Hyperstack local NVMe storage refers to NVMe storage located within the GPU node. This storage type offers the lowest possible latency because it is placed inside the system chassis, eliminating the need for network-based communication.
Hyperstack local NVMe storage refers to NVMe storage located within the GPU node. This storage type offers the lowest possible latency because it is placed inside the system chassis, eliminating the need for network-based communication. However, it's important to note that in the event of a system or disk failure, any data stored on this disk will be lost when the customer's Virtual Machine (VM) is rescheduled to a new node. Therefore, it's essential to consider the specific workload demands and potential risks before utilizing this storage option for critical data.
Use cases
Local NVMe storage is most suitable for high-performance read-and-write tasks that do not demand strict data redundancy.
Presentation
Local NVMe is presented to the VM as a physical disk and will be used by the disk on which the operating system is located.
Features
NVMe (Non-Volatile Memory express) passthrough enables a virtual machine (VM) to access and utilize a physical NVMe storage device directly. This direct communication between the VM and NVMe storage offers high-speed, low-latency access, bypassing the hypervisor.
On your virtual machine, you'll see this storage device represented as /dev/nvme0n1
.
Available in CANADA-1
region for the 8x A100-80GB-PCIe w/ NVLink GPU flavor. NVMe passthrough is currently exclusively available for contracted customers. For inquiries, please reach out to our team at [email protected].
Capacity
Hypervisor Local NVMe capacity depends on the VM's GPU count and system type. Systems have fixed NVMe disks, allocated proportionally to GPU usage.
For example, a 1 x 2TB NVMe system with 8 GPUs allocates 1TB of NVMe storage for 4 GPUs and 2TB for 8 GPUs.
Please refer to the following page for details on hardware systems and their disk configurations:
Supported operating systems
Any operating system capable of supporting VirtIO-type devices, including Windows and Linux.
Datacenter availability
Available in all regions on all GPU-enabled systems.
Pricing
Customers are billed based on the virtual machine flavors used, with no additional costs.