Smart IOPS Unobtanium™ T50 SSD and H3 Platform’s GPU-centric appliance bring compute, networking and ultra-high-IOPS flash together for NVIDIA Storage-Next™ and SCADA-enabled AI infrastructure
MILPITAS, Calif. and NEW TAIPEI CITY, Taiwan, Aug. 6, 2026 /PRNewswire/ — Smart IOPS, Inc., a provider of high-performance enterprise storage solutions, and H3 Platform, a pioneer in AI and composable IT infrastructure, today announced world’s first one billion IOPS capable AI Compute Storage solution. The solution integrates Smart IOPS Unobtanium™ T50 solid-state device with a new H3 Platform appliance built for the recently announced NVIDIA AI infrastructure, NVIDIA networking technologies.
The solution is being engineered for the emerging class of data-intensive AI Inference applications that require GPUs to access very large datasets through fine-grained, highly parallel I/O. It is designed to support NVIDIA’s Storage-Next initiative and the NVIDIA SCADA (Scaled Accelerated Data Access) programming model. NVIDIA Storage-Next and NVIDIA SCADA enable GPU-initiated storage access for workloads whose working sets extend beyond local GPU memory. By shortening the path between GPU compute and NVMe storage, the companies aim to reduce CPU control-path overhead, keep accelerators supplied with data and improve infrastructure utilization for AI inference and data analytics.
At the center of the announcement is the Smart IOPS AI Compute Storage™ or ACS device, Unobtanium™ T50. The PCIe Gen6 x4, NVMe 2.0 ACS device in an E3.S form factor is slated to deliver up to 50 million random-read IOPS and up to 10 million random-write IOPS at a highly granular 512-byte block size, together with up to 28 GB/s sequential reads and 24 GB/s sequential writes. Four such T50 devices can provide an aggregate device-level target of up to 200 million random-read IOPS—to align with the I/O capability of next-generation PCIe Gen6 x16 GPUs like NVIDIA Rubin. Actual system performance will depend on workload, software, topology, configuration and other system-level factors.
A pathbreaking attribute of the Smart IOPS TruRandom® controller architecture is its use of broadly available TLC NANDs operated in pseudo-SLC mode to deliver the extreme IOPS. Because the standard TLC NAND can be sourced from multiple qualified suppliers, Smart IOPS ACS device benefits from the flash industry’s manufacturing scale, supply-chain resilience, and cost efficiencies. ACS devices, however, can also support specialized low-latency NANDs, where a workload or customer requirement is justified over the cost.
H3 Platform is developing an air-cooled appliance that brings these storage devices into a deployable GPU-centric system. The planned platform combines four NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, four NVIDIA ConnectX-8 and 20 E3.S 2T storage slots, each designed to support up to 60 watts. This allows up to a billion IOPS to be delivered from 20 storage slots each supporting up to 50 million IOPS. The appliance is designed to provide balanced connectivity across compute, networking, and storage resources, giving customers a practical platform for evaluating and deploying GPU-initiated, small-block storage workloads in production environments.
“AI infrastructure is entering a phase in which storage must behave less like a passive repository and more like an active extension of the compute fabric,” said Ashutosh Das, CEO of Smart IOPS. “With the Unobtanium™ T50 ACS device, our goal is to deliver 50 million IOPS in a compact PCIe Gen6 x4 E3.S form factor using standard TLC NAND. This approach combines extreme performance with flash-industry’s economies of scale making the ACS devices suitable for large scale deployment. H3 Platform complements that innovation with the system architecture needed to place high-IOPS storage close to NVIDIA AI infrastructure and networking.”
The planned H3 Platform delivers the highest IOPS performance ever achieved in the smallest datacenter footprint by integrating Smart IOPS ACS devices across 20 E3.S slots. The Air-cooled appliance design combines GPUs, high-speed network cards, and E3.S storage device slots that support up to 60W of power and thermal requirement per slot to handle the demands of high-performance storage devices like Smart IOPS ACS.
“The next wave of AI systems must be designed as an integrated data path—not as disconnected compute, network and storage components,” said Brian Pan, CEO of H3 Platform. “Our appliance is being developed to combine NVIDIA accelerated computing and networking with 20 high-power E3.S slots in an efficient, air-cooled platform. Smart IOPS’ ACS device is a strong complement to this design, helping us turn the concept of GPU-accelerated memory/storage into a system customers can evaluate, optimize and deploy.”
