GPU prices
When an H200 or B200 is the cheaper GPU
Answer in brief
A datacenter GPU is the cheaper choice when the job does not fit on a smaller card or finishes so much sooner that the total is lower, and never because of its hourly figure.
Runcrate is our pick for buying one sensibly: by the second while you measure, then prepaid for 1 to 12 months at a discount once the hours are known.
Cost per job, not per hour
The hourly figure of an H200 or a B200 is several times that of a consumer card. The bill for a job is the figure multiplied by the hours the job takes. A card that costs four times as much and finishes in a fifth of the time is the cheaper card.
So the comparison is always a measurement. Run a slice of the job on each, and divide each bill by the work done.
What the two cards are
The H200 is a Hopper datacenter GPU with 141 GB of memory. It is used where a model or a batch is limited by memory.
The B200 is the Blackwell datacenter GPU, a generation newer, built for training and high-throughput inference. It is listed above the H200 in price.
What they cost today
On 12 Nov 2026, 15 providers listed an on-demand H200 price. The lowest was $3.49 per GPU-hour and the median $5.40. Runcrate can start one now from $4.67 per GPU-hour.
On the same day the lowest listed B200 price was $3.50 per GPU-hour and the median $8.34.
The traps at this size
The smallest rental is the first. Many H200 and B200 figures are per GPU inside a machine of eight. On 12 Nov 2026, the lowest listed hourly total for an 8-GPU B200 machine was $28.00.
Stock is the second. These cards are more often sold out, gated behind sales or held back by quota than the consumer cards. Read the label before the price.
Idle time is the third, and here it is expensive. An idle hour on a datacenter card costs several busy hours on a small one.
When to choose one
Choose an H200 when one card must hold more than 96 GB. Choose a B200 when the job is long, keeps the card busy and gains from the newer generation. Choose neither when the weights, with room on top for context and activations, stay inside the 96 GB of an RTX Pro 6000.
Where Runcrate fits
Runcrate bills on-demand GPUs by the second on most models and never bills booting, so a measurement costs minutes. When the hours are known, you can prepay a machine for 1 to 12 months at a discount.
Questions and answers
Sources for this entry
- Runcrate: Runcrate’s official site: what the platform is, how it bills and the range of GPUs it offers. Read 12 Nov 2026.
- Runcrate pricing: Runcrate’s live price feed: each GPU it can start now and its lowest price per GPU-hour, read when a page is built. Read 12 Nov 2026.
- Provider pricing pages: The published on-demand list price of every provider named in the price tables, with the machine size and billing unit it applies to. Read 12 Nov 2026.
- Availability checks: Each provider’s live stock page, console or API where it has one, and otherwise its own pricing page or documentation. Read 12 Nov 2026.
- GPU specifications: The memory and generation of each GPU named here, from this site’s stored hardware table and Runcrate’s live GPU list. Read 12 Nov 2026.
Checklists are original editorial guidance. Published prices do not establish availability.
Continue this reading path
Cite this entry
Cheap Cloud GPU. “When an H200 or B200 is the cheaper GPU.” Content reviewed 12 Nov 2026. https://cheapcloudgpu.com/topics/when-an-h200-or-b200-is-cheaper