GPU × use-case guide · Text-to-speech

Is the H200 SXM a good GPU for text-to-speech?

The H200 SXM is a hopper NVIDIA GPU with 141 GB HBM3e (4800 GB/s), 990 TFLOPS BF16 / 1979 TFLOPS FP8, and a 700 W TDP. Text-to-speech workloads care most about low-latency streaming synthesis; small models, so value cards win. Here's how the H200 SXM measures up.

What models fit on a single H200 SXM?

Weights only, reserving ~25% of the 141 GB for KV cache, activations and fragmentation. ✓ = fits on one card.

Largest single-card fit: Mixtral 8x7B (MoE) at BF16, Qwen 2.5 72B at FP8, Qwen 2.5 72B at INT4. Bigger models need tensor-parallel across 8 cards.

H200 SXM for text-to-speech, specifically

Text-to-speech is throughput-bound rather than VRAM-bound, so the H200 SXM's 4800 GB/s of bandwidth and 990 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.

H200 SXM pricing across providers

Verdict

With 141 GB, the H200 SXM is a data-center-class card that comfortably handles text-to-speech for models up to Mixtral 8x7B (MoE) at full precision on a single card — a strong pick if your budget supports ~$3.49/hr.