Apple SSD speed dependence on NAND capacity and package count

Facts

• Ars Technica (June 2022) reports that the M2 13-inch MacBook Pro 256 GB uses one 256 GB NAND chip where the M1 model uses a pair of 128 GB chips. [source]
• Its stated mechanism is interleaving: with more physical chips the controller reads and writes several at once, so fewer chips limit peak performance. [source]
• Max Tech's Blackmagic sustained read fell from about 2,900 MB/s (M1) to 1,446 MB/s (M2); write fell from 2,215 to 1,463 MB/s. [source]
• The same article says the 512 GB and 1 TB M2 versions looked similar to M1 speed. [source]
• Ars (Jan 2023): the M1 Pro 512 GB used four 128 GB NAND chips (per iFixit), while the 9to5Mac M2 Pro 512 GB appeared to use two 256 GB chips. [source]
• Measured on those two 512 GB machines: read 4,900 MB/s (M1 Pro) versus 2,973 MB/s (M2 Pro), a 40 percent drop; write 3,951 versus 3,154.5 MB/s, a 20 percent drop. [source]
• The 2023 article says the 512 GB M2 Pro still has two chips, so it is much faster than the one-chip 256 GB M2 MacBook Pro and Air. [source]
• A MacRumors Mac Studio poster says the base M4 Max 512 GB is a single card with four 128 GB NAND chips, while the earlier M2 Max 512 GB was a single card with two 256 GB chips. [source]
• The same poster says soldering eight 1 TB NAND chips from a donor card gave a working 8 TB drive. [source]
• Chip count per capacity is not fixed across generations. M2 256 GB parts used one chip. A MacRumors poster says the 256 GB M3 MacBook Air has two chips and the M3 256 GB parts do not show the slowdown. [source]
• Tom's Hardware's headline says an extra NAND chip makes the M3 MacBook Air 256 GB SSD "drastically faster" than its predecessor; the cached page body was empty, so only the headline is confirmed. [source]
• MyWorkDrive (July 2024) reports directory loads and file writes were visibly slower on the entry M2 MacBook Air than on the M1 Air and ties it to a single NAND chip. [source]
• A MacRumors poster guessed the 256 GB M4 Mac mini has one chip; another poster in the same thread points out that Apple Silicon M3 256 GB parts had two, and no teardown was posted in the cached page. The M4 mini chip count is unconfirmed. [source]
• Sustained versus burst matters: one MacRumors poster says the Blackmagic 5 GB test gives slower figures (4,300 write, 5,200 read) than a smaller-chunk test on the same machine. [source]
• Capacity-only rules ("bigger is faster") misstate the cause: two machines with the same 512 GB capacity differed by 40 percent read speed because of chip count (see Mechanism). [source]
• Forum view: a 256 GB Mac is "approximately half" the speed of upgraded ones. Counter in the same thread: M3 256 GB has two chips and is not slowed, and the difference is "negligible for the majority of users" on small-block workloads. Neither side measured an LLM weight-streaming load. [source]
• Apple publishes no NAND package count per SKU; counts come only from teardowns and forum posts. [source]
• No source measures sequential 1 MB to 100 MB expert-sized reads at queue depth across 256 GB, 512 GB and 1 TB configurations of one chip generation. [source]
• Whether a larger Studio or Pro card with more packages scales linearly beyond four packages is unmeasured. [source]
• Apple Silicon SSD throughput tracks the number of NAND packages (interleaving), not capacity itself. [source]
• M2 13-inch MacBook Pro 256 GB: one 256 GB chip versus two 128 GB on M1; sustained read about 1,446 versus 2,900 MB/s. [source]
• M1 Pro 512 GB (four 128 GB chips) read 4,900 MB/s versus M2 Pro 512 GB (two 256 GB chips) read 2,973 MB/s. [source]
• The same capacity can have a different chip count in different generations, so a streaming projection must name chip generation and chip count, not only capacity. [source]
• A MacRumors poster reports the Mac Studio M4 Max 512 GB base uses four 128 GB NAND chips. [source]
• Existing ssd-bandwidth-growth dossier's capacity statement is correct but incomplete: its forum-only claim is supported by teardown-level evidence for 2022 and 2023 laptops, not for the Mac Studio M4 Max. [source]
• The Apple SoC performs the SSD controller role, so the chip-count effect cannot be offset by a faster controller in the drive. [source]