On memory, yes: Nemotron 3.5 Bolt 74B-A9B fits 25 of 54 Macs in the LLMCheck index (runtime caveat below). It needs about 45 GB at a 4K context, 46 GB at 64K. Smallest and cheapest current Mac at 15+ tok/s: the Mac mini M5 Pro 48 GB ($2,299, raised GPU limit). Best measured: 71.7 tok/s on an M5 Max (community).
Stock llama.cpp runs it from b11755, but NVIDIA's 24 GB GGUF falls back to the CPU on its NVFP4 layers (use a Q4_K_M file); Ollama does not support it on Apple Silicon (PR #19229) and stock mlx-lm cannot load it.
One row per Mac configuration, grouped by product line. The build is the one each speed is for; the context column is the longest prompt that fits inside macOS’s default GPU memory limit with a 16-bit KV cache (“raised” where the limit has to be raised first; “full” when the whole window fits). Speeds are decode tok/s from the LLMCheck index, each with its provenance mark. For large mixture-of-experts models (over 40B total), the formula’s typical error is about 34% (the median gap to 70 measured plain-decode runs; it tends to undershoot (median −33%), and an estimate scaled from the model’s own measured run is usually closer; method).
| Mac | Verdict | Build | Speed | Max context | Model page |
|---|---|---|---|---|---|
| MacBook Air | |||||
| MacBook Air M5 16 GB | Doesn't fit | — | — | — | — |
| MacBook Air M5 24 GB | Doesn't fit | — | — | — | — |
| MacBook Air M5 32 GB | Doesn't fit | — | — | — | — |
| MacBook Air M1 8 GB refurb | Doesn't fit | — | — | — | — |
| MacBook Air M4 16 GB used | Doesn't fit | — | — | — | — |
| MacBook Air M2 16 GB refurb | Doesn't fit | — | — | — | — |
| MacBook Air M3 16 GB refurb | Doesn't fit | — | — | — | — |
| MacBook Air M1 16 GB refurb | Doesn't fit | — | — | — | — |
| MacBook Air M4 24 GB used | Doesn't fit | — | — | — | — |
| MacBook Air M2 24 GB refurb | Doesn't fit | — | — | — | — |
| MacBook Air M3 24 GB refurb | Doesn't fit | — | — | — | — |
| MacBook Pro | |||||
| MacBook Pro M6 16 GB | Doesn't fit | — | — | — | — |
| MacBook Pro M5 Pro 24 GB | Doesn't fit | — | — | — | — |
| MacBook Pro M6 24 GB | Doesn't fit | — | — | — | — |
| MacBook Pro M6 32 GB | Doesn't fit | — | — | — | — |
| MacBook Pro M5 Pro 48 GB | Raise GPU limit command ↓ | 4-bit | estimated31 tok/s | — | M5 Pro page |
| MacBook Pro M5 Max 64 GB | Runs | 4-bit | community58 tok/s | 256K (full) | M5 Max page |
| MacBook Pro M5 Pro 64 GB | Runs | 4-bit | estimated31 tok/s | 256K (full) | M5 Pro page |
| MacBook Pro M5 Max 128 GB | Runs | 5-bit | community64 tok/s | 256K (full) | M5 Max page |
| MacBook Pro M3 Max 64 GB refurb | Runs | 4-bit | estimated39 tok/s | 256K (full) | M3 Max page |
| MacBook Pro M3 Max 128 GB refurb | Runs | 4-bit | estimated39 tok/s | 256K (full) | M3 Max page |
| Mac mini | |||||
| Mac mini M6 16 GB | Doesn't fit | — | — | — | — |
| Mac mini M5 Pro 24 GB | Doesn't fit | — | — | — | — |
| Mac mini M6 24 GB | Doesn't fit | — | — | — | — |
| Mac mini M6 32 GB | Doesn't fit | — | — | — | — |
| Mac mini M5 Pro 48 GB | Raise GPU limit command ↓ | 4-bit | estimated31 tok/s | — | M5 Pro page |
| Mac mini M5 Pro 64 GB | Runs | 4-bit | estimated31 tok/s | 256K (full) | M5 Pro page |
| Mac mini M2 Pro 16 GB refurb | Doesn't fit | — | — | — | — |
