SP7K-NVMe G5 Series · PCIe Storage Architecture
Your data, held as memory
The SP7K-NVMe G5 holds your data as memory on the virtual memory map, so any model can train or infer against it in place — no export, no second copy. Built on Virtual Storage Architecture and operating within the OmniBUS® matrix, it's the storage and data fabric foundation of the AI Habitat: the applications you run today and the AI you run tomorrow draw on the same data.
A data platform for the Intelligence Economy
Storage that does more than store — consistent under saturation, continuous through growth.
Performance uncertainty
Conventional arrays degrade under saturation, force redesigns to scale, and separate memory, caching, replication, and protection into layers that create variability under load.
One operating model
A hardware-level interconnect, virtualized storage memory, dynamic protection policies, and unified block/file management — positioned as a strategic asset, not a commodity storage platform.
Architecture overview
NVMe, SAS, SATA, and Fibre Channel — consolidated into a single virtual storage memory layer.
In place of Ethernet or InfiniBand for internal operations, the system uses PCIe direct memory and I/O transactions to cut protocol overhead and synchronization latency — preserving low latency and high parallelism under heavy load. SAN and NAS services share one namespace and one management framework.
Predictable under AI load
The decision point isn't a benchmark number. It's whether data stays reachable as AI and production workloads share it.
Stable sub-millisecond I/O under full-load conditions, via direct peer-to-peer clustering that reduces synchronization overhead — so inference and training run against live data without stealing headroom from the production workload beside them.
Deterministic I/O avoids the latency spikes and cache-thrashing common in conventional platforms during high saturation, which is what keeps data-in-place AI dependable when many models and applications reach the same volumes at once.
Capacity & growth
From a single 2U appliance to an 8 exabyte matrix — without a second architecture.
Scale-up begins at 2.94 PB, expands to 7.86 PB, and extends to 8 EB through OmniBUS®-based clustering — the same operating model from first deployment to hyperscale.
Key capabilities
A general-purpose platform built to participate cleanly in existing operations.
Business benefits
Five reasons this shows up as a leadership decision, not a procurement line item.
Performance consistency
Avoids the latency spikes and cache-thrashing common in conventional storage at high saturation.
Infrastructure consolidation
SAN and NAS in one system, reducing storage silos and administrative complexity across teams.
Elastic scale
Grows from a single 2U configuration to large clustered environments without architectural change.
Availability & resilience
Multi-parity protection, hot-swappable components, and high-availability clustering for continuous operation.
Broad compatibility
Support for major protocols and operating environments across backup, virtualization, and containers.
Ideal use cases
Where storage performance is a business enabler, not a background utility.
Model specifications
SP7K-NVMe 200, 400 and 600 — feature, protocol and capacity matrix, from the 2026 specification sheet.
| Feature | SP7K-NVMe 200 | SP7K-NVMe 400 | SP7K-NVMe 600 |
|---|---|---|---|
| Host Ports | IB 100/200/400 Gb/s, 4–8 ports. Ethernet 10/25/40/100/200/400/800 GbE/s, 4–8 ports. FC 16/32 Gb/s, 4–8 ports. OmniBUS® PCIe 5.0, 1,025 Gb/s. | IB 100/200/400 Gb/s, 8–16 ports. Ethernet 10/25/40/100/200/400/800 GbE/s, 8–16 ports. FC 16/32 Gb/s, 8–16 ports. OmniBUS® PCIe 5.0, 1,025 Gb/s. | IB 100/200/400 Gb/s, 8–16 ports. Ethernet 10/25/40/100/200/400/800 GbE/s, 16–32 ports. FC 16/32 Gb/s, 16–32 ports. OmniBUS® PCIe 5.0, 1,025 Gb/s. |
| Drives | 24 NVMe + 240 SAS/SATA | 60 NVMe + 240 SAS/SATA | 120 NVMe + 480 SAS/SATA |
| Encryption | AES-256, TLS 1.2, TLS 1.3, IPSec | ||
| Mgmt Console | Web HTTPS unified SAN/NAS | ||
| 10GbE Ports | 4 Base-T | 4 Base-T | 8 Base-T |
| iSCSI | Bonding supported | ||
| FCP | Up to 32 Gb/s | ||
| IB SRP | Up to 400 Gb/s | ||
| IPoIB | Supported | ||
| AFP / WebDav / FTP / Telnet | Supported | ||
| Cisco / UNH / Microsoft initiators | Supported | ||
| Network Mgmt | Multiple portals, portal groups, multiple targets per port, initiator access control, TOE NIC support, LAG, syslog. | ||
| HBA Support | Intel®, Mellanox® (boot), Qlogic® | ||
| RAID | 0, 1, 3, 5, 6, 50, 60, 550, 560(3P) | ||
| CIFS / SMBv3 | NAS file sharing | ||
| NFSv4.1 / S3 | NAS file sharing | ||
| iSER | Enhances iSCSI performance | ||
| NVMeoF | Supported | ||
| NDMP | Backup and recovery | ||
| RoCE | Low-latency RDMA | ||
| LDAP / AD / CHAP | User authentication | ||
| Content Adaptation | Content inspection | ||
| iSCSI RFC 3720 | Supported | ||
| Applications | Bacula, Commvault, Veeam, Zerto, VMware, Hyper-V, Proxmox, KVM, Hanna, Vaultsafe, MinIO | ||
| Storage Virt. | Full, all protocols | ||
| Strip Size | 4K, 8K, 16K, 32K, 64K, 128K | ||
| PCIe 5.0 Switched | Yes, 1 Tb/s | ||
| PCIe 5.0 Direct | Yes, 1 Tb/s | ||
| Expansion Unit | Yes, 1 TB/s w/OmniBUS® | ||
| In-Storage Memory | Yes, 1 TB/s w/OmniBUS® | ||
| Write Mode | Write pass-through and write back | ||
| Cluster Backbone | Yes, 1 TB/s w/OmniBUS® | ||
| PCIe Backward Compat. | Yes, 4.0 / 3.0 / 2.0 | ||
| Availability | 99.99%–99.999999% | ||
| Storage Virtualization | Third-party and physical storage, independent of type | ||
| Discovery / iSNS | Supported | ||
Hardware specifications
Head-unit platform, capacity envelope, data services and environmental limits.
