EDGE OSDEEPSEEK · 671B PARAMETERS
DeepSeek-R1 hardware and VRAM requirements
DeepSeek’s 671B reasoning model, built on DeepSeek-V3 and still one of the most searched open-weight models. DeepSeek-R1 is a mixture-of-experts model with 37B of its 671B parameters work on each token. At FP8 it needs about 706 GB of GPU memory for one user, and about 715 GB to serve a team of 16 with 8K context each.
Total parameters
671B
Active per token
37B
Context length
128K tokens
Released as
FP8 (native)
Minimum VRAM
410 GB at 4-bit
For a team of 16
715 GB at FP8
VRAM needed to run DeepSeek-R1
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 (native) Released | 671 GB | 706 GB | 715 GB | 742 GB | 816 GB |
| 4-bit Estimated | 389 GB | 410 GB | 419 GB | 446 GB | 520 GB |
| BF16 Upcast | 1.34 TB | 1.41 TB | 1.42 TB | 1.45 TB | 1.52 TB |
Full fine-tuning in mixed precision needs about 10.81 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-R1 on-premise
For a team of 16 using DeepSeek-R1 at FP8 with 8K context each (about 715 GB), smallest fit first.
NVIDIA DGX B200
1.44 TB per system · 581 GB to spare.
AMD Instinct MI325X
256 GB per GPU · Split across 4 GPUs in one server, 207 GB to spare.
NVIDIA H200 NVL
141 GB per GPU · Split across 8 GPUs in one server, 300 GB to spare.
AMD Instinct MI300X
192 GB per GPU · Split across 8 GPUs in one server, 668 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-R1 for your team?
We will size the GPU server for your users and data, and set up DeepSeek-R1 on it inside your own building. Free, no obligation.
DeepSeek-R1 at a glance
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-R1 GPU questions
How do I run DeepSeek R1 locally?
To run DeepSeek-R1 locally you need about 410 GB of GPU memory at 4-bit for one user (715 GB for a team of 16), system RAM of at least 389 GB so the weights can be loaded, around 389 GB of fast NVMe storage for the model files, and an inference engine such as vLLM or SGLang. A team of 16 fits on NVIDIA DGX B200.
Is DeepSeek R1 free?
Yes. DeepSeek released the R1 weights under the MIT license, so you can download and run them without license fees. Running the full model still needs several hundred gigabytes of GPU memory.
How big is DeepSeek R1?
DeepSeek-R1 has 671B parameters in total, of which 37B work on each token. The weights take about 389 GB at 4-bit.
How much VRAM to run DeepSeek R1?
DeepSeek-R1 needs about 706 GB of GPU memory at FP8 for one user with 8K context: 671 GB for the weights plus the KV cache and runtime overhead. Serving 16 people at once with 8K context each needs about 715 GB.
How many GPUs to run DeepSeek R1?
For 16 users with 8K context at FP8 it needs about 715 GB. That fits on NVIDIA DGX B200.
How much RAM do I need to run DeepSeek R1?
System RAM should be at least as large as the model files so they can be loaded: about 389 GB at 4-bit, or 671 GB at FP8. For fast responses, though, the model has to fit in GPU memory (VRAM), not just RAM; running from RAM and CPU is many times slower.
Can DeepSeek R1 process images?
No. DeepSeek-R1 is a text model: it takes text in and produces text out.
Can you fine-tune DeepSeek R1?
LoRA fine-tuning at FP8 needs roughly 816 GB with 4K sequences; it trains small adapter layers on a frozen copy of the model. Full fine-tuning needs about 10.81 TB, which calls for a multi-GPU training cluster.
What about the smaller DeepSeek R1 distilled models?
DeepSeek also released distilled models from 1.5B to 70B parameters, based on Qwen2.5 and Llama 3 models. They are separate, much smaller models that need far less memory than the full 671B DeepSeek-R1 covered on this page.
How are these numbers calculated?
Weights are parameters × bytes per parameter. The KV cache uses 68.6 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 GitHub: deepseek-ai/DeepSeek-R1. 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-R1 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.