Apparently overclocking have no impact on crunch times. I, tbh, can't figure out why. I'm all ears if someone has a theory...
It depends on the application, the host hardware and the system loading.
Some projects respond to the cpu being overclocked and some the gpu.
Some applications respond better to core clock overclocking and some memory overclocking.
Some applications respond better to higher PCIE bandwidth and run faster on more lanes.
So the answer is . . . . . it depends.
Keith has covered most of the aspects that come into play on modern PCs, but there's another factor when using a "development board" with an ARM SOC on it...
The Raspberry Pi (and lots of other machines based on various ARM CPUs) bring memory contention into the equation in a big way. There's no Level 3 cache, and even "small" programs like the OPN1 project at WCG (their only non-Android ARM program at present) slow down dramatically once lots of Level 2 cache misses start happening.
I've not tried it myself, but I suspect that an Einstein program could become very memory-bound (especially if you run more than one at once!) It could well be that your machine is having to wait for memory much/most of the time, even when not overclocked...
This is just a thought - as I said, I've not tested on Einstein; this is based on performance tests carried out using the aforementioned OPN1 program on a Pi4 clocked at various rates up to 2 GHz. From a throughput standpoint, it was never really worth running four copies, and the third copy was of arguable benefit at the highest clock rate.
You may have already said it but what are the crunch times for a RPi4? I have several sitting around doing nothing
Edit: Nevermind I see it on your profile. I have an old GPU that does a runtime of 1533 and an RPI seems to do it in about 25,000.
I have several pi's I can dedicate to it but if a gpu can do ~15-16 in the time it takes a Pi to do one is there much point? I know every bit counts but for the price of a few Pi's I can get another GPU to do way more. I may be wrong here, advise, please.
You may have already said it but what are the crunch times for a RPi4? I have several sitting around doing nothing
Edit: Nevermind I see it on your profile. I have an old GPU that does a runtime of 1533 and an RPI seems to do it in about 25,000.
I have several pi's I can dedicate to it but if a gpu can do ~15-16 in the time it takes a Pi to do one is there much point? I know every bit counts but for the price of a few Pi's I can get another GPU to do way more. I may be wrong here, advise, please.
For me, the process is much more important then a result.
I have "a farm" with about 25 computers of all kinds including RPIs. Not all of them are crunching here. Some of them periodically need my attention if something goes wrong. i tend to them like little pets.
I read somehwere .... For people using Raspberry Pi 4th generation, speed crunching may be greatly impacted by the «heat» of the Pi4 itself. If they are not properly cooled and temperature is not staying low, they will throttle the CPU themselves and lower their «crunch power».... and even if they have been overcloked.
Quote:Apparently overclocking
)
It depends on the application, the host hardware and the system loading.
Some projects respond to the cpu being overclocked and some the gpu.
Some applications respond better to core clock overclocking and some memory overclocking.
Some applications respond better to higher PCIE bandwidth and run faster on more lanes.
So the answer is . . . . . it depends.
Thomas, Quote: Quote:Ap
)
Thomas,
Keith has covered most of the aspects that come into play on modern PCs, but there's another factor when using a "development board" with an ARM SOC on it...
The Raspberry Pi (and lots of other machines based on various ARM CPUs) bring memory contention into the equation in a big way. There's no Level 3 cache, and even "small" programs like the OPN1 project at WCG (their only non-Android ARM program at present) slow down dramatically once lots of Level 2 cache misses start happening.
I've not tried it myself, but I suspect that an Einstein program could become very memory-bound (especially if you run more than one at once!) It could well be that your machine is having to wait for memory much/most of the time, even when not overclocked...
This is just a thought - as I said, I've not tested on Einstein; this is based on performance tests carried out using the aforementioned OPN1 program on a Pi4 clocked at various rates up to 2 GHz. From a throughput standpoint, it was never really worth running four copies, and the third copy was of arguable benefit at the highest clock rate.
Cheers - Al.
You may have already said it
)
You may have already said it but what are the crunch times for a RPi4?I have several sitting around doing nothingEdit: Nevermind I see it on your profile. I have an old GPU that does a runtime of 1533 and an RPI seems to do it in about 25,000.
I have several pi's I can dedicate to it but if a gpu can do ~15-16 in the time it takes a Pi to do one is there much point? I know every bit counts but for the price of a few Pi's I can get another GPU to do way more. I may be wrong here, advise, please.
LunaorBust wrote: You may
)
For me, the process is much more important then a result.
I have "a farm" with about 25 computers of all kinds including RPIs. Not all of them are crunching here. Some of them periodically need my attention if something goes wrong. i tend to them like little pets.
Hi guys, I read
)
Hi guys,
I read somehwere .... For people using Raspberry Pi 4th generation, speed crunching may be greatly impacted by the «heat» of the Pi4 itself. If they are not properly cooled and temperature is not staying low, they will throttle the CPU themselves and lower their «crunch power».... and even if they have been overcloked.
Raspberry Pi 4 Cases, Temperature and CPU Throttling Under Load - Martin Rowan
Hope this help.
Best regards.
Jeff
Hi! For those in need of
)
Hi!
For those in need of cooling an overclocked Raspberry Pi 4 I can recommend this solution: https://www.conrad.com/p/joy-it-armor-case-block-active-sbc-housing-compatible-with-raspberry-pi-active-cooling-black-2140232
My Pi's running Einstein@Home never exceeds 50 degrees celcius with this type of cooling.
There's also variants available for Raspberry 2 & 3.
Regards,
Thomas :-)