EDGE OSQWEN · 35B PARAMETERS
Qwen3.6-35B-A3B hardware and VRAM requirements
Alibaba’s small mixture-of-experts model: 35B parameters with only 3B active per token, so it runs fast on a single GPU. Qwen3.6-35B-A3B is a mixture-of-experts model with 3B of its 35B parameters work on each token. At FP8 it needs about 38 GB of GPU memory for one user, and about 42 GB to serve a team of 16 with 8K context each.
Total parameters
35B
Active per token
3B
Context length
262K native
Released as
BF16, FP8, NVFP4 (4-bit)
Minimum VRAM
24 GB at NVFP4
For a team of 16
42 GB at FP8
VRAM needed to run Qwen3.6-35B-A3B
GPU memory by precision, number of concurrent users and context length. Each user needs their own KV cache, so memory grows with users and context.
| Precision | Weights only | 1 user, 8K | 16 users, 8K | 16 users, 32K | LoRA fine-tune |
|---|---|---|---|---|---|
| BF16 Released | 70 GB | 75 GB | 78 GB | 87 GB | 80 GB |
| FP8 Released | 35 GB | 38 GB | 42 GB | 50 GB | 44 GB |
| NVFP4 (4-bit) 22.1 GB NVIDIA build | 22 GB | 24 GB | 28 GB | 36 GB | 30 GB |
Full fine-tuning in mixed precision needs about 565 GB: weights, gradients and optimizer states at roughly 16 bytes per parameter. LoRA trains small adapter layers on a frozen copy of the model instead, so it needs far less.
GPUs that can run Qwen3.6-35B-A3B on-premise
For a team of 16 using Qwen3.6-35B-A3B at FP8 with 8K context each (about 42 GB), smallest fit first.
NVIDIA L40S
48 GB per GPU · 1.5 GB to spare.
NVIDIA H100 NVL
94 GB per GPU · 43 GB to spare.
NVIDIA RTX PRO 6000 Blackwell
96 GB per GPU · 45 GB to spare.
NVIDIA DGX Spark
128 GB per system · 67 GB to spare.
NVIDIA H200 NVL
141 GB per GPU · 85 GB to spare.
AMD Instinct MI300X
192 GB per GPU · 131 GB to spare.
AMD Instinct MI325X
256 GB per GPU · 189 GB to spare.
NVIDIA DGX B200
1.44 TB per system · 1.25 TB to spare.
Weighing two of these? Compare AI hardware side by side, for example H100 vs H200, RTX PRO 6000 vs RTX 5090 or DGX Spark vs Mac Studio.
Running Qwen3.6-35B-A3B for your team?
We will size the GPU server for your users and data, and set up Qwen3.6-35B-A3B on it inside your own building. Free, no obligation.
Qwen3.6-35B-A3B at a glance
Compare Qwen3.6-35B-A3B with other models
FROM SIZING TO RUNNING
Sizing the hardware is the easy part.
Getting several GPUs to work as one system for a whole company is where most projects slow down. That is the part Vanik Edge OS does for you.
STEP 01 · THIS TOOL
Size it
Work out the GPU memory your models need and which GPUs or servers can hold them.
STEP 02 · VANIK EDGE OS
Cluster it
Edge OS joins your GPU servers into one cluster, splits a large model across them and keeps it running if a server fails. NVIDIA and AMD servers work side by side.
STEP 03 · VANIK EDGE OS
Run it
Your team gets AI apps with single sign-on and an audit trail, managed from one console, with your data staying inside your building.
Qwen3.6-35B-A3B GPU questions
Can I run Qwen 3.6 35B-A3B locally?
Yes, on data-center GPUs. At NVFP4 it needs about 24 GB of GPU memory for one user, more than consumer graphics cards or laptops hold. A team of 16 fits on NVIDIA L40S.
How big is Qwen 3.6 35B-A3B?
Qwen3.6-35B-A3B has 35B parameters in total, of which 3B work on each token. The weights take about 22 GB at NVFP4.
How much VRAM to run Qwen 3.6 35B-A3B?
Qwen3.6-35B-A3B needs about 38 GB of GPU memory at FP8 for one user with 8K context: 35 GB for the weights plus the KV cache and runtime overhead. Serving 16 people at once with 8K context each needs about 42 GB.
Is Qwen 3.6 35B-A3B open source?
Yes. Its weights are published under the Apache 2.0 license, so you can download, self-host and use it commercially.
Can Qwen 3.6 35B-A3B run on a single GPU?
Yes. At FP8 with one user and 8K context it needs about 38 GB, which fits on a single NVIDIA L40S or NVIDIA H100 NVL.
Why does Qwen 3.6 35B-A3B need so much memory if only 3B parameters are active?
It is a mixture-of-experts model. Only 3B parameters work on each token, which keeps it fast, but a different set of experts is picked for every token, so all 35B parameters must stay loaded in GPU memory.
Can I fine-tune Qwen 3.6 35B-A3B?
LoRA fine-tuning at FP8 needs roughly 44 GB with 4K sequences; it trains small adapter layers on a frozen copy of the model. Full fine-tuning needs about 565 GB, which calls for a multi-GPU training cluster.
How are these numbers calculated?
Weights are parameters × bytes per parameter. The KV cache uses 20.0 KB per token per user. We add about 5% of the weights plus 1 GB for runtime overhead and keep 10% of each GPU free. Figures are planning estimates; real usage depends on the inference engine and settings.
VRAM for similar models
Specifications from Hugging Face: Qwen/Qwen3.6-35B-A3B. Memory and bandwidth are manufacturer figures; GPU counts are rounded to 1, 2, 4 or 8 per server. All figures are planning estimates. Last checked November 10, 2026.
Run Qwen3.6-35B-A3B inside your own building
Vanik Edge OS runs open-weight models on NVIDIA DGX, AMD Instinct and other GPU servers on your premises, so your data never leaves.