| Mac mini M4 16 GB used | Doesn't fit | — | — | — | — |
| Mac mini M4 Pro 24 GB used | Doesn't fit | — | — | — | — |
| Mac mini M4 24 GB used | Doesn't fit | — | — | — | — |
| Mac mini M2 Pro 32 GB refurb | Doesn't fit | — | — | — | — |
| Mac mini M4 Pro 48 GB used | Raise GPU limit command ↓ | Q4_K_M | estimated28 tok/s | — | M4 Pro page |
| Mac mini M4 Pro 64 GB used | Runs | Q4_K_M | estimated28 tok/s | 256K (full) | M4 Pro page |
| iMac | |||||
| iMac M6 16 GB | Doesn't fit | — | — | — | — |
| iMac M6 24 GB | Doesn't fit | — | — | — | — |
| iMac M6 32 GB | Doesn't fit | — | — | — | — |
| Mac Studio | |||||
| Mac Studio M5 Max 36 GB | Doesn't fit | — | — | — | — |
| Mac Studio M5 Max 48 GB | Raise GPU limit command ↓ | Q4_K_M | estimated55 tok/s | — | M5 Max page |
| Mac Studio M5 Max 64 GB | Runs | 4-bit | community58 tok/s | 256K (full) | M5 Max page |
| Mac Studio M5 Ultra 96 GB | Runs | Q4_K_M | estimated79 tok/s | 256K (full) | — |
| Mac Studio M5 Max 128 GB | Runs | 5-bit | community64 tok/s | 256K (full) | M5 Max page |
| Mac Studio M5 Ultra 256 GB | Runs | Q4_K_M | estimated79 tok/s | 256K (full) | — |
| Mac Studio M5 Ultra 512 GB | Runs | Q4_K_M | estimated79 tok/s | 256K (full) | — |
| Mac Studio M2 Max 32 GB refurb | Doesn't fit | — | — | — | — |
| Mac Studio M1 Ultra 64 GB refurb | Runs | Q4_K_M | estimated55 tok/s | 256K (full) | — |
| Mac Studio M4 Max 64 GB used | Runs | 4-bit | estimated52 tok/s | 256K (full) | — |
| Mac Studio M2 Max 64 GB refurb | Runs | Q4_K_M | estimated39 tok/s | 256K (full) | — |
| Mac Studio M1 Ultra 128 GB refurb | Runs | Q4_K_M | estimated55 tok/s | 256K (full) | — |
| Mac Studio M2 Ultra 128 GB refurb | Runs | Q4_K_M | estimated55 tok/s | 256K (full) | — |
| Mac Studio M4 Max 128 GB used | Runs | 4-bit | estimated52 tok/s | 256K (full) | — |
| Mac Studio M2 Ultra 192 GB refurb | Runs | Q4_K_M | estimated55 tok/s | 256K (full) | — |
| Mac Studio M3 Ultra 256 GB used | Runs | 4-bit | community57 tok/s | 256K (full) | M3 Ultra page |
| Mac Studio M3 Ultra 512 GB refurb | Runs | 4-bit | community57 tok/s | 256K (full) | M3 Ultra page |
Context memory is computed from its config.json (hybrid Mamba-2: attention on 8 of 88 layers; Mamba layers keep a constant state): about 0.5 GB of 16-bit KV cache at 64K tokens. A q8 cache (llama.cpp -ctk q8_0 -ctv q8_0, Ollama OLLAMA_KV_CACHE_TYPE=q8_0) halves that part. Runtimes also grow a compute buffer with context, so a fit within about 1 GB of the limit is optimistic.
macOS lets the GPU wire only part of unified memory by default (Metal’s recommended working set; M5 and M6
Macs run macOS 26 or later, whose default is higher), and Nemotron 3.5 Bolt 74B-A9B needs about 45 GB. On the Macs marked
Raise GPU limit it fits in RAM but not inside that default. The default
is a performance threshold, not a hard wall: past it, the runtime moves part of it to the CPU or shortens the context (llama.cpp, Ollama), runs it slowly (MLX), or stops with an out-of-memory error. One command raises the limit until the next restart
(sudo sysctl iogpu.wired_limit_mb=0 restores the default sooner); it leaves macOS a reserve, and other
apps get less memory while the model is loaded.
