Business Wire

KAYTUS

21.5.2024 12:18:30 CEST | Business Wire | Press release

Share
KAYTUS Introduces KR1280V2 - a 1U Server with 32 E1.S SSDs, Boosting I/O Performance for LLM Training

KAYTUS, a leading IT infrastructure provider, introduced at ISC High Performance 2024 its high-density computing server KR1280V2. It supports 32 E1.S SSDs, achieving an impressive flash memory density of 512TB in a 1U form factor. Compared to the last generation, KR1280V2 adopts low latency storage and network design, doubling the high concurrency I/O performance for LLM training scenarios, and improving model training efficiency.

This 1U2S flagship server is powered by the latest Intel, AMD, or ARM processors. The KR1280V2 features an optimized cooling strategy to support CPUs with the highest TDP across platforms, without configuration restrictions. It also introduces a novel front I/O configuration, enhancing scalability with 8 PCIe 5.0 expansion cards and improving maintenance efficiency in data centers. The server’s drive tray type has been optimized to offer an innovative storage configuration of 12 front SFF drives, increasing storage density by 20%.

Supports 32 E1.S SSDs without performance degradation, delivering tens of millions IOPS

Powered by 5th Gen Intel® Xeon® processors and boasting remarkable computing density, the KR1280V2 supports up to 32 all-flash E1.S SSDs directly connected to CPUs within a 1U form factor. This configuration harnesses the full I/O rate of PCIe 5.0, maximizing throughput between CPUs and storage, achieving up to 160GB/s storage throughput. With load balancing across two OCP 3.0 cards, it doubles network bandwidth, catering to diverse data storage and processing needs while enhancing overall computing performance by 60%.

With ultimate flash memory density, the KR1280V2 effectively doubles LLMs' high concurrency performance

The KR1280V2 achieves tens of millions of IOPS performance in distributed storage scenarios, marking a remarkable 77% improvement. Deploying KR1280V2 servers for large-scale LLMs training ensures a twofold increase in high-concurrency I/O performance compared to the previous generation. As the number of GPUs and workload increases, the KR1280V2's latency curve becomes smoother and more predictable, demonstrating a notable 13% improvement in data latency also compared to the prior generation. These significant advancements enhance data preparation efficiency, model training, and data management while reducing wasted storage space. Consequently, it delivers robust storage capacity and performance for innovative LLM applications and better supports Java-based business scenarios requiring massive data, low latency, and high concurrency, such as in finance, cloud services, enterprises, and other data center applications.

Optimized cooling technologies and exceptional business reliability and sustainability

In terms of heat dissipation, the KR1280V2 supports multiple cooling methods, including EVAC heatsink air cooling, cold-plated liquid cooling, and single-phase immersion liquid cooling. It achieves an 8% to 12% power reduction through fan partition regulation, one-click BIOS optimization, and other techniques across 17 energy-efficient scenarios. Additionally, the KR1280V2 employs numerous advanced technologies to enhance stability and security. It introduces the exceptional MUPR2.0 early detection memory fault function, which utilizes a more comprehensive memory scrubbing technique to self-check and fix memory faults, achieving a broader range of error correction and a higher level of system protection. Consequently, the system's downtime rate is 69% lower than the industry average.

To learn more about the KAYTUS KR1280V2 high-density server, please visit the website.

About KAYTUS

KAYTUS is a leading provider of IT infrastructure solutions, delivering a diverse range of cutting-edge, 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

To view this piece of content from cts.businesswire.com, please give your consent at the top of this page.

View source version on businesswire.com: https://www.businesswire.com/news/home/20240521588996/en/

About Business Wire

Business Wire
Business Wire
101 California Street, 20th Floor
CA 94111 San Francisco

http://businesswire.com
DK

Subscribe to releases from Business Wire

Subscribe to all the latest releases from Business Wire by registering your e-mail address below. You can unsubscribe at any time.

