System Requirements#

This page provides detailed hardware, software, and platform requirements to help you set up and run the microservice efficiently.

Supported Platforms#

The multilevel-video-understanding microservice itself is platform-agnostic — it only needs an OpenAI-compatible model serving. In practice the platform sensitivity comes entirely from the model serving (vLLM: llm-scaler). To run on a different platform, simply deploy llm-scaler on that platform and point the microservice at its endpoint; nothing else changes.

Two platforms have been validated:

  • Intel® Core™ Ultra (Panther Lake, PTL 358H) with integrated GPU sharing system RAM — the on-device profile this Get Started guide is written around (no discrete accelerator or remote inference cluster required).

  • Intel® Xeon™ (e.g. GOLD 6530) + Intel® Arc™ Pro B60 Graphics — a discrete-GPU serving host.

Operating Systems

  • Ubuntu 24.04 LTS or later

Hardware — Platform #1: Intel® Core™ Ultra (PTL 358H), on-device iGPU

  • Intel® Core™ Ultra processor with integrated GPU (validated on Panther Lake, PTL 358H), sharing system RAM

  • Validated kernel: 7.0.0-38-generic

  • At least 64 GB RAM — default deployment target for Qwen3.5-35B-A3B (FP8, 60k context)

  • At least 32 GB swap — so the model load and KV cache can spill under peak memory pressure

  • At least 128 GB disk space (model weights + Hugging Face cache)

Note

To fit smaller-memory hosts, lower MAX_MODEL_LEN or use awq / sym_int4 quantization in docker/set_env.sh.

Hardware — Platform #2: Intel® Xeon™ + Intel® Arc™ Pro B60, discrete-GPU serving host

  • Intel® Xeon™ processor (validated on Intel® Xeon™ GOLD 6530)

  • Intel® Arc™ Pro B60 Graphics — at least 2 GPUs are required to serve Qwen3.5-35B-A3B

  • At least 64 GB RAM

  • At least 128 GB disk space (model weights + Hugging Face cache)

Note

On this platform, deploy vLLM (llm-scaler) on 2x B60 host and modify related env variables before deplyment:

export GPU_MEM_UTIL=0.9
export TENSOR_PARALLEL_SIZE=2

Software Requirements#

Required Software:

  • Docker 24.0

  • Python 3.10

Validation#

  • Ensure all required software are installed and configured before proceeding to Get Started.

Supporting Resources#