KAYTUS Launches Commercial CXL 2.0 Server, KR2280V3
December 29 2024 - 8:02PM
Business Wire
KAYTUS, a leading provider of IT infrastructure, launched the
KR2280V3, the commercially available server compliant with the CXL
2.0 standard. Housed in a 2U chassis, the KR2280V3 supports front-
or mid-mounted CXL 2.0 E3. S modules, offering up to 10 front and
18 rear configuration options.
The KR2280V3 accommodates up to 45 small 2.5-inch drives, 22
large 3.5-inch drives, or 48 of the latest 1TB E3.S SSD modules. It
supports AIC CXL memory expansion devices with up to 24 DDR5 memory
modules, boosting single-machine memory capacity by 75%.
Additionally, 16 8,000MT/s MRDIMM modules enhance memory bandwidth
by 25% compared to RDIMMs, enabling superior computational
performance and scalability.
Benchmark tests indicate that the KR2280V3, leveraging CXL
technology, excels in AI and data-driven workloads. Under
sub-100-nanosecond latency conditions, the system's memory capacity
and bandwidth double, resulting in a 76% improvement in BERT model
inference performance, a 94% increase in Redis in-memory database
throughput, and a 24% enhancement in vector data processing
scenarios compared to conventional local memory configurations.
With its CXL advanced memory expansion and cutting-edge design, the
KR2280V3 sets a new benchmark for AI and data-centric
computing.
CXL Technology: Breaking the ‘Memory Wall’
Rapid advancements in cloud computing, artificial intelligence,
and high-performance computing have fuelled the evolution of
processor capabilities and increased core counts, driving memory
systems toward greater capacity and higher bandwidth. In 2017, the
average memory per processing core ranged from 3-4GB; by 2021, this
figure rose to 8-10GB.
Although overall system memory bandwidth has grown, per-core
bandwidth improvements remain constrained due to the increasing
number of CPU cores. Concurrently, the exponential growth of AI
model parameters has outpaced memory performance advancements,
creating a widening performance gap.
CXL technology addresses this challenge by enabling shared,
coherent memory access between CPUs and accelerators or memory
expansion devices, supporting rack-level memory decoupling. It
effectively boosts memory bandwidth, expands memory capacity and
enables memory pooling for shared access, enhancing overall system
efficiency.
KAYTUS: Pioneers CXL Technology Deployment
KAYTUS is always committed to continuous innovation and
performance optimization. Its V3 server series supports CXL
technology on both Intel and AMD platforms, enabling high-speed
interconnects and memory pooling. Performance testing of the
KR2280V3 with CXL memory demonstrates significant improvements:
memory capacity-sensitive workloads achieve higher CPU utilization
and concurrency without substantial performance loss, while
bandwidth-sensitive tasks benefit from optimized memory tiering for
enhanced throughput.
Performance testing indicates that in AI scenarios, integrating
local and CXL memory significantly boosts bandwidth and application
throughput, accelerating BERT inference speeds by 76%. For
in-memory database applications, CXL memory extends database
capacity, increasing throughput by 94% compared to traditional
non-CXL architectures and by 116% when optimized memory tiering
strategies are applied. For vector data applications, balancing
local and CXL memory usage improves performance by 24% over
standard local memory configurations. This advancement enhances
computing efficiency across AI inference, deep learning, and
memory-intensive tasks such as genome mapping.
CXL BOX: Unlocking the Memory Resource Pooling
Efficiency
As data volumes grow and demand for computational capacity
intensifies, memory bandwidth and capacity requirements continue to
rise. In response, KAYTUS goes beyond integrating CXL into its
products by offering the CXL BOX memory resource pooling solution,
designed to meet the needs of users requiring high memory
capacity
This solution integrates CXL switching units with JBOM memory
pooling to create a multi-host, fine-grained (64GB/module) shared
memory resource pool. Specifically designed for cloud computing and
distributed storage scenarios, it enhances memory resource
utilization while reducing idle memory caused by workload
fluctuations.
About KAYTUS
KAYTUS is a leading provider of IT infrastructure solutions,
delivering a diverse range of innovative, open, and eco-friendly
products for cloud, AI, edge computing, and other emerging
applications. With a customer-centric approach, KAYTUS is agile and
responsive to user needs through its adaptable business model.
Discover more at KAYTUS.com
View source
version on businesswire.com: https://www.businesswire.com/news/home/20241230276424/en/
Media contact media@kaytus.com