Parakeet-TDT-0.6B-v3 VRAM & GPU Requirement Calculator
Interactive hardware sizing for Parakeet-TDT-0.6B-v3 (0.6B by NVIDIA NeMo). Estimated 8K context footprint: 1.0 GB (INT4) • 1.3 GB (FP8) • 1.9 GB (FP16).
Non-autoregressive encoder/DiT (0 KV cache)
GPUs
Memory Requirements 1.89 GB
Requires 1 GPUs (based on memory capacity)
1.2 GB
All model weights
0 GB
Conversation history cache
0.6 GB
Concurrent processing cache
0.09 GB
Temporary computation cache
Performance Metrics
Visualize 714 tok/s →Max Streams (This GPU)
107 streams
Duplex Architecture
Stream / Turn-Based
Estimated Latency
210 ms
Max QPS
4.8 queries/s
Estimated using memory bandwidth and tensor compute throughput across 1x NVIDIA RTX 4090 (24GB).
Quick Deploy CLI Command
python -c "import nemo.collections.asr as nemo_asr; asr = nemo_asr.models.ASRModel.from_pretrained('nvidia/parakeet-tdt-0.6b-v3')"Flags reflect your selected precision, context length, and tensor parallel GPU count.Open Parakeet-TDT-0.6B-v3 spec page →
Parakeet-TDT-0.6B-v3 Architecture & Memory Profile
Total Parameters
0.6B
Architecture
Dense Transformer
Developer / Provider
NVIDIA NeMo
Recommended Precision
FP16
Parakeet-TDT-0.6B-v3 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 | 1.9 GB | 2.5 GB | 5.0 GB |
| FP8 | 1.3 GB | 1.9 GB | 4.4 GB |
| INT8 | 1.3 GB | 1.9 GB | 4.4 GB |
| INT4 | 1.0 GB | 1.6 GB | 4.1 GB |
Recommended GPU Hardware for Parakeet-TDT-0.6B-v3 (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 | 107 streams |
| NVIDIA RTX 5090 (32GB) | 1× | 32 GB | 146 streams |
| Apple M4 Max (128GB) (128GB) | 1× | 128 GB | 512 streams |
| NVIDIA A100 (80GB) (80GB) | 1× | 80 GB | 378 streams |
| NVIDIA H100 (80GB) | 1× | 80 GB | 378 streams |
| NVIDIA H200 (141GB) | 1× | 141 GB | 512 streams |
| Huawei Ascend 910B (64GB) | 1× | 64 GB | 301 streams |
Parakeet-TDT-0.6B-v3 Deployment FAQ
How much VRAM is needed to run Parakeet-TDT-0.6B-v3?
For single-stream inference, Parakeet-TDT-0.6B-v3 uses about 1.9 GB in FP16, 1.3 GB in FP8, and 1.0 GB in INT4 including resident acoustic codecs and stream buffers.
Can I run Parakeet-TDT-0.6B-v3 on RTX 4090 or H100 GPUs?
Running Parakeet-TDT-0.6B-v3 in FP16 precision requires at least 1x RTX 4090 (24GB) or 1x NVIDIA H100 (80GB) GPUs.
What precision should I choose for Parakeet-TDT-0.6B-v3 in production?
For Parakeet-TDT-0.6B-v3 (0.6B by NVIDIA NeMo), FP16 provides the best balance between inference speed, VRAM footprint, and output quality.