F5-TTS (0.34B) VRAM & GPU Requirement Calculator
Interactive hardware sizing for F5-TTS (0.34B) (0.34B by SWivid). Estimated 8K context footprint: 1.1 GB (INT4) • 1.3 GB (FP8) • 1.6 GB (FP16).
Non-autoregressive encoder/DiT (0 KV cache)
GPUs
Memory Requirements 1.6 GB
Requires 1 GPUs (based on memory capacity)
0.83 GB
All model weights
0 GB
Conversation history cache
0.67 GB
Concurrent processing cache
0.1 GB
Temporary computation cache
Performance Metrics
Visualize 1260 tok/s →Max Streams (This GPU)
85 streams
Duplex Architecture
Stream / Turn-Based
Estimated Latency
119 ms
Max QPS
8.4 queries/s
Estimated using memory bandwidth and tensor compute throughput across 1x NVIDIA RTX 4090 (24GB).
Quick Deploy CLI Command
f5-tts_infer-gradio --port 7860
Flags reflect your selected precision, context length, and tensor parallel GPU count.Open F5-TTS spec page →
F5-TTS (0.34B) Architecture & Memory Profile
Total Parameters
0.34B
Architecture
Dense Transformer
Developer / Provider
SWivid
Recommended Precision
FP16
F5-TTS (0.34B) VRAM Matrix by Precision & Concurrent Streams
Precomputed audio-stream VRAM requirements including unquantized acoustic/codec submodules and runtime workspace.
| Precision | 1 Stream | 4 Streams | 16 Streams |
|---|---|---|---|
| FP16 | 1.6 GB | 2.4 GB | 5.6 GB |
| FP8 | 1.3 GB | 2.1 GB | 5.2 GB |
| INT8 | 1.3 GB | 2.1 GB | 5.2 GB |
| INT4 | 1.1 GB | 1.9 GB | 5.1 GB |
Recommended GPU Hardware for F5-TTS (0.34B) (FP16)
Minimum GPU count and maximum concurrent real-time voice streams per node.
| GPU Hardware | Min GPUs | Total VRAM Pool | Max Streams |
|---|---|---|---|
| NVIDIA RTX 4090 (24GB) | 1× | 24 GB | 85 streams |
| NVIDIA RTX 5090 (32GB) | 1× | 32 GB | 115 streams |
| Apple M4 Max (128GB) (128GB) | 1× | 128 GB | 478 streams |
| NVIDIA A100 (80GB) (80GB) | 1× | 80 GB | 296 streams |
| NVIDIA H100 (80GB) | 1× | 80 GB | 296 streams |
| NVIDIA H200 (141GB) | 1× | 141 GB | 512 streams |
| Huawei Ascend 910B (64GB) | 1× | 64 GB | 236 streams |
F5-TTS (0.34B) Deployment FAQ
How much VRAM is needed to run F5-TTS (0.34B)?
For single-stream inference, F5-TTS (0.34B) uses about 1.6 GB in FP16, 1.3 GB in FP8, and 1.1 GB in INT4 including resident acoustic codecs and stream buffers.
Can I run F5-TTS (0.34B) on RTX 4090 or H100 GPUs?
Running F5-TTS (0.34B) in FP16 precision requires at least 1x RTX 4090 (24GB) or 1x NVIDIA H100 (80GB) GPUs.
What precision should I choose for F5-TTS (0.34B) in production?
For F5-TTS (0.34B) (0.34B by SWivid), FP16 provides the best balance between inference speed, VRAM footprint, and output quality.