The most cost effective way to export for MacOS

[d.healey]

Looks like Ryzen might not be the best choice for a passive system due to power consumption - https://hattedsquirrel.net/2020/12/power-consumption-of-ryzen-5000-series-cpus/

This stuff gives me a headache.

[d.healey]

So I went with a Ryzen 5900 in the end (thanks Lindon). I'll probably switch to one of the APUs when they become available to the public. I made a semi-passive build as Christoph suggested, with an ITX motherboard and a humongous heatsink.

I've power limited (65W) and undervolted the CPU to keep things nice and cool. Idle temps are around 43C and I think I can improve on this with more tweaking. I've set the fans to only turn on when the temps are above 45, but I think I'll increase this to 50.

The graphics card uses more power than the CPU at idle and produces a fair bit of heat, so switching to an APU in the future should help with that.

[Natan]

@d-healey Sir, How much this Cost?
And can you Share the details of the PC?

[Lindon]

@d-healey - good grief - what a heat sink!

[d.healey]

@Natan

And I got this case - https://www.scan.co.uk/products/cooler-master-masterbox-nr200p-white-mini-tower-w-tempered-glass-window-2x-120mm-fans-usb-32-gen1-mi

The graphics card is a Palit 1050TI. I don't know if it's still possible to buy it (certainly not at a reasonable price) I've had this one for about 5 years.

[Dan Korneff]

@d-healey said in The most cost effective way to export for MacOS:

The graphics card is a Palit 1050TI. I don't know if it's still possible to buy it (certainly not at a reasonable price) I've had this one for about 5 years.

I'm using an RTX3090 here. Lightning fast for 3D renders, and I mine Bitcoin when it's not in use

[d.healey]

@dustbro said in The most cost effective way to export for MacOS:

RTX3090

Is it passive?

[Dan Korneff]

@d-healey unfortunately not. It's has a cooling system that rivals a commercial rooftop HVAC

[d.healey]

@Christoph-Hart What kind of temps do you see with your semi-passive system? Now I've put my system into its new case the temps are pretty firm at 52C but it's a little toasty for my taste.

[Christoph Hart]

A CPU can handle 100C so I wouldn‘t worry.

[d.healey]

I changed the CPU governor from schedutil to conservative and now I'm at 45C.

[d.healey]

I just did a benchmark with the new CPU compiling HISE (debug build so linking time is negligable).

8 threads: 1.50 minutes
10 threads: 1.51 minutes
20 threads: 1.37 minutes

So my recommendation is not to spend extra on a 5900x if your main use is software development.

[d.healey]

Digging up this thread because I'm going to have a go building an Alder Lake system and see how it compares with the Ryzen.

My updated build times with the Ryzen 5900x since I did more tweaking to my system settings over the last few months

8 threads: 1m 5s
10 threads: 0m 57s
12 threads: 0m 51s
20 threads: 0m 47s
22 threads: 0m 46s

The extra cores/threads still show diminishing returns.

[d.healey]

I've been using the Alder Lake system for a while now but today Debian got Linux 5.18 which adds some 12th gen scheduler improvements so I wanted to compare the 12600k compile times against the Ryzen.

8 threads: 1m 5s
10 threads: 1m 2s
12 threads: 0m 57s
14 threads: 0m 55s

The results aren't bad at all. 8 threads compiled in exactly the same time as on the Ryzen. The performance doesn't improve much though until 12 threads where the 12600 is trailing behind the Ryzen a little, and at 14 threads we're seeing diminishing returns.

So in my test the Alder Lake was a little slower than the Ryzen at anything above 8 threads. However there is one great benefit (for me) with the Alder Lake, it runs much much cooler than the Ryzen and I've got it installed in a completely passively cooled case, there are 0 fans, 0 noise, 0.1 dust. Compiling on the Ryzen always pushed the temps up to 70-80 and revved up the fans. I'll happily wait an extra 10 seconds in silence :)

[Christoph Hart]

@d-healey Are these times for a full rebuild of HISE develop in Release mode? If yes, this looks awesome (I think I'm at about 2 minutes on my 9900K system).

[orange]

@d-healey 1 minutes? Very good results.

About 2 minutes on 10th gen i9 10900K system (128 GB RAM) and 2 minutes on M1 Max (96 GB RAM) here.

[d.healey]

@Christoph-Hart This is the debug compiling only, so it doesn't include optimized linking because linking is single threaded and my original tests on the Ryzen were to compare the result with different thread counts. However I'm now using the Mold linker which I think supports multi-thread linking so I'll test the full build time on the Alder Lake with a release build and see what I get. I think Mold is only available on GNU/Linux (possibly MacOS).

[d.healey]

Release build with 14 threads takes 4m 18.529s. I image the i9 would be considerably faster.