Most reliable way to measure Hashrate?

[phelix]

There is a lot of discussion and some bragging about hashrates - I wonder how much measuring error there is in this.

Do the miners all use the same way to calculate the hashrate or are there miners that just spit out higher values?

For how long a time should you average? My miners' speeds sometimes seem to fluctuate in slow waves of 10 minutes or so.

The -a flag in phoenix seems to help, I have set it to 64 now.

Should stales be counted? Depends?  So measure at deepbit with 60min window?

[willphase]

the hashrate from your miner software will be the raw speed at which it's running through the SHA1 keyspace.  It is theoretically possible to be very very lucky for every single share you get, and solve it faster than this average, or very very unlucky and have to exhaust the whole keyspace before submitting the share.

In addition, packets can be lost on the internet, shares can be rejected because the block that you were solving mid-share was solved and so the pool isn't interested in your share any more, and you might just be running against a buggy pushpool that is sending duplicate shares to clients, meaning your share is being rejected because someone else has solved it.

All in all, the best most accurate stats are therefore likely to come from your pool, because that's what will be determining your payment rate.

However, if you're interested in tweaking your setup to get more hash/sec from e.g. boosting clock speed, or playing with memory speed, then the rate from your miner is a good indicator of this raw hashing speed.

Will

[keybaud]

The pool hash rate is an estimate, whereas the client hash rate is your GPU's hash rate.

Do you actually mean a different measure and not hash rate?

[phelix]

The pool hash rate is an estimate, whereas the client hash rate is your GPU's hash rate.

Do you actually mean a different measure and not hash rate?

well, I mean hash rate but measured in a way so that it makes sense to fine tune.

For example if you start several miners on one gpu you might get a higher hash rate reportet from the miner instances alltogether but because the stale shares increase it might be less efficient.

So I guess it's miner hash rate in most cases and pool hashrate to check including influence of stales and network.