ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sat Jul 11, 2020 7:36 pm

bensimmo wrote:
Tue Jul 07, 2020 3:16 pm
Just because I was reading the thread

New processor run (in a laptop)

Code: Select all

root@G3ntleGiraffe:~/pichart-34# ./pichart-openmp -t "Ubuntu WSL2 Win10 i5-9300H"
pichart -- Raspberry Pi Performance OPENMP version 34

Prime Sieve          P=14630843 Workers=8 Sec=0.196982 Mops=4743.21
Merge Sort           N=16777216 Workers=16 Sec=0.307859 Mops=1307.92
Fourier Transform    N=4194304 Workers=8 Sec=0.182756 Mflops=2524.53
Lorenz 96            N=32768 K=16384 Workers=8 Sec=0.0630457 Mflops=51093.5

The Ubuntu WSL2 Win10 i5-9300H has Raspberry Pi ratio=149.345
Making pie charts...done.
It's amazing how well the new multi-core processors work. Thanks for posting!

I'm continuing to be surprised by the qemu-user-static package. After thinking about how it was possible to run 64-bit Intel-compatible binaries on the 4B under a 32-bit operating system, I decided it should also be possible to run 64-bit ARM binaries on a Pi Zero.

Therefore, I compiled a statically-linked version of the Pi pie chart program on a nearby 64-bit ARM-based single-board computer and copied the binary over to the unsuspecting Pi Zero. My hopes were nearly dashed to the ground by

Code: Select all

$ ./pichart-serial 
-bash: ./pichart-serial: cannot execute binary file: Exec format error
$ file pichart-serial 
pichart-serial: ELF 64-bit LSB executable, ARM aarch64, version 1 (GNU/Linux), statically linked, for GNU/Linux 3.7.0, stripped
$ cat /proc/cpuinfo 
processor   : 0
model name  : ARMv6-compatible processor rev 7 (v6l)
BogoMIPS    : 997.08
Features    : half thumb fastmult vfp edsp java tls 
CPU implementer : 0x41
CPU architecture: 7
CPU variant : 0x0
CPU part    : 0xb76
CPU revision    : 7

Hardware    : BCM2835
Revision    : 900093
Serial      : 0000000076f68901
Model       : Raspberry Pi Zero Rev 1.3
However, after a little study I created the magical file qemu-aarch64 with

Code: Select all

:qemu-aarch64:M::\x7f\x45\x4c\x46\x02\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\xb7\x00:\xff\xff\xff\xff\xff\xff\xff\x00\xff\xff\xff\xff\xff\xff\xff\xff\xfe\xff\xff\xff:/usr/bin/qemu-aarch64-static:
typed

Code: Select all

$ sudo cat qemu-aarch64 > /proc/sys/fs/binfmt_misc/register
and then obtained

Code: Select all

$ ./pichart-serial -t "Zero AARCH64"
pichart -- Raspberry Pi Performance Serial version 34

Prime Sieve          P=14630843 Workers=2 Sec=92.516 Mops=10.0991
Merge Sort           N=16777216 Workers=2 Sec=94.1978 Mops=4.27455
Fourier Transform
qemu: uncaught target signal 11 (Segmentation fault) - core dumped
Segmentation fault
The above disaster suggests there may be a need for a 64-bit version of the Pi Zero after all, because the 32-bit processor seems not able to properly run 64-bit ARM code through emulation. Who would have known molasses would ruin the taste of a raspberry pie in such a way.

Edit: It seems the aged molasses which comes with Raspbian Stretch is much better. Correct results using an older version of qemu-aarch64-static are given in the posts

viewtopic.php?p=1694822#p1694822

and

viewtopic.php?p=1694866#p1694866

that appear below.
Last edited by ejolson on Sun Jul 12, 2020 6:37 am, edited 14 times in total.

User avatar
dickon
Posts: 1550
Joined: Sun Dec 09, 2012 3:54 pm
Location: Home, just outside Reading

Re: A Pi Pie Chart

Sat Jul 11, 2020 7:49 pm

The results of that are going to be highly entertaining. It's going to be dog slow, is my prediction. Don't get me wrong: I love binfmt, and qemu, and I use them a lot (mostly for fixing up systems, or to speed up installing lots of packages, which is a very slow operation over NFS), but quick it isn't...

