$300USD saving worth only to change 3 drives to SSD. But I think you should measure economy for each drive separately, so each new drive should save valuable money of its cost. This way there is no sense to chage any drive while it is working. I myself use even old 1Gb-4Gb drives that survive these days. Any new drive will not save the same value as it cost.
Here are some further power meter results. I've set up a Q8400 system on a third meter. Stock frequency is 8x333 = 2.66GHz but I'm running at 8x420 = 3.36GHz. For stability, I'm running with a Vcore increase of +100mV. For ease of comparison. I've created a list below showing the measured power draws for all three machines I've tested (all running Linux) at various stages of bootup, running at idle and running at full load.
I've also had a lot more time to accumulate ABP1 results so that I now have good average crunch times which I've used to recalculate credit per watt figures. I've calculated the average time for a task based on the 20 most recently completed results. In all cases the variation between the results for a given host was quite minor - of the order of +/- a few minutes maximum in a 4 hour crunch time.
These figures show the efficiency of the Q8400. I had always imagined it would perform better in credit/watt terms but I had no inkling of how much better. I was quite surprised to see how little extra power draw there was at idle and even more so when looking at the full load draw. For a while it was sitting on just 113 watts but all of today it has been sitting on 117 watts.
I guess the next thing to do is to remove the +100mV Vcore tweak on the Q8400 and back off the overclock until stability returns. That should see the credit per watt increase a little I would hope - slightly longer crunch times but enough power reduction to more than compensate.
$300USD saving worth only to change 3 drives to SSD. But I think you should measure economy for each drive separately, so each new drive should save valuable money of its cost. This way there is no sense to chage any drive while it is working. I myself use even old 1Gb-4Gb drives that survive these days. Any new drive will not save the same value as it cost.
The calculation of $USD328 was for doing away with 50 drives and saving 5 watts per machine. The 50 machines would have to boot off CD or USB and load BOINC from a network share. I already have plenty of network drive space for 50 BOINC instances. It's only 100GB for 2GB per BOINC host.
These figures show the efficiency of the Q8400. I had always imagined it would perform better in credit/watt terms but I had no inkling of how much better.
While for most real-world application mixes I think most consumers get very little benefit from the extra two cores in a quad, for BOINC processing they really are quite wonderful. You get very nearly double the output of useful work out of nearly the same investment in all of the non-CPU components of a box, and as for power, you nearly get to cut the penalty of all the overhead in the box by a factor of two also.
Of course whether it's really a good economy deal the depends much on what current pricing or special deals are available. But my impression is that ever since Intel made its first really big price cut on the Q6600, there has usually been an economy quad thats a pretty good deal for BOINC.
Aside from CPU price, about the only downside I'm aware of is the possibility of needing a more expensive heatsink fan or of having to back off on the overclock because of the much higher power dissipation of the quad. If one goes far into extreme overvolt and overclock territory one might need a significantly more capable and expensive power supply. But at the more moderate levels Gary is exploring, the system power consumption is well in the economy power supply range. Anyway, I think current fashion and the range of supply offerings mean most of use have substantionally over-sized supplies.
$300USD saving worth only to
)
$300USD saving worth only to change 3 drives to SSD. But I think you should measure economy for each drive separately, so each new drive should save valuable money of its cost. This way there is no sense to chage any drive while it is working. I myself use even old 1Gb-4Gb drives that survive these days. Any new drive will not save the same value as it cost.
Here are some further power
)
Here are some further power meter results. I've set up a Q8400 system on a third meter. Stock frequency is 8x333 = 2.66GHz but I'm running at 8x420 = 3.36GHz. For stability, I'm running with a Vcore increase of +100mV. For ease of comparison. I've created a list below showing the measured power draws for all three machines I've tested (all running Linux) at various stages of bootup, running at idle and running at full load.
I've also had a lot more time to accumulate ABP1 results so that I now have good average crunch times which I've used to recalculate credit per watt figures. I've calculated the average time for a task based on the 20 most recently completed results. In all cases the variation between the results for a given host was quite minor - of the order of +/- a few minutes maximum in a 4 hour crunch time.
These figures show the efficiency of the Q8400. I had always imagined it would perform better in credit/watt terms but I had no inkling of how much better. I was quite surprised to see how little extra power draw there was at idle and even more so when looking at the full load draw. For a while it was sitting on just 113 watts but all of today it has been sitting on 117 watts.
Average task time . . . . . 15542 secs . . . . . . . 14115 secs . . . . . . . . . 15542 secs
credit per watt . . . . 500/(15542/3600x79) . . . 500/(14115/3600x87) . . . 1000/(15542/3600x117)
calculated value . . . . . . 1.466 . . . . . . . . . . 1.466 . . . . . . . . . . . . 1.980
I guess the next thing to do is to remove the +100mV Vcore tweak on the Q8400 and back off the overclock until stability returns. That should see the credit per watt increase a little I would hope - slightly longer crunch times but enough power reduction to more than compensate.
Cheers,
Gary.
RE: $300USD saving worth
)
The calculation of $USD328 was for doing away with 50 drives and saving 5 watts per machine. The 50 machines would have to boot off CD or USB and load BOINC from a network share. I already have plenty of network drive space for 50 BOINC instances. It's only 100GB for 2GB per BOINC host.
Cheers,
Gary.
RE: These figures show the
)
While for most real-world application mixes I think most consumers get very little benefit from the extra two cores in a quad, for BOINC processing they really are quite wonderful. You get very nearly double the output of useful work out of nearly the same investment in all of the non-CPU components of a box, and as for power, you nearly get to cut the penalty of all the overhead in the box by a factor of two also.
Of course whether it's really a good economy deal the depends much on what current pricing or special deals are available. But my impression is that ever since Intel made its first really big price cut on the Q6600, there has usually been an economy quad thats a pretty good deal for BOINC.
Aside from CPU price, about the only downside I'm aware of is the possibility of needing a more expensive heatsink fan or of having to back off on the overclock because of the much higher power dissipation of the quad. If one goes far into extreme overvolt and overclock territory one might need a significantly more capable and expensive power supply. But at the more moderate levels Gary is exploring, the system power consumption is well in the economy power supply range. Anyway, I think current fashion and the range of supply offerings mean most of use have substantionally over-sized supplies.