Skip to content

GLM-5.3 Flash on M5 Ultra 512 GB: VRAM, speed and fit

GLM-5.3 Flash at FP8 runs on the M5 Ultra 512 GB with 1M of context at about 33 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on M5 Ultra 512 GB · FP8 · Ollama / LM Studio · Chat

GPU memory
447 GB
of 512 GB
Context
1M
auto
Speed
33 tok/s
21 when full
First reply
3.4 s
2K prompt

Where the memory goes

Model
306 GB
Context
12.5 GB
Overhead
687 MB
OS
128 GB
Free
65.0 GB
Memory
447 GB of 512 GB
Context
1M tokens · auto
Speed
33 tok/s · 21 when full
First reply
3.4 s for a 2K prompt
Running cost
$5.06/mo at home · $574/mo on a cloud 8× A40

Validation: Validated with conditions

What GLM-5.3 Flash on M5 Ultra 512 GB was checked against
CheckTargetResultStatus
Fits in GPU memory384 GB usable319 GB neededPass
Headroom5% free65.0 GB free (17%)Pass
Context for chat16K or more1MPass
Writing speed15 tok/s or more33 tok/sPass
First replyunder 3 s, 2K prompt3.4 sCheck

Targets: a reply you can read along with, a first reply the use case tolerates, and 5% of memory left free. Calculated from published specs, not benchmarked.

The setup

Model
GLM-5.3 Flash
Parameters
321B, 18B active
Precision
FP8
Context
1M, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
M5 Ultra 512 GB
GPU can use
384 GB of 512 GB

Running cost

At home
$5.06/mo
231 W at the wall
Renting instead
$574/mo
8× A40 at $4.72/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 21, 2026.

How to run it

llama.cpp
llama-server -hf unsloth/GLM-5.3-Flash-GGUF
vLLM
vllm serve zai-org/GLM-5.3-Flash
MLX (Mac)
mlx_lm.chat --model mlx-community/GLM-5.3-Flash-4bit

GLM-5.3 Flash on similar machines

All machines: GLM-5.3 Flash VRAM requirements. Everything that runs here: what can the M5 Ultra 512 GB run?

Other models that run on M5 Ultra 512 GB

Questions

Can GLM-5.3 Flash run on M5 Ultra 512 GB?

GLM-5.3 Flash at FP8 runs on the M5 Ultra 512 GB with 1M of context at about 33 tokens a second.

Report VR-02CJU9X, calculated Nov 22, 2026 from published specs, not benchmarked.