2 min read
Add as a preferred source on Google

Nvidia H200 Rental Index Rises 20% in a Month to $6.20

The index settled at 4 p.m. ET (20:00 UTC) Oct. 24, up 8.01% from the previous day and 20.16% over one month.

A graphite accelerator module rests on a sculpted rising platform / TokenPost.ai
A graphite accelerator module rests on a sculpted rising platform / TokenPost.ai

A transaction-based index for Nvidia H200 rentals rose 20.16% over one month, settling at $6.20 per GPU-hour on Oct. 24. The increase gives a reading of completed on-demand rentals, a distinct measure from a provider’s posted rate.

At its 4 p.m. ET (20:00 UTC) settlement, the index was up 8.01% from the previous day. It calculates a volume-weighted average across a rolling one-hour window and adjusts for outliers. Reserved and long-term contracts are excluded.

The index covers one part of the rental market, so its reading does not represent what every customer pays for H200 capacity. Nebius lists its H200 on-demand rate at $5.40 per GPU-hour, effective Oct. 1, up from $4.50. The company also offers preemptible capacity starting at $0.79 per GPU-hour. Its spot prices can change as often as every 15 minutes with capacity and demand.

The Nebius rate is one provider’s listed price, while the index reflects completed rentals. The two figures are not directly interchangeable or a single market-wide rate.

The H200 has 141 GB of HBM3e memory and 4.8 terabytes per second of bandwidth per GPU. The H100 has 80 GB of HBM3 memory and 3.35 terabytes per second of bandwidth. Those specifications distinguish the two GPUs, but they do not explain the rental-price increase.

TokenPost’s earlier H100 rental benchmark coverage also examined differences between completed-rental readings and advertised prices. The measures reflect different inputs and should be read separately.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…