Igor on ARM M1/2 in Windows virtualization?

Share icon

Hello,

before I start with what looks like crazy afternoon, may be someone tried already... If yes, can you comment, please? 

Supposedly, we can run Parallels on Mac M1/M2 devices to virtualize ARM for Windows; Windows 11 ARM is supposedly free as some kind of technical preview or insider preview or something; and Windows 11 ARM supposedly CAN run SOME Windows x86 applications in emulation (kind of Rosetta 2 for Windows?). 

So this will run virtual Windows in Mac, in which will emulate x86 processor for x86 Windows applications... Sounds crazy. But web pages I found talk about usable gaming in this setup, which seems bonkers. 

Anyone tried to run Igor Pro in this crazy setup? I just wonder if it works and how well. If no one tried yet, I may try it sometimes on my little test mac mini M1, 8GB RAM, 256GM SSD... What can go wrong? 

OK, 

nothing went wrong, this craziness actually works... Igor starts and runs. Need to do more testing with my own code and use to see how reliable it is. Seems reasonable from my quick test. 

I installed current Parallels on Mac mini M1 with 8GB RAM. No issues. I installed WIndows 11Home ARM (Parallels have button for that, so this is automatic procedure). WIndows ARM64 started, I updated Windows with current updates. Installed Igor 9.01 Windows x64 version in demo mode. Visual C/C++ installation failed, which caused Igor to fail to start (not to install, though), had to install manually Visual C/C++ ARM64 version, then IP9 starts without issues.  

So, Igor starts and runs. I tested benchmark 2.02 which came with IP 9.01. 

VM environment: 4CPUs, 3GBRAM, 1.5GB graphics RAM. No limits on resources. These are default Parallels settings for this mini, sounds kind of limited. 

test on Microsoft Windows 11 Home (21H2)10.0.22000.856 using 9.01 and 21 passes; 
Windows ARM64, Parallels virtual machine on mac Mini M1, 8GB RAM 
  Create new graph  time:   159.71ms, relative speed=   1.87
  big data update  time:    123.73ms, relative speed=   1.16
  curve fit  time:  1.41ms, relative speed=     1.34
  user curve fit  time:     16.66ms, relative speed=    2.13
  double complex fft  time:     178.50µs, relative speed=  8.25
  single complex fft  time:     164.88µs, relative speed=  7.21
  double real fft  time:    94.53µs, relative speed=   5.97
  single real fft  time:    87.42µs, relative speed=   5.63
  5 pass smooth  time:  217.78µs, relative speed=  2.49
  Sort 8192 points  time:   22.51ms, relative speed=    1.70
  WaveStats  time:  542.81µs, relative speed=  0.49
  simple eqn  time:     698.54µs, relative speed=  2.00
  exp eqn  time:    913.51µs, relative speed=  2.20
  sqrt eqn  time:   778.51µs, relative speed=  2.27
  sin eqn  time:    583.37µs, relative speed=  1.99
  User fit fctn  time:  464.33µs, relative speed=  2.78
  MatrixOp eqn  time:   52.60µs, relative speed=   0.45
  **** done ****
  total test time=   10.1046

 For comparison, here is IP9 same benchmark test running directly in that Mini. Therefore IP9.01 (x86) for macOS running through Rosetta 2:

**** test on macOS12.5.0 using 9.01 and 21 passes; Mac mini M1 w/8GB RAM
  Create new graph  time:   98.74ms, relative speed=    3.03
  big data update  time:    27.38ms, relative speed=    5.23
  curve fit  time:  174.87µs, relative speed=  10.79
  user curve fit  time:     3.88ms, relative speed=     9.14
  double complex fft  time:     156.89µs, relative speed=  9.39
  single complex fft  time:     133.11µs, relative speed=  8.92
  double real fft  time:    91.86µs, relative speed=   6.14
  single real fft  time:    82.74µs, relative speed=   5.95
  5 pass smooth  time:  215.12µs, relative speed=  2.52
  Sort 8192 points  time:   4.70ms, relative speed=     8.13
  WaveStats  time:  81.79µs, relative speed=   3.26
  simple eqn  time:     223.00µs, relative speed=  6.26
  exp eqn  time:    286.66µs, relative speed=  7.01
  sqrt eqn  time:   258.88µs, relative speed=  6.83
  sin eqn  time:    189.91µs, relative speed=  6.12
  User fit fctn  time:  127.14µs, relative speed=  10.17
  MatrixOp eqn  time:   28.23µs, relative speed=   0.85
  **** done ****
  total test time=   4.02395

Looks kind of related to cpu and memory available to the Igor...  

And for comparison, here is my Macbook Pro with i7, 16GB RAM :

**** test on macOS12.5.0 using 9.01 and 21 passes; 2017 MBP, i7, 16GB RAM
  Create new graph  time:   182.13ms, relative speed=   1.64
  big data update  time:    46.79ms, relative speed=    3.06
  curve fit  time:  288.72µs, relative speed=  6.53
  user curve fit  time:     3.80ms, relative speed=     9.34
  double complex fft  time:     395.58µs, relative speed=  3.72
  single complex fft  time:     344.88µs, relative speed=  3.44
  double real fft  time:    179.11µs, relative speed=  3.15
  single real fft  time:    180.28µs, relative speed=  2.73
  5 pass smooth  time:  144.96µs, relative speed=  3.74
  Sort 8192 points  time:   5.87ms, relative speed=     6.51
  WaveStats  time:  104.18µs, relative speed=  2.56
  simple eqn  time:     166.36µs, relative speed=  8.39
  exp eqn  time:    225.11µs, relative speed=  8.93
  sqrt eqn  time:   204.29µs, relative speed=  8.66
  sin eqn  time:    174.21µs, relative speed=  6.68
  User fit fctn  time:  123.69µs, relative speed=  10.45
  MatrixOp eqn  time:   53.32µs, relative speed=   0.45
  **** done ****
  total test time=   6.93114

Ouch, that mac mini with 8GB RAM is faster than my i7 MBP and IP9 in ARM64 bit virtual machine is only slightly slower. 

August 12, 2022 at 09:38 am - Permalink