Apple / unified

Apple M2 Ultra

192 GB decides what fits. 800 GB/s decides how fast it runs once it does.

Computed for 192 GB, the largest configuration. The Apple M2 Ultra is also sold with 64 or 128 GB, and what fits changes with it.

Run it with 64 GB →Run it with 128 GB →

Open in the calculator →Best models for 192 GB, on every card that size →

Find one: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.

Quick answer

The Apple M2 Ultra fits 265 of 320 sized models entirely in its 192 GB; the largest widely used one is DeepSeek-V4-Flash-Vision-Exp (188.1 GB at Q4_K_M). Memory decides what fits; its 800 GB/s of bandwidth decides how fast it answers.

Largest popular model that fitsDeepSeek-V4-Flash-Vision-Exp305B params · Q4_K_M · needs 188.1 GB~48 tok/s Faster than you readOpen in the calculator →
Most downloaded that fitsQwen3.8-27B28B params · Q4_K_M · needs 18.6 GB~39 tok/s Faster than you readOpen in the calculator →

At 8,192 tokens of context with the whole model in device memory. Speeds are estimates from memory bandwidth, not benchmarks run on this card.

Apple M2 Ultra · 192 GB · 800 GB/s · Q4_K_M where published · 8,192 tokensSpecificationpublished dataDecode speedestimate

265 of 320 fit entirely

265 of the 320 models the engine can size fit entirely in device memory at Q4_K_M where it is published, otherwise the nearest published format, and 8,192 tokens. Unified memory is one pool, so there is no second memory tier to spill into; the remaining 55 do not run at this context.

83%of the models the engine can size fit entirely

192GB of device memory

800GB/s memory bandwidth

0run with system memory

55do not run at 8,192 tokens

Featured models · Q4_K_M at 8,192 tokens, including offload

Model	Needs	Verdict	Decode	Calculator
Qwen3.5-2B2.3B parameters	2.32 GB	fits	~136 tok/s	Open →
NVIDIA-Nemotron-3-Nano-4B-BF164.0B parameters	3.46 GB	fits	~108 tok/s	Open →
Qwen3-VL-8B-Instruct8.8B parameters	7.04 GB	fits	~74 tok/s	Open →
Qwen3.5-9B9.7B parameters	7.05 GB	fits	~82 tok/s	Open →
gemma-4-12B-it12B parameters	9.54 GB	fits	~60 tok/s	Open →
gpt-oss-20b21B parameters	13.8 GB	fits	~111 tok/s	Open →
Qwen3.8-27B28B parameters	18.6 GB	fits	~39 tok/s	Open →
Qwen3.6-27B28B parameters	18.6 GB	fits	~39 tok/s	Open →
Kimi-Linear-48B-A3B-Instruct49B parameters	30.4 GB	fits	~122 tok/s	Open →
Qwen3.8-Flash-Next180B parameters	111.6 GB	fits	~95 tok/s	Open →
DeepSeek-V4-Flash-Vision-Exp305B parameters	188.1 GB	fits	~48 tok/s	Open →
GLM-5.3-Flash321B parameters	198.3 GB	short by 6.32 GB	does not run	Open →

These examples are selected from prominent labs using the catalogue’s latest Hugging Face 30-day downloads and repository-creation freshness signal, with newer releases guaranteed a place. Offloaded rows assume 32 GB of system RAM, and a speed is only shown for a row that runs. Fitting in memory is not the same as loading: whether the runtime and version you have supports each architecture and format on this machine has not been tested here. Each “Open” link carries the same model, format, context and RAM into the calculator.

Fits entirely in Apple M2 Ultra (192 GB) memory — 265 of 320 sized models

Every model that fitswhole model resident · offload off

Your configuration

The most demanding model that fits is DeepSeek-V4-Flash-Vision-Exp at Q4_K_M: 188.1 GB of the 192 GB, leaving 3.87 GB spare.

Fully resident at 8,192 tokens (or the model’s own maximum, where that is shorter), offload off, at 800 GB/s. Each row is a run of the engine for this configuration; the rows start with current, prominent releases and “fits” is memory, not a tested runtime. Older or less prominent models remain available through this search and “Show all”.

Find a model in this list

Showing 40 of 265 models that fit.