sudo sysctl iogpu.wired_limit_mb=43008sudo sysctl iogpu.wired_limit_mb=43008Weights are 74B parameters at each format’s effective size (the index’s GB-per-billion table), plus about 2 GB for the runtime and a short context; longer contexts add the KV cache computed from the model’s config.json at 16 bits. It is a mixture-of-experts model: all 74B parameters sit in memory although only 9B are read per token. The first row is the catalog’s own figure (about 45 GB at a 4K context), the one every verdict on this page uses. “Smallest Mac” is the smallest memory size among the Macs Apple sells new that holds it at a 4K context inside that Mac’s default GPU limit. Estimates, not measurements.
| Build | Weights | RAM at 4K | at 32K | at 64K | Smallest Mac |
|---|---|---|---|---|---|
| As catalogued | — | 45 GB | 45 GB | 46 GB | 64 GB |
| Q2_K | 25.2 GB | 27 GB | 27 GB | 28 GB | 36 GB |
| Q3_K_M | 36.3 GB | 38 GB | 38 GB | 39 GB | 64 GB |
| Q4_K_M | 42.2 GB | 44 GB | 44 GB | 45 GB | 64 GB |
| Q5_K_M | 52.5 GB | 55 GB | 55 GB | 55 GB | 96 GB |
| Q6_K | 60.7 GB | 63 GB | 63 GB | 63 GB | 96 GB |
| Q8_0 | 78.4 GB | 80 GB | 81 GB | 81 GB | 128 GB |
| F16 | 148.0 GB | 150 GB | 150 GB | 151 GB | 256 GB |
Every published (filled square) and community (hollow circle) run the index holds for Nemotron 3.5 Bolt 74B-A9B: 0 published, 13 community, each linked to its source. LLMCheck runs no tests of its own. One row is one measurement session; on each chip the table above shows the lower median of the eligible runs (plain decode at 5 bits or fewer), and a binned or “not an anchor” run is shown but never used as a reference.
| Chip | Runtime | Quant | Decode | Context | Date | Source |
|---|---|---|---|---|---|---|
| M3 Ultra | MLX | IQ2_XXS | community74.5 tok/s | 1k | 2026-10-18 | omlx.ai not an anchor |
| M3 Ultra | MLX | 4-bit | community56.5 tok/s | 1024-token prompt, 96 generated | 2026-10-15 | github.com |
| M3 Ultra | llama.cpp | Q4_K_M | community41.61 tok/s | tg128 | 2026-10-16 | github.com not an anchor |
| M5 Max | MLX | IQ2_XXS | community73 tok/s | 1k | 2026-10-23 | omlx.ai not an anchor |
| M5 Max | MLX | 4-bit | community71.7 tok/s | 128-token greedy sample | 2026-10-16 | huggingface.co |
| M5 Max | MLX | 5-bit | community63.6 tok/s | 64-token greedy sample | 2026-10-16 | huggingface.co |
| M5 Max | MLX | 4-bit | community57.9 tok/s | 4k | 2026-10-20 | omlx.ai |
| M5 Max | MLX | 8-bit | community37 tok/s | — | 2026-10-15 | huggingface.co not an anchor |
| M3 Max | MLX | IQ2_XXS | community50.8 tok/s | 1k | 2026-10-22 | omlx.ai not an anchor |
| M3 Max | MLX | 4-bit | community44 tok/s | generation length not stated | 2026-10-16 | blog.bytebunkerlabs.ai not an anchor |
| M5 Pro | MLX | IQ2_XXS | community39 tok/s | 512-token coding answers, median of 5 at T 0.6 | 2026-10-19 | github.com not an anchor |
| M5 Pro | llama.cpp | IQ2_XXS + Q8_0 | community24.6 tok/s | random-text prompts, greedy, -c 262144 | 2026-10-16 | huggingface.co not an anchor |
| M4 Pro | MLX | IQ2_XXS | community34.6 tok/s | 512-token coding answers, median of 5 at T 0.6 | 2026-10-19 | github.com not an anchor |
Before you start: Stock llama.cpp runs it from b11755, but NVIDIA's 24 GB GGUF falls back to the CPU on its NVFP4 layers (use a Q4_K_M file); Ollama does not support it on Apple Silicon (PR #19229) and stock mlx-lm cannot load it.