Latest releases from Business Wire

Reply and IEO Launch Collaboration to Co-Develop and Train Domain-Specific Large Language Models for Oncology4.6.2026 10:10:00 CEST | Press release

Reply [EXM, STAR: REY] and the European Institute of Oncology (IEO) have launched a collaboration focused on the co-development and training of domain-specific Large Language Models for oncology. The initiative brings together Reply’s expertise in building frontier generative models grounded in enterprise knowledge with the IEO’s clinical expertise and data assets to develop models tailored to highly complex oncology settings. This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20260604488955/en/ Reply and the European Institute of Oncology (IEO) have launched a collaboration focused on the co-development and training of domain-specific Large Language Models for oncology. As a first step, the IEO’s clinical teams and Information Systems department are working with a multidisciplinary Reply team with deep expertise in healthcare and LLM specialisation to define and prioritise the use cases that will guide development, while also map

UK AI Minister Kanishka Narayan Joins The AI Summit London as 10th Anniversary Agenda Expands Across Enterprise AI, Innovation and Policy4.6.2026 10:00:00 CEST | Press release

London Tech Week's flagship AI event will feature 300 speakers, live AI demonstrations, startup pitch competitions, executive training and the National AI Awards The AI Summit London today revealed the full agenda for its highly anticipated 10th anniversary edition, delivering its most expansive and future-focused programme to date. Taking place 10–11 June at Tobacco Dock as the headline AI event of London Tech Week 2026, the Summit expects more than 5,000 attendees, 300 speakers, 100+ sponsors and exhibitors, and a special executive presentation from Kanishka Narayan, Minister for AI and Online Safety, Department for Science, Innovation and Technology, across two days dedicated to the commercial realities and transformative potential of artificial intelligence. The 10th anniversary has been designed to move the conversation beyond experimentation and into enterprise-wide execution. New in 2026, the programme introduces The AI Impact Arena, delivering focused briefings and live demos,

Corona Launches 2026 Beach 100 Guide, Invites The World To Explore The Outdoors This Summer4.6.2026 10:00:00 CEST | Press release

Ahead of World Oceans Day, the global beach authority returns with a list of new locations, over 300,000 Tripadvisor experiences, and dedicated beach grants as part of its “This is Living” platform Corona, the world’s most valuable beer brand1, today announced the return of its annual global beach guide, the 2026 Corona Beach 100, ahead of World Oceans Day on June 8. The highly anticipated second edition spotlights 100 breathtaking and varied ocean-front destinations around the world while celebrating the natural ecosystems and local communities that help bring them to life. This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20260604412899/en/ 2026 Corona Beach 100 - Legzira Beach, Morocco Connected to Corona’s “This Is Living” platform, which invites consumers to step outside and live life to the fullest this summer, the 2026 guide features 27 new beaches, hundreds of localized Tripadvisor travel experiences inspired by the world

Lantern Expands Its Platform for Solving the Data Trust Problem with Strategic Acquisition and Key Executive Hires4.6.2026 10:00:00 CEST | Press release

Lantern Acquires Freyda to Create the First End-to-End Data Assurance Platform for Private Markets Lantern Limited (“Lantern”), the data assurance platform for private markets, announced today a strategic acquisition, expanded platform capabilities, and key leadership additions to address one of private markets biggest operational challenges: transforming fragmented, disconnected data across multiple sources into trusted, actionable intelligence. Private markets run on data, yet much of that data remains siloed, trapped in PDFs, and processed through highly manual workflows across both fund managers and investors. While GPs have increasingly modernized how they manage and report fund information internally, LPs still spend billions of dollars each year manually extracting, normalizing, and reconciling data from capital notices, quarterly statements, and other fund documents. With the acquisition of Freyda Limited (“Freyda”), one of the leading AI document intelligence platforms for ins

KAYTUS Launches MotusAI Enterprise Token Management Platform4.6.2026 09:02:00 CEST | Press release

KAYTUS, a leading provider of AI infrastructure and liquid cooling solutions, today announced the launch of enterprise-grade token management capabilities within MotusAI, enabling organizations to rapidly build, deploy, and operate secure on-premises AI token platform. This pioneering infrastructure enhancement transforms traditional GPU hardware into a unified, governable, and highly efficient internal token service ecosystem. By enabling rapid private deployment, MotusAI provides enterprises with a cost-competitive foundation for accelerating the scale-up of enterprise-grade AI agents while ensuring strong data privacy. Why Enterprises Must Build a Private Token Management Platform Now? As enterprise AI evolves from isolated model-based tools to interconnected AI agent systems that enable deep workflow automation, tokens are emerging as the fundamental “computing currency” of the modern organization. While standard LLM queries typically involve predictable, one-off inference requests

In our pressroom you can read all our latest releases, find our press contacts, images, documents and other relevant information about us.

Visit our pressroom
World GlobeA line styled icon from Orion Icon Library.HiddenA line styled icon from Orion Icon Library.Eye