MBP 14" 8/14/32GB/1TB - Java build time comparison

[tomstone74]

Hello,

have my refurbed MBP 14" 8c/14c/32GB/1TB for one week now and thought I share my initial personal impression in terms of Java build times using the RxJava (https://github.com/ReactiveX/RxJava) project available on Github as an example. With > 10K tests being executed during a full build, I wouldn't consider that project as small .

Machines​

I have three machines to work with in this Java build time comparison:

• MBP 14" M1 Pro 8c/14c/32GB/1TB (private use)
  • MacOS 12.6.1
  • Power supply: Battery only
  • JDK: Temurin 11.0.17 arm64 (native Apple Silicon)
• MBP 16" M1 Max 10c/32c/64GB/1TB (daily job use)
  • MacOS 12.6.1
  • Power supply: Battery only
  • JDK: Temurin 11.0.17 arm64 (native Apple Silicon) + tests with x64 (non-native)
• DeskMini X300 6c(12t)/64GB/1TB (private use)
  • Windows 11
  • JDK: Temurin 11.0.17 x64
  • CPU: AMD 5600G
  • Memory: Crucial Ballistix SO-DIMM Kit 64GB, DDR4-3200, CL16-18-18-36
  • Storage: WD Black SN750 NVMe 1TB
  • Fan: Noctua NH-L9a-AM4

Test Runs / Results​

A test run basically contained of, thus a full build from scratch due to cleaning first:

gradlew clean
gradlew build

Run#	MBP 14" M1 Pro 8c	MBP 16" M1 Max 10c	DM X300 6c/12t
#1	3m46s	4m2s	6m59s (***)
#2	3m41s	4m1s	5m57s (***)
#3	3m38s	4m5s	7m35s (***)
#4	3m41s	3m20s (*)	5m37s (*)(***)
#5	3m37s	3m17s (*)	5m4s (*)(***)
#6	3m38s	3m18s (*)	5m13s (*)(***)
#7	-	4m32s (*)(**)
#8	-	4m33s (*)(**)
#9	-	4m39s (*)(**)

(*) Excluded development directory from Microsoft Defender background scans. No MS Defender installed on the 14".
(**) Using Non-native Apple Silicon Temurin 11.0.17 x64
(***) Partly audible fan noise > 2000 rpm

I would consider this test as purely CPU-bound where 16GB/32GB/64GB RAM does not really matter.

Interesting to see:

• The consistent results on the MBPs vs. some shakiness on the DeskMini
• The obvious impact from background antivirus stuff covering your development directory, which makes the MBP 16" M1 Max even slower than the 14" M1 Pro, which does not have Microsoft Defender installed (#1-3)
• The MBP 16" M1 Max being faster than the MBP 14" M1 Pro (#4-6) - heck, costs a ton more - but not THAT much
• The impact on using a non-native x64 JDK on the MBP 16" (#7-9 vs. #4-6) giving you ~ 30% slower build times and making it slower than the MBP 14" M1 Pro with a arm64 JDK
• Both MBPs, dead silent, no fans kicking, whatever these builds are kicking off vs. the DeskMini partly > 2000rpm, audible, although already using a silent Noctua fan

Simply just the right decision - at least for me personally - going smaller and "cheaper" with the M1 Pro 8c/14c, although with a 32GB RAM bump to have more head room moving forward with containers etc ... vs. the pure base model.

The 14" M1 Pro is a beauty and a beast of machine, with some "wobbling" of the Touch ID key (compared to the MBP 16") though, although functional, but perhaps some sort of replacement needed at some point.

 

[maflynn]

Thanks for the info, but I'm a little confused.

Why is the 14" MBP beating out the 16" MBP when the 16" has more cores and ram?

 

[tomstone74]

If you mean:

• #1-3: I forgot that the MBP 16" M1 Max had Microsoft Defender installed and didn't have the development directory of the builds excluded from background scans. #4-6 is more representative in this regard, cause here I had the development directory excluded from background scans.
• #7-9: This was with a non-native x64 JDK on the MBP 16" Apple Silicon, showing the impact when going through Rosetta 2

Hope this makes sense.

 

[maflynn]

Hope this makes sense.

Are you not running the same the test the same exact way on all machines?

Its hard to make an apples to apples comparison when one computer is running a different version of the software and/or other configurations (like antivirurs being active)

 

[tomstone74]

Are you not running the same the test the same exact way on all machines?

Its hard to make an apples to apples comparison when one computer is running a different version of the software and/or other configurations (like antivirurs being active)

Ideally yes (of course), but I'm not allowed to uninstall MS defender from my day job MBP 16" and I won't install it on my private 14" just for this test purpose . So, the best I could do was to exclude my entire software development directory on disk from MS defender background/real-time scans, where also RxJava is checked out.

Even if the 2 additional cores would give you in theory a ~ 33% decreased build time (which I think will be only in theory, not strictly scaling linearly in practice), the M1 Pro 8c/14c is plenty of power, saving quite some bucks.