Accelerated (GPU) instance profiles - Gen 4

The Gen 4 accelerated (GPU) family of profiles provides on-demand, cost-effective access to the latest generation accelerators for IBM Cloud® Virtual Servers for Virtual Private Cloud (VPC). These accelerators help to reduce the processing time that is required for compute intensive workloads such as AI, machine learning, inferencing, and more.

NVIDIA HGX B300 accelerated virtual server profiles are available for select customers. Create a support case if you are interested in purchasing and using this offering.

NVIDIA B300 instance profile

The Blackwell-based Accelerated virtual server profiles are built atop the NVIDIA B300 accelerator. These accelerators are tuned for AI workloads, including inferencing, fine-tuning, and large-scale training. The solution is paired with the 6th Generation Intel Xeon Scalable processors.

Operating systems

  • Linux

Processor generation

  • Intel® Xeon® 6776P Processor

Accelerator

  • NVIDIA B300 SXM6 (288 GB) based on the Blackwell Ultra chip

Availability

Status: Select Availability

The following table lists the available regions and universal zones for the NVIDIA B300 accelerated virtual server profiles.

Supported regions and zones
Region Universal zone
Washington DC (us-east) us-east-wdc07-a

For more information about regions and universal zones, see Regions. You can review the assigned zone mapping for an account on the VPC Infrastructure Overview page in the Endpoint section. The zone mapping shows how the zone corresponds to the universal zone name that represents the physical location.

Capabilities

  • Core type: Dedicated
  • Dedicated host: No
  • Hyperthreading: Yes (SMT-2)
  • Secure boot: No
  • Confidential computing: No
  • Live migration: No
  • Instance storage: Yes
  • NVLink: Yes (1.8 TBps)
  • NVIDIA GPUDirect Capable: Yes
  • NIC capabilities:
    • Max single NIC throughput: up to 200 Gbps VPC traffic and 32 Gbps external
    • Bandwidth Pooling: Yes
  • Volume bandwidth allocation method: pooled by default; it can be updated to weighted.

VM configuration

  • Hardware type: q35
  • Cloud networking: virtio
  • Block boot volume: virtio
  • Block data volumes: virtio
  • Instance storage: NVMe

Instance profiles

The following table lists the NVIDIA B300 accelerated virtual server profile.

Accelerated NVIDIA B300 profile options
Instance profile vCPU / Cores / NUMA Memory (GiB) Bandwidth cap (Gbps) Accelerators Instance storage (GB)
gx4d-232x3840x8b300 232 / 116 / 2 3840 200 8x NVIDIA B300 (288 GB) 4 x 7.68 TB

This large profile likely requires that you open a support ticket to request a quota increase. Review your quota levels, and determine whether the account that's provisioning the resource requires a change to the quotas. This server uses vCPU, RAM, instance storage, and GPU quotas.

Limits

An instance has a limit for the number of volumes and virtual network interfaces that can be attached. This limit is based on the size of the instance.

Accelerated NVIDIA B300 limits for vCPU, maximum volumes, and maximum network interfaces
Number of vCPUs Max volumes Max vNICs
49+ 12 15

Boot volume profiles

In the current release of Block Storage for VPC offering, only first-generation volumes from the tiered and custom volume profile families can be used as boot volumes for the NVIDIA B300 instances.