DAW benchmarks on new Mac M4 chip

[Lerian]

Thanks for your interesting point of view, I am curious about the benchmark between Nvidia GPU and Apple GPU

4090 kind of crashes the apple gpu, both in 3D rendering and LLM processing (except apple's MLX).

M4 Ultra is supposed to marginally beat 4090, but at that time 5090 will be here too, and if the difference will be the same as 3090-4090, it will have double the performance of 4090 in LLM, and at least 150% in 3D.



Also Apple claims are not entirely true.

[PieBerger]

2nd tier M4 Mini; 16GB RAM, 512GB storage, is only £799 retail in the UK, that is insanely good value (for Apple) and I could even justify paying the tax for extra RAM and or storage at that price.

[keyman_sam]

I've just got a mac mini m4 pro and it is also silent in use so far.
The mac mini also only cost me £1600, so cheaper than an rtx 4090 without the noise.

Can you do some benchmarks on logic? Do you notice any noise under heavy load?

[jonljacobi]

I've just got a mac mini m4 pro and it is also silent in use so far.
The mac mini also only cost me £1600, so cheaper than an rtx 4090 without the noise.

Can you do some benchmarks on logic? Do you notice any noise under heavy load?

Check out the video in the first message in this thread.

[jens]

4090 kind of crashes the apple gpu, both in 3D rendering and LLM processing (except apple's MLX).

M4 Ultra is supposed to marginally beat 4090, but at that time 5090 will be here too, and if the difference will be the same as 3090-4090, it will have double the performance of 4090 in LLM, and at least 150% in 3D.

The other week I put together a new desktop machine at work. I had purchased a RTX4060 for it and was shocked at how huge the f**king thing is. It's bigger than a whole Mac Mini - and needs a lot more juice too (also has two f**king fans for crying out loud) - and a 4090 certainly isn't more modest than that. So these benchmarks just prove how amazingly efficent and powerful the Apple chips are.

[machinesworking]

Never fails, any talk about here turns into a shit ton of garbage mac vs pc nonsense.

[jonljacobi]

I believe the original ARM was observed still running on the trace power left in a machine when the main power was pulled.
https://arstechnica.com/gadgets/2022/09 ... irst-chip/
Obviously, they've come a long way since then.

[Kypresso]

I've just got a mac mini m4 pro and it is also silent in use so far.
The mac mini also only cost me £1600, so cheaper than an rtx 4090 without the noise.

Can you do some benchmarks on logic? Do you notice any noise under heavy load?

Sorry, i dont use logic.

[jens]

I believe the original ARM was observed still running on the trace power left in a machine when the main power was pulled.
https://arstechnica.com/gadgets/2022/09 ... irst-chip/
Obviously, they've come a long way since then.

Yes (well, almost - iIrc they made a mistake with their test-board so that the CPU wasn't actually supplied with its own power but was running fine nonetheless).

They had no money, so in order to keep the production costs as low as possible they wanted a plastic casing for the die which is why it was important that it would't get too hot... in the end it was several times as fast as all the competition per clock-cycle.

[Lerian]

The other week I put together a new desktop machine at work. I had purchased a RTX4060 for it and was shocked at how huge the f**king thing is. It's bigger than a whole Mac Mini - and needs a lot more juice too (also has two f**king fans for crying out loud) - and a 4090 certainly isn't more modest than that. So these benchmarks just prove how amazingly efficent and powerful the Apple chips are.

They are amazing in the sense that everything is amazing at doing very few things well. They have the new ARM design, which means very specialized type of processing, having parts of the chips designed specifically for some types of jobs. H264 is hard coded in the chip, and so are other flows. But if you want to do something new, that the chip wasn't designed for, its struggling. That's why the difference in size and consumption - the nvidia can do a broader range of jobs, faster. And so are the x86/64 chips. Apple M is a very specialized, consumer-targeted chip.

[keyman_sam]

I've just got a mac mini m4 pro and it is also silent in use so far.
The mac mini also only cost me £1600, so cheaper than an rtx 4090 without the noise.

Can you do some benchmarks on logic? Do you notice any noise under heavy load?

Sorry, i dont use logic.

What daw do you use and how is the noise level under load?
That’s the only thing holding me back from upgrading from my MBA M1.

[jonljacobi]

The other week I put together a new desktop machine at work. I had purchased a RTX4060 for it and was shocked at how huge the f**king thing is. It's bigger than a whole Mac Mini - and needs a lot more juice too (also has two f**king fans for crying out loud) - and a 4090 certainly isn't more modest than that. So these benchmarks just prove how amazingly efficent and powerful the Apple chips are.

They are amazing in the sense that everything is amazing at doing very few things well. They have the new ARM design, which means very specialized type of processing, having parts of the chips designed specifically for some types of jobs. H264 is hard coded in the chip, and so are other flows. But if you want to do something new, that the chip wasn't designed for, its struggling. That's why the difference in size and consumption - the nvidia can do a broader range of jobs, faster. And so are the x86/64 chips. Apple M is a very specialized, consumer-targeted chip.

I'm just like ARM... I do very few things well! Or was that a few things very well. Just kidding... I would love to see benchmarks comparing the efficiency per cycle. But aren't NVidia's current chips RISC as well? I believe they're also releasing an ARM design next year.

But I'm going to test an M4 and see if I should sell my M1 Studio and upgrade. The numbers are pretty outstanding.

[Scotty]

I am interested in your take on this but don't wish to hijack the thread. I'll send you a PM.

They are amazing in the sense that everything is amazing at doing very few things well. They have the new ARM design, which means very specialized type of processing, having parts of the chips designed specifically for some types of jobs. H264 is hard coded in the chip, and so are other flows. But if you want to do something new, that the chip wasn't designed for, its struggling. That's why the difference in size and consumption - the nvidia can do a broader range of jobs, faster. And so are the x86/64 chips. Apple M is a very specialized, consumer-targeted chip.

[Gam456]

M4 Ultra is supposed to marginally beat 4090, but at that time 5090 will be here too, and if the difference will be the same as 3090-4090, it will have double the performance of 4090 in LLM, and at least 150% in 3D.

4090 ---> 28.3 TFLOPS
M4 Max -----> 4.6 TFLOPS

Do you really believe Apple can fill the gap ?

[jens]

The other week I put together a new desktop machine at work. I had purchased a RTX4060 for it and was shocked at how huge the f**king thing is. It's bigger than a whole Mac Mini - and needs a lot more juice too (also has two f**king fans for crying out loud) - and a 4090 certainly isn't more modest than that. So these benchmarks just prove how amazingly efficent and powerful the Apple chips are.

They are amazing in the sense that everything is amazing at doing very few things well. They have the new ARM design, which means very specialized type of processing, having parts of the chips designed specifically for some types of jobs. H264 is hard coded in the chip, and so are other flows. But if you want to do something new, that the chip wasn't designed for, its struggling. That's why the difference in size and consumption - the nvidia can do a broader range of jobs, faster. And so are the x86/64 chips. Apple M is a very specialized, consumer-targeted chip.

Hang on - I thought a GPU was just about the most specialized thing ever - which is outstanding at exactly one thing: massive parallel processing?

H.264 (and .265) is hardcoded into Intel CPUs as well, so I'm not really sure why you brought that one up. (In both cases that doesn't mean they couldn't compute the codecs otherwise by the way - it's just a lot more efficient this way.)


Don't get me wrong: your reply sounded good and expertish - it's just that it wasn't really.