In that case, it's not too bad to rely on the readout from poclbm if you freeze all other variables. You can surely spare a few minutes for the crunching to stabilize =)
I prefer to use poclbm for this since it seems to be stable, steady and has low rates of stales/corrupts (it doesn't matter if your rate is higher if you end up with stale or corrupted shares). Make sure to not have anything else running that could affect the GPU usage (even newer browsers use some GPU for instance) while you're testing and set -f to 1 (or 0) to give the GPU priority.
Topic: What is a good MHps measure? (Read 3230 times)
Sure. But you are not going to wait the day to see if the last 5MHz added to core really increased the performance, or? Especially when a short lag in network connectivity might fully distort your measurement.
I should better have asked: what is your process of finding the optimal settings for your mining HW?
I should better have asked: what is your process of finding the optimal settings for your mining HW?
Average it over a day. This way it won't be so jumpy.
Miners,
just put together my first mining rig (3*6950) and was playing around to get reasonable MHps rates out of it.
I'm mining at deepbit with guiminer and started first using the gfx cards untouched. I got some 270-280 each. Increasing core from 800 to 880 with MSI afterburner moves the performance to the 310-320 range, giving an overall performance of ~950 MHps (which is also reported as my rate at deepbit).
Now I'd try to tweak further to get the kickin' numbers from the mining HW comparison list that gives a max of ~365 for locked HD6950s. Alas, I don't see which measure to use for reliably detecting MHps.
Increasing core or voltage values further with afterburner, I see the numbers displayed by guiminer (which in fact are poclbm values) fluctuating between 180 and 320 with no clear trend indicating an increased performance. After a while (~3 min) the values stabilize near the old level of ~320 (but still fluctuating between ~290-340). As an effect, it's impossible to reliably determine if the last tweak was helpful.
Even more strange are the values reported by deepbit (averaged over last 5 minutes): every now and then the MHps rate varies quite significantly in a range between 700 and 1000 with sporadic outliers going down to 300 and peaking up to 1150. And this really confuses me: how can the mean value reach 1150 while guiminer was not reporting values > 950 the previous 10 minutes?
I might be missing something, but the observed issues indicate (at least for me) that there is a lack of reliable MHps measures. My best guess so far is just listening to the fans and correlate noise and performance
Or, how do you measure your MHps values?
just put together my first mining rig (3*6950) and was playing around to get reasonable MHps rates out of it.
I'm mining at deepbit with guiminer and started first using the gfx cards untouched. I got some 270-280 each. Increasing core from 800 to 880 with MSI afterburner moves the performance to the 310-320 range, giving an overall performance of ~950 MHps (which is also reported as my rate at deepbit).
Now I'd try to tweak further to get the kickin' numbers from the mining HW comparison list that gives a max of ~365 for locked HD6950s. Alas, I don't see which measure to use for reliably detecting MHps.
Increasing core or voltage values further with afterburner, I see the numbers displayed by guiminer (which in fact are poclbm values) fluctuating between 180 and 320 with no clear trend indicating an increased performance. After a while (~3 min) the values stabilize near the old level of ~320 (but still fluctuating between ~290-340). As an effect, it's impossible to reliably determine if the last tweak was helpful.
Even more strange are the values reported by deepbit (averaged over last 5 minutes): every now and then the MHps rate varies quite significantly in a range between 700 and 1000 with sporadic outliers going down to 300 and peaking up to 1150. And this really confuses me: how can the mean value reach 1150 while guiminer was not reporting values > 950 the previous 10 minutes?
I might be missing something, but the observed issues indicate (at least for me) that there is a lack of reliable MHps measures. My best guess so far is just listening to the fans and correlate noise and performance
Or, how do you measure your MHps values?
Jump to: