Apple Unveils New M1 Pro, Monster M1 Max SoCs
There have been rumors of a new M1 processor practically since Apple kicked the M1 out the door. Today, Apple finally took the wraps off what it had planned. The M1 Pro and M1 Max pack a lot of firepower, at least as far as specs are concerned. We’ll see how they perform in the not-too-distant future.
The M1 Pro packs up to 10 CPU cores (8 high-performance, 2 high efficiency) and pairs it with up to 16 GPU cores. There are 33.7 billion transistors in the product and the SoC offers 200GB/s of memory bandwidth across a 256-bit memory bus. The eight high performance cores offer a total of 24MB of L2 cache, with a 192KB instruction cache and 128KB data caches. The data cache sizes for the M1 Pro and M1 Max are scaled up to match the increased core count.
The M1 Pro’s GPU offers 2048 execution units in 16 graphics cores, with 5.2TFLOPs of processing and an 82 gigapixel fill rate. We’ll have to wait for benchmarks to see how well the onboard GPU scales, but Apple has some predictions of its own (more on those shortly):
The M1 Max is a 57 billion transistor SoC with a 32-core GPU, up to 64GB of onboard RAM, a 512-bit memory interface, 400GB/s of memory bandwidth, a 10-core CPU, up to 32 GPU cores, and support for four external displays. Apple is claiming that its new chip can outperform a Core i9 by 3x – 20x in machine learning performance.
The M1 Max’s beefed-up transistor count is largely spent on the GPU core, since the CPU core configuration is the same between the two chips. Apple also predicts it can deliver the equivalent performance of a 160W laptop in a 60W power envelope:
Overall, Apple states a 2.5x – 3x GPU performance improvement at equivalent power. Apple’s previous MacBook Pro systems used GPUs based on AMD’s RDNA silicon, and while either Ampere or RDNA2 would likely score better, they wouldn’t close a claimed 2.5x – 3x efficiency gap relative to RDNA. The company has pulled off a real improvement to overall efficiency if its benchmark figures prove true.
The M1 Pro and M1 Max double the maximum number of high performance cores and offer 3x – 4x the memory bandwidth depending on the model you choose. The memory loadout and high bandwidth are intended to keep the GPU fed. The implication of Apple’s presentation is that its CPU and GPU efficiency are high enough that it can combine two power-hungry parts together on the same silicon without taking a hefty power penalty.
We’re not going to prejudge the CPUs before they can be tested, but Apple clearly believes it can keep scaling its CPUs until it replaces the entire x86 product line. The performance claims here are strong enough to still be impressive after accounting for typical corporate marketing and cherry-picked tests. Apple is essentially claiming that the M1 Pro and M1 Max can deliver far more performance in a mobile system than we’ve seen previously from any x86 CPU or SoC. The impact of translation software, where applicable, will always exact a penalty, but Apple has beefed up its SoC design a great deal in just 12 months. The M1 Pro and M1 Max are likely to threaten AMD, Intel, and Nvidia in markets that the original M1 CPU couldn’t touch and offering eight high-performance cores puts the M1 Pro and M1 Max on par with the latest SoCs from AMD and Intel.