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sat Jul 11, 2020 7:51 pm

dickon wrote:
Sat Jul 11, 2020 7:49 pm
The results of that are going to be highly entertaining. It's going to be dog slow, is my prediction. Don't get me wrong: I love binfmt, and qemu, and I use them a lot (mostly for fixing up systems, or to speed up installing lots of packages, which is a very slow operation over NFS), but quick it isn't...
As the results come in, I'll be updating the previous post. The update for Prime Sieve is now in place and about 5 times faster than a 486 processor running native code!

viewtopic.php?p=1552972#p1552972

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sat Jul 11, 2020 8:10 pm

The entertainment seems to have ended with a segmentation fault. I've updated the post accordingly and changed my optimistic conclusion. Maybe I need to set the Cortex-A72 option in QEMU. Any help getting the rest of the benchmarks to run would be appreciated.
Last edited by ejolson on Sat Jul 11, 2020 8:58 pm, edited 2 times in total.

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sat Jul 11, 2020 8:58 pm

dickon wrote:
Sat Jul 11, 2020 7:49 pm
It's going to be dog slow, is my prediction.
To get help I showed this thread to the canine coder who immediately complained that dogs are not slow. After looking at the results, Fido coughed and then asked, do you want me to debug all the segfaults in your code? It would be quicker if you rewrote it in Rust. I nervously adjusted my social-distancing mask and remarked, I thought you said dogs were not slow.

The developer explained, software emulation is usually more strict than real hardware in catching memory bounds violations. Rather than proper n-level meta-programming such as available in FidoBASIC, you used parallel processing macros that make things impossible to read or debug. The rest sounded like barking, though I think I made out something about segfaults being the reason C exists.
Last edited by ejolson on Sat Jul 11, 2020 9:33 pm, edited 1 time in total.

User avatar
dickon
Posts: 1550
Joined: Sun Dec 09, 2012 3:54 pm
Location: Home, just outside Reading

Re: A Pi Pie Chart

Sat Jul 11, 2020 9:03 pm

I asked a cat. It confirmed I was right.

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sat Jul 11, 2020 9:25 pm

dickon wrote:
Sat Jul 11, 2020 9:03 pm
I asked a cat. It confirmed I was right.
That's what I thought. The pichart program has an option -s to skip the tests and simply draw a pie chart based on the reference data included in the program. This also caused a segmentation fault until I inserted some debugging code and now it runs fine. The printf statements must have led to some sort of quantum-mechanical effect, as the pichart.svg graph is also completely correct.

Would it be possible for that cat to look?
Last edited by ejolson on Sun Jul 12, 2020 2:25 am, edited 2 times in total.

User avatar
dickon
Posts: 1550
Joined: Sun Dec 09, 2012 3:54 pm
Location: Home, just outside Reading

Re: A Pi Pie Chart

Sat Jul 11, 2020 9:35 pm

It isn't sure, but is wondering if there's some weird alignment issue cropping up somewhere.

Then it left to catch rodents in my back garden -- it farms them -- so I think that's the last I'll see of it for the evening.

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sat Jul 11, 2020 9:59 pm

dickon wrote:
Sat Jul 11, 2020 9:35 pm
It isn't sure, but is wondering if there's some weird alignment issue cropping up somewhere.

Then it left to catch rodents in my back garden -- it farms them -- so I think that's the last I'll see of it for the evening.
I tried the exact same binary that was segfaulting using qemu-aarch64-static on the Pi Zero with the QEMU emulator on an x86 PC running Debian Linux

