DeepSeek V2 Lite
DeepSeek · moe · 15.7B parameters · 32,768 context
Parameters
15.7B
Context Window
32K tokens
Architecture
MoE
Best GPU
H100 SXM
Intelligence Brief
DeepSeek V2 Lite is a 15.7B parameter Mixture-of-Experts (64 experts, 6 active) model from DeepSeek, featuring Multi-Head Attention (MHA) with 27 layers and 2,048 hidden dimensions. With a 32,768 token context window, it supports code, math, multilingual. For self-hosted inference, H100 SXM delivers optimal throughput at $1794/month.
Architecture Details
Memory Requirements
BF16 Weights
31.4 GB
FP8 Weights
15.7 GB
INT4 Weights
7.8 GB
GPU Compatibility Matrix
DeepSeek V2 Lite is compatible with 82% of GPU configurations across 41 GPUs at 3 precision levels.
GPU Recommendations
FP8 · 1 GPU · tensorrt-llm
100/100
score
Throughput
1.1K tok/s
Latency (ITL)
1.0ms
Est. TTFT
0ms
Cost/Month
$1794
Cost/M Tokens
$0.65
FP8 · 1 GPU · tensorrt-llm
100/100
score
Throughput
1.1K tok/s
Latency (ITL)
1.0ms
Est. TTFT
0ms
Cost/Month
$1794
Cost/M Tokens
$0.65
BF16 · 1 GPU · vllm
100/100
score
Throughput
864.0 tok/s
Latency (ITL)
1.2ms
Est. TTFT
0ms
Cost/Month
$465
Cost/M Tokens
$0.20
Deployment Options
API Deployment
No API pricing available
Single GPU
H100 SXM
$1794/mo
Min VRAM: 16 GB
Multi-GPU
A10G x2
897.4 tok/s
TP· $569/mo
API Pricing Comparison
No API pricing data available for this model.
Performance Estimates
Throughput by GPU
VRAM Breakdown (H100 SXM, FP8)
Precision Impact
bf16
31.4 GB
weights/GPU
fp8
15.7 GB
weights/GPU
~1.1K tok/s
int4
7.8 GB
weights/GPU
Capabilities
Features
Supported Frameworks
Supported Precisions
Where to Deploy DeepSeek V2 Lite
Self-Hosted Infrastructure
Similar Models
OctoCoder 15B
15.5B params · dense
Quality: 50
StarCoder2 15B
15.5B params · dense
Quality: 42
from $0.30/M
CodeGen2 16B
16B params · dense
Quality: 50
DeepSeek MoE 16B
16.4B params · moe
Quality: 50
Nemotron 15B
15B params · dense
Quality: 72
from $0.30/M
Frequently Asked Questions
How much VRAM does DeepSeek V2 Lite need for inference?
DeepSeek V2 Lite requires approximately 31.4 GB of VRAM at BF16 precision, 15.7 GB at FP8, or 7.8 GB at INT4 quantization. Additional VRAM is needed for KV-cache (221184 bytes per token) and activations (~0.50 GB).
What is the best GPU for DeepSeek V2 Lite?
The top recommended GPU for DeepSeek V2 Lite is the H100 SXM using FP8 precision. It achieves approximately 1.1K tokens/sec at an estimated cost of $1794/month ($0.65/M tokens). Score: 100/100.
How much does DeepSeek V2 Lite inference cost?
DeepSeek V2 Lite inference costs vary by provider and GPU setup. Use our calculator for detailed cost estimates across all providers.