IndexTTS-2.5 (0.8B) VRAM & GPU Requirement Calculator
Interactive hardware sizing for IndexTTS-2.5 (0.8B) (0.8B by Bilibili IndexTeam). Estimated 8K context footprint: 4.0 GB (INT4) • 4.4 GB (FP8) • 5.2 GB (FP16).
Fixed 30s audio window / codec frame cap
GPUs
Memory Requirements 5.22 GB
Requires 1 GPUs (based on memory capacity)
3.8 GB
All model weights
0.09 GB
Conversation history cache
1.15 GB
Concurrent processing cache
0.17 GB
Temporary computation cache
Performance Metrics
Visualize 536 tok/s →Max Streams (This GPU)
30 streams
Duplex Architecture
Stream / Turn-Based
Estimated Latency
280 ms
Max QPS
3.6 queries/s
Estimated using memory bandwidth and tensor compute throughput across 1x NVIDIA RTX 4090 (24GB).
Quick Deploy CLI Command
python webui.py --model_dir checkpoints --fp16
Flags reflect your selected precision, context length, and tensor parallel GPU count.Open IndexTTS-2.5 spec page →
IndexTTS-2.5 (0.8B) Architecture & Memory Profile
Total Parameters
0.8B
Architecture
Dense Transformer
Developer / Provider
Bilibili IndexTeam
Recommended Precision
FP16
IndexTTS-2.5 (0.8B) 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 | 5.2 GB | 7.1 GB | 14.7 GB |
| FP8 | 4.4 GB | 6.1 GB | 13.2 GB |
| INT8 | 4.4 GB | 6.1 GB | 13.2 GB |
| INT4 | 4.0 GB | 5.6 GB | 12.4 GB |
Recommended GPU Hardware for IndexTTS-2.5 (0.8B) (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 | 30 streams |
| NVIDIA RTX 5090 (32GB) | 1× | 32 GB | 43 streams |
| Apple M4 Max (128GB) (128GB) | 1× | 128 GB | 195 streams |
| NVIDIA A100 (80GB) (80GB) | 1× | 80 GB | 119 streams |
| NVIDIA H100 (80GB) | 1× | 80 GB | 119 streams |
| NVIDIA H200 (141GB) | 1× | 141 GB | 215 streams |
| Huawei Ascend 910B (64GB) | 1× | 64 GB | 94 streams |
IndexTTS-2.5 (0.8B) Deployment FAQ
How much VRAM is needed to run IndexTTS-2.5 (0.8B)?
For single-stream inference, IndexTTS-2.5 (0.8B) uses about 5.2 GB in FP16, 4.4 GB in FP8, and 4.0 GB in INT4 including resident acoustic codecs and stream buffers.
Can I run IndexTTS-2.5 (0.8B) on RTX 4090 or H100 GPUs?
Running IndexTTS-2.5 (0.8B) in FP16 precision requires at least 1x RTX 4090 (24GB) or 1x NVIDIA H100 (80GB) GPUs.
What precision should I choose for IndexTTS-2.5 (0.8B) in production?
For IndexTTS-2.5 (0.8B) (0.8B by Bilibili IndexTeam), FP16 provides the best balance between inference speed, VRAM footprint, and output quality.