Model	Format	Needs	Spare	Decode	Downloads / 30d	Calculator
Qwen3.8-27BQwen · 28B params	Q4_K_M	18.6 GB	173.4 GB	~39 tok/s	6.9M	Open →
DeepSeek-V4-Flash-0731DeepSeek · 304B params	Q4_K_M	187.5 GB	4.48 GB	~48 tok/s	4.5M	Open →
Qwen3.8-Flash-NextNEWQwen · 180B params	Q4_K_M	111.6 GB	80.4 GB	~95 tok/s	1.4M	Open →
gemma-4-26B-A4B-itGoogle · 26B params	Q4_K_M	16.8 GB	175.2 GB	~108 tok/s	13M	Open →
DeepSeek-V4-Flash-Vision-ExpNEWDeepSeek · 305B params	Q4_K_M	188.1 GB	3.87 GB	~48 tok/s	915.3K	Open →
gemma-4-31B-itGoogle · 31B params	Q4_K_M	22.2 GB	169.8 GB	~30 tok/s	9.9M	Open →
Qwen3.5-9BQwen · 9.7B params	Q4_K_M	7.05 GB	185.0 GB	~82 tok/s	9M	Open →
Qwen3.5-4BQwen · 4.7B params	Q4_K_M	3.98 GB	188.0 GB	~108 tok/s	7.8M	Open →
Qwen3.6-35B-A3BQwen · 36B params	Q4_K_M	23.1 GB	168.9 GB	~123 tok/s	3.3M	Open →
gemma-4-12B-itGoogle · 12B params	Q4_K_M	9.54 GB	182.5 GB	~60 tok/s	1.9M	Open →
Inkling-Smallthinkingmachines · 266B params	Q4_K_M	165.5 GB	26.5 GB	~66 tok/s	657.4K	Open →
Qwen3.6-27BQwen · 28B params	Q4_K_M	18.6 GB	173.4 GB	~39 tok/s	2.5M	Open →
Qwen3.5-2BQwen · 2.3B params	Q4_K_M	2.32 GB	189.7 GB	~136 tok/s	4.9M	Open →
gemma-4-E4B-itGoogle · 8.0B params	Q4_K_M	5.86 GB	186.1 GB	~103 tok/s	4.4M	Open →
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16NVIDIA · 32B params	Q4_K_M	20.3 GB	171.7 GB	~32 tok/s	530K	Open →
NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA · 4.0B params	Q4_K_M	3.46 GB	188.5 GB	~108 tok/s	3.4M	Open →
gemma-4-E2B-itGoogle · 5.1B params	Q4_K_M	4.02 GB	188.0 GB	~139 tok/s	3M	Open →
Qwen3.5-0.8BQwen · 873M params	Q4_K_M	1.46 GB	190.5 GB	~155 tok/s	2.6M	Open →
DeepSeek-V4-FlashDeepSeek · 284B params	Q4_K_M	175.9 GB	16.1 GB	~48 tok/s	1.1M	Open →
Qwen3-VL-8B-InstructQwen · 8.8B params	Q4_K_M	7.04 GB	185.0 GB	~74 tok/s	14.6M	Open →
MiniMax-M2.7MiniMaxAI · 229B params	Q4_K_M	141.2 GB	50.8 GB	~58 tok/s	1.1M	Open →
Qwen3.5-27BQwen · 28B params	Q4_K_M	18.6 GB	173.4 GB	~39 tok/s	1.9M	Open →
North-Micro-Vision-InstructCohereLabs · 2.5B params	Q4_K_M	2.99 GB	189.0 GB	~122 tok/s	180.7K	Open →
Hy3Tencent · 299B params	Q4_K_M	181.2 GB	10.8 GB	~41 tok/s	304.9K	Open →
Qwen3.5-35B-A3BQwen · 36B params	Q4_K_M	23.1 GB	168.9 GB	~123 tok/s	1.6M	Open →
NVIDIA-Nemotron-3-Super-120B-A12B-BF16NVIDIA · 124B params	Q4_K_M	76.9 GB	115.1 GB	~9.3 tok/sSlow	1.2M	Open →
granite-4.2-8bIBM · 8.8B params	Q4_K_M	7.49 GB	184.5 GB	~70 tok/s	128.4K	Open →
GLM-4.7-Flashzai-org · 31B params	Q4_K_M	20.4 GB	171.6 GB	~31 tok/s	1.8M	Open →
granite-4.1-3bIBM · 3.4B params	Q4_K_M	3.72 GB	188.3 GB	~107 tok/s	521.8K	Open →
LFM2.5-2.6BLiquidAI · 2.7B params	Q4_K_M	2.61 GB	189.4 GB	~125 tok/s	108.1K	Open →
Qwen3-VL-4B-InstructQwen · 4.4B params	Q4_K_M	4.72 GB	187.3 GB	~94 tok/s	3.5M	Open →
granite-4.1-30bIBM · 29B params	Q4_K_M	20.4 GB	171.6 GB	~31 tok/s	301.5K	Open →
Qwen3.5-122B-A10BQwen · 125B params	Q4_K_M	77.9 GB	114.1 GB	~80 tok/s	512.6K	Open →
Qwen3-VL-2B-InstructQwen · 2.1B params	Q4_K_M	3.05 GB	189.0 GB	~119 tok/s	2.8M	Open →
granite-4.2-3bIBM · 3.7B params	Q4_K_M	3.72 GB	188.3 GB	~107 tok/s	50.6K	Open →
LFM2.5-230MLiquidAI · 230M params	Q4_K_M	1.05 GB	190.9 GB	~166 tok/s	88.6K	Open →
Qwen3-0.6BQwen · 752M params	Q4_K_M	2.22 GB	189.8 GB	~134 tok/s	29.7M	Open →
Qwen3-Coder-NextQwen · 80B params	Q4_K_M	50.0 GB	142.0 GB	~106 tok/s	596.3K	Open →
gpt-oss-20bOpenAI · 21B params	Q4_K_M	13.8 GB	178.2 GB	~111 tok/s	6.6M	Open →
MiniMax-M2.5MiniMaxAI · 229B params	Q4_K_M	141.2 GB	50.8 GB	~58 tok/s	444.8K	Open →

Sized at the format most people actually download, not at FP16. The catalogue ordered by downloads →

Find the Apple M2 Ultra: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.

The Apple M2 Ultra, model by model

One page per model: whether it fits this device, at which formats, and how fast.

Under the hood

The specification behind every figure

What the manufacturer publishes for this device, and the pages it was read from.

Manufacturer specification

Memory scope	unified-system
Capacity	192 GB
Published options	64 GB, 128 GB, 192 GB
Bandwidth	800 GB/s
Memory type	unified memory
Bus width	Not published
FP32 peak	Not published
Dense matrix peak	Not published without sparsity
Power	Not published

Source ledger

Mac Pro (2023) - Tech Specs ↗

Retrieved 2026-08-30

Caveats

Apple's 64/128/192GB capacities are unified-memory capacity options; no VRAM capacity is claimed. The page also lists a configurable 76-core GPU, but does not publish a separate bandwidth value for it.

Published capacity is a hardware ceiling, not guaranteed free runtime memory. The calculator shows the runtime reserve separately rather than folding it into a single number.

Run the diagnostic on the Apple M2 Ultra →