Nvidia Pushes for 100M IOPS SSDs to Eliminate AI GPU Bottlenecks

Nvidia's Push for Ultra-Fast SSDs

Nvidia is collaborating with partners to develop solid-state drives (SSDs) capable of achieving an unprecedented 100 million input/output operations per second (IOPS) for small-block workloads. This initiative aims to address the storage performance bottlenecks faced by AI training and inference systems 1

1

Wallace C. Kuo, CEO of Silicon Motion Inc. (SMI), revealed this ambitious goal in an exclusive interview with Tom's Hardware. The target of 100 million IOPS represents a significant leap from current PCIe 5.0 x4 SSDs, which top out at around 2-3 million IOPS for both 4K and 512B random reads 1

1

The Need for Speed in AI Storage

Source: Tom's Hardware

Modern AI accelerators, such as Nvidia's B200, feature high-bandwidth memory (HBM3E) with bandwidth around 8 TB/s. This significantly exceeds the capabilities of current storage subsystems in both overall throughput and latency. AI models typically perform small, random fetches, making 512B blocks more suitable for their latency-sensitive patterns 1

1

Kioxia's 'AI SSD' Development

In response to these emerging demands, Kioxia is developing an 'AI SSD' based on its XL-Flash memory. This drive aims to surpass 10 million 512K IOPS, a significant improvement over current SSDs. Kioxia plans to release this drive during the second half of 2026, potentially aligning with the rollout of Nvidia's Vera Rubin platform 1

1

2

A Kioxia spokesperson told TechPowerUp, "We're taking our ultra-fast XL-Flash memory chips, which use single-level cells, and pairing them with a completely new controller... We're targeting over 10 million IOPS, and we plan to have samples ready by the second half of 2026" 2

2

Source: TechRadar

Related Stories

Nvidia and Kioxia Collaborate on Revolutionary 100 Million IOPS SSD for AI Servers

Kioxia Unveils PCIe 6.0 SSDs Targeting 10 Million IOPS for AI Workloads

Micron Unveils World's First PCIe 6.0 SSD, Revolutionizing Data Center Storage

Micron Unveils High-Performance SSDs Optimized for AI Data Centers

Challenges and Future Technologies

Achieving 100 million IOPS on a single drive with conventional NAND while maintaining cost-effectiveness and power efficiency poses significant challenges. SMI's CEO believes that a new type of memory might be necessary to reach this goal 1

1

Several companies, including Micron and SanDisk, are developing new types of non-volatile memory. However, the commercial viability of these technologies remains uncertain 1

1

Kioxia's Comprehensive Approach

Kioxia is not only focusing on high-IOPS SSDs but also developing a range of products to meet diverse storage needs. Their CM9 series, currently sampling to customers, aims to match the speed and reliability requirements of high-end GPUs used in AI. The LC9 series offers massive 122TB capacities for large databases 2

2

The company is also preparing future flash memory generations using two methods: adding more layers for increased capacity and blending new CMOS designs with older cell structures to manage investment costs 2

2