GPU × use-case guide · Speech-to-text

Is the T4 a good GPU for speech-to-text?

The T4 is a turing NVIDIA GPU with 16 GB GDDR6 (300 GB/s), 65 TFLOPS BF16 / 65 TFLOPS FP8, and a 70 W TDP. Speech-to-text workloads care most about encoder throughput and real-time factor; models are small so mid-range cards are plenty. Here's how the T4 measures up.

What models fit on a single T4?

Weights only, reserving ~25% of the 16 GB for KV cache, activations and fragmentation. ✓ = fits on one card.

Largest single-card fit: none (needs multi-GPU) at BF16, Llama 3.1 8B at FP8, Qwen 2.5 14B at INT4. Bigger models need tensor-parallel across 8 cards.

T4 for speech-to-text, specifically

Speech-to-text is throughput-bound rather than VRAM-bound, so the T4's 300 GB/s of bandwidth and 65 TFLOPS matter more than raw capacity. Because it's latency-sensitive, run it at low batch sizes on a fast framework (vLLM ≈ 85% MFU) rather than maximising batch. Size it precisely on the calculator.

T4 pricing across providers

Verdict

At 16 GB, the T4 is a value card best suited to smaller speech-to-text models (up to Qwen 2.5 14B with INT4 quantisation). For larger models you'll want more VRAM or multi-GPU.