Those CPU cores are a bit less powerful per MHz clock, but the board more than offsets this by the 1.5GHz clock rate, so all in all it should do a bit better on BOINC than the Pi2 ....if you ever let it do that :-)
So how long are tasks taking on the new Raspberry Pi2 as apposed to the old Pi's?
Obviously there can be four at a time, but how much faster is the neon ARM7 versus the ARM6 for Einstein@home.
So how long are tasks taking on the new Raspberry Pi2 as apposed to the old Pi's?
Obviously there can be four at a time, but how much faster is the neon ARM7 versus the ARM6 for Einstein@home.
When executing 4 tasks in parallel, a RPi2 clocked at 1GHz (the raspi-config overclocking config option for the RPi2) a task takes ca. 70k sec. I have not yet tried to overclock it any higher.
When overclocking the RPi1 to 1GHz, I usually get a runtime just under 90k s. However, not all RPi1 will overclock to 1GHz, so a more realistic runtime for the average, moderately overclocked RPi1 is probably closer to 110k s. That would make the RPi2 about 4 x 110/70 =~ 6 times more productive in Einstein@Home, which matches well the advertised speed-up.
HBE
EDIT: In terms of efficiency, very roughly the RPi2 should deliver ca 3.5k credits / kWh , so if we assume a cost of 0.25EUR per kWh, this would mean 7EUR cents will "buy" you 1000 credits. To earn 1 million credits with a single RPi2, your cost will be ca 100 EUR (ca 30 EUR for the PI and 70EUR for electricity), and it would take you 8.5 years (assuming constant cost of electricity during the next decade which is unlikely ;-) ). A PC with a reasonably fast GPU can easily generate ca 100 times more credits per day at no more than 50 times the power consumption, so there still is at least a factor of 2 in efficiency that the RPi2 is behind even moderately efficient GPUs. Comparing to standard desktop PCs without GPUs, the RPi2 and similar boards should now be quite comparable in efficiency, tho, if not superior, depending on the CPU you compare it to.
So how long are tasks taking on the new Raspberry Pi2 as apposed to the old Pi's?
Obviously there can be four at a time, but how much faster is the neon ARM7 versus the ARM6 for Einstein@home.
New one around 80k seconds with no overclock (at the moment). Only done a few tasks so not a big sample yet: Pi2
Old one around 111-115k secs with a "medium" overclock: Pi B
I bought 3 Pi2's to replace 5 B's and a B+. I have run down most of the B's and retired them. I also have a couple of Parallella's that are a bit quicker but only dual core.
After I've got a few more samples I will try the fftw plan and see if that makes any difference.
After I've got a few more samples I will try the fftw plan and see if that makes any difference.
I have very little confidence in those. I'm currently generating new wisdom in exhaustive mode (will take some days to complete) and in parallel to E@H tasks to better simulate the conditions the FFT plans would actually be executed in.
After I've got a few more samples I will try the fftw plan and see if that makes any difference.
I have very little confidence in those. I'm currently generating new wisdom in exhaustive mode (will take some days to complete) and in parallel to E@H tasks to better simulate the conditions the FFT plans would actually be executed in.
Cheers
HB
I had better wait then. If you like PM me and I can give them a try when you have new wisdom files.
Those CPU cores are a bit
)
Those CPU cores are a bit less powerful per MHz clock, but the board more than offsets this by the 1.5GHz clock rate, so all in all it should do a bit better on BOINC than the Pi2 ....if you ever let it do that :-)
Cheers
HB
I'm not so much into
)
I'm not so much into Raspberry Pi stuff and such, but I'm sure into woodworking. So here is a video that might interest you guys ;)
Raspberry pi computer holder
So how long are tasks taking
)
So how long are tasks taking on the new Raspberry Pi2 as apposed to the old Pi's?
Obviously there can be four at a time, but how much faster is the neon ARM7 versus the ARM6 for Einstein@home.
RE: So how long are tasks
)
When executing 4 tasks in parallel, a RPi2 clocked at 1GHz (the raspi-config overclocking config option for the RPi2) a task takes ca. 70k sec. I have not yet tried to overclock it any higher.
When overclocking the RPi1 to 1GHz, I usually get a runtime just under 90k s. However, not all RPi1 will overclock to 1GHz, so a more realistic runtime for the average, moderately overclocked RPi1 is probably closer to 110k s. That would make the RPi2 about 4 x 110/70 =~ 6 times more productive in Einstein@Home, which matches well the advertised speed-up.
HBE
EDIT: In terms of efficiency, very roughly the RPi2 should deliver ca 3.5k credits / kWh , so if we assume a cost of 0.25EUR per kWh, this would mean 7EUR cents will "buy" you 1000 credits. To earn 1 million credits with a single RPi2, your cost will be ca 100 EUR (ca 30 EUR for the PI and 70EUR for electricity), and it would take you 8.5 years (assuming constant cost of electricity during the next decade which is unlikely ;-) ). A PC with a reasonably fast GPU can easily generate ca 100 times more credits per day at no more than 50 times the power consumption, so there still is at least a factor of 2 in efficiency that the RPi2 is behind even moderately efficient GPUs. Comparing to standard desktop PCs without GPUs, the RPi2 and similar boards should now be quite comparable in efficiency, tho, if not superior, depending on the CPU you compare it to.
RE: I'm not so much into
)
Nice!!! Refreshing to see things like this being made the old fashioned way, without a 3D-printer ;-).
Cheers
HB
RE: So how long are tasks
)
New one around 80k seconds with no overclock (at the moment). Only done a few tasks so not a big sample yet: Pi2
Old one around 111-115k secs with a "medium" overclock: Pi B
I bought 3 Pi2's to replace 5 B's and a B+. I have run down most of the B's and retired them. I also have a couple of Parallella's that are a bit quicker but only dual core.
After I've got a few more samples I will try the fftw plan and see if that makes any difference.
BOINC blog
RE: After I've got a few
)
I have very little confidence in those. I'm currently generating new wisdom in exhaustive mode (will take some days to complete) and in parallel to E@H tasks to better simulate the conditions the FFT plans would actually be executed in.
Cheers
HB
RE: RE: After I've got a
)
I had better wait then. If you like PM me and I can give them a try when you have new wisdom files.
BOINC blog
Will do that, thx. HB
)
Will do that, thx.
HB
I think that EaH client only
)
I think that EaH client only uses 12M R2C FFTW functions, so you can do an exhaustive wisdom only for this type so save a lot of time:
fftw-wisdom -v -x -o wisdom_eah rof12582912Furthermore, this host has a FreeScale iMX6Q (4 x Cortex-A9 @ 1GHz - this board).
Thank you,