I have not ordered any yet either but am thinking about it, I would want a case with a fan too though a cluster case with a huge pc type fan might work as well. BTW I read your blog every once in awhile and love it!!
Raspberry Pi OS 12 (based upon Debian Bookworm) is now out. I've upgraded 4 of my farm to it and they seem to be working away. it says:
Linux *** 6.1.0-rpi4-rpi-v8 #1 SMP PREEMPT Debian 1:6.1.54-1+rpt2 (2023-10-05) aarch64
I'm not sure if that means they have an rpi4 specific kernel image. You'll get the 7.20.5 version of boinc if installing the repo version.
One thing I did find annoying was raspi-config no longer has an option to set the GPU memory so I ended up adding the following to the end of /boot/config.txt to limit it to 16MB (the min).
Does anyone know if booting from a USB is a bottleneck on speed? (Compared to booting from an SSD) I'm running BOINC from a Raspberry Pi 4.
You mean for BOINC performance, does it matter if the filesystem that BOINC is running on (booting itself is not so relevant here) is on an SD card or on an SSD drive? No, unless the BOINC project app you are using is written in a completely brain-dead style (e.g. checkpointing super-excessively despite user configuration that tells it to checkpoint only every so often), there will be no significant difference.
So I have set up my first Raspi 5 for E@H, running on all 4 cores, 64 bit Bookworm.
It's more than 3 times, almost 4 times as fast as my ole Raspi 3Bs, in line with all the published benchmarks, so no surprises here. I have the "official" fan installed on the Raspi5, which is pretty quiet IMHO. With it the CPU temp levels off at ca 58°C , well below the thermal throttling threshold (my Raspi 3s also got a fan so they also never throttle).
Nice. I haven't measured the power draw yet but I'm sure it's less than three times that of the Raspi 3B under E@H load, so some progress here wrt. E@H on Raspis. Every little bit helps :-)
Somehow I expected the performance jump from Pi 4 -> Pi 5 to be more significant. My Raspberry Pi 4 (no fan, 90%-100% CPU time utilization, I do have a cheap heatsink attached to the CPU) is doing 15k-16k CPU seconds per task. Congrats on getting your hands on a Pi 5. They seem to be completely out of stock where I live.
I assume the Jetson nano uses a custom CUDA version of the app, so it's mostly not running on the A57 cores but on the NVIDIA GPU with how many ??...128 cores?? Not apples to apples compared to the Raspi5 (also at > twice the price point). Also the Jetson will probably need to run just one instance of the (CUDA) app at a time for the quoted 650 seconds runtime, right? Execution times quoted for the Raspis here were for 4 threads running in parallel.
I guess people here will try to build apps compiled specifically for the Raspi 5s CPU, would be interesting to see if that helps a bit.
mikey wrote: <snipped> I
)
Thanks Mikey.
MarksRpiCluster
Raspberry Pi OS 12 (based
)
Raspberry Pi OS 12 (based upon Debian Bookworm) is now out. I've upgraded 4 of my farm to it and they seem to be working away. it says:
Linux *** 6.1.0-rpi4-rpi-v8 #1 SMP PREEMPT Debian 1:6.1.54-1+rpt2 (2023-10-05) aarch64
I'm not sure if that means they have an rpi4 specific kernel image. You'll get the 7.20.5 version of boinc if installing the repo version.
One thing I did find annoying was raspi-config no longer has an option to set the GPU memory so I ended up adding the following to the end of /boot/config.txt to limit it to 16MB (the min).
[all]
gpu_mem=16
MarksRpiCluster
Does anyone know if booting
)
Does anyone know if booting from a USB is a bottleneck on speed? (Compared to booting from an SSD) I'm running BOINC from a Raspberry Pi 4.
Does anyone know if booting
)
You mean for BOINC performance, does it matter if the filesystem that BOINC is running on (booting itself is not so relevant here) is on an SD card or on an SSD drive? No, unless the BOINC project app you are using is written in a completely brain-dead style (e.g. checkpointing super-excessively despite user configuration that tells it to checkpoint only every so often), there will be no significant difference.
HB
So I have set up my first
)
So I have set up my first Raspi 5 for E@H, running on all 4 cores, 64 bit Bookworm.
It's more than 3 times, almost 4 times as fast as my ole Raspi 3Bs, in line with all the published benchmarks, so no surprises here. I have the "official" fan installed on the Raspi5, which is pretty quiet IMHO. With it the CPU temp levels off at ca 58°C , well below the thermal throttling threshold (my Raspi 3s also got a fan so they also never throttle).
Nice. I haven't measured the power draw yet but I'm sure it's less than three times that of the Raspi 3B under E@H load, so some progress here wrt. E@H on Raspis. Every little bit helps :-)
Raspi 5: https://einsteinathome.org/host/13163354/tasks/4/0
Raspi 3: https://einsteinathome.org/host/12654103/tasks/4/0
Happy crunching
HB
Thanks Bikeman for sharing
)
Thanks Bikeman for sharing that, have you seen my post in the ARM crunching thread? I have the same observations as you regarding the temps.
And now I am testing Asteroids@Home, and the results are interesting to say the least, soon I will report on it as well :)
my blog about raspberry pis and diy life
Somehow I expected the
)
Somehow I expected the performance jump from Pi 4 -> Pi 5 to be more significant. My Raspberry Pi 4 (no fan, 90%-100% CPU time utilization, I do have a cheap heatsink attached to the CPU) is doing 15k-16k CPU seconds per task. Congrats on getting your hands on a Pi 5. They seem to be completely out of stock where I live.
For what it is worth I added
)
For what it is worth I added a new pi 5 8 gig to my pi array today. We will see how it "goes".
I too was VERY underwhelmed
)
I too was VERY underwhelmed by your runtimes. I don't see the 3 - 4X improvement over the Pi4.
I average right around 9200 seconds for my Pi4 on BRP4 tasks.
I average right around 650 seconds on my Jetson Nano on BRP4 tasks which is only using Cortex-A57 cores.
I assume the Jetson nano uses
)
I assume the Jetson nano uses a custom CUDA version of the app, so it's mostly not running on the A57 cores but on the NVIDIA GPU with how many ??...128 cores?? Not apples to apples compared to the Raspi5 (also at > twice the price point). Also the Jetson will probably need to run just one instance of the (CUDA) app at a time for the quoted 650 seconds runtime, right? Execution times quoted for the Raspis here were for 4 threads running in parallel.
I guess people here will try to build apps compiled specifically for the Raspi 5s CPU, would be interesting to see if that helps a bit.