Question 1

What AI models can run on M3 Ultra (192 GB)?

Accepted Answer

The M3 Ultra (192 GB) (144 GB VRAM) can run models including Llama-2-7b-chat, Llama-3.2-1B-Instruct-QLORA_INT4_EO8, MediPhi-Instruct, Llama-3.2-3B-Instruct-SpinQuant_INT4_EO8, Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8 and more — 10 compatible models indexed so far. Compatible models are determined by whether the model's memory footprint fits within the GPU's 144 GB VRAM, potentially using quantization (Q8, Q6, or Q4) to reduce memory usage.

Question 2

What is the cheapest cloud option for M3 Ultra (192 GB)?

Accepted Answer

We don't currently have cloud pricing data for the M3 Ultra (192 GB). This GPU may not be widely available from major cloud providers, or pricing data hasn't been synced yet.

Question 3

Is M3 Ultra (192 GB) good for fine-tuning?

Accepted Answer

Yes, the M3 Ultra (192 GB) is excellent for fine-tuning. With 144 GB of VRAM, it can handle full fine-tuning of models up to ~30B parameters and LoRA/QLoRA fine-tuning of much larger models. It's a top-tier choice for training workloads.

Question 4

How much VRAM does M3 Ultra (192 GB) have?

Accepted Answer

The M3 Ultra (192 GB) has 144 GB of VRAM (LPDDR5). It also offers 819 GB/s memory bandwidth, 27.2 TFLOPS of FP16 compute.

Question 5

What quantization formats work on M3 Ultra (192 GB)?

Accepted Answer

The M3 Ultra (192 GB) supports the following quantization formats: FP16 (full precision), Q8 (8-bit), Q6 (6-bit), Q4 (4-bit). FP16 works for models up to ~40B parameters. Q4 quantization lets models up to ~489B parameters (by VRAM estimate) fit on the M3 Ultra (192 GB).

M3 Ultra (192 GB)

Specifications

Cloud pricing

Compatible models

Compare this GPU

Frequently asked questions

VRAM	144 GB
Memory type	LPDDR5
Bandwidth	819 GB/s
FP16 TFLOPS	27.2
FP8 support	No