Cheapest Cloud GPU to Host Whisper Large v3
Whisper Large v3 is OpenAI's state-of-the-art speech-to-text model supporting 99 languages. At FP16 (4.5 GB), it runs on a single L40S with capacity for 10+ concurrent transcription workers. The model excels at long-form audio transcription with word-level timestamps.
VRAM Sizing Matrix
Memory consumption at different quantization levels for Whisper Large v3. Includes model weights only — KV-cache adds 2-32 GB depending on context length and batch size.
| Precision | VRAM Required | Fits On | Quality Loss |
|---|---|---|---|
| FP16 / BF16 | 4.5 GB | Single GPU with headroom | None (full precision) |
| FP8 | 2.5 GB | Single GPU | <1% perplexity delta |
| INT4 / AWQ | 1.8 GB | Single consumer GPU | 1-3% quality degradation |
Hardware Bottleneck Advice
KV-cache expansion, tensor parallelism constraints, and production considerations for Whisper Large v3.
KV-Cache & Context Scaling
1.55B encoder-decoder model. The encoder processes audio mel spectrograms; the decoder generates text tokens. At FP16 (4.5 GB), the L40S (48 GB) can run 10+ concurrent transcription workers.
Tensor Parallelism
TP=1. The 1.55B parameter count is small enough for single-GPU serving. L40S is ideal for high-throughput concurrent batching of multiple audio streams.
Live Cloud Pricing for L40S
Filtered to L40S-class hardware — the recommended GPU for hosting Whisper Large v3.
| Provider | GPU & VRAM | Interconnect | Spot Rate | On-Demand | Monthly | Status | Action | |
|---|---|---|---|---|---|---|---|---|
Community | PCIe 4.0 (64 GB/s) | $0.69 / hr | $1.73 / hr | $422 / mo | Instant | |||
Cloud | PCIe 4.0 (64 GB/s) | $1.09 / hr | $2.73 / hr | $667 / mo | Instant | |||
Bare Metal | PCIe 4.0 (64 GB/s) | $1.19 / hr | $2.97 / hr | $728 / mo | Instant | |||
Dedicated | PCIe 4.0 (64 GB/s) | $1.49 / hr | $3.73 / hr | $912 / mo | Instant |
Production Runbook
Ready-to-deploy vLLM / SGLang launch command for Whisper Large v3.
vllm serve openai/whisper-large-v3 --max-model-len 448 --dtype autoRelated hosting guides