Code: Select all

$ grep "model name" /proc/cpuinfo 
model name  : AMD Athlon(tm) II X2 255 Processor
model name  : AMD Athlon(tm) II X2 255 Processor
$ file pichart-serial 
pichart-serial: ELF 64-bit LSB executable, ARM aarch64, version 1 (GNU/Linux), statically linked, for GNU/Linux 3.7.0, stripped
$ qemu-aarch64-static ./pichart-serial -s
pichart -- Raspberry Pi Performance Serial version 34
The Pi B+ has Raspberry Pi ratio=1
Making pie charts...done.
$ md5sum pichart-serial 
dd2cf1e1feaccbc63dbbb0aa4ddb733d  pichart-serial
Then I spun up another non-virtual Zero in the Pi cloud

viewtopic.php?f=36&t=279176

and obtained

Code: Select all

$ grep "model name" /proc/cpuinfo
model name  : ARMv6-compatible processor rev 7 (v6l)
$ file pichart-serial
pichart-serial: ELF 64-bit LSB executable, ARM aarch64, version 1 (GNU/Linux), statically linked, for GNU/Linux 3.7.0, stripped
$ qemu-aarch64-static ./pichart-serial -s
pichart -- Raspberry Pi Performance Serial version 34
The Pi B+ has Raspberry Pi ratio=1
Making pie charts...qemu: uncaught target signal 11 (Segmentation fault) - core dumped
Segmentation fault
$ md5sum pichart-serial 
dd2cf1e1feaccbc63dbbb0aa4ddb733d  pichart-serial
If QEMU was catching a segmentation violation in the pichart code, then the same emulator on x86 should produce an identical error. While this looks like a bug with qemu-aarch64-static in Raspberry Pi OS, similar things can result from using uninitialized memory. Just like an engine computer that suddenly starts working after being taken apart and back together

viewtopic.php?p=1693585#p1693585

segfaults that disappear when printf is inserted are difficult to debug.
Last edited by ejolson on Sun Jul 12, 2020 6:40 am, edited 4 times in total.

User avatar
dickon
Posts: 1550
Joined: Sun Dec 09, 2012 3:54 pm
Location: Home, just outside Reading

Re: A Pi Pie Chart

Sat Jul 11, 2020 10:04 pm

The aarch64 binary runs happily on my amd64 router / rootfs provider / monitoring machine, with a result of 'My Computer has Raspberry Pi ratio=1.92935' (whilst doing ~4Mb/s Wireguard traffic). So yeah, it's an odd bug that it doesn't work on aarch64 on armhf.

Have you tried gdb?

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sat Jul 11, 2020 10:34 pm

dickon wrote:
Sat Jul 11, 2020 10:04 pm
The aarch64 binary runs happily on my amd64 router / rootfs provider / monitoring machine, with a result of 'My Computer has Raspberry Pi ratio=1.92935' (whilst doing ~4Mb/s Wireguard traffic). So yeah, it's an odd bug that it doesn't work on aarch64 on armhf.

Have you tried gdb?
It seems to work on Raspbian Stretch. So, on with the show!

This time running on a Pi B+ we have

Code: Select all

$ file pichart-serial 
pichart-serial: ELF 64-bit LSB executable, ARM aarch64, version 1 (GNU/Linux), statically linked, for GNU/Linux 3.7.0, stripped
$ grep "model name" /proc/cpuinfo 
model name  : ARMv6-compatible processor rev 7 (v6l)
$ qemu-aarch64-static ./pichart -t "B+ AARCH64"
Error while loading ./pichart: No such file or directory
$ qemu-aarch64-static ./pichart-serial -t "B+ AARCH64"
pichart -- Raspberry Pi Performance Serial version 34

Prime Sieve          P=14630843 Workers=1 Sec=129.672 Mops=7.20532
Merge Sort           N=16777216 Workers=1 Sec=129.893 Mops=3.09987
Fourier Transform    N=4194304 Workers=2 Sec=269.462 Mflops=1.7122
Lorenz 96            N=32768 K=16384 Workers=2 Sec=1422.02 Mflops=2.26525

The B+ AARCH64 has Raspberry Pi ratio=0.0856662
Making pie charts...done.
Last edited by ejolson on Sun Jul 12, 2020 5:46 am, edited 4 times in total.

User avatar
dickon
Posts: 1550
Joined: Sun Dec 09, 2012 3:54 pm
Location: Home, just outside Reading

Re: A Pi Pie Chart

Sat Jul 11, 2020 10:43 pm

Next question: have you tried to run it in a chroot, rather than a static binary?