Why High IOPS Matters for AI Infrastructure
AI workloads are increasingly combining intensive computation with irregular, data-dependent access patterns. Graph neural networks, vector search, recommendation systems, retrieval-augmented generation, embedding stores and other sparse analytics can generate large numbers of small, random requests from many GPU threads. Conventional CPU-orchestrated I/O can add synchronization, queueing and control-path overhead, while bandwidth-oriented SSD specifications alone may not describe how well a system serves these fine-grained accesses.
Storage-Next and SCADA address this shift by enabling GPUs to initiate and orchestrate storage activity at scale. High-IOPS, low-latency SSDs are relevant because they can service more concurrent fine-grained requests, helping use PCIe bandwidth efficiently and reducing the risk that expensive GPU resources wait for data. The combined Smart IOPS and H3 Platform solution is intended to provide an integrated environment for this new storage architecture.
Smart IOPS ACS — Unobtanium™ T50
Attribute
Preliminary specification
Random read
Up to 50 million IOPS
Random write
Up to 10 million IOPS
Block size
512 bytes
Sequential read
Up to 28 GB/s
Sequential write
Up to 24 GB/s
Capacity
9 TB, 18 TB and 36 TB
Media
TLC NAND operated in pseudo-SLC mode; support for
specialized low-latency SLC NANDs
Form factor
E3.S 2T
Interface
PCIe Gen6 x4; NVMe 2.0
Planned availability
Evaluation samples: Q1 2027; production: Q2 2027
Preliminary specifications; subject to change without notice.
H3 Platform High IOPS Appliance
Attribute
Preliminary specification
GPU
4 × NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs
Networking
4 × NVIDIA ConnectX-8
SSD bays
20 × E3.S 2T slots for high-IOPS SSDs
Per-slot power
Up to 60 W per E3.S slot
Cooling
Air-cooled
Planned
availability
Evaluation systems: Q1 2027; production: Q2 2027
Preliminary specifications; subject to change without notice.
Availability
Smart IOPS and H3 Platform expect evaluation samples and systems to be available in the first quarter of 2027, with production availability planned for the second quarter of 2027. Product specifications, configurations and schedules are preliminary and may change without notice. Customer qualification, ecosystem software readiness and NVIDIA validation or approval are separate processes and are not implied by this announcement.
About Smart IOPS
Smart IOPS provides enterprise-class SSD solutions designed for the extreme demands of AI-driven and data-intensive workloads. The company’s SSD controller architecture is powered by TruRandom® technology to deliver high random-I/O performance, bandwidth and consistent quality of service. Smart IOPS solutions help cloud service providers, high-performance computing environments and enterprise data centers address storage I/O bottlenecks across AI/ML, analytics, databases, caching and other demanding applications. For more information, visit https://www.smartiops.com
About H3 Platform
Founded in 2014 and headquartered in New Taipei City, Taiwan, H3 Platform is a global pioneer in AI and composable IT infrastructure solutions. With more than a decade of expertise in PCIe switch fabric and CXL technologies, H3 Platform develops solutions for enterprise memory scaling, resource allocation and composable infrastructure. The company is advancing GPU-accelerated memory and storage architectures that establish direct, GPU-centric data paths, helping AI-native data centers improve accelerator utilization, scale demanding workloads and reduce total cost of ownership. For more information, visit https://www.h3platform.com
Forward-Looking Statements
This press release contains forward-looking statements regarding products under development, expected performance, interoperability, customer benefits and availability. Words such as “aim,” “design,” “develop,” “expect,” “intend,” “plan,” “target,” “will” and similar expressions identify forward-looking statements. These statements are based on current expectations and assumptions and are subject to risks and uncertainties that could cause actual results to differ materially, including development and validation results, component and software availability, ecosystem readiness, customer qualification, manufacturing schedules and changes in market requirements. Smart IOPS and H3 Platform undertake no obligation to update these statements except as required by law.
Trademarks
NVMe is a trademark of NVM Express, Inc. TruRandom® and Unobtanium™ are trademarks and/or claimed marks of Smart IOPS, Inc. All other company names, product names and service names may be trademarks of their respective owners.
View original content:https://www.prnewswire.com/news-releases/smart-iops-and-h3-platform-unveil-ai-compute-storage-solution-designed-to-deliver-up-to-one-billion-random-iops-302845255.html
SOURCE Smart IOPS, Inc.; H3 Platform
Technology4 days ago
Coin Market4 days ago
Technology5 days ago
Technology5 days ago
Technology3 days ago
Coin Market4 days ago
Coin Market4 days ago