AMD Validated Reference Architecture: Data Storage for GPU-Accelerated AI

Learn More

Velocity • Durability

Book a Briefing
Contact Sales
Book a Briefing
Book A Demo
Contact Sales
Contact Sales
AI SOLUTIONS / OPEN HARDWARE

Standard servers. Standard drives. No hardware mark-up.

HYDRA is software. It runs on commodity servers from AIC, Dell, Supermicro, or any other roadmap-certified platform, with single-port NVMe and SATA HDDs from qualified vendors. Nothing proprietary anywhere in the stack, and you buy the metal at your own buying power.

✓ Zero hardware mark-up ✓ Multi-vendor by design ✓ No proprietary parts
ANY CERTIFIED PARTS · ONE PLATFORM
SERVERS SupermicroDellAIC NVME FLASH PhisonKioxiaSolidigm HDD CAPACITY Western DigitalSeagateWestern Digital VDURA True parallel file system Software-defined Shared-nothing SCALE Performance SCALE Capacity independently, online
Buy what ships, from whoever can ship it. The platform does not change.
Zero
HARDWARE MARK-UP
VDURA sells software and support. Servers and media come through your own channel.
Multi-platform
CERTIFIED TODAY
AIC, Dell, and Supermicro, with more being certified. Same software on all of them.
~10%
OF CLUSTER SPEND
Where storage lands with mixed flash and HDD, against the 20 to 30% all-flash forces.
Days
TIME TO CAPACITY
Software-defined deployment lets capacity land in any colocation in days rather than quarters.
THE LOCK-IN PROBLEM

Proprietary hardware is a decision you keep paying for.

Appliance storage bundles the software you want with metal you did not choose, at a price you cannot see. The cost is not only the mark-up. It is the lead times, the coupled lifecycles, and the fact that every expansion has exactly one supplier.

THE MARK-UP
You pay a storage premium on general-purpose parts
An appliance bundles the software you want with servers and drives you did not select, priced as one opaque number. The components underneath are the same ones on the open market.
VDURA: Open hardware at your buying power, with zero mark-up applied by VDURA.
COUPLED LIFECYCLES
The software refresh waits on the hardware refresh
When features arrive only with the next generation of controller, your upgrade cadence is set by a vendor roadmap rather than by what your cluster needs.
VDURA: Software upgrades independently of the hardware lifecycle. New capability lands on the servers you already own, and refreshes happen when the economics say so. Features arrive in software, not new boxes.
WHY IT IS POSSIBLE

Commodity hardware, by design.

Most systems need special hardware because they push availability down into the box. HYDRA is shared-nothing, so availability is a property of the cluster. That single decision is what makes standard parts sufficient.

SHARED NOTHING
Availability comes from the cluster
Every node owns its own data and metadata, so there is no controller pair to buy in matched, failover-capable sets.
SOFTWARE EC
Protection runs in software
Client-side dual-parity (n+2) erasure coding runs across nodes. Firmware-based RAID controllers and their write-hole edge cases are simply absent.
SINGLE PORT
Standard drives are sufficient
Failover is not a drive-level concern, so single-port NVMe works — removing one of the largest hidden premiums in enterprise storage.
MIXED FLEET
Any media, one namespace
NVMe flash and SATA HDD live in a single global namespace, so you ride both cost curves independently as prices move. Buy only the resource that is short.
SUPPLY-CHAIN INSURANCE

Open hardware at your buying power.

VDURA sells software and support. You source servers and media through your own channel, on your own paper, at your own negotiated rates. When the component market moves, you move with it instead of waiting for a refreshed appliance SKU.

✓ Servers from AIC, Dell, or Supermicro, or any roadmap-certified platform, on your own paper
✓ NVMe from Phison, Kioxia, or Solidigm and HDDs from WD or Seagate through your existing channel
✓ Quarterly TLC, QLC, and HDD pricing from OEM sources, with zero mark-up applied by VDURA
✓ Mixed vendors within one cluster, so a single supply constraint does not stop an expansion
✓ A SaaS business model: you buy software and support, not a bundled box
✓ Buy what ships, from whoever can ship it

In a constrained supply chain, flexibility is resilience: source what is available, scale what you need.

CERTIFIED COMPONENTS
Server platforms AIC, Dell, Supermicro
Node form factor 1U/2U, minimum config 6U
Flash Single-port U.2 NVMe, TLC or QLC
Capacity media SATA HDD, CMR, SMR, or HAMR
NVMe vendors Phison, Kioxia, Solidigm
HDD vendors Western Digital, Seagate
Networking vendors AMD, NVIDIA, Broadcom
Networking InfiniBand NDR/NDR200, 400/200/100 GbE
Scale 6 nodes to thousands
Proprietary components None
LIFECYCLE AND SUPPORT

Open hardware without a support gap.

10-Year Drive Replacement
VDURACare Premier includes no-cost drive replacement for 10 years, on commodity media you bought through your own channel.
One Contract, 24×7
Hardware, software, and support on one contract, with 24×7 expert response and OEM support for compute nodes.
Independent Upgrades
Software moves forward on its own cadence, decoupled from the hardware refresh cycle.
Custom Terms
Customized support offerings are available where a deployment has specific requirements.
WHAT THIS MEANS FOR THE BUSINESS

Storage stops being the line item that crowds out GPUs.

Every Dollar Saved Buys A GPU
Storage lands closer to 10% of cluster spend rather than 20 to 30%. The difference funds accelerators on the floor.
Procurement Velocity
Standard servers are available now, from multiple suppliers. Capacity lands in a colocation in days rather than quarters.
Supply Chain Resilience
Multiple certified vendors per component class means one allocation problem does not become your outage.

Model what open hardware does to your three-year storage cost.