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sat Jul 11, 2020 11:09 pm

dickon wrote:
Sat Jul 11, 2020 10:43 pm
Next question: have you tried to run it in a chroot, rather than a static binary?
That was my original plan. After the Pi pie chart program runs stably, I'll try chroot to the full 64-bit Raspberry Pi OS and after that maybe an initial RAM file system designed to pivot root straight into the beta release. That's for a different thread.

Right now, I'm remembering other binaries in Raspbian Buster that never worked quite right and wondering whether there was a general tool-chain problem related to the 8.3.0 version of gcc used as the system compiler. Alternatively, this could be another manifestation of the rumored bit rot that has been happening to 32-bit code for some time.

viewtopic.php?f=66&t=249873

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sun Jul 12, 2020 12:40 am

I've just copied the qemu-aarch64-static binary from Raspbian Stretch over to Buster Raspberry Pi OS and it seems to be working fine. If the lid on the jar is loose, the molasses may accumulate ants and even some bugs.

The new results are

Code: Select all

$ ./qemu-aarch64-static ./pichart-serial -t "Zero AARCH64"
pichart -- Raspberry Pi Performance Serial version 34

Prime Sieve          P=14630843 Workers=2 Sec=88.1097 Mops=10.6042
Merge Sort           N=16777216 Workers=1 Sec=88.203 Mops=4.56507
Fourier Transform    N=4194304 Workers=1 Sec=175.492 Mflops=2.62903
Lorenz 96            N=32768 K=16384 Workers=1 Sec=944.574 Mflops=3.41024

The Zero AARCH64 has Raspberry Pi ratio=0.128165
Making pie charts...done.
The Pi Zero emulating a 64-bit ARM architecture Linux computer is happily twice as fast as a 486DX-66 which, for comparison, has a ratio of 0.0585114 and much less RAM. Compared with native 32-bit code, the slowdown when executing 64-bit binaries on the Zero is about 11 fold.

I've updated the original post in which the Buster executable for qemu-aarch64-static led to a segmentation fault.

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Fri Jul 17, 2020 4:07 pm

I've started a build of gcc version 10.1 on the Zero yesterday in preparation for compiling the latest aarch64 static QEMU usermode binary. The build, of course, is not being done inside any emulator. Even so, it's been running for about 10 hours and is still at stage 1.

If the state governors had been equally cautious about advancing from phase 1 to phase 2, the coronavirus cases in the States might not have doubled from 32K a day to 64K. On the other hand, since that's still in the realm of 8-bit computing, things could always go faster.

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Fri Jul 17, 2020 5:30 pm

Woohoo! The build of gcc on the Zero has advanced to stage 2. Since the Pi doesn't include any casinos or bars to open, there hopefully won't be any (power) spikes that require a return to stage 1.

I'm a worried about the beaches though, because isn't sand made from silicon?

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sat Jul 18, 2020 6:40 pm

ejolson wrote:
Fri Jul 17, 2020 5:30 pm
Woohoo! The build of gcc on the Zero has advanced to stage 2. Since the Pi doesn't include any casinos or bars to open, there hopefully won't be any (power) spikes that require a return to stage 1.

I'm a worried about the beaches though, because isn't sand made from silicon?
The bad news is that the Zero has remained in stage 2 while the infection rates continue to increase. The good news is that compiling gcc is unrelated to social distancing so progress to stage 3 should happen on schedule rather than suddenly reverting back to stage 1 as the politicians realize that wrong decisions under the present conditions noticeably affect people and need to be corrected.

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sun Jul 19, 2020 3:33 am

As in real life, things are staying in stage 2 for a long time. I hate to think how long stage 3 will last after that. Could the Zero represent the new normal? I wonder if the performance difference between the Zero and the 4B when compiling gcc is similar to the ratio given by the Pi pie chart.

jahboater
Posts: 5799
Joined: Wed Feb 04, 2015 6:38 pm
Location: West Dorset

Re: A Pi Pie Chart

Sun Jul 19, 2020 3:40 am

I would just leave the Zero for the weekend!
50 hours plus is the norm :(

I always find it amazing that such a tiny and cheap computer can complete such a huge task at all.
Pi4 8GB running PIOS64

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sun Jul 19, 2020 4:19 am

jahboater wrote:
Sun Jul 19, 2020 3:40 am
I would just leave the Zero for the weekend!
50 hours plus is the norm :(

I always find it amazing that such a tiny and cheap computer can complete such a huge task at all.
Yes, I just read your post at

viewtopic.php?p=1659200#p1659200

and am now expecting the requirement to wear a face mask may last yet another year before moving on to stage 3.

