Apple M3 Ultra: 512 GB, what it runs and what it costs
Specs checked · Aliteq Team · How we compute this
The short answer
The Apple M3 Ultra has 512 GB of VRAM and 819.2 GB/s of memory bandwidth (2025). At Q4_K_M quality or better with an 8K context it runs 46 of the open models we track; the largest that fits with room to spare is MiniMax-M3 (MoE). We don't see it for rent on the providers we track right now.
- VRAM
- 512 GB
- Bandwidth
- 819.2 GB/s
- Board power
- —
- Launched
- 2025
Source: Apple specification · bandwidth not yet confirmed by a second source, so we don't estimate speed from it
Buy it or rent it
Not for rent on the providers we track right now. See every card that is.
Which AI models run on the Apple M3 Ultra
Computed by our VRAM engine for an 8K context: the best quality each model fits at in 512 GB. Speeds are upper-bound estimates from memory bandwidth, not benchmarks.
| Model | Best fit | VRAM used | ~tok/s |
|---|---|---|---|
| DeepSeek V4.1 Flash (MoE) | Q5_K_M | 505 GB99% · tight | — |
| GLM-5.3 (MoE) | Q5_K_M | 500 GB98% · tight | — |
| MiniMax-M3 (MoE) | Q8_0 | 424 GB83% | — |
| Qwen3 235B-A22B (MoE) | F16 | 440 GB86% | — |
| MiniMax-M2.5 (MoE) | F16 | 429 GB84% | — |
| Qwen3.8-Flash-Next (MoE) | F16 | 336 GB66% | — |
| Qwen3.5 122B-A10B (MoE) | F16 | 234 GB46% | — |
| NVIDIA Nemotron 3 Super 120B-A12B (MoE) | F16 | 231 GB45% | — |
| gpt-oss 120B (MoE) | F16 | 219 GB43% | — |
| Llama 4 Scout 17B-16E (MoE) | F16 | 204 GB40% | — |
| GLM-4.5-Air (MoE) | F16 | 200 GB39% | — |
| Qwen3-Next 80B-A3B (MoE) | F16 | 150 GB29% | — |
Too big for 512 GB (rent a bigger card instead): what it costs to run DeepSeek V4 Pro (MoE), what it costs to run Kimi K3 (MoE).
Check a model and context of your own in the VRAM calculator.
Cards to compare it with
- Apple M2 Ultra · 192 GB
- Apple M4 Max (16-core CPU / 40-core GPU) · 128 GB
- DGX Spark (GB10 Grace Blackwell) · 128 GB
- AMD Ryzen AI Max+ 395 (Strix Halo,) · 128 GB
- Apple M2 Max · 96 GB
- RTX PRO 6000 Blackwell Max-Q Workstation Edition · 96 GB
Every card by VRAM tier: all GPUs for gaming and local AI.