TRUSTA Launches Upgraded AI Scaler Toolkit at SuperComputing 2026

icon币界网
Share
AI summary iconSummary
TRUSTA, a brand under ADATA, launched its upgraded AI Scaler Toolkit at SuperComputing 2026 in Chicago. The toolkit leverages vLLM’s KV cache offloading to extend AI working memory from the GPU to DRAM and TRUSTA enterprise SSDs, reducing response times and enabling cost-effective local AI deployment. It also integrates the Strata inference engine with Intel AMX, boosting prefill speed by up to 25%. This AI and crypto update highlights a significant development in the crypto news landscape.
CoinDesk reports:

Taipei, November 18 /PRNewswire/ — TRUSTA, the enterprise storage brand under storage manufacturer ADATA, made its global debut today at SuperComputing 2026 (SC26), the premier event for high-performance computing. This year’s SC26 is expected to attract over 450 exhibitors and more than 18,000 attendees. TRUSTA is showcasing solutions designed to help businesses scale AI efficiently, covering software, memory, and storage. Highlights include the upgraded AI Scaler Toolkit, which leverages vLLM’s KV cache offloading to extend AI working memory from GPUs to DRAM and TRUSTA enterprise SSDs. Long-context data is reloaded from the SSD rather than recomputed, significantly reducing response latency and enabling enterprises to deploy AI locally at lower costs despite limited GPU resources. The toolkit also integrates the Strata inference engine with Intel AMX acceleration, boosting prompt processing (prefill) speeds by up to 25%*.

Enhanced AI Scaler Toolkit: vLLM KV Cache Offloading to SSD for Enterprise AI

As AI shifts from model training to inference, storage demands are growing rapidly. According to Counterpoint Research, enterprise SSDs accounted for 48% of global NAND flash capacity shipments in the second quarter of 2026, nearly double the 26% recorded in the same period last year. During inference, large language models (LLMs) maintain KV caches for each token they process, and long documents, multi-turn conversations, and AI agents can quickly cause cache sizes to exceed GPU memory capacity. Once evicted, the cache must be recomputed, leading to longer wait times for users. The upgraded AI Scaler Toolkit integrates vLLM—a widely adopted open-source engine for enterprise AI services—and offloads its KV cache in three tiers: GPU memory for active requests, DRAM as a fast cache, and TRUSTA PCIe Gen5 enterprise SSDs as a high-capacity, low-cost layer that retains data across reboots. Internal TRUSTA testing shows:

  • Faster long-document processing: Long documents are reloaded from SSD instead of being recalculated, significantly reducing wait times before AI responses.
  • Context is preserved after restart: After a server reboot or restart, the conversation can resume from the KV cache stored on the SSD, rather than recalculating from scratch.

This tool also offers a one-stop experience: one-click installation, support for multiple models and inference engines (vLLM, llama.cpp, and Strata), and an interface for allocating GPU, DRAM, and SSD resources, adjusting parameters, and testing models.

Strata with Intel AMX: Prompt processing speed increased by up to 25%

For deployment scenarios that require loading ultra-large models into limited GPU memory while efficiently utilizing DRAM and SSD, this tool includes a lightweight Strata engine. On Intel Xeon processors, Intel AMX enables the CPU to directly process model data in DRAM, boosting prompt processing speed by up to 25%*.

High-performance enterprise SSDs and DDR5 R-DIMMs: Strengthening AI infrastructure

In addition to the AI Scaler Toolkit, TRUSTA also showcased PCIe Gen5 enterprise SSDs and DDR5 R-DIMMs. These high-performance, high-capacity, and energy-efficient memory and storage solutions meet the diverse demands of AI servers and HPC environments.

TD7P51 ECO / T7P5 PCIe Gen5 Enterprise SSD: Offers capacities up to 15.36TB in standard U.2, E1.S, and E3.S form factors, with endurance up to 3 DWPD. Flexible Data Placement (FDP) technology enhances storage efficiency and extends drive lifespan. With power consumption under 20W, it delivers 1.7x higher energy efficiency compared to similar products, helping data centers reduce power usage and support the global trend toward green data centers.

Enterprise-grade DDR5 R-DIMM: Speeds up to 6,400 MT/s with capacities up to 128 GB per module. Built-in ECC automatically detects and corrects data errors, helping AI servers maintain stability and reliability under sustained high loads.

TRUSTA says: "As enterprise AI adoption continues to grow, effectively integrating compute, memory, and storage resources has become critical to improving AI deployment efficiency and controlling costs. Enterprises no longer need to continuously purchase expensive GPUs to adopt AI. By better utilizing CPUs, memory, and SSDs, they can run AI faster and more cost-effectively. With the upgraded AI Scaler Toolkit, enterprise-grade SSDs, and memory technologies, TRUSTA helps businesses maximize the potential of their existing hardware to build more flexible, efficient, and scalable AI computing environments, enabling efficient AI expansion."

*Test Notes: Based on internal testing conducted by TRUSTA in November 2026, the TRUSTA version of Strata with AMX enabled in the AI Scaler Toolkit was compared against upstream Strata v0.2.11. The test platform consisted of an Intel Xeon w5-3535X processor with a single NVIDIA RTX 6000 Ada GPU, running a Mixture-of-Experts (MoE) LLM. Only the prefill phase was measured. Each test was run at least three times, and the median result was taken. For a single input of approximately 8K tokens, speed improved by about 11%; for multi-turn conversations with approximately 4K new tokens per turn, speed improved by about 25%; for single inputs exceeding approximately 12K tokens, the difference was negligible. Actual performance may vary depending on hardware configuration, model, and input length.

Event Information

Event: Supercomputing 2026 (SC26)

Date: November 17–19, 2026

Location: McCormick Place West, Chicago, USA

Booth number: 7214

About TRUSTA

TRUSTA is an enterprise storage brand under ADATA Technology, focused on the AI server and data center markets. Guided by the brand ethos “Together, Empowering Tomorrow,” TRUSTA builds on trust and collaboration to deliver stable, high-speed, energy-efficient, and scalable storage solutions that meet the evolving needs of data-driven enterprises. Leveraging ADATA’s extensive global distribution network, robust R&D capabilities, and advanced manufacturing resources, TRUSTA is poised to lead the next wave of storage innovation in the AI era, helping organizations worldwide build a solid foundation for future enterprise computing.

Global Media Contact – Brand Marketing & Media Relations ADATA Technology Co., Ltd. Phone: +886-2-8228-0886 Fax: +886-2-8227-1760 Email: [email protected]

Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information. Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.