So the next burning question is are the slots open in the back? So we can get real GPU in there. Followed by how slow will the bandwidth congestion slow the GPU processing?
And when it might become available in the US.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
So the next burning question is are the slots open in the back? So we can get real GPU in there. Followed by how slow will the bandwidth congestion slow the GPU processing?
And when it might become available in the US.
Tom M
From Jeff Geerling's testing he couldn't get a GPU working on the Pi. There have been a few carrier boards that offer PCIe x1 slots when using a CM4.
There isn't any OpenCL available for the Pi so it wouldn't be usable here. They could try porting the AMD OpenCL across but that might take years to happen.
As for the PCIe x1 slots I think Jeff ended using a 1x to 16x riser to be able to plug his RX6700XT card in.
I have two Raspberry Pi 4s that I allow 100% access 100% of the time. One has 4GB and runs Raspbian 32. The other one has 8GB and I upgraded to the new Raspberry Pi 64 bit OS. (I had been running Ubuntu 64).
Anyway, until yesterday both CPUs were crunching numbers, the 64-bit one finished all assigned jobs and isn't receiving any new ones. The 32 bit one (using on 1/2 of its 64 bit ALU and 1/2 of its 30 GPRs due to the 32bit OS) is still merrily chugging away on P2030-201606... jobs, while the 64 bit one is no longer receiving the same functional job in the ARCH64 version. I guess it would be OK it received 32bit jobs rather than no jobs?
I have resorted to running yoyo@home jobs and keeping track separately.
I have 4 computers of various flavors running Einstein@home jobs, which is convenient since that all show up on the same status page. Leading-edge all AMD desktop, Laptop with Intel APU/ nVidia 1070 GPU (both GPUs turning out jobs), and the two Raspberry Pis mentioned.
Why am I no longer receiving ARCH64 jobs?
P.S. I could run the same jobs as the other Pi (32-bit jobs) if they were offered, but I would prefer to take full advantage of the hardware I have.
You need to enable beta applications on the Pi4 to get the new AARCH64 BRP4 1.61 app.
The old 1.60 app was producing too many invalids against the new 1.33 BRPX64 app on Intel gpus so Bernd deprecated the old 1.60 and made a new 1.61 app.
It is validating against the new Intel application and is faster too as a benefit.
Binary Radio Pulsar Search (Arecibo) v1.60 aarch64-unknown-linux-gnu
vs
Binary Radio Pulsar Search (Arecibo) v1.61 aarch64-unknown-linux-gnu
vs
My other Pi4 32Bit Raspian Buster running
Binary Radio Pulsar Search (Arecibo) Anonymous platform
=====================================================
My own FP benchmarks show that number-crunching under 64 bit OS with the 64Bit ALU and double the GPRs to be over 50% faster and I was expecting the results here to be the same.
I don't know if the amount of work done by the 3 jobs above is the same, but the rewards are the same 62 per job for the 1st last one and the pre-beta version for the 64-bit platform.
None of the jobs I have run under the new beta have been validated yet, so I don't know what its reward will be. I was hoping you would be using the 64 bit ALU. The runtimes so far seem to be a bit longer in 1.61 than 1.60, or the one running on my Pi4 with the 32 bit OS.
I made my computers visible. My computers were not hidden on purpose. I guess that was the default setting. ;)
Is there a security reason I should have left them hidden?
BTW, Einstein's ratings have the 64 bit one a 130% FP and 135% integer when compared to the same hardware running the 32 bit OS. I am discounting the extra 4GB as I suspect all of the memory isn't utilized anyway. As I imagine there is a compiler in the middle, I don't know if it uses all 30 GPRs.
Does anybody do Assy anymore? I gave up assy after my 6th or so assy language! LoL (Although I did use 8080 Assy language to learn what C was up to. Microsoft Quick C compiled to assy then got linked. I understood the assy language! ;) I also wrote a program to translate Intel assy into Zilog assy (without the Z80 enhancements) in BASIC. As I also did DEC-10 Assy I like the operand order of the Z80 better, and I could at least hand optimize the results.) (I also purchased a compiler for Microsoft Quick basic, so it was compiled code as well). I never bothered to learn assy for the VAX or Intel since the i8086. ...or the 680xx in my 3 Amigas.
I made my computers visible. My computers were not hidden on purpose. I guess that was the default setting. ;)
Is there a security reason I should have left them hidden?
no security reason at all to hide them. there is no PII in the public information about one's hosts.
it's just that European law requires explicit consent to share any data, and the project is Europe based and must follow the rule. that's why it's the default behavior.
https://www.tomshardware.com/
)
https://www.tomshardware.com/news/raspberry-pi-compute-module-4-four-pcie-slots
So the next burning question is are the slots open in the back? So we can get real GPU in there. Followed by how slow will the bandwidth congestion slow the GPU processing?
