VDURA: Velocity meets durability

VDURA GPU Storage Calculator

Feed every GPU cycle with precision storage tiers.

Size storage against AMD Instinct GPU Accelerator reference specifications. Enter your GPU footprint to see how many VDURA flash nodes are required to meet or exceed published production and high-resolution targets.

Interactive sizing

Model flash storage nodes for your GPU cluster.

Adjust the GPU count to show how many VDURA flash storage nodes exceed AMD Instinct GPU Accelerator production and high-resolution specifications.

Active GPUs

Typical VDURA scalable units are planned in blocks of 256 GPUs.

Scalable units --

Production targets

Read -- GB/s

Write -- GB/s

High-Res targets

Read -- GB/s

Write -- GB/s

GPU reference - production

AMD Instinct production spec

Spec read throughput: -- GB/s
Spec write throughput: -- GB/s
VDURA flash nodes required: -- nodes
Delivered read throughput: -- GB/s
Delivered write throughput: -- GB/s

Based on AMD Instinct GPU Accelerator production guidance and partner reference deployments.

GPU reference - high resolution

AMD Instinct high-resolution spec

Spec read throughput: -- GB/s
Spec write throughput: -- GB/s
VDURA flash nodes required: -- nodes
Delivered read throughput: -- GB/s
Delivered write throughput: -- GB/s

Mirrors AMD Instinct GPU Accelerator high-intensity workload recommendations.

Exceeds reference specs

VDURA exceeds AMD Instinct Accelerator throughput targets in a compact footprint.

Compare published reference requirements against VDURA’s delivered throughput for a 4 scalable-unit (1,024 GPU) deployment. VDURA provides additional read and write headroom while using fewer 1U flash nodes than hyperscaler designs.

GPU reference design targets

Baseline throughput required to keep 1,024 GPUs operational.

Production read target (AMD Instinct) --

Production write target (AMD Instinct) --

High-Res read target (AMD Instinct) --

High-Res write target (AMD Instinct) --

Targets sourced from AMD Instinct GPU Accelerator production and high-resolution reference designs.

VDURA delivered headroom

Policy-driven flash tiers exceed spec requirements with fewer nodes.

VDURA flash nodes (Production) -- --

VDURA flash nodes (High-Res) -- --

Headroom vs spec (Production) --

Headroom vs spec (High-Res) --

VDURA delivered read (Production) --

VDURA delivered write (Production) --

VDURA delivered read (High-Res) --

VDURA delivered write (High-Res) --

Each VDURA flash node is a compact 1U appliance. A 69-node high-res configuration occupies just 69U, leaving ample rack space for compute and networking. Specification targets sourced from AMD Instinct GPU Accelerator storage reference guidance.

Watch & learn

Hear how VDURA and AMD keep GPUs fully utilized.

Explore a joint session with AMD and an in-depth look at the data pipeline pressures that VDURA solves for modern AI factories.

Bridging the Gap of AI: VDURA and AMD

AI Pipeline Pressure Presentation

Next steps

Ready for a deep dive on your GPU roadmap?

Share the scenario JSON or bring your telemetry to a working session. VDURA architects will validate node counts, director design, and data services tailored to your workload mix.