GRAID separates RAID computation from the path used to move application data. NVMe drives remain connected to the host PCIe fabric while SupremeRAID handles RAID computation and presents protected virtual NVMe devices to the operating system.
SupremeRAID provides protected block storage. Filesystems, shared storage services and application-level availability are designed separately.
GRAID NVMe storage is suited to applications that need high-performance block storage with drive-level protection. Broadberry selects the RAID policy, SSDs and supporting infrastructure around the application's actual I/O and recovery requirements.
| Workload | Typical I/O | Design Priority |
| Transactional Databases | Small random reads, durable writes and logs | Latency, write policy and recovery |
| AI & Analytics | Parallel reads, ingest and temporary output | Sustained bandwidth and CPU headroom |
| Virtualisation | Mixed random I/O from concurrent VMs | Queue depth, latency and failure behaviour |
| Media & Scientific Computing | Large streams, scratch files and burst writes | Sequential throughput and usable capacity |
Fast Local Storage for Data-Intensive Processing
AI preprocessing, analytics and data-processing workloads can combine parallel reads, ingest and temporary writes. GRAID provides protected NVMe block storage while offloading RAID computation from the host CPU.
SupremeRAID does not create an AI cluster filesystem. Shared datasets and multi-node access require a separate supported storage architecture.
The right RAID level depends on usable capacity, write behaviour, failure tolerance and recovery requirements. Broadberry selects the RAID configuration around the requirements of the workload.
NVMe performance depends on the complete system. Broadberry evaluates drives, PCIe topology, CPU resources and networking together rather than sizing around component specifications alone.
Before recommending a GRAID configuration, Broadberry evaluates the workload, RAID requirements, data path and recovery targets that determine the complete storage design.
Workload Assessment
Model block sizes, read/write
patterns, queue depth and
concurrency.
Capacity Planning
Calculate usable capacity,
RAID overhead, spares,
temporary data & growth.
RAID Design
Select RAID level and drive-
group layout around protection
and recovery requirements.
Infrastructure Design
Size NVMe, CPU, memory,
PCIe and network
together.
Validation
Test representative
workloads, degraded operation
and rebuild behaviour.
Lifecycle Support
Plan firmware, driver and
software updates, expansion
and ongoing support.
What is GRAID NVMe RAID?
GRAID SupremeRAID is a software-defined NVMe RAID platform that uses a GPU to perform RAID computation while NVMe drives remain connected to the host PCIe fabric.
How does GPU-assisted NVMe RAID work?
SupremeRAID uses a dedicated GPU for RAID computation and control. Storage data does not need to pass through card-attached storage ports, allowing NVMe devices to use the host PCIe fabric.
What does the GPU do in a GRAID server?
The GPU handles RAID computation and control. It is not an application accelerator and does not add GPU compute capacity to AI applications.
Which RAID levels does GRAID support?
SupremeRAID Ultra supports RAID 0, 1, 5, 6 and 10. The appropriate configuration depends on usable capacity, write behaviour and recovery requirements.
Does GRAID provide a filesystem or NAS?
No. SupremeRAID provides protected block devices. Filesystems, databases, hypervisors and network storage services run above that block-storage layer.
Is GRAID suitable for AI workloads?
GRAID can provide high-performance protected NVMe storage for AI data preparation, analytics and other processing. It does not itself provide a shared AI cluster filesystem.
Does GRAID replace backup?
No. RAID protects against specified drive failures. Independent backup is still required for deletion, corruption, ransomware and other recovery scenarios.
How should a GRAID NVMe system be sized?
Start with usable capacity, daily writes, application block sizes, queue depth, concurrency, RAID requirements and recovery targets. Then size NVMe, PCIe, CPU and network resources around the complete application data path.
Tell us about your application, usable-capacity target, I/O requirements and preferred RAID policy. Broadberry can configure the server platform, SupremeRAID, NVMe SSDs and networking as one qualified storage system.
Extensive Testing
Customization Service
Call Our US Sales Team Now