Geo News
Community curated by people like you
LatestAICryptoHealthWorld AffairsUS Politics
SK hynix and SanDisk Release First High Bandwidth Flash (HBF) Standard Specification for AI Systems
00

SK hynix and SanDisk Release First High Bandwidth Flash (HBF) Standard Specification for AI Systems

Aug 4, 2026

SK hynix and SanDisk release the first open High Bandwidth Flash (HBF) specification through the Open Compute Project to resolve AI inference bottlenecks. The standard defines 8-high and 16-high NAND stacks up to 512GB, utilizing UCIe interfaces to deliver bandwidths from 0.4 TB/s to 3.0 TB/s. Positioned between HBM and SSDs, HBF targets read-intensive workloads like Llama 3.1 model weights. While Google and Tenstorrent back the consortium, major players like NVIDIA and AMD have yet to adopt the technology.

HBF specification release

  • ▪The HBF specification defines memory packages with capacities up to 512GB using either 8-high or 16-high NAND die stacks.
  • ▪The HBF specification adopts the Universal Chiplet Interconnect Express (UCIe) standard as its physical and protocol-level host interface.
  • ▪SK hynix and SanDisk released the first standard technical specification for High Bandwidth Flash (HBF) through the Open Compute Project (OCP) on August 4, 2026.

Performance tiers bandwidth capacity

  • ▪The highest HBF performance grade of 3.0 TB/s is designed to compete with the memory bandwidth of a single HBM4 stack, which ranges from 2.0 TB/s to over 3.3 TB/s.
  • ▪The HBF specification defines three performance grades with read bandwidth ranging from approximately 0.4 TB/s to 3.0 TB/s.

AI inference workload applications

  • ▪A simulation by SanDisk using the Llama 3.1 405B model showed that HBF-based system performance was within 2.2% of a system with unlimited HBM capacity.
  • ▪HBF is designed as an intermediate memory tier between High Bandwidth Memory (HBM) and enterprise SSDs to store read-intensive data like pretrained AI model weights.

SanDisk Gen 1 targets

  • ▪SanDisk is targeting a first-generation HBF product with 1.6 TB/s of read bandwidth and 512GB of capacity using a 16-die stack.
  • ▪SanDisk previously targeted the second half of 2026 for its first HBF samples, with commercialization targeted for 2027.

Consortium member participation

  • ▪The HBF consortium was launched in February 2026, six months prior to the release of the first standard specification.
  • ▪Google and Tenstorrent are participating in the HBF consortium alongside founding members SK hynix and SanDisk.

Adoption indicators dependencies

  • ▪Major technology companies including AMD, Broadcom, Intel, NVIDIA, Marvell, Micron, Qualcomm, Samsung, and Western Digital have expressed no official interest in HBF.
  • ▪The commercial success of HBF depends on whether accelerator vendors integrate required interfaces, software frameworks support efficient data placement, and sampled devices meet latency and power targets.

8 sources

News
SK hynix Unveils First HBF Standard Specifications with Sandisk, Presenting AI Memory Solutions at ‘FMS 2026’ | SK hynix Newsroom
View source article
Tomshardware
New HBF spec outlines tech that can give GPUs terabytes of extra memory — Sandisk and SK hynix unveil spec with up to 16-Hi NAND stacks, 3 TB/s bandwidth, UCIe
View source article
Videocardz
Sandisk and SK hynix release first High Bandwidth Flash specification - VideoCardz.com
View source article
Trendforce
[News] SK hynix, SanDisk Debut HBF Standard to Challenge AI Memory Bottlenecks with Google, Tenstorrent Support
View source article
Hothardware
SK Hynix And Sandisk Unleash High-Bandwidth Flash To Fix AI Bottlenecks
View source article

Featured stories

View more in AI accelerators

TSMC weighs multibillion-dollar Texas campus expansion for AI chips

Oct 1, 2026 · 2 sources

OpenAI and Synopsys partner to develop AI model for chip design

Sep 30, 2026 · 3 sources

DeepSeek releases software tools for Huawei AI chips to challenge Nvidia

Sep 30, 2026 · 4 sources

Accelevation raises $540 million in US IPO

Sep 29, 2026 · 2 sources

Story comments

Loading comments…

Related Projects

SK hynixGoogleSanDisk

Topics

AI acceleratorsMemory technologyAI hardwareCompute, chips & AI infrastructure

Featured stories

View more in AI accelerators

TSMC weighs multibillion-dollar Texas campus expansion for AI chips

Oct 1, 2026 · 2 sources

OpenAI and Synopsys partner to develop AI model for chip design

Sep 30, 2026 · 3 sources

DeepSeek releases software tools for Huawei AI chips to challenge Nvidia

Sep 30, 2026 · 4 sources

Accelevation raises $540 million in US IPO

Sep 29, 2026 · 2 sources