The index has no verified Ollama tag for Nemotron 3.5 Bolt 74B-A9B. Start from its listing and copy the exact name of a build that fits: Hugging Face repo → · model card →
GGUF builds run in LM Studio or llama.cpp; MLX builds run in LM Studio or mlx-lm.
On the Mac mini M5 Pro (64 GB), the smallest current Mac that runs it inside the default GPU limit, the index figure is ~31 tok/s (estimated) at 4-bit; keep the context at or under 256K there, or use a q8 KV cache.
Step-by-step: the Ollama install guide.
Mac mini M5 Pro (48 GB) — smallest current Mac that runs it and cheapest at 15+ tok/s, $2,299 new (Apple list price, checked 2026-11-14), ~31 tok/s (estimated) once the GPU memory limit is raised. Amazon does not stock the Mac mini M5 Pro (48 GB): Apple builds it to order.
On memory, yes (its Mac runtime carries a caveat: Stock llama.cpp runs it from b11755, but NVIDIA's 24 GB GGUF falls back to the CPU on its NVFP4 layers (use a Q4_K_M file); Ollama does not support it on Apple Silicon (PR #19229) and stock mlx-lm cannot load it). Nemotron 3.5 Bolt 74B-A9B needs about 45 GB of unified memory at a short (4K-token) context, so it runs on 25 of the 54 Mac configurations in the LLMCheck index; 21 of them hold it inside macOS’s default GPU memory limit and 4 only after raising that limit. The smallest Mac Apple sells new that runs it is the Mac mini M5 Pro (48 GB).
About 45 GB at a 4K context: the weights plus about 2 GB for the runtime. Context adds KV cache: about 46 GB in total at 32K and 46 GB at 64K with the default 16-bit cache (about 46 GB at 64K with a q8 cache). These are estimates from the model’s config.json, not measurements.
On an M5 Max, the chip the leaderboard quotes it on, the index figure is 64 tok/s (community run) at 5-bit. The best measured figure is 71.7 tok/s on an M5 Max 128 GB (community run, 4-bit). Every Mac’s figure is in the table on this page with its provenance: published, community run or estimated.
Among Macs Apple sells new, the Mac mini M5 Pro (48 GB), $2,299 new (Apple list price, checked 2026-11-14), is the cheapest that reaches 15 tok/s with Nemotron 3.5 Bolt 74B-A9B, at ~31 tok/s (estimated), after raising the GPU memory limit.
On 4 of the 54 configurations, yes: Mac mini M5 Pro 48 GB, Mac Studio M5 Max 48 GB, MacBook Pro M5 Pro 48 GB and 1 more. macOS lets the GPU use only part of unified memory by default (about 37 GB on the Mac mini M5 Pro (48 GB)), and Nemotron 3.5 Bolt 74B-A9B needs about 45 GB. Running sudo sysctl iogpu.wired_limit_mb=43008 in Terminal raises the limit on a 48 GB Mac until the next restart. From 64 GB up, every Mac Apple sells new holds it inside the default limit.
Q4_K_M (about 42 GB of weights) is the common 4-bit GGUF choice. Q8_0 (about 78 GB) is closer to full quality and needs a 128 GB Mac at the default GPU limit. Q3_K_M and Q2_K shrink it to about 36 and 25 GB at a quality cost. The index’s reference figure (64 tok/s on an M5 Max) is for its 5-bit build.
Its own window is 256K tokens. At macOS’s default GPU limit, the Mac mini M5 Pro (64 GB) holds about 256K (the full window).
LLMCheck. (2026). Can I run Nemotron 3.5 Bolt 74B-A9B on a Mac? llmcheck.net. https://llmcheck.net/can-i-run/nemotron-35-bolt-74b-a9b/ (updated ). CC BY 4.0.