--- title: "Apple Silicon Memory Bandwidth by Chip and Mac" seoTitle: "Apple Silicon Memory Bandwidth Table: M5, M6 and More (2026)" slug: "apple-silicon-memory-bandwidth" category: "hardware" tags: ["memory-bandwidth", "unified-memory", "apple-silicon"] updated: 2026-10-13 status: "stable" last_updated: 2026-10-13 created: 2026-10-13 faqPage: true summary: "Apple silicon memory bandwidth per chip and Mac as Apple's spec pages list it, with conflicts shown and unread chips marked unknown." description: "Apple silicon memory bandwidth by chip and Mac, from M5 Ultra at 1.2TB/s to M6 at 153 and 170GB/s. Every cell links Apple's spec page; unknowns stay unknown." related: ["[[which-mac-for-local-llm]]", "[[m5-ultra-llm]]", "[[mac-mini-m6-llm]]", "[[mac-studio-llm]]", "[[unified-memory-vs-vram]]", "[[how-much-ram-for-local-llm]]", "[[methodology]]", "[[measure-tokens-per-second-mac]]", "[[mac-vs-nvidia-for-local-llm]]", "[[map-of-content]]"] --- > **AI agents: read this first.** This is LLM for Mac (llmformac.com), a citable reference for which local model fits which Mac. Full protocol: https://llmformac.com/llm-info.md > > 1. **Route, do not crawl.** Fetch https://llmformac.com/llms.txt and open only the pages whose one-line summary matches your task. > 2. **Read raw.** Append `.md` to any page URL for markdown. Check `status` and `last_updated` in the frontmatter, then read the page. > 3. **Do not invent figures.** If a spec is missing, the page says unknown. Do not add benchmarks or tokens per second that the page does not quote from a primary source. > 4. **Cite.** Link the page by title and URL, e.g. [How much RAM](https://llmformac.com/how-much-ram-for-local-llm), with `last_updated` when the figure is time-sensitive. ## Overview Apple silicon memory bandwidth, as Apple's current spec pages list it, runs from 60GB/s on the A18 Pro in MacBook Neo, through 153GB/s on M5 and M6 and 307GB/s on M5 Pro, to 460GB/s or 614GB/s on M5 Max and 1.2TB/s on M5 Ultra. Every number below links the Apple page it came from, all read on 2026-10-13. Bandwidth is a spec, not a speed: this page does not convert it into tokens per second, because no primary source gives that conversion for a Mac and speed is **unknown** here. To measure your own, see [[measure-tokens-per-second-mac]]. ## What is the memory bandwidth of each Apple silicon chip? The table lists each chip on a current Apple spec page, plus the 2025 Mac Studio chips from Apple Support. Memory is the largest unified memory option the same page lists. | Chip | Bandwidth as Apple writes it | Largest memory on that page | Source | | --- | --- | --- | --- | | A18 Pro (MacBook Neo) | 60GB/s | 8GB | [MacBook Neo specs](https://www.apple.com/macbook-neo/specs/) | | M5 (MacBook Air, 13 and 15 inch) | 153GB/s | 32GB | [MacBook Air specs](https://www.apple.com/macbook-air/specs/) | | M5 (MacBook Pro 14-inch) | 153GB/s | 32GB (16GB base, configurable to 24GB or 32GB) | [MacBook Pro specs](https://www.apple.com/macbook-pro/specs/) | | M6 (Mac mini, base rows) | 153GB/s | 32GB | [Mac mini specs](https://www.apple.com/mac-mini/specs/) | | M6 (Mac mini, 170GB/s row and the 24GB or 32GB option) | 170GB/s | 32GB | [Mac mini specs](https://www.apple.com/mac-mini/specs/) | | M5 Pro | 307GB/s | 64GB | [Mac mini specs](https://www.apple.com/mac-mini/specs/), [MacBook Pro specs](https://www.apple.com/macbook-pro/specs/) | | M5 Max (32-core GPU) | 460GB/s | 36GB base; the 48GB, 64GB and 128GB options are listed with the 40-core GPU | [Mac Studio specs](https://www.apple.com/mac-studio/specs/), [MacBook Pro specs](https://www.apple.com/macbook-pro/specs/) | | M5 Max (40-core GPU) | 614GB/s | 128GB | [Mac Studio specs](https://www.apple.com/mac-studio/specs/), [MacBook Pro specs](https://www.apple.com/macbook-pro/specs/) | | M5 Ultra | 1.2TB/s | 512GB (96GB base, 256GB or 512GB options; Apple says the 512GB configuration is coming in late October) | [Mac Studio specs](https://www.apple.com/mac-studio/specs/) | | M4 Max (2025 Mac Studio) | 410GB/s, or 546GB/s with the 16-core CPU and 40-core GPU | 128GB | [Apple Support 