Kioxia AiSAQ Software-Optimized RDMA SSD Solution

Brand: Kioxia | Category: Enterprise SSDs

SKU: KIOX-KIOXIAAISAQRDMA | Part #: KIOXIA-AISAQ-RDMA | MPN: KIOXIA-AISAQ-RDMA

Contact for Pricing — Request a Quote

Request a Quote Contact Us

About the Kioxia AiSAQ Software-Optimized RDMA SSD Solution

The Kioxia AiSAQ Software-Optimized RDMA SSD Solution from Kioxia is enterprise-grade Enterprise SSDs hardware built for data centers. Kioxia AiSAQ offloads vector database indices from DRAM to NVMe SSDs via RDMA, enabling petabyte-scale AI inference and RAG pipelines at a fraction of the memory footprint. Available in brand new condition. Kioxia enterprise ssds products are trusted by enterprises and data centers worldwide. Available from Omnixon Global. Contact our sales team to request a quote.

Technical Specifications

BrandKioxia
CategoryEnterprise SSDs
SKUKIOX-KIOXIAAISAQRDMA
Part NumberKIOXIA-AISAQ-RDMA
ConditionNew
Solution NameAiSAQ (AI Storage Acceleration with Queue)
Launch DateQ1 2025
Primary TechnologySoftware-Optimized RDMA SSD Vector Offload
Target WorkloadsVector database ANN search, RAG inference, semantic similarity search, AI recommendation
Supported Vector Index FormatsDiskANN, HNSW (disk-resident variants), IVF-Flat with SSD paging
RDMA Transport CompatibilityRoCE v2, InfiniBand (with verbs API), iWARP
Host Interface (Underlying SSDs)PCIe Gen 4 x4 / PCIe Gen 5 x4, NVMe 1.4 / NVMe 2.0
DRAM Reduction vs. In-Memory IndexUp to 90% reduction for billion-scale vector datasets
Supported Vector DimensionalityUp to 1536 dimensions (validated); extensible via configuration
Query Latency (P99, 1B vectors, SSD-resident)< 10 ms with prefetch optimization enabled
QPS Throughput ScalingLinear scaling with additional NVMe SSD nodes in disaggregated topology
I/O Pattern OptimizationCustom random small-block read scheduler with ANN-graph-aware prefetching
Operating System SupportLinux kernel 5.15+; RHEL 8/9, Ubuntu 20.04/22.04 validated
Software IntegrationPython SDK, C++ API, gRPC service interface; compatible with LangChain and LlamaIndex RAG frameworks
Underlying SSD Form Factors SupportedU.2 (2.5-inch), E1.S, E3.S, EDSFF E1.L
Maximum Index Scale (Validated)Up to 100 billion vectors per storage cluster
Concurrent Client ConnectionsUp to 1024 simultaneous RDMA sessions per storage node
Power Profile (per storage node, typical)Varies by SSD density; baseline NVMe SSD active draw 5–25 W per drive
Security FeaturesNVMe TCG Opal 2.0 self-encrypting drive support; in-flight data encryption via RDMA transport layer
Deployment ModelSoftware package deployable on standard x86 servers with RDMA-capable NICs and Kioxia NVMe SSDs
Management InterfaceREST API dashboard, CLI toolset, Prometheus-compatible metrics exporter