Snapdragon ARM processor with Stata

[Said Jafar]

Dear members,

I'm considering running antient Stata 13 MP on a Lenovo ThinkPad T14 equipped with a Snapdragon ARM processor (running Windows 11), and I'm trying to understand whether this setup is practically usable.
Has anyone here had experience running Stata 13 MP (or a similar version) on this or a comparable ARM-based Windows device?

I read somewhere that some other software can be used with Snapdragon under emulation. If this is the case for Stata 13, then how well Stata runs (performance, speed, responsiveness) under emulation? Any shared experience or guidance would be very much appreciated.

(I also understand newer Stata versions may have better support, but due to licensing limitations not much added value for me, I’ll be sticking with version 13 MP for now.)

[Erik Ruzek]

I can confirm that v16 runs on the Snapdragon processor. I don't see why v13 wouldn't. Microsoft has a really good emulation layer for x86/64 applications, called Prism, which handles everything behind the scenes.

[Said Jafar]

Thank you Erik Ruzek. If it works for v16, then shall also work for v13.

Could you say, how large and computationally intensive are the datasets you use? And, could you say, do you observe a lag in performance, speed, responsiveness?

[Erik Ruzek]

Sure. Lag is miniscule to non-existent. I don't think dataset size should make any difference. All the work is in translating the x86/64 code into ARM64 code and optimization, which is what Prism does. To give you a sense of speed, I ran this speed test code that George Ford created on a fairly powerful Intel-based laptop I had from my employer and my personal Snapdragon based machine. Both tests were run on Stata 16 MP (4 cores). Results are below:

Intel Mobile i7-11800H @ 2.3Ghz, turbo to 4.05Ghz
variable	p50	mean	sd	min	max
replace	0.0055	0.0052	0.001398	0.003	0.007
regress	0.0305	0.031	0.005228	0.024	0.038
predict	0.01	0.011	0.005754	0.006	0.026
correl	0.0225	0.0208	0.00478	0.014	0.029
bootstrap	7.194	7.0282	0.773574	6.028	8.279
mvtest	0.1045	0.1059	0.007415	0.096	0.122
xtile	0.5985	0.6008	0.028161	0.565	0.67
expand_drop	0.056	0.0571	0.006658	0.046	0.069
arfima	3.649	3.7136	0.180164	3.512	4.035
eigenv	0.229	0.2427	0.04434	0.219	0.366

Lenovo Yoga Slim 7x Snapdragon X1E78100 3.42 Ghz
variable	p50	mean	sd	min	max
replace	0.005	0.005	0.000471	0.004	0.006
regress	0.034	0.0341	0.001197	0.033	0.037
predict	0.009	0.0129	0.010203	0.008	0.041
correl	0.0205	0.0225	0.005874	0.02	0.039
bootstrap	11.0765	11.1298	0.142998	11.014	11.406
mvtest	0.1485	0.1569	0.016974	0.144	0.186
xtile	1.047	1.0558	0.031871	1.025	1.13
expand_drop	0.064	0.0642	0.001814	0.06	0.066
arfima	5.715	5.7719	0.15724	5.635	6.177
eigenv	0.22	0.2232	0.007361	0.219	0.241

These results tell me that, on average, this chip is competitive with really good mobile x86/64 chips. However, without a Stata compiled for ARM64, we don't know just how much faster this chip might be than Intel/AMD. When Stata Corp recompiled for Apple Silicon (ARM-based), they saw a 30-35% boost compared to Intel Macs at that time.

[Said Jafar]

Thank you very much, Erik Ruzek. for very comprehensive response. This answers my question. Time to say good bye to 5-6 hours battery life and move to Snapdragons.

[George Ford]

That's great information. I was shying away from SnapDragon for speed worries, but they appear unwarranted, except perhaps for large scale bootstrapping, which I wouldn't do on a laptop anyway.

[Erik Ruzek]

I agree with both of you. I love this laptop. The worst thing is that some arguably critical software, such as Cisco VPN or Box, do not have an ARM native Windows app (yet) and for some reason, you cannot run them via emulation. Cisco VPN and Box are the only critical ones that I use that cannot be emulated. Both companies are working on native ARM versions, but you are at the mercy of the company for this, unfortunately.