SwissMicros Calculator Forum

StatsDoctor wrote: ↑

Thu May 23, 2019 6:15 am

It is interesting that for the iPhone, running the programs using a power supply yielded only slightly better results than running the programs using the battery. For the DM42, running the programs using a power supply yielded results that were much better than running the programs using the battery.

Nothing surprising about that at all.

The DM42 is designed specifically to push the CPU clock frequency from 25MHz up to 80MHz when external power is applied.

Thanks, grsbanks. I was not fully aware of this. I have seen a few other posts (not directed to me) where a recommendation was made to apply external power to the DM42, but the reason for doing this was not given. Now I understand the reason.

This is a really cool feature of the DM42!

A couple of weeks ago, I was reading an old PPC Notes (or something similar), and there was a letter/article about someone doing a large prime number search on their TI-59, and it took about 5 hours. I did the same on my DM42 using a Rabin-Miller prime test wrapped in an outer loop and the whole thing took about 15 seconds. I just tried it again on external power, and it was only about 5.

Dave Britten wrote: ↑

Sat May 25, 2019 3:10 pm

A couple of weeks ago, I was reading an old PPC Notes (or something similar), and there was a letter/article about someone doing a large prime number search on their TI-59, and it took about 5 hours. I did the same on my DM42 using a Rabin-Miller prime test wrapped in an outer loop and the whole thing took about 15 seconds. I just tried it again on external power, and it was only about 5.

That's it. I'm officially old. I had an SR-52 and then upgraded to a TI-59 when it was first available (1977 or so?) I was thrilled! I can't even imagine what will be available in another 40 years (not that I'll be here to see it)!

Michael must have send me the DM42 of David, it is clear now for me, why my unit was affected by the "timing issue affecting a small number of boards".


Free42Linux decimal on PC with Ryzen 7 2700X and Debian 9.9:
MATH 3'030'000
TRIG(DEG) 1'880'000
TRIG(RAD) 2'416'000

Free42Windows decimal, same PC, running Windows 2000 in VirtualBox:
MATH 16'064'000
TRIG(DEG) 1'352'000
TRIG(RAD) 2'342'000

Android Free42 on Phone with Snapdragon 820 and LineageOS 14.1:
MATH 4'810'000
TRIG(DEG) 636'000
TRIG(RAD) 1'056'000

DM42 connected with USB
MATH 75'350
TRIG(DEG) 10'925
TRIG(RAD) 17'200

DM42 on battery
MATH 31'000
TRIG(DEG) 4'250
TRIG(RAD) 6'925

Dani R. wrote: ↑

Sat May 25, 2019 9:35 pm

Free42Linux decimal on PC with Ryzen 7 2700X and Debian 9.9:
MATH 3'030'000
TRIG(DEG) 1'880'000
TRIG(RAD) 2'416'000

Free42Windows decimal, same PC, running Windows 2000 in VirtualBox:
MATH 16'064'000
TRIG(DEG) 1'352'000
TRIG(RAD) 2'342'000

Is that a typo in your MATH benchmark data, or is the 32-bit Windows build, running in a VM, really five times faster than the 64-bit Linux build, running on the metal on the same machine?

Thomas Okken wrote: ↑

Sun May 26, 2019 6:14 am

Is that a typo in your MATH benchmark data, or is the 32-bit Windows build, running in a VM, really five times faster than the 64-bit Linux build, running on the metal on the same machine?

I have a AMD Ryzen 5 2600 in the Office with native Windows 10. Tomorrow I can check Free42Windows decimal on this PC.

Yes I have double checked the program and the results, this is not a typo, Free42Windows decimal on 32-bit Windows 2000 Professional in a VM delivers me anytime a benchmark greater than 16'000'000 for MATH. But Free42Linux "only" greater than 3'000'000 for MATH.

Even worse, same Machine:
Free42Windows decimal running with wine on Debian 9.9:
MATH 363'000
TRIG(DEG) 294'000
TRIG(RAD) 346'000

So, wine is not the solution. Sometime I will set up a VM with Windows 8.1 64-bit.
(edit in bold)

Dani R. wrote: ↑

Sun May 26, 2019 11:30 am

Thomas Okken wrote: ↑

Sun May 26, 2019 6:14 am

Is that a typo in your MATH benchmark data, or is the 32-bit Windows build, running in a VM, really five times faster than the 64-bit Linux build, running on the metal on the same machine?

I have a Ryzen 7 2600 in the Office with native Windows 10. Tomorrow I can check Free42Windows decimal on this PC.

Yes I have double checked the program and the results, this is not a typo, Free42Windows decimal on 32-bit Windows 2000 Professional in a VM delivers me anytime a benchmark greater than 16'000'000 for MATH. But Free42Linux "only" greater than 3'000'000 for MATH.

Maybe its a 32-bit versus 64-bit thing. It would be interesting to compare the performance of Free42 Linux builds for the three different ABIs (x86-64, x32, and i386) supported by the 64-bit x86 Linux kernel.

Free42Windows Decimal on PC with AMD Ryzen 5 2600 running Windows 10 Enterprise LTSC
MATH 5'532'000
TRIG(DEG) 1'058'000
TRIG(RAD) 1'706'000

Free42Windows Decimal, same PC, running Windows 2000 in VMware Workstation Pro
MATH 13'554'000
TRIG(DEG) 1'172'000
TRIG(RAD) 2'030'000

Yes, Free42Windows Decimal is also here faster in the VM against the same version running on the metal of the same machine, even the TRIG benchmark.
(edit in bold)

Very weird. Is Free42 making some kind of blocking syscall that's much slower/more complex on newer operating systems?