EDGE OSDEEPSEEK · 685B PARAMETERS
DeepSeek-V3.2 hardware and VRAM requirements
DeepSeek’s 685B flagship, the most downloaded 70B+ open-weight model on Hugging Face in early November 2026. DeepSeek-V3.2 is a mixture-of-experts model with 37B of its 685B parameters work on each token. At FP8 it needs about 721 GB of GPU memory for one user, and about 730 GB to serve a team of 16 with 8K context each.
Total parameters
685B
Active per token
37B
Context length
128K
Released as
FP8, NVFP4 (4-bit)
Minimum VRAM
419 GB at NVFP4
For a team of 16
730 GB at FP8
VRAM needed to run DeepSeek-V3.2
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 |
|---|---|---|---|---|---|
| FP8 Released | 685 GB | 721 GB | 730 GB | 761 GB | 833 GB |
| NVFP4 (4-bit) NVIDIA NVFP4 build (size estimated) | 397 GB | 419 GB | 428 GB | 459 GB | 531 GB |
| BF16 Upcast | 1.37 TB | 1.44 TB | 1.45 TB | 1.48 TB | 1.55 TB |
Full fine-tuning in mixed precision needs about 11.03 TB: 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 DeepSeek-V3.2 on-premise
For a team of 16 using DeepSeek-V3.2 at FP8 with 8K context each (about 730 GB), smallest fit first.
NVIDIA DGX B200
1.44 TB per system · 566 GB to spare.
AMD Instinct MI325X
256 GB per GPU · Split across 4 GPUs in one server, 191 GB to spare.
NVIDIA H200 NVL
141 GB per GPU · Split across 8 GPUs in one server, 285 GB to spare.
AMD Instinct MI300X
192 GB per GPU · Split across 8 GPUs in one server, 652 GB to spare.
NVIDIA DGX Spark
128 GB per system · Needs more than 2 linked systems.
NVIDIA L40S
48 GB per GPU · Needs more than one 8-GPU server.
NVIDIA H100 NVL
94 GB per GPU · Needs more than one 8-GPU server.
NVIDIA RTX PRO 6000 Blackwell
96 GB per GPU · Needs more than one 8-GPU server.
Some of these setups need several GPUs working as one. Buying the GPUs is the easy part. Vanik Edge OS splits the model across them, keeps it running if a server fails and manages it all from one console. See how Edge OS clusters GPU servers
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 DeepSeek-V3.2 for your team?
We will size the GPU server for your users and data, and set up DeepSeek-V3.2 on it inside your own building. Free, no obligation.
DeepSeek-V3.2 at a glance
Compare DeepSeek-V3.2 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.
DeepSeek-V3.2 GPU questions
Is DeepSeek V3.2 free?
The weights are free: DeepSeek publishes DeepSeek-V3.2 under the MIT license, so you can download and run it without license fees. Running it yourself still needs several hundred gigabytes of GPU memory.
Is DeepSeek V3.2 open source?
Yes. DeepSeek-V3.2’s weights are released under the MIT license, so you can self-host and use it commercially.
Is DeepSeek V3.2 multimodal?
No. DeepSeek-V3.2 is a text model: it takes text in and produces text out.
How much VRAM to run DeepSeek V3.2?
DeepSeek-V3.2 needs about 721 GB of GPU memory at FP8 for one user with 8K context: 685 GB for the weights plus the KV cache and runtime overhead. Serving 16 people at once with 8K context each needs about 730 GB.
Can I run DeepSeek V3.2 locally?
Yes, on data-center GPUs. At NVFP4 it needs about 419 GB of GPU memory for one user, more than consumer graphics cards or laptops hold. A team of 16 fits on NVIDIA DGX B200.
How many GPUs does DeepSeek V3.2 need?
For 16 users with 8K context at FP8 it needs about 730 GB. That fits on NVIDIA DGX B200.
Why does DeepSeek V3.2 need so much memory if only 37B parameters are active?
It is a mixture-of-experts model. Only 37B parameters work on each token, which keeps it fast, but a different set of experts is picked for every token, so all 685B parameters must stay loaded in GPU memory.
Can I fine-tune DeepSeek V3.2?
LoRA fine-tuning at FP8 needs roughly 833 GB with 4K sequences; it trains small adapter layers on a frozen copy of the model. Full fine-tuning needs about 11.03 TB, which calls for a multi-GPU training cluster.
How are these numbers calculated?
Weights are parameters × bytes per parameter. The KV cache uses 76.2 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 NVIDIA: DeepSeek-V3.2 model card. 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 DeepSeek-V3.2 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.