And when it might become available in the US.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Tom M
)
From Jeff Geerling's testing he couldn't get a GPU working on the Pi. There have been a few carrier boards that offer PCIe x1 slots when using a CM4.
There isn't any OpenCL available for the Pi so it wouldn't be usable here. They could try porting the AMD OpenCL across but that might take years to happen.
As for the PCIe x1 slots I think Jeff ended using a 1x to 16x riser to be able to plug his RX6700XT card in.
You can find his YouTube channel HERE
MarksRpiCluster
Anybody hear the reason why
)
Anybody hear the reason why the FGRP5 ARM 1.16 app has been pulled?
I have two Raspberry Pi 4s
)
I have two Raspberry Pi 4s that I allow 100% access 100% of the time. One has 4GB and runs Raspbian 32. The other one has 8GB and I upgraded to the new Raspberry Pi 64 bit OS. (I had been running Ubuntu 64).
Anyway, until yesterday both CPUs were crunching numbers, the 64-bit one finished all assigned jobs and isn't receiving any new ones. The 32 bit one (using on 1/2 of its 64 bit ALU and 1/2 of its 30 GPRs due to the 32bit OS) is still merrily chugging away on P2030-201606... jobs, while the 64 bit one is no longer receiving the same functional job in the ARCH64 version. I guess it would be OK it received 32bit jobs rather than no jobs?
I have resorted to running yoyo@home jobs and keeping track separately.
I have 4 computers of various flavors running Einstein@home jobs, which is convenient since that all show up on the same status page. Leading-edge all AMD desktop, Laptop with Intel APU/ nVidia 1070 GPU (both GPUs turning out jobs), and the two Raspberry Pis mentioned.
Why am I no longer receiving ARCH64 jobs?
P.S. I could run the same jobs as the other Pi (32-bit jobs) if they were offered, but I would prefer to take full advantage of the hardware I have.
You need to enable beta
)
You need to enable beta applications on the Pi4 to get the new AARCH64 BRP4 1.61 app.
The old 1.60 app was producing too many invalids against the new 1.33 BRPX64 app on Intel gpus so Bernd deprecated the old 1.60 and made a new 1.61 app.
It is validating against the new Intel application and is faster too as a benefit.
Pi4 8GB running Raspberry Pi
)
Pi4 8GB running Raspberry Pi Bullseye 64Bit OS.
Binary Radio Pulsar Search (Arecibo) v1.60 aarch64-unknown-linux-gnu
vs
Binary Radio Pulsar Search (Arecibo) v1.61 aarch64-unknown-linux-gnu
vs
My other Pi4 32Bit Raspian Buster running
Binary Radio Pulsar Search (Arecibo) Anonymous platform
=====================================================
My own FP benchmarks show that number-crunching under 64 bit OS with the 64Bit ALU and double the GPRs to be over 50% faster and I was expecting the results here to be the same.
I don't know if the amount of work done by the 3 jobs above is the same, but the rewards are the same 62 per job for the 1st last one and the pre-beta version for the 64-bit platform.
None of the jobs I have run under the new beta have been validated yet, so I don't know what its reward will be. I was hoping you would be using the 64 bit ALU. The runtimes so far seem to be a bit longer in 1.61 than 1.60, or the one running on my Pi4 with the 32 bit OS.
The project pays a fixed
)
The project pays a fixed
)
The project pays a fixed amount of credit for each sub-project regardless of runtimes. BRP4 pays 62 credits.
My change from the 1.60 to the 1.61 application resulted in a 10-30% improvement in runtimes.
Since your computers are hidden I am unable to look at your results to see if they are similar to mine.
I made my computers
)
I made my computers visible. My computers were not hidden on purpose. I guess that was the default setting. ;)
Is there a security reason I should have left them hidden?
BTW, Einstein's ratings have the 64 bit one a 130% FP and 135% integer when compared to the same hardware running the 32 bit OS. I am discounting the extra 4GB as I suspect all of the memory isn't utilized anyway. As I imagine there is a compiler in the middle, I don't know if it uses all 30 GPRs.
Does anybody do Assy anymore? I gave up assy after my 6th or so assy language! LoL (Although I did use 8080 Assy language to learn what C was up to. Microsoft Quick C compiled to assy then got linked. I understood the assy language! ;) I also wrote a program to translate Intel assy into Zilog assy (without the Z80 enhancements) in BASIC. As I also did DEC-10 Assy I like the operand order of the Z80 better, and I could at least hand optimize the results.) (I also purchased a compiler for Microsoft Quick basic, so it was compiled code as well). I never bothered to learn assy for the VAX or Intel since the i8086. ...or the 680xx in my 3 Amigas.
Ronald McNichol wrote:I
)
no security reason at all to hide them. there is no PII in the public information about one's hosts.
it's just that European law requires explicit consent to share any data, and the project is Europe based and must follow the rule. that's why it's the default behavior.
_________________________________________________________________________