Apple's M5 Max Chip Achieves a New Record in First Benchmark Result

[chucker23n1]

The table is missing the New M5-Pro,
It’s only showing the M5-Max.

The article isn’t about the M5 Pro.

 

[attila]

Really hope they come out with a M5 Ultra with 1TB+ memory Studio.. literally everyone wants to run local LLMs. Come on Apple!

No, not literally everyone. Many, perhaps!

 

[chars1ub0w]

More memory bandwidth helps only if you have enough cores to saturate it, and even then only makes a difference int workloads that require more bandwidth.

No. As others have pointed out elsewhere in this thread, CPU cores don't make use of the extra bandwidth provided by Max vs. Pro. Only the GPU cores do. So if the code is CPU-code, one can ignore the extra bandwidth, as the Pro and Max models will perform the same.

 

[cptcaveman]

As I posted in another thread, I think the drop is for two reasons:

(1) the M5 is the limit of the M1 architecture, and 12 p-cores had become unwieldy to manage in a laptop's thermal and noise envelope (the M4 Max, which I own, can get quite noisy when fully pushed, although thankfully it is silent 97% of the time);

Even if they could swing 12 p-cores somewhat acceptably, they probably see the writing on the wall as the 2nm shrink will provide ever-diminishing returns.

(2) the new chiplet architecture, surely inspired by AMD and Intel, will allow them to finally take better advantage of the traditional power and thermal advantages of the desktop. In other words, expect the Mac Studio to have more super cores. They might give us 12 super cores, 6 p-cores, etc. There will be a real distinction again b/t desktop and laptop.

And whether it's the M5 Studio, or M6, or the next Ultra, I'm also expecting the GPU to be unleashed with higher power draw and more cores. That's my prediction.

The M4->M5 GPU gain was 40% or so (raster performance), yet only 20% for the M4 Max to M5 Max. I suspect it's due to power draw limitations of the laptop architecture. Going forward, Apple may now have the flexibility to create its own quasi-GPU cards, and my sense is they are positioning to challenge Nvidia on both the graphics and AI front.

Not really. Could well be:

• ~15–25% overall performance improvement vs M5 Max
• ~20–30% better power efficiency from the 2 nm node
• Slightly improved thermals thanks to vapor chamber heat spreading
• Higher sustained clocks under long workloads (less throttling)
• Potential for a slightly thinner chassis because vapor chambers distribute heat more evenly than traditional heat pipes