GPU VRAM calculator

Run Qwen3.5-9B locally: hardware requirements

Qwen3.5-9B needs about 22.3 GB of GPU memory for one person at 32K context, and about 2.0 GB more for each extra person at 64K. For 80 users working as AI agents, plan on 53.3 GB, which fits one A100.

Alibaba (Qwen), China. From its files on Hugging Face, checked 2026-11-05.

Context per person: more context lets the model work with longer conversations and more files, and needs more GPU memory for every person. As a guide: 8K for short chat, 32K for chat over documents, 64K for AI agents, whose instructions, tools and files all count.

80 users means about 16 people working at the same moment. That share is a planning assumption; if you know yours, choose the total that matches it.

Work it out for your team: Qwen/Qwen3.5-9B

On Hugging Face

Alibaba (Qwen), China · 9.7B parameters · License: apache-2.0

What you need for 16 active users

1 × A100

It holds the model and up to 27 active users at 64K context.

The server

128 GB RAM
48 CPU cores
250 GB NVMe

Rules of thumb

GPU memory in total
53.3 GB
Weights
19.3 GB (published files)
Conversation cache, all active
32.0 GB
Engine (vLLM)
3.0 GB

Your options by card

80 users means about 16 using the model at any moment. The others are not waiting; they are simply doing something else. If more become active than a build handles, they queue. 1 card too small for this model is left out.

CardMemoryActive users, one serverFor your 16 active
A10080 GB271 × A100
H10080 GB271 × H100
RTX PRO 6000 Blackwell96 GB341 × RTX PRO 6000 Blackwell
H200 NVL141 GB561 × H200 NVL
B200180 GB741 × B200
GeForce RTX 309024 GB122 servers, each with 2 × GeForce RTX 3090
GeForce RTX 409024 GB122 servers, each with 2 × GeForce RTX 4090
GeForce RTX 508016 GB44 servers, each with 2 × GeForce RTX 5080
GeForce RTX 509032 GB192 × GeForce RTX 5090
L40S48 GB342 × L40S
RTX 4000 Ada20 GB173 × RTX 4000 Ada
RTX 6000 Ada48 GB342 × RTX 6000 Ada
RTX A500024 GB233 × RTX A5000
RTX PRO 4000 Blackwell24 GB233 × RTX PRO 4000 Blackwell
RTX PRO 4500 Blackwell32 GB192 × RTX PRO 4500 Blackwell
RTX PRO 5000 Blackwell48 GB342 × RTX PRO 5000 Blackwell

These are memory limits. A card can hold more people than it can answer quickly, so test your own workload before you buy. More than one card in a server needs an engine that splits the model.

Qwen3.5-9B by team size

AI agents, 1 user in 5 active at a time, 64K context, as published.

Total usersAt onceGPU memoryWhat you need
20429.3 GB1 × GeForce RTX 5090
801653.3 GB1 × A100
1202469.3 GB1 × A100
500100222 GB2 × H200 NVL
  • Qwen3-VL-8B-InstructAlibaba (Qwen) · 8.8B · 164 GB for 16 at once
  • Qwen3-8BAlibaba (Qwen) · 8.2B · 109 GB for 16 at once
  • Qwen3-14BAlibaba (Qwen) · 15B · 131 GB for 16 at once
  • Qwen3.5-4BAlibaba (Qwen) · 4.7B · 43.8 GB for 16 at once
  • Qwen3-4BAlibaba (Qwen) · 4.0B · 101 GB for 16 at once
  • Qwen3.8-27BAlibaba (Qwen) · 28B · 120 GB for 16 at once

All models

How this is worked out

GPU memory = weights + a conversation cache for each person + the serving engine, from this model's files on Hugging Face. These are memory limits, not speed. The full method and its limits.

Questions

How much GPU memory does Qwen3.5-9B need?

Qwen3.5-9B needs about 22.3 GB of GPU memory for one person at 32K context, and about 2.0 GB more for each extra person at 64K. For 80 users working as AI agents, plan on 53.3 GB, which fits one A100.

How many active users can Qwen3.5-9B handle on one 96 GB GPU?

By memory, about 34 active users at once at 64K context on a card such as the RTX PRO 6000 Blackwell. Speed can set a lower limit, so load-test before you buy.

Can Qwen3.5-9B run on a 24 GB GPU such as the RTX 4090?

Yes as published for one person at 32K context: it needs about 22.3 GB.

What license does Qwen3.5-9B use?

Hugging Face lists its license as apache-2.0. Check the license on the model's page before you use it.

Running a model on your own servers?

NeueCode is private autonomous intelligence for enterprise: it takes on your organization's tasks on your own GPUs, with nothing leaving your network, a person approving every action, and all of it on record. It runs each model on one card, so read the single-card figures above.

What you need to run NeueCode