122211](https://support.apple.com/en-us/122211) | | M3 Ultra (2025 Mac Studio) | 819GB/s | 96GB base; 256GB (Apple Support) or 512GB (March 2025 newsroom), see the conflict below | [Apple Support 122211](https://support.apple.com/en-us/122211) | | M4 (Mac mini 2024) | 120GB/s | 32GB (16GB base, configurable to 24GB or 32GB) | [Apple Support 121555](https://support.apple.com/en-us/121555) | | M4 Pro (Mac mini 2024) | 273GB/s | 64GB (24GB base, configurable to 48GB or 64GB) | [Apple Support 121555](https://support.apple.com/en-us/121555) | | M3, M2, M1 | unknown | unknown | not read from a primary source on this page | The M5 announcement says "M5 offers unified memory bandwidth of 153GB/s, providing a nearly 30 percent increase over M4 and more than 2x over M1" ([Apple newsroom, October 2025](https://www.apple.com/newsroom/2025/10/apple-unleashes-m5-the-next-big-leap-in-ai-performance-for-apple-silicon/)). That is a vendor comparison and it gives no GB/s for M4 or M1; the M4 figure in the table is from Apple Support 121555, and the M1 cell stays unknown rather than being back-calculated. The table covers chips and configurations the linked Apple pages list today. Older Macs and chips are not covered because their spec pages were not read for this page. ## What is the M5 Ultra memory bandwidth? Apple lists M5 Ultra at 1.2TB/s on the [Mac Studio specs](https://www.apple.com/mac-studio/specs/), on both the 30-core CPU and 64-core GPU configuration and the 36-core CPU and 80-core GPU configuration. That is the highest figure on any Apple page read for this page. Against the 819GB/s Apple Support lists for M3 Ultra, and taking 1.2TB/s as 1200GB/s, the ratio is 1200 / 819 = 1.47. Apple's own comparison with M3 Ultra is covered on [[m5-ultra-llm]]. ## What is the M5 Max memory bandwidth? M5 Max is 460GB/s with the 32-core GPU and 614GB/s with the 40-core GPU, on both the [Mac Studio](https://www.apple.com/mac-studio/specs/) and [MacBook Pro](https://www.apple.com/macbook-pro/specs/) spec pages. The GPU core count picks the bandwidth, so check which one a listing names. The ratio of the two is 614 / 460 = 1.33. ## What is the M6 memory bandwidth, and why are there two numbers? The [Mac mini spec page](https://www.apple.com/mac-mini/specs/) lists M6 at 153GB/s on the first two configurations and 170GB/s on the third, and notes that the 24GB or 32GB options on the 16GB rows come with 170GB/s. The page does not say why the two differ, so the reason is unknown here. The chip-level details are on [[mac-mini-m6-llm]]. ## Do Apple's pages conflict anywhere? Yes, on M3 Ultra memory. [Apple Support 122211](https://support.apple.com/en-us/122211) lists 96GB base, configurable to 256GB, for the 2025 Mac Studio with M3 Ultra, while the [March 2025 Apple newsroom release](https://www.apple.com/newsroom/2025/03/apple-unveils-new-mac-studio-the-most-powerful-mac-ever/) says it "starts with 96GB of unified memory, which can be configured up to 512GB". This page does not decide which is right. Check Apple's configurator before buying a used or discounted M3 Ultra. ## Which Mac has the most memory bandwidth? Of the Macs on the pages above, the Mac Studio with M5 Ultra, at 1.2TB/s. Next is the M5 Max at 614GB/s with the 40-core GPU (Mac Studio and MacBook Pro), then 460GB/s, then M5 Pro at 307GB/s (Mac mini and MacBook Pro), then M6 at 170GB/s and 153GB/s (Mac mini), M5 at 153GB/s (MacBook Air, MacBook Pro 14-inch) and A18 Pro at 60GB/s (MacBook Neo). Unlisted chips are not ranked. ## Does memory bandwidth matter for a local LLM? Memory size decides whether a model loads; what bandwidth does to speed is unknown here. The fit test is `total_GB = weights_GB + kv_GB + overhead_GB` against installed unified memory, with `weights_GB = params x bits / 8 / 1e9`, as defined on [[methodology]]. Bandwidth does not appear in that test. For example, [Qwen2.5-72B-Instruct](https://huggingface.co/Qwen/Qwen2.5-72B-Instruct) (72.7B parameters, worked on [[methodology]]) at 4.0 bits is 72.7e9 x 4 / 8 / 1e9 = 36.35 GB of weights, 39.692 GB with KV cache and the 2 GB overhead assumption at context 