HPC Inventory for Short-Term Training Bursts

AI researchers and organizations often need to utilize HPC resources for training bursts that may last only 1 to 2 weeks. Unfortunately, given the height of demand and the scarcity of cutting-edge GPU chips, many cloud providers are requiring at least 6-month reservations alongside large upfront payments. This results in many organizations either being overprovisioned and operating at a significant loss, or outright unable to perform the desired work.

To address this, VALDI actively curates a catalog of inventory that is available for short-term training bursts. You can check this page regularly to see an up-to-date list of our short-term inventory. If you're interested in reserving any of the servers below, please email contact@valdi.ai.

Upcoming and Available Inventory

Last updated May 15, 2024

H100 HGX SXM5 IB (Norway)
Quantity available	Up to 128 in 8x increments
Interconnect	InfiniBand
Location	Norway
Deployment type	Virtualized
Date available	May 20, 2024
Min reservation	1 week
Max reservation	4 weeks

H100 SXM5 NVLink (United States)
Quantity available	Up to 128 in 8x increments
Interconnect	Ethernet
Location	United States
Deployment type	Virtualized
Date available	Immediately
Min reservation	1 week
Max reservation	2 weeks

H100 HGX SXM5 IB (Sweden)
Quantity available	Up to 512 in 8x increments
Interconnect	InfiniBand
Location	Sweden
Deployment type	Virtualized
Date available	Immediately
Min reservation	1 week
Max reservation	1 week (can extend in weekly increments)