you should try running the FGRP5 CPU project. they r-released an linux aarch64 app for that since last you tested (need to enable beta too). but be aware it uses a good amount of RAM. might not be able to run all 4 cores.
would be cool to get relative performance numbers for that app vs the BRP4 app you ran now. same with SD vs nvme
Didn't know that. I started crunching on the SD card with two FGRP5 tasks, and a few minutes after starting RAM usage is 1.7GB. I'll leave it at two cores for now, maybe the RAM usage will rise as the tasks continue. If not, I'll later switch to more cores.
I also run some R-Pi's though am very much the novice at it but with Keith Myers help I too am crunching along very nicely. I have R-Pi model 4's though and noticed you didn't try running yours thru an SSD instead of just the SD card. I know they don't make a "hat" for one but do you think it would make any difference or is the throughput limited thru the USB ports sooo much that it's not a worthwhile test?
I think what you did is extremely helpful to the crunching community and want to say "Thank you" for doing it!!
Write speeds comparing USB 2.0 and MicroSD cards are about equivalent. Only when using the USB 3.0 ports on the Pi4 would there be an advantage. Would like to see that experiment also.
The main reason why it makes sense on the Pi5 is the the two lane PCI Gen. 2 interface on the Pi5.
I looked at your NVMe hat from your blog. Does it power the NVMe hat and drive from only the PCIe ribbon cable? I didn’t see any other connections from the board to the hat.
OK, I crunched FGRP5 tasks with and without the NVME HAT... and there's basically no difference
SD Card (all runtimes in seconds):
AVG 18976
MIN 17601
MAX 21378
TPD 9.106 (tasks per day with two cores running)
NVMe:
AVG 18701
MIN 17050
MAX 21780
TPD 9.240
I was crunching FGRP on two cores. In light of this, that 30% for the Arecibo tasks seems implausible. Or maybe it shows up only when all four cores are stressed? I would need to do more testing. And what is more, yesterday the NVMe drive died, so I need to get a new one.
Hi, it's me again! This
)
Hi, it's me again!
This time I tested if using an NVMe HAT on a Raspberry Pi 5 makes it crunch BOINC faster.
And for Einstein@Home it does, switching from an SD card to an NVMe drive makes the Pi5 crunch around 30% faster.
Here's the blog post with the benchmarks:
https://stfn.pl/blog/50-pi5-nvme-performance-in-boinc/
my blog about raspberry pis and diy life
stfn wrote:Hi, it's me
)
you should try running the FGRP5 CPU project. they r-released an linux aarch64 app for that since last you tested (need to enable beta too). but be aware it uses a good amount of RAM. might not be able to run all 4 cores.
would be cool to get relative performance numbers for that app vs the BRP4 app you ran now. same with SD vs nvme
_________________________________________________________________________
Didn't know that. I started
)
Didn't know that. I started crunching on the SD card with two FGRP5 tasks, and a few minutes after starting RAM usage is 1.7GB. I'll leave it at two cores for now, maybe the RAM usage will rise as the tasks continue. If not, I'll later switch to more cores.
my blog about raspberry pis and diy life
Yes, I'd like to see FGRP5
)
Yes, I'd like to see FGRP5 tasks on the Pi4 and Pi5 also. Here are my Pi4 and Pi5 tasks for comparison on FGRP5.
Pi4 FGRP5 tasks
Pi5 FGRP5 tasks
No NVMe drive though. The tower cooler precludes fitting a NVMe Hat device.
stfn wrote:Hi, it's me
)
I also run some R-Pi's though am very much the novice at it but with Keith Myers help I too am crunching along very nicely. I have R-Pi model 4's though and noticed you didn't try running yours thru an SSD instead of just the SD card. I know they don't make a "hat" for one but do you think it would make any difference or is the throughput limited thru the USB ports sooo much that it's not a worthwhile test?
I think what you did is extremely helpful to the crunching community and want to say "Thank you" for doing it!!
Write speeds comparing USB
)
Write speeds comparing USB 2.0 and MicroSD cards are about equivalent. Only when using the USB 3.0 ports on the Pi4 would there be an advantage. Would like to see that experiment also.
The main reason why it makes sense on the Pi5 is the the two lane PCI Gen. 2 interface on the Pi5.
stfn wrote: Hi, it's me
)
I looked at your NVMe hat from your blog. Does it power the NVMe hat and drive from only the PCIe ribbon cable? I didn’t see any other connections from the board to the hat.
_________________________________________________________________________
From the mounting
)
From the mounting instructions and pictures of the hat, it looks like power and data is only through the ribbon cable.
mikey wrote: I think what
)
Thank you :)
Yes, the NVMe hats do not require additional power, everything comes through the ribbon cable.
my blog about raspberry pis and diy life
OK, I crunched FGRP5 tasks
)
OK, I crunched FGRP5 tasks with and without the NVME HAT... and there's basically no difference
SD Card (all runtimes in seconds):
AVG 18976
MIN 17601
MAX 21378
TPD 9.106 (tasks per day with two cores running)
NVMe:
AVG 18701
MIN 17050
MAX 21780
TPD 9.240
I was crunching FGRP on two cores. In light of this, that 30% for the Arecibo tasks seems implausible. Or maybe it shows up only when all four cores are stressed? I would need to do more testing. And what is more, yesterday the NVMe drive died, so I need to get a new one.
https://stfn-public.s3.eu-west-1.amazonaws.com/blog/benchmarks/einstein_benchmark_data3.ods
my blog about raspberry pis and diy life