KV Cache Size Calculator
Loading model configurations...
Model:
Data Type:
float16 (FP16)
bfloat16 (BF16)
float32 (FP32)
int8 (INT8)
DeepSeek V4 KV Precision (bytes / dimension):
Paper default — FP8 NoPE / BF16 RoPE / FP4 indexer
Custom…
NoPE bytes/dim:
RoPE bytes/dim:
Indexer bytes/dim:
Number of Tokens:
Calculate KV Cache Size
Reverse Calculator: Find Maximum Tokens
Model:
Data Type:
float16 (FP16)
bfloat16 (BF16)
float32 (FP32)
int8 (INT8)
DeepSeek V4 KV Precision (bytes / dimension):
Paper default — FP8 NoPE / BF16 RoPE / FP4 indexer
Custom…
NoPE bytes/dim:
RoPE bytes/dim:
Indexer bytes/dim:
GPU RAM Size (GB):
Calculate Maximum Tokens