Note this is a Pi Zero compiling the C, C++ and Fortran language options with a non-encrypted iSCSI root mounted over the USB Ethernet gadget. That's a lot faster than NFS root, but still as slow as a slow thing on a slow day in Slowville.

viewtopic.php?p=1589747#p1589747

jahboater
Posts: 5799
Joined: Wed Feb 04, 2015 6:38 pm
Location: West Dorset

Re: A Pi Pie Chart

Sun Jul 19, 2020 9:17 am

ejolson wrote:
Sun Jul 19, 2020 4:19 am
Note this is a Pi Zero compiling the C, C++ and Fortran language options with a non-encrypted iSCSI root mounted over the USB Ethernet gadget. That's a lot faster than NFS root, but still as slow as a slow thing on a slow day in Slowville.
On PC's with lots of memory I do the entire download, build, install in /tmp which is a tmpfs memory disk.
It takes 1/2 an hour or less, I don't know how much of that fast time is due to the CPU speed or the very fast disk.
(Sadly the 8GB on the Pi4 isn't quite big enough to do that ... )

For slow disks, one might hope that make -jN where N is more than the number of CPU's would mean that there is always likely to be a runnable process even if some of them are held up waiting for I/O. The downside of that is more time spent context switching.

My guess is that the Pi Zero is simply CPU limited (and the the load averages will be more than one).
Pi4 8GB running PIOS64

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Sun Jul 19, 2020 6:17 pm

jahboater wrote:
Sun Jul 19, 2020 9:17 am
My guess is that the Pi Zero is simply CPU limited (and the the load averages will be more than one).
The Zero has a load average of

Code: Select all

$ uptime
 19:14:46 up 2 days, 15:55,  1 user,  load average: 1.00, 1.02, 1.00
which seems pretty reasonable. I think it is possible accessing the iSCSI network block device through a USB Ethernet gadget actually takes less CPU than the SD card. At any rate, the Zero has advanced on to stage 3, so pending a second wave, there may be hope for the world as well.

For those who have just joined, earlier posts in this thread indicated that the static binary for user-mode aarch64 emulation in 32-bit Raspberry Pi OS Buster segfaults. To work around this an older 64-bit emulator from 32-bit Raspbian Stretch was copied over and used to run the 64-bit ARM Pi pie chart executable on the Zero. While it worked, this is not an optimal solution.

The current goal is to build the latest version of qemu-aarch64-static from source using gcc version 10.1 to see if any performance improvements result when running the Pi pie chart program under 64-bit emulation on the Zero or at least whether it crashes. For luck, gcc is first being built from scratch on the Zero to make sure no ARMv7-only instructions creep in by mistake.

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Tue Jul 21, 2020 12:50 am

For some reason stage 3 didn't seem to last as long as stage 2. This good news may be related to new vaccines and antiviral treatments. In conclusion, the time needed to build gcc version 10.1 on the Zero with the configure options

Code: Select all

$ ../gcc-10.1.0/configure -v --prefix=/usr/local/gcc-10.1 \
    --enable-languages=c,c++,fortran --with-cpu=arm1176jzf-s \
    --with-fpu=vfp --with-float=hard --build=arm-linux-gnueabihf \
    --host=arm-linux-gnueabihf --target=arm-linux-gnueabihf
compared to building a similar compiler on the 4B with the configure options

Code: Select all

$ ../gcc-10.1.0/configure -v --prefix=/usr/local/gcc-10.1 \
    --enable-languages=c,c++,fortran --with-cpu=cortex-a72 \
    --with-fpu=neon-fp-armv8 --with-float=hard --build=arm-linux-gnueabihf \
    --host=arm-linux-gnueabihf --target=arm-linux-gnueabihf
may be summarized as

Code: Select all

Time in seconds to build gcc version 10.1
                User     System    Elapsed
Zero       236938.58    9893.42     268352
Pi 4B       43742.06    2962.62      13464
Relative        5.42       3.34      19.93
In particular, it took 74.54 hours to build the C, C++ and Fortran compilers on the
Zero. Compared to the build time of 70.78 hours for just the C and C++ compilers reported at

viewtopic.php?p=1659200#p1659200

it would appear there is no significant penalty for running the Pi Zero with an iSCSI root mounted through the USB Ethernet gadget.

