Step-Audio-2-mini (8B) VRAM & GPU Requirement Calculator
Interactive hardware sizing for Step-Audio-2-mini (8B) (8B by StepFun). Estimated 8K context footprint: 6.2 GB (INT4) • 10.4 GB (FP8) • 18.9 GB (FP16).
GPUs
Memory Requirements 18.92 GB
Requires 1 GPUs (based on memory capacity)
17 GB
All model weights
1 GB
Conversation history cache
0.8 GB
Concurrent processing cache
0.12 GB
Temporary computation cache
Performance Metrics
Visualize 54 tok/s →Total Throughput
54 tokens/s
Throughput Per User
54 tokens/s/user
Estimated Latency
2801 ms
Max QPS
0.4 queries/s
Estimated using memory bandwidth and tensor compute throughput across 1x NVIDIA RTX 4090 (24GB).
Quick Deploy CLI Command
vllm serve stepfun-ai/Step-Audio-2-mini --max-model-len 8192
Flags reflect your selected precision, context length, and tensor parallel GPU count.Open Step-Audio-2-mini spec page →
Step-Audio-2-mini (8B) Architecture & Memory Profile
Total Parameters
8B
Architecture
Dense Transformer
Developer / Provider
StepFun
Recommended Precision
FP16
Step-Audio-2-mini (8B) VRAM Matrix by Precision & Context Length
Precomputed single-user memory requirements (model weights + KV cache + runtime overhead).
| Precision | 8K Context | 32K Context | 128K Context |
|---|---|---|---|
| FP16 | 18.9 GB | 21.9 GB | 33.9 GB |
| FP8 | 10.4 GB | 11.9 GB | 17.9 GB |
| INT8 | 10.4 GB | 11.9 GB | 17.9 GB |
| INT4 | 6.2 GB | 6.9 GB | 9.9 GB |
Recommended GPU Hardware for Step-Audio-2-mini (8B) (FP16)
Minimum GPU count for 8K context at batch size 1 and estimated generation throughput.
| GPU Hardware | Min GPUs | Total VRAM Pool | Est. Speed |
|---|---|---|---|
| NVIDIA RTX 4090 (24GB) | 1× | 24 GB | 54 tok/s |
| NVIDIA RTX 5090 (32GB) | 1× | 32 GB | 95 tok/s |
| Apple M4 Max (128GB) (128GB) | 1× | 128 GB | 22 tok/s |
| NVIDIA A100 (80GB) (80GB) | 1× | 80 GB | 103 tok/s |
| NVIDIA H100 (80GB) | 1× | 80 GB | 175 tok/s |
| NVIDIA H200 (141GB) | 1× | 141 GB | 255 tok/s |
| Huawei Ascend 910B (64GB) | 1× | 64 GB | 64 tok/s |
Step-Audio-2-mini (8B) Deployment FAQ
How much VRAM is needed to run Step-Audio-2-mini (8B)?
At 8K context and batch size 1, Step-Audio-2-mini (8B) uses about 18.9 GB in FP16, 10.4 GB in FP8, and 6.2 GB in INT4/AWQ quantization.
Can I run Step-Audio-2-mini (8B) on RTX 4090 or H100 GPUs?
Running Step-Audio-2-mini (8B) in FP16 precision requires at least 1x RTX 4090 (24GB) or 1x NVIDIA H100 (80GB) GPUs.
What precision should I choose for Step-Audio-2-mini (8B) in production?
For Step-Audio-2-mini (8B) (8B by StepFun), FP16 provides the best balance between inference speed, VRAM footprint, and output quality.