Parallella was a good idea and an amazing implementation at the time... However...
It had already been overtaken and soon became a dud with little follow-on support.
Instead, at that sort of scale and electrical power, we have the RaspberryPi, FPGAs, and RISC-V.
As I'm sure you're already well experienced with:
A higher power dud that similarly was quickly overtaken and is now abandoned was the Intel "Knights Landing" "Xeon Phi" "MIC" attempt. There are various of those once very expensive CPU cards now available for a $100 or so.
For high compute, AVX looks to have won for small scale SIMD / WIP whereas the AMD and nVidia big GPUs rule supreme for big SIMD arrays.
Have you a new optimised app experiment in mind?? ;-)
Have you a new optimised app experiment in mind?? ;-)
Well, would be good, but first I should get sources and study them. It seems there are GW ones available currently, no binary pulsar search ones. And GW seems too demanding to run on anything small.
Turing Pi have announced they are working on a V2 known as the Turing Pi2. The Turing Pi used 7 Raspberry Pi Compute Modules (up to CM 3+). It was a mITX board. The Turing Pi will hold 4 CM4 modules and also provide 2 x mini PCI, 2 x Sata III, a layer 2 managed switch, USB ports and is expected to use a standard PC power connector. See Turing Pi
Over 111 pages what was consensus of Parallella perspectives for E@h ?
Parallella was a grand idea but the author/designer "disappeared" into DARPA, and it just remains as a patented design now for an array of RISC processors. FFT would definitely have been it's main purpose for E@H.
I had some big ideas once, involving a LEGO model of the dataflows, but it would require a processor array far bigger in size than was actually implemented in hard silicon.
Raistmer* wrote:
What FFT sizes are usual for E@h tasks?
Oooooh .... IIRC 222 is about right, that being about twenty minutes of the data stream from one of the GW interferometers.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Now here is a fan and a heat sync for an odroid n2 + you will have to scroll down to view the heat sync and the fan. I have two odroids. one running einstein and the other world community.
Please need some help and guidance with what to do with the loss of the beta 1.47 NEON application that was used for BPR4 tasks that the new Android beta application caused to force the incompatible 1.06 application onto my Raspberry Pi 3B+
Please need some help and guidance with what to do with the loss of the beta 1.47 NEON application that was used for BPR4 tasks that the new Android beta application caused to force the incompatible 1.06 application onto my Raspberry Pi 3B+
Over 111 pages what was
)
Over 111 pages what was consensus of Parallella perspectives for E@h ?
What FFT sizes are usual for E@h tasks?
Parallella was a good idea
)
Parallella was a good idea and an amazing implementation at the time... However...
It had already been overtaken and soon became a dud with little follow-on support.
Instead, at that sort of scale and electrical power, we have the RaspberryPi, FPGAs, and RISC-V.
As I'm sure you're already well experienced with:
A higher power dud that similarly was quickly overtaken and is now abandoned was the Intel "Knights Landing" "Xeon Phi" "MIC" attempt. There are various of those once very expensive CPU cards now available for a $100 or so.
For high compute, AVX looks to have won for small scale SIMD / WIP whereas the AMD and nVidia big GPUs rule supreme for big SIMD arrays.
Have you a new optimised app experiment in mind?? ;-)
Happy crunchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ML1 wrote: Have you a
)
Well, would be good, but first I should get sources and study them. It seems there are GW ones available currently, no binary pulsar search ones. And GW seems too demanding to run on anything small.
Turing Pi have announced they
)
Turing Pi have announced they are working on a V2 known as the Turing Pi2. The Turing Pi used 7 Raspberry Pi Compute Modules (up to CM 3+). It was a mITX board. The Turing Pi will hold 4 CM4 modules and also provide 2 x mini PCI, 2 x Sata III, a layer 2 managed switch, USB ports and is expected to use a standard PC power connector. See Turing Pi
MarksRpiCluster
Raistmer* wrote: Over 111
)
Parallella was a grand idea but the author/designer "disappeared" into DARPA, and it just remains as a patented design now for an array of RISC processors. FFT would definitely have been it's main purpose for E@H.
I had some big ideas once, involving a LEGO model of the dataflows, but it would require a processor array far bigger in size than was actually implemented in hard silicon.
Oooooh .... IIRC 222 is about right, that being about twenty minutes of the data stream from one of the GW interferometers.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
The Raspberry Pi foundation
)
The Raspberry Pi foundation have released a fan to fit inside the official Pi case. See https://www.raspberrypi.org/products/raspberry-pi-4-case-fan/?resellerType=home for all the details.
MarksRpiCluster
PorkyPies wrote: The
)
Now here is a fan and a heat sync for an odroid n2 + you will have to scroll down to view the heat sync and the fan. I have two odroids. one running einstein and the other world community.
Please need some help and
)
Please need some help and guidance with what to do with the loss of the beta 1.47 NEON application that was used for BPR4 tasks that the new Android beta application caused to force the incompatible 1.06 application onto my Raspberry Pi 3B+
All tasks erroring out now.
Keith Myers wrote: Please
)
Maybe this will help:
Thanks. I'll skip the
)
Thanks. I'll skip the anonymous platform for now. They fixed the NEON scheduler issue and I just got 1.47 tasks again.