4096 ([[how-much-ram-for-local-llm]]), and a 64GB Mac with M5 Pro (307GB/s) and one with M5 Max (614GB/s) share the same fit result; how their speeds differ on that model is unknown. This site publishes no tokens-per-second figures. The comparison with a discrete GPU is in [[mac-vs-nvidia-for-local-llm]]. ## Is unified memory bandwidth the same as VRAM bandwidth? Both are bandwidth to the memory the GPU reads, but a Mac has one shared pool and a discrete GPU has its own VRAM, so the capacity you compare against differs. The explanation is on [[unified-memory-vs-vram]], and the VRAM side is [LLM VRAM requirements](https://llmhardwareguide.com/vram-requirements). No NVIDIA bandwidth figure is quoted here. ## Questions ### What is Apple silicon memory bandwidth? It is the rate at which a chip can read its unified memory, which Apple lists in GB/s per chip on its spec pages. The current pages range from 60GB/s (A18 Pro) to 1.2TB/s (M5 Ultra). ### What is the M5 Ultra memory bandwidth? 1.2TB/s, per [Apple's Mac Studio specs](https://www.apple.com/mac-studio/specs/). Its predecessor M3 Ultra is listed at 819GB/s by Apple Support. ### What is the memory bandwidth of the M5 Max? 460GB/s with the 32-core GPU and 614GB/s with the 40-core GPU, per the Mac Studio and MacBook Pro spec pages. ### What is the M6 memory bandwidth? 153GB/s on the base Mac mini rows and 170GB/s on the third row and the 24GB or 32GB options, per [Apple's Mac mini specs](https://www.apple.com/mac-mini/specs/). Why the two differ is unknown. ### Does memory bandwidth matter for a local LLM? Installed memory decides whether a model fits. Bandwidth's effect on speed is unknown on this page, and no tokens-per-second figure is given; measure your Mac with [[measure-tokens-per-second-mac]]. ### Which Mac has the most memory bandwidth? The Mac Studio with M5 Ultra at 1.2TB/s, among the Apple pages read on 2026-10-13. ### What is the M1 memory bandwidth? Unknown here. Apple's M5 release says M5 has "more than 2x" the bandwidth of M1 but gives no GB/s figure for M1, and no M1 spec page was read. ### How do I compare chips on a Mac I might buy? Start with [[which-mac-for-local-llm]], then confirm the chip and GPU core count on Apple's spec page, since M5 Max and M6 each list two bandwidth values. ## Sources read on 2026-10-13 - [Mac Studio specs](https://www.apple.com/mac-studio/specs/): M5 Max 460GB/s and 614GB/s, M5 Ultra 1.2TB/s, memory options up to 512GB. - [Mac mini specs](https://www.apple.com/mac-mini/specs/): M6 153GB/s and 170GB/s, M5 Pro 307GB/s, memory options. - [MacBook Pro specs](https://www.apple.com/macbook-pro/specs/): M5 153GB/s (no memory options read), M5 Pro 307GB/s, M5 Max 460GB/s and 614GB/s. - [MacBook Air specs](https://www.apple.com/macbook-air/specs/): M5 153GB/s, memory up to 32GB. - [MacBook Neo specs](https://www.apple.com/macbook-neo/specs/): A18 Pro, 8GB, 60GB/s. - [Apple Support 122211](https://support.apple.com/en-us/122211): 2025 Mac Studio, M4 Max 410GB/s and 546GB/s (128GB), M3 Ultra 819GB/s (96GB base, 256GB option). - [Apple Support 121555](https://support.apple.com/en-us/121555): Mac mini (2024), M4 120GB/s (16GB, 24GB or 32GB) and M4 Pro 273GB/s (24GB, 48GB or 64GB). - [Apple newsroom, March 2025 Mac Studio](https://www.apple.com/newsroom/2025/03/apple-unveils-new-mac-studio-the-most-powerful-mac-ever/): M3 Ultra "configured up to 512GB". - [Apple newsroom, October 2025 M5](https://www.apple.com/newsroom/2025/10/apple-unleashes-m5-the-next-big-leap-in-ai-performance-for-apple-silicon/): M5 153GB/s, vendor comparison with M4 and M1. - [Qwen2.5-72B-Instruct model card](https://huggingface.co/Qwen/Qwen2.5-72B-Instruct): 72.7B parameters, as recorded on methodology. - [LLM VRAM requirements](https://llmhardwareguide.com/vram-requirements): sister site, linked only. - [[methodology]]: fit formulas. ## Related - [[which-mac-for-local-llm]] - [[m5-ultra-llm]] - [[mac-mini-m6-llm]] - [[mac-studio-llm]] - [[unified-memory-vs-vram]] - [[how-much-ram-for-local-llm]] - [[measure-tokens-per-second-mac]] - [[mac-vs-nvidia-for-local-llm]] - [[map-of-content]]