Enterprise

OCI Compute BM.GPU.H100.8 Brings 8 NVIDIA H100 GPUs to the Cloud

Oracle Cloud Infrastructure (OCI) has launched a new bare metal compute shape that packages eight NVIDIA H100 GPUs, twin Intel 56-core Xeon CPUs, 2TB of DDR5, 16x NVMe SSDs, and 400Gb networking into an AI dream in the cloud. OCI Compute BM.GPU.H100.8 is generally available now.

Oracle Cloud Infrastructure (OCI) has launched a new bare metal compute shape that packages eight NVIDIA H100 GPUs, twin Intel 56-core Xeon CPUs, 2TB of DDR5, 16x NVMe SSDs, and 400Gb networking into an AI dream in the cloud. OCI Compute BM.GPU.H100.8 is generally available now.

The new OCI Compute BM.GPU.H100.8 shapes essentially leverage Oracle’s vast bare metal cloud expertise to deliver a cloud version of the highly dense 8-way H100 GPU servers often found on-prem. But where something like the Dell PowerEdge XE9680 might take a little time to deliver and hits as CAPEX, OCI Compute BM.GPU.H100.8 can be made more immediately available and follows the traditional cloud OPEX model.

OCI Compute BM.GPU.H100.8

Component Specifications
CPU 4th Gen Intel Xeon processors, 2x 56 Cores
GPU 8x NVIDIA H100 80GB Tensor Core
Memory 2TB DDR5
Local Storage 16x 3.84 TB NVMe
Cluster Networking 8x 400 Gb/sec

The OCI Compute BM.GPU.H100.8 specs more or less model the on-prem versions of these powerful GPU servers, albeit with twice as many SSDs (though low capacity) typically found on servers like this. Oracle has also indicated it plans to offer the new NVIDIA L40S GPUs in shapes that are coming in early 2024. The BM.L40S.4 shape will include four L40S GPUs with twin Intel Xeon 56-core CPUs, 1TB of memory, 15.36TB of NVMe storage and 400GB/s of network bandwidth.

While many organizations will make due just fine with a single one of these instances, OCI Supercluster allows them to scale from a single node up to 50,000 H100 GPUs. This option will be available later this year in the London and Chicago regions, with more to come in the future.

OCI AI Infrastructure

Engage with StorageReview

Newsletter | YouTube | Podcast iTunes/Spotify | Instagram | Twitter | TikTok | RSS Feed 

Brian Beeler

Brian is located in Cincinnati, Ohio and is the chief analyst and President of StorageReview.com.

Recent Posts

OpenNebula 7.0 “Phoenix” Ushers in a New Era for Sovereign, AI-Ready, and Edge-Enabled Cloud Infrastructure

OpenNebula has released version 7.0 “Phoenix,” marking a significant leap forward for organizations building sovereign, AI-ready, and edge-enabled cloud environments.…

2 days ago

NVIDIA Unveils Helix Parallelism Enabling 32x Faster AI Inference with Multi-Million Token Contexts

NVIDIA Helix Parallelism boosts real-time LLM performance on Blackwell GPUs, scaling multi-million-token AI with 32x efficiency gains. (more…)

2 days ago

Hypertec TRIDENT iGW610R-G6: Immersion-Born 1U GPU Server for High-Density AI and HPC

Hypertec TRIDENT iGW610R-G6, a 1U server, supports up to four full-height GPUs in a single-phase immersion environment. That's up to…

3 days ago

IBM Power11 Servers Launch with Enhanced AI and Security Features

IBM Power11 servers deliver unprecedented AI performance, hybrid-cloud flexibility, and robust resiliency, ensuring seamless, secure operations for enterprise workloads. (more…)

5 days ago

Dell Collaborates With CoreWeave to Ship the First NVIDIA GB300 NVL72

Dell and CoreWeave deliver the first NVIDIA GB300 NVL72 system, setting a new benchmark in AI performance and scalability for…

6 days ago

Dell Technologies Achieves NVIDIA Cloud Provider Program Certification for PowerScale

Dell PowerScale earns NVIDIA Cloud Provider program certification. (more…)

6 days ago