It will be nice to compare with dual-channel RAM interleaving on/off, and Hyper-Threading on/off.
Suggestions for setting up a computer for /e/OS building (and of course other Android Open Source versions...)
Also using cache or not?
Without caching, it’s quite sure that performances will decrease
But you may try, this world of ours is full of surprises!
As an example of unexpected, I remember an attempt on a dual E5xx Xeon with more than 256GB RAM (quad channel) was not as fast as a simple i7 with 64GB.
Culprit: the NUMA interleaving. Also Ubuntu is not very good at SMP tuning, should have needed some kernel tuning.
I also have a project to attempt building on a ppc64 (IBM E9xx Power9), but this will take time and I don’t have much for now
The conclusions (at least for my hardware or similar hardware):
(I) Fast M.2 SSD does not improve build performance over SATA/600 SSD
(II) Cache is slightly decreasing build performance
(III) Without SMT/HT: performance decrease about 15%
(IV) More RAM: minimal improvement
So a fast CPU with many cores is a plus, cache and more RAM than 32GB have minimal impact.
Would be interesting if and how 128GB and a ramdisk perform
That match my feelings: compiling speed is mostly related to pure, raw, computing performance.
Thanks @ronnz98 for your work and feedback
Would be nice from compiler to be able to use GPU!
I suppose this would not have much of an impact
or if money does not play a role:
here some 2022 benchmark numbers and core/mem-ratio +storage behaviour comments - https://groups.google.com/g/android-building/c/Q76dYAhwrkA
Didn’t yet build on a machine with that many core/threads (>32), so I maxed them out before anything else. Core to mem ratio as 1:2 feels true. Cache on nvme didn’t matter much. With 4-cores, you can even build on sata hdd to keep them cores busy.
If price isn’t the limit, I personally go for https://ark.intel.com/content/www/us/en/ark/products/series/232643/intel-xeon-cpu-max-series.html
In the past, I built some with a bi-Xeon Gold 5xxx (2 * 32 cores), came to the conclusion that a high number of cores (or sockets ?) may need some OS or compiler tuning … I saw performance increase staling with more than ~40 cores.
When it comes to high level performance, with the costs associated, I now prefer to use cloud VM with Docker method. It can be up and building within minutes, and my home keeps silent
For example, CCX63 at Hetzner, costs less than a good beer to produce a ROM image.
ah cool. When I have too much money I in the future I will setup a build monster
Just saw a good offer if anyone is interested:
and btw if anyone in Germany is looking for an upgrade, I sell my old hardware as upgrade kit:
CPU: AMD Ryzen 9 5950X AM4
RAM: Kingston FURY DIMM 64 GB DDR4-3600 (2x 32 GB)
CPU Cooler: be quiet! Shadow Rock Slim
Mainboard: AM4 ASUS PRIME X570-P
I wish I could build on this https://www.ibm.com/products/power-e1080
And yes, it can run Linux