heder_bg.jpg

UnifabriX CXL™-based
Smart Memory Node™

Product1.png
Based on its novel RPU™ (Resource Processing Unit™), UnifabriX Smart Memory Node enables data center operators to unleash the speed, density, and scale of their infrastructure.

UnifabriX CXL™-based products achieve exceptional performance and elasticity in bare-metal and virtualized environments over a wide range of applications, including the most demanding HPC, AI, and Database workloads. Working with UnifabriX, organizations can gain greater insight and control over their costs and performance targets.

THE DRAM CHALLENGE

DRAM is expensive, badly utilized and not efficiently allocated

Only ~40% of
DRAM is used in production clusters

quotes

DRAM makes up 50% of the server cost for Azure

quotes

In Azure, up to 25% of memory is stranded

quotes
google logo
Microsoft logo
Microsoft logo

STRANDED MEMORY LEADS TO GROSS INEFFICIENCIES

CXL opens the door to true composable infrastructure, but the journey to performance is not trivial

diagram of Unused standard CPUs vs. CPUs utilizing a CXL chip

THE PERFORMANCE CHALLENGE

CPU sockets are memory-bandwidth-limited, and not CPU-cores-limited

Running memory intensive algorithms result in underutilization of the available CPU cores – the DRAM bandwidth is exhausted while cores remain idle, and performance doesn’t reach its potential

Performance plateus before utilizing all cores
bg.jpg

UnifabriX is the first company in the CXL arena to focus on performance

UnifabriX Smart Memory Node

Performance Acceleration for CXL-based Systems

Introducing UnifabriX Smart Memory Node

Holistic approach

Workload-aware

Memory acceleration techniques

Memory mapping

Holistic approach to realizing CXL’s potential – system-wide acceleration

Workload-aware and media-aware approach to datacenter optimization

Memory acceleration techniques provide efficient memory expansion for the most demanding performance hungry applications

Memory mapping, pooling, sharing, and optimizing at rack-scale

SYSTEM-WIDE ACCELERATION

SYSTEM-WIDE ACCELERATION diagram

Improving CPU utilization

Improving CPU utilization and increasing system-wide bandwidth

Super-scaling performance

Super-scaling performance, bandwidth, capacity, and elasticity

Programmable management

Programmable management

Plug-n-Play deployment

Plug-n-Play deployment: Standard Rack-mount installation

Works with any CXL

Works with any CXL-enabled OS/hypervisor

Seamless integration

Seamless integration with orchestration layers; Application/Workload independent with native user experience

PERFORMANCE UNLOCKED

UnifabriX Smart Memory Node increases CPU available memory bandwidth, reaching the full potential of all CPU cores

UnifabriX Smart memory node adds memory bandwidth, enabling utilization of all cores
High-performance

High-performance

- High-performance CXL Smart Memory Node.
- Rack-mount 2U/4U options.
- Serviceable EDSFF E3 media bays.

Scalable media capacity

Scalable media capacity

Scalable media capacity up to 128TB Flexible media types:
- DDR5/DDR4
- NVMe

High throughput

High throughput

- High throughput NVMe-oM™ (NVMe over Memory).
- Modular connectivity up to 20 CXL FE/BE ports.
- Node stacking via BE Fabric.
- High-bandwidth low-latency server connectivity.

Standard compliant

Standard compliant

Standard compliant: CXL 1.1, CXL 2.0, PCIe Gen5 CXL 3.0 ready

Server-grade RAS

Server-grade RAS

Server-grade RAS Multi-layer enterprise-class security

TECHNOLOGY ACCESS

UnifabriX CXL™-based
Smart Memory Expansion Card™

Get your hands on state-of-the-art CXL technology, powered by UnifabriX Resource
Processing Unit™ (RPU™). Start gaining more from your infrastructure today!

UnifabriX CXL™-based Smart Memory Expansion Card™

Contact us

Thanks for submitting!

UnifabriX Ltd.

Haifa, Israel