14" M5 Max users - Local ML / AI workload / Full load experiences?

[novagamer]

I'm wondering if anyone with a 14'" M5 Max can tell me how it's performing under full load, e.g. all 40 GPU cores saturated for hours for intense inference time or even training use.

Noise is not a concern for me at all.

I have a 16" M4 Max and I'm running into areas where I think the better GPU cores on the M5 Max will help but I also need a travel sized laptop for in-person demos (which won't be full-load for hours at a time, but I'd still be doing deep work in a hotel for example). At home I'd keep both machines and probably use them in parallel for a period of time.

Does anyone have anecdotal experience they can share? It's very difficult to find this information, everyone just runs a benchmark or two and I'm looking for real work – does it throttle to the point where performance degrades below what the M4 Max can do? Does it CRASH? (my 16" shut OFF when I pushed it too hard once recently but that may have been due to OOM / swap taking down the machine).

If the Studio is in fact delayed until fall then I need a second machine here within the next month or so.

I also don't care that much about MLX, I'm using MPS for almost everything right now for specific reasons I can't get into.

I'd really appreciate any advice that goes beyond "get the 16" I know the 16 has more thermal headroom and I'm OK with tradeoffs. At home / docked I can get some kind of cooling solution to put it in but flying all over the place with the 16" I have is cumbersome.

 

[PBG4 Dude]

nvm

 

[bnumerick]

The closest thing I'm using mine for at the moment is for Topaz AI video upscaling with their Proteus model. I've run 8-9 hour upscales on it, and as long as the high power mode is on, it is generally stable after the initial burst. I have the 16" model too, and there's about a 15% performance penalty with the 14". It's about the same speed as the 60-core M3 Ultra in this process and is ~30% faster than the 16" M4 Max.

At home / docked I can get some kind of cooling solution to put it in but flying all over the place with the 16" I have is cumbersome.

I don't think this will really help. It does not actually thermal throttle because of heat. It's throttling due to power constraints. The 14" chip doesn't have the same power envelope that the 16" does. With that said, it seems quite a bit better than the M1-M3 series (I didn't own the 14" M4 max).

 

[novagamer]

The closest thing I'm using mine for at the moment is for Topaz AI video upscaling with their Proteus model. I've run 8-9 hour upscales on it, and as long as the high power mode is on, it is generally stable after the initial burst. I have the 16" model too, and there's about a 15% performance penalty with the 14". It's about the same speed as the 60-core M3 Ultra in this process and is ~30% faster than the 16" M4 Max.



I don't think this will really help. It does not actually thermal throttle because of heat. It's throttling due to power constraints. The 14" chip doesn't have the same power envelope that the 16" does. With that said, it seems quite a bit better than the M1-M3 series (I didn't own the 14" M4 max).

Thanks, that helps a lot even though the workloads are different. Appreciate it.