Whisper-large-v3 (1.55B) VRAM & GPU Requirement Calculator

Interactive hardware sizing for Whisper-large-v3 (1.55B) (1.55B by OpenAI). Estimated 8K context footprint: 1.8 GB (INT4) • 2.5 GB (FP8) • 4.1 GB (FP16).

Fixed 30s audio window / codec frame cap
GPUs

Memory Requirements 4.12 GB

Requires 1 GPUs (based on memory capacity)

3.1 GB

All model weights

0.07 GB

Conversation history cache

0.83 GB

Concurrent processing cache

0.12 GB

Temporary computation cache

Performance Metrics
Visualize 276 tok/s →
Max Streams (This GPU)
52 streams
Duplex Architecture
Stream / Turn-Based
Estimated Latency
543 ms
Max QPS
1.8 queries/s

Estimated using memory bandwidth and tensor compute throughput across 1x NVIDIA RTX 4090 (24GB).

Quick Deploy CLI Command
vllm serve openai/whisper-large-v3 --task transcription
Flags reflect your selected precision, context length, and tensor parallel GPU count.Open Whisper-large-v3 spec page →

Whisper-large-v3 (1.55B) Architecture & Memory Profile

Total Parameters
1.55B
Architecture
Dense Transformer
Developer / Provider
OpenAI
Recommended Precision
FP16

Whisper-large-v3 (1.55B) VRAM Matrix by Precision & Concurrent Streams

Precomputed audio-stream VRAM requirements including unquantized acoustic/codec submodules and runtime workspace.

Precision1 Stream4 Streams16 Streams
FP164.1 GB5.3 GB9.9 GB
FP82.5 GB3.6 GB7.8 GB
INT82.5 GB3.6 GB7.8 GB
INT41.8 GB2.7 GB6.7 GB

Recommended GPU Hardware for Whisper-large-v3 (1.55B) (FP16)

Minimum GPU count and maximum concurrent real-time voice streams per node.

GPU HardwareMin GPUsTotal VRAM PoolMax Streams
NVIDIA RTX 4090 (24GB)1×24 GB52 streams
NVIDIA RTX 5090 (32GB)1×32 GB73 streams
Apple M4 Max (128GB) (128GB)1×128 GB324 streams
NVIDIA A100 (80GB) (80GB)1×80 GB199 streams
NVIDIA H100 (80GB)1×80 GB199 streams
NVIDIA H200 (141GB)1×141 GB358 streams
Huawei Ascend 910B (64GB)1×64 GB157 streams

Whisper-large-v3 (1.55B) Deployment FAQ

How much VRAM is needed to run Whisper-large-v3 (1.55B)?
For single-stream inference, Whisper-large-v3 (1.55B) uses about 4.1 GB in FP16, 2.5 GB in FP8, and 1.8 GB in INT4 including resident acoustic codecs and stream buffers.
Can I run Whisper-large-v3 (1.55B) on RTX 4090 or H100 GPUs?
Running Whisper-large-v3 (1.55B) in FP16 precision requires at least 1x RTX 4090 (24GB) or 1x NVIDIA H100 (80GB) GPUs.
What precision should I choose for Whisper-large-v3 (1.55B) in production?
For Whisper-large-v3 (1.55B) (1.55B by OpenAI), FP16 provides the best balance between inference speed, VRAM footprint, and output quality.