DeepSeek AI
DeepSeek V4 Flash (120B)
Fast V4 variant optimized for cost-sensitive deployments
Model Summary
Family
DeepSeek
Version
4.0
Parameters
120B (est.)
Parameter counts for closed models are estimates; vendors rarely publish exact sizes.
VRAM Requirements by Quantization
Memory needed to serve DeepSeek V4 Flash (120B) for inference, including a 20% overhead for activations and KV cache.
| Precision | VRAM needed | Smallest single GPU that fits |
|---|---|---|
| INT4 (4-bit) | 67.06 GB | NVIDIA H100 SXM5 (80 GB) |
| INT8 (8-bit) | 134.11 GB | NVIDIA H200 SXM5 (141 GB) |
| FP16 (16-bit) | 268.22 GB | NVIDIA B300 SXM (288 GB) |
| FP32 (32-bit) | 536.44 GB | Multi-GPU required |
Recommended GPU Configurations
Cheapest on-demand configurations to serve DeepSeek V4 Flash (120B) at 8-bit (134 GB VRAM).
1x AMD Instinct MI300X
192 GB total VRAM · CDNA 3
~$6.00/h
1x AMD Instinct MI325X
256 GB total VRAM · CDNA 3
~$8.00/h
2x NVIDIA A100 80GB
160 GB total VRAM · Ampere
~$9.00/h
Quick GPU Planning
Use the calculator pre-filled with this exact version to estimate memory, speed, and compute requirements in a few clicks.
Access Pre-filled CalculatorFrequently Asked Questions
How much VRAM do you need to run DeepSeek V4 Flash (120B)?
With an estimated 120B parameters, DeepSeek V4 Flash (120B) needs roughly 134 GB of VRAM in 8-bit (INT8), 67 GB in 4-bit, and 268 GB in FP16, including a 20% overhead for activations and KV cache.
Which GPUs can run DeepSeek V4 Flash (120B)?
At 8-bit quantization, the most cost-effective option is 1x AMD Instinct MI300X (192 GB combined VRAM, around $6.00/hour on-demand). Higher-end cards like the NVIDIA B200 or AMD MI355X reduce the GPU count needed.
Can DeepSeek V4 Flash (120B) run on a single GPU?
Yes. In 8-bit, a single NVIDIA H200 SXM5 (141 GB) fits the model.
How much does it cost to serve DeepSeek V4 Flash (120B) in the cloud?
Renting 1x Instinct MI300X costs on the order of $6.00/hour, i.e. about $4,380/month running 24/7. Actual prices vary by provider and commitment; spot and reserved capacity can be significantly cheaper.
Deploy on a GPU cloud
Rent 1x Instinct MI300X by the hour instead of buying hardware.
