Every memory option compared, 800 GB/s memory bandwidth. Updated 2 November 2026.
M1 Ultra / M2 Ultra are sold with 64, 128, 192 GB of memory. Counting the 31 models tracked here: 64 GB runs 28 models well; 128 GB runs 30 models well; 192 GB runs 30 models well. Memory bandwidth is 800 GB/s, about 6.7 times the memory speed of a base M4 (120 GB/s), so the same model generates words proportionally faster. Apple Silicon shares one pool of memory between the processor and the GPU, and macOS lets the GPU use about two thirds of it on Macs up to 36 GB and about three quarters above that, so the memory you buy decides which models load at all. Not every size is sold on every chip: M1 Ultra: 64, 128 GB; M2 Ultra: 64, 128, 192 GB.
| Model | 64 GB | 128 GB | 192 GB |
|---|---|---|---|
| Qwen 3.5 122B A10B | No | Great 41/s | Great 41/s |
| gpt-oss 120B | No | Great 77/s | Great 77/s |
| Llama 3.3 70B | Good 10/s | Good 10/s | Good 10/s |
| Wan 2.2 14B | Great about 14 min | Great about 14 min | Great about 8 min |
| LTX-2.3 | Great about 2 min | Great about 2 min | Great about 1 min |
| Qwen 3.6 35B A3B | Great 120/s | Great 120/s | Great 120/s |
| Qwen 3.5 35B A3B | Great 125/s | Great 125/s | Great 125/s |
| Gemma 4 31B | Great 21/s | Great 21/s | Great 21/s |
| DeepSeek R1 32B | Great 21/s | Great 21/s | Great 21/s |
| Qwen3 Coder 30B | Great 115/s | Great 115/s | Great 115/s |
| Qwen 3.6 27B | Great 23/s | Great 23/s | Great 23/s |
| Gemma 4 26B A4B | Great 89/s | Great 89/s | Great 89/s |
| Z-Image Turbo 6B | Great about 27 s | Great about 27 s | Great about 15 s |
| Wan 2.2 5B | Great about 7 min | Great about 7 min | Great about 4 min |
| gpt-oss 20B | Great 95/s | Great 95/s | Great 95/s |
| ACE-Step 1.5 | Great about 1 min | Great about 1 min | Great about 47 s |
| FLUX.2 klein 4B | Great about 19 s | Great about 19 s | Great about 11 s |
| Phi-4 14B | Great 47/s | Great 47/s | Great 47/s |
| Gemma 4 12B | Great 53/s | Great 53/s | Great 53/s |
| Stable Diffusion XL | Great about 11 s | Great about 11 s | Great about 7 s |
| Qwen3 TTS 1.7B | Great about 30 s | Great about 30 s | Great about 17 s |
| Gemma 4 E4B | Great 131/s | Great 131/s | Great 131/s |
| Qwen 3.5 9B | Great 64/s | Great 64/s | Great 64/s |
| Qwen3 VL 8B | Great 69/s | Great 69/s | Great 69/s |
| DeepSeek R1 8B | Great 81/s | Great 81/s | Great 81/s |
| Qwen2.5 Coder 7B | Great 89/s | Great 89/s | Great 89/s |
| Gemma 4 E2B | Great 190+/s | Great 190+/s | Great 190+/s |
| Whisper large-v3 turbo | Great about 42 s | Great about 42 s | Great about 25 s |
| Qwen 3.5 4B | Great 128/s | Great 128/s | Great 128/s |
| Kokoro TTS | Great about 6 s | Great about 6 s | Great about 4 s |
Text models show words a second. Image, video and voice models show the time per job (image, 4 second clip, minute of speech or similar).
As an Amazon Associate I earn from qualifying purchases.
About 61 GB is usable by the GPU. The largest chat or coding model that fits is Llama 3.3 70B (45.6 GB), generating about 10 words a second with 15.2 GB to spare. The quickest is Gemma 4 E2B at about 190+ words a second. Closest miss: gpt-oss 120B, which needs about 68.8 GB (8 GB more than this configuration can give the GPU).
Find M2 Ultra Macs with 64 GB on Amazon
About 122 GB is usable by the GPU. The largest chat or coding model that fits is Qwen 3.5 122B A10B (85.5 GB), generating about 41 words a second with 36.1 GB to spare. The quickest is Gemma 4 E2B at about 190+ words a second. Closest miss: DeepSeek V3 671B, which needs about 424.7 GB (303 GB more than this configuration can give the GPU).
Find M2 Ultra Macs with 128 GB on Amazon
About 182 GB is usable by the GPU. The largest chat or coding model that fits is Qwen 3.5 122B A10B (85.5 GB), generating about 41 words a second with 96.9 GB to spare. The quickest is Gemma 4 E2B at about 190+ words a second. Closest miss: DeepSeek V3 671B, which needs about 424.7 GB (242 GB more than this configuration can give the GPU).
For everyday chat, 64 GB runs 28 of 31 models well. 128 GB to 192 GB unlocks nothing new in this list. The extra memory only gives longer conversations and more room for other apps.
Yes, from 64 GB of memory, at about 21 words a second.
64 GB runs gpt-oss 20B at about 95 words a second; 128 GB runs gpt-oss 20B at about 95 words a second; 192 GB runs gpt-oss 20B at about 95 words a second.
You are already at the top end. Nothing popular is out of reach except the giant server models.