- Processors
- 2 or 4 Intel Xeon 64-bit, 8–112 cores each
- Cache
- 64GB to 4TB volatile + 8TB persistent
- Power Supply
- Head: 1,280W Titanium redundant; JBOD: 1,000W Titanium
- Cooling
- 5 hot-swap fans
- Dimensions
- 17.2" W × 3.5" H × 27.76" D (437 × 89 × 705 mm)
- Weight
- Net: 37 lbs; Gross: 63 lbs
- Environmental
- Operating: 10–35°C, 8–90% RH; Non-op: -40–60°C, 5–95% RH
- OS Compatibility
- Windows 10/11/Server 2019/2022, Red Hat, CentOS, Rocky, SuSE, Solaris, Oracle Linux, VMware
- HDD Sizes
- 18TB, 20TB, 22TB, 24TB, 30TB
- NVMe/SSD Sizes
- 3.84TB, 7.68TB, 15.36TB, 30.72TB, 61.44TB, 122.88TB
- Global Hot Spare
- Minimum 1 per drive type, up to 256
- OS Certified
- Windows 11/10, Windows Server 2019/2022, Red Hat, CentOS, Rocky, SuSE, Solaris, Oracle Linux, VMware
- Compatible
- Classic HCP Tier 2 Storage, HCP-CS, Veeam Backup & Replication, Commvault Backup, Bacula Backup, Hanna controller in-memory volumes w/OmniBUS®
- Controllers
- Single, Dual Active/Active, 3–1024 cluster mode
- Options
- Immutable snapshots up to 1024 per volume, thin provisioning, encryption, compression/dedupe, replication, iSCSI port bonding, perpetual license
- Backplane
- End-to-end PCIe 5.0 with NVMe Gen 5.0 / SAS/SATA support, SES-II, SGPIO
- HDD Bays
- Interface NVMe options, SAS 12Gb/s / SATA3 6Gb/s
- NVMe Bays
- 24 on head unit, PCIe 5.0
- Capacity
- NVMe: 15.6 PB with 122TB drives; SAS/SATA: 13.2 PB
- HDD Capacity
- 13.2 PB with 30TB SAS/SATA drives
- JBOD Expansion (NVMe)
- Up to 2 JBODs (36 NVMe Gen 5 drives each)
- JBOD Expansion (HDD)
- Up to 10 JBOD units (44 HDD × 30TB each)
- Write Mode
- Write Back / Write Through
- Copy / Mirror / Replication
- Local, remote, intra, cross-host, reflective, replicated, duplication, volume copy
- Monitoring
- Throughput, IOPS, CPU, idle time, channel stats, volume history, logs, client connections, interface status, drive status
- Alerts
- Mail, LED, audio; triggers for HDD failure, fan failure, PSU failure, remote mirror break
- SW License
- Perpetual license for firmware and all system software
- Hot-swap PSU
- 1,280W Titanium-level redundant; 1,000W Titanium-level redundant for 32-drive chassis
- Power Source
- AC 110V~240V full range, 50–60Hz, 13A–6.5A max, C13 power cords
- Operating Temp.
- 10°C~35°C (50°F~95°F)
- Non-operating Temp.
- -40°C~60°C (-40°F~140°F)
- Operating Humidity
- 8%~90% non-condensing
- Non-operating Humidity
- 5%~95% non-condensing
Not just speed. Consistency — at whatever scale the business becomes.
Predictable latency, scalable growth, unified operations, and resilience designed for mission-critical use — a candidate for consolidation, modernization, and long-term architectural planning.
SP7K-NVMe G5 · Virtual Storage Architecture · OmniBUS® PCIe Fabric
Go deeper
Explore everything on Memory Platform
Explore all platforms