Apple M1 vs AMD Ryzen 9 8945H vs Apple M1 Max

Apple M1

The Apple M1 is a System on a Chip (SoC) from Apple that is found in the late 2020 MacBook Air, MacBook Pro 13, and Mac Mini. It offers 8 cores divided in four performance cores and four power-efficiency cores. The big cores offer 192 KB instruction cache, 128 KB data cache, and 12 MB shared L2 cache. According to Apple the performance of these cores should be better than anything on the market (in late 2020). The four efficiency cores are a lot smaller and offer only 128 KB instruction cache, 64 KB data cache, and 4 MB shared cache. The efficiency cores (E cluster) clock with 600 - 2064 MHz, the performance cores (P cluster) with 600 - 3204 MHz.

The M1 is available in two TDP variants, a passive cooled 10 Watt variant for the MacBook Air and an active cooled faster variant for the MacBook Pro 13 and Mac Mini. Those should offer a better-sustained performance according to Apple.

The integrated graphics card in the M1 offers 8 cores (7 cores in the entry MacBook Air) and a peak performance of 2.6 teraflops. Apple claims that it is faster than any other iGPU at the time of announcement.

Furthermore, the SoC integrates a fast 16 core neural engine with a peak performance of 11 TOPS (for AI hardware acceleration), a secure enclave (e.g., for encryption), a unified memory architecture, Thunderbolt / USB 4 controller, an ISP, and media de- and encoders.

The Apple M1 includes 16 billion transistors (up from the 10 billion of the A12Z Bionic and therefore double the amount of a Tiger Lake-U chip like the i7-1185G7) and is manufactured in 5nm at TSMC.

AMD Ryzen 9 8945H

The Ryzen 9 8945H is a rebadged Ryzen 9 7940HS with higher clock speeds on the Ryzen AI NPU. To be more specific, this is a powerful Hawk Point family laptop chip that has eight Zen 4 cores running at 4.0 GHz to 5.2 GHz. All of its cores are SMT-enabled for a total of 16 processing threads.

Compared to the 8945HS, the H seems to be a local variant for China and offers the same specifications.

The Radeon 780M serves as the integrated graphics adapter.

Architecture & Features

Hawk Point family chips are powered by the Zen 4 architecture, much like Phoenix and Dragon Range family chips are. That's not to say there is no difference between the three. With Hawk Point, AMD is betting big on generative AI; these chips are promised to deliver an up to 40% increase in generative AI performance over 7040 series APUs making apps like DaVinci Resolve and Adobe Photoshop even more powerful.

Unlike Zen 3, Zen 4 features AVX512 support and, thanks to a plethora of other improvements including larger caches/registers/buffers across the board, is slated to bring a double-digit IPC improvement over the former.

Elsewhere, the 8945H has 16 MB of L3 cache and support for super-fast RAM (up to LPDDR5x-7500 or DDR5-5600; no ECC support). The processor is compatible with USB 4 and thus with Thunderbolt; PCIe support is limited to the 4.0 spec for a throughput of 1.97 GB/s per lane. 20 lanes are available.

OS support is limited to 64-bit editions of Windows 11 and Windows 10 and of course to Linux. Note that the chip isn't overclockable and neither is it user-replaceable as it gets soldered down for good (FP7, FP7r2, FP8 socket interfaces).

Performance

The 8945H is an R9 7940HS in disguise, so it's realistic to expect the chip to be just slightly faster than the Core i9-12900HK and also the Ryzen 7 7840HS, as far as multi-thread performance is concerned.

Your mileage may vary depending on how high the CPU power limits are and how competent the cooling solution of your system is.

Graphics

The Radeon 780M has 12 CUs (768 shaders) running at up to 2,800 MHz. This is a very fast iGPU, as of late 2023. It will let you use up to 4 monitors with resolutions as high as SUHD 4320p, and it is also capable of HW-encoding and HW-decoding the most widely used video codecs including AVC, HEVC and AV1. More importantly, it is fast enough for proper 1080p gaming as long as one is fine with low to medium detail settings.

Your mileage may vary depending on how high the CPU power limits are, how competent the cooling solution of the system is, how fast the RAM of the system is (there is no dedicated VRAM here).

Power consumption

This Ryzen 9 series chip has a long-term power limit (default TDP) of 35 W to 54 W, giving laptop makers a choice between longer battery life and higher performance. Either way, an active cooling solution is a must for a system powered by this chip.

The 8945H is built with TSMC's 4 nm process for high, as of late 2023, energy efficiency.

Apple M1 Max

The Apple M1 Max is a System on a Chip (SoC) from Apple that is found in the late 2021 MacBook Pro 14 and 16-inch models. It offers all 10 cores available in the chip divided in eight performance cores (P-cores with 600 - 3220 MHz) and two power-efficiency cores (E-cores with 600 - 2064 MHz). There is no Turbo Boost for single cores or short burst periods. The cores are similar to the cores in the Apple M1.

The big cores (codename Firestorm) offer 192 KB instruction cache, 128 KB data cache, and 24 MB shared L2 cache (up from 12 MB in the M1). The four efficiency cores (codename Icestorm) are a lot smaller and offer only 128 KB instruction cache, 64 KB data cache, and 4 MB shared cache. CPU and GPU can both use the 48 MB SLC (System Level Cache). The efficiency cores (E cluster) clock with 600 - 2064 MHz, the performance cores (P cluster) with 600 - 3228 MHz.

The unified memory (32 or 64 GB LPDDR5-6400) next to the chip is connected by a 512 bit memory controller (200 GB/s bandwidth) and can be used by the GPU and CPU. This is the main difference to the M1 Pro and the CPU performance is quite similar.

The biggest difference to the M1 Pro is the bigger integrated GPU with 24 or 32 cores (up from 16).

Furthermore, the SoC integrates a fast 16 core neural engine, a secure enclave (e.g., for encryption), a unified memory architecture, Thunderbolt 4 controller, an ISP, and media de- and encoders (including two ProRes engines).

The M1 Pro is manufactured in 5 nm at TSMC and integrates 57 billion transistors. The peak power consumption of the chip was advertised around 30W for CPU intensive tasks.