate something is up.

[Calavera]

I don't know why everybody is giving you a hard time.  The reporting functions could easily differ between applications (accidentally).

However, the variance on share success rate could be high, and you'd have to look at the stale shares for each card too.

So either take more measurements or see if you can take an average over a longer period (for instance if your mine shows the total shares per miner you could note the change over a few days).

[bcpokey]

Most people are probably giving OP a hard time because it doesn't really make sense. Deepbit has a large variance even averaged over long periods of time. Here are my last 7 solved blocks on Deepbit:

02 Jun 2011 12:38:30
202 / 86798
3615.630 MH/s
+ 0.11636213 BTC
02 Jun 2011 12:34:32
55 / 31106
2693.103 MH/s
+ 0.08840738 BTC
02 Jun 2011 12:33:05
228 / 109837
3281.351 MH/s
+ 0.10379016 BTC
02 Jun 2011 12:28:09
314 / 149217
3327.463 MH/s
+ 0.10521589 BTC
02 Jun 2011 12:21:26
455 / 236195
3047.642 MH/s
+ 0.09631872 BTC
02 Jun 2011 12:10:49
1383 / 728234
2691.448 MH/s
+ 0.09495574 BTC
02 Jun 2011 11:34:20
944 / 472862
3129.525 MH/s
+ 0.09981771 BTC

My current Hashrate is roughly 3200Mhash/sec give or take as reported by miners. You can see deepbit sees wild variation from between 2600MHash to 3600Mhash/sec. A 1GHash/sec difference.

The best data would be to mine for 24 hours on polcbm, and look at the statistics page and view your rewards. Average it out if you wish. Then mine for 24 hours on phatk, and look at statistics again and view your rewards. If you really see over 24 hours a consistently lower payout block after block then you might be on to something. But just looking at the averaged MHash rate isn't really going to work.

[datguywhowanders]

I would even say that 24 hours probably isn't enough to test it out. Due to the inherent varying nature of Bitcoin, using anything less than a 7 day average is really just an estimate, and probably not a very good one at that.

[padrino]

I would even say that 24 hours probably isn't enough to test it out. Due to the inherent varying nature of Bitcoin, using anything less than a 7 day average is really just an estimate, and probably not a very good one at that.

My numbers come from an average over about 2 days for each, it's a good point that I need to push it out further, thus far it's held for a 48 hour average but I'll stretch it out longer.

To address a post further up, I was wondering the same thing about how the rates are calculated, with any shared pool there is more to one's effeciency than simply raw churn given the frequent fetches, etc.

[bcpokey]

I would even say that 24 hours probably isn't enough to test it out. Due to the inherent varying nature of Bitcoin, using anything less than a 7 day average is really just an estimate, and probably not a very good one at that.

I agree and disagree. For a formalized study of the differences between the two I would say yes. But I was mostly suggesting a simple pattern seeking "once over" of his history to start. Deepbit solves around 75-90 blocks a day depending on its luck, which is a sizable number to work with. He is thinking he sees ~10% difference between the two miners, which is a significant difference. Were he to look at all the points from each he should see a significant and consistent shift upward between the two data sets. As well as a significantly different average once outliers are thrown away.

To really get into the meat of the matter though I agree you need to go big on your datapoints.