howmanygpus.ai

howmanygpus.ai — Daily normalized cloud GPU rental price panel across 37 GPU models (B300 SXM 288GB down to RTX 4070 12GB), where each model page holds a provider-level offer table — provider name, source pricing URL, pricing tier (on-demand / spot / community), hourly USD price, 30-day change — roughly 107 provider x model offer points per daily pass. Layered on top: a market-median index ($1.20/hr as of 2026-10-02), per-model min/median/provider-count aggregates, a GPU specification catalog (VRAM, memory bandwidth, TFLOPS, NVLink, price/hour) spanning 130+ LLM model configs, and roofline-ca…

howmanygpus.ai offers (Price, Alternative) — ~107 provider x GPU-model offer observations per daily pass (the sum of the per-model provider counts printed in the 37-model panel — H100 SXM5 80GB and L40S 48GB at 12 providers each, H200 SXM at 8, A100 SXM4 at 7), one snapshot per day, plus 37 model detail pages, per-model min/median/30-day aggregates, and a GPU spec catalog covering 130+ LLM model configurations..

GPU procurement benchmarking across on-demand vs spot tiers; total-cost modelling for LLM inference (the native product mechanic converts GPU hourly prices into tokens-per-dollar and cost-per-million-token economics); sizing and capacity planning for N users at an SLA, including fine-tuning VRAM (full/LoRA/QLoRA with ZeRO/FSDP) and edge/Jetson Orin-Thor feasibility; monitoring the long-tail older-card rental market where 30-day moves run +58% to +182% (RTX 3090 24GB, RTX A5000, RTX 4070 Ti Super); cross-provider spread and arbitrage detection; corroborating specialist GPU index vendors on the sub-$1/hr consumer and previous-gen datacenter SKUs they omit.

Coverage spans US; Application Software, IT Services; price, alternative; equities.