To make sure the new compilers are working properly and return to the topic of this thread, the Pi pie chart program was compiled with gcc version 10.1 and run on each of the respective Pi computers. The results were

Code: Select all

Pi Pie Chart (larger is better)
          Prime     Sort      FFT    Lorenz   Ratio
Zero      87.78    37.65    26.82     71.50    1.41
Pi 4B   1697.53   547.48   347.77   4850.33   31.42
Relative  19.34    14.54    12.97     67.84   22.28
I found it interesting how the relative performance of 22.28 between the Zero and the 4B on the Pi pie chart program was similar to the relative times of 19.93 that it took to build the compilers. You don't see many computer vendors selling a line of computers in which the fastest is 20 times faster than the slowest.

For reference the two runs of Pi pie chart program were

Code: Select all

$ ./pichart-serial -t "Zero gcc 10.1"
pichart -- Raspberry Pi Performance Serial version 34

Prime Sieve          P=14630843 Workers=2 Sec=10.6445 Mops=87.7754
Merge Sort           N=16777216 Workers=1 Sec=10.6933 Mops=37.6548
Fourier Transform    N=4194304 Workers=1 Sec=17.2035 Mflops=26.8185
Lorenz 96            N=32768 K=16384 Workers=1 Sec=45.0499 Mflops=71.5035

The Zero gcc 10.1 has Raspberry Pi ratio=1.4089
Making pie charts...done.
and

Code: Select all

$ ./pichart-openmp -t "Pi4B gcc 10.1"
pichart -- Raspberry Pi Performance OPENMP version 34

Prime Sieve          P=14630843 Workers=4 Sec=0.550404 Mops=1697.53
Merge Sort           N=16777216 Workers=8 Sec=0.735467 Mops=547.479
Fourier Transform    N=4194304 Workers=8 Sec=1.32665 Mflops=347.772
Lorenz 96            N=32768 K=16384 Workers=4 Sec=0.664125 Mflops=4850.33

The Pi4B gcc 10.1 has Raspberry Pi ratio=31.42
Making pie charts...done.

jahboater
Posts: 5799
Joined: Wed Feb 04, 2015 6:38 pm
Location: West Dorset

Re: A Pi Pie Chart

Tue Jul 21, 2020 2:29 am

ejolson wrote:
Tue Jul 21, 2020 12:50 am
The Zero gcc 10.1 has Raspberry Pi ratio=1.4089
Which matches the clock speed ratio 1000/700 near enough.

Perhaps for interest, run the Pi4 at 2000Mhz (say) and check you get 2000/1500 extra performance?
You need over_voltage=5 by the way.
Pi4 8GB running PIOS64

ejolson
Posts: 5426
Joined: Tue Mar 18, 2014 11:47 am

Re: A Pi Pie Chart

Tue Jul 21, 2020 3:12 am

jahboater wrote:
Tue Jul 21, 2020 2:29 am
ejolson wrote:
Tue Jul 21, 2020 12:50 am
The Zero gcc 10.1 has Raspberry Pi ratio=1.4089
Which matches the clock speed ratio 1000/700 near enough.

Perhaps for interest, run the Pi4 at 2000Mhz (say) and check you get 2000/1500 extra performance?
You need over_voltage=5 by the way.
I'd be more likely to explore the other limit and try under clocking.

In either case, I'm pretty sure the 4B will not scale linearly, because merge sort and the fast Fourier transform are memory bound. As demonstrated by

viewtopic.php?p=1644489#p1644489

the processor speed is relatively much faster than memory on the 4B. The situation appears reversed with the Zero where the processor is so slow that memory speed is never an issue.

Return to “General discussion”