Just wondering if the amount and speed of memory has an effect on how fast SL will process audio. I suppose it could be for either RAM or GPU memory, but I’m especially curious about the GPU memory. I’ve been looking at the AMD Strix Halo APUs which allow for up to 128 GB memory where a large amount can be assigned to the GPU.
I think the speed of transfer between the CPU’s RAM and the GPU’s RAM could be important. The upcoming Framework Desktop could be very interesting in this regard.
1 Like
It certainly could. I’m actually looking at laptops for now though, to use basically as a desktop relacement-light.
I really like the Framework approach though and I was looking at their larger laptop but they didn’t have a compelling CPU last time I looked. Maybe in the future.
I decided I would not be able to wait for the first trusted reports to appear about hands-on experience with the Framework Desktop as DAW, so I’ve gone and built another Windows 11 PC, but based on the spec and the detailed rationale around design decisions, I imagine the Framework Desktop will be an amazing machine for Steinberg products.
Next in line would be ARM-based laptops, for one simple reason: heat dissipation.
It will be a while before either AMD or Intel can catch up in this area, and IMHO, heat is the killer in a laptop for applications that need full throttle on demand. Framework have effectively designed a tiny desktop around a laptop platform, thus circumventing (pun?) the heat issue.
Yeah as soon as iLok and a couple of apps are supported natively on Win on ARM I’ll get an ARM laptop. Since I have to wait I’m looking at Strix Halo for now.
I thought you were considering coming over to the Dark Side and getting a MBP? That’s no longer in the cards?
EDIT: Maybe because of the lack of GPU support?
One does not exclude the other!
I’m in fact considering getting a MacBook Pro and a Win laptop. There are reasons for why.
3 Likes
Indeed. I still keep Windows Server 2025 running in my studio - sometimes I need it. Win11 not so much, but “reasons” there too 
I’m interested to hear your thoughts if/when you make all that happen :).
2 Likes
I took a look at memory usage yesterday and from what I can see the results pose some interesting questions (for me).
If I run SL standalone the RAM usage is really insignificant. I loaded a 3:30 tune and while the unmix song process was running usage got up to about 3.8GB GPU memory, and an additional 1.6GB RAM. This was of course with my GPU selected for AI workloads (Radeon VII). Using the CPU however RAM increased by 6.6GB during the process.
In Nuendo however activating SL as an editor or whatever it’s called took up an additional 600MB, and then the processing memory demands were the same as standalone. But, I didn’t take into account that each track that gets SL enabled eats up more system RAM. My 32GB total RAM is normally at about 16GB usage on a 1hr TV show and I managed to get that up to 24GB by enabling SL on about 10 dialog tracks leaving only 8GB unused.
Getting back to AMD Strix Halo then it poses an interesting question: How much memory to allocate to the GPU? From what I can tell some manufacturers set the default to 512MB, with the idea that software can deal with effectively “unified” memory. With a computer with 32GB of memory though it gets tricky, right? Allocate 16GB to the GPU? Then SL standalone would be fine, but I would likely run out of “RAM” in Nuendo. 8GB to the GPU could be right on the edge.
Unfortunately all options for Strix Halo (mobile) that I currently see are either 32GB or 128GB! An HP g1a ultra with 128GB is about 60% more than a ZFlow with 32GB.
I guess I could get one and just try.
The other device I was looking at was the ASUS ProArt with an HX370 and an NVidia dedicated GPU. But then it’s not the “full” cores…
So much simpler to buy Apple.
Any additional input?
1 Like