Samsung's HBM4E memory passes NVIDIA qualification tests
Samsung's new 12-layer HBM4E high-bandwidth memory has reportedly passed qualification testing at NVIDIA, clearing a key technical hurdle for use in NVIDIA's next generation of AI GPU accelerators. This memory delivers up to 48 GB per stack and bandwidth of 3.6 TB/s, with future versions expected to reach 64 GB per stack and 4 TB/s. Samsung began shipping HBM4E samples in May but hasn't confirmed the news itself.
- HBM4E offers up to 48 GB per stack, 3.6 TB/s bandwidth
- Qualification tested by NVIDIA and hyperscale data centers
- Samsung shipped samples in May, hasn't confirmed qualification
- Future HBM4E could reach 64 GB and 4 TB/s per stack
- Mass production expected to ramp up through 2027
Sources covering this
In this story
More in Chips
SpaceX in talks to finance $40 billion Nvidia chip order
SpaceX is working with Apollo Global Management and several banks to line up $40 billion in financing for an enormous order of Nvidia AI…
Intel confirms continued partnership with Musk's Terafab
Intel will remain involved with Elon Musk's Terafab chip manufacturing venture, even as Musk discusses a possible sublease for TSMC to…
Carl Zeiss and ASML Propose Hyper-NA EUV Lithography
Researchers from Carl Zeiss SMT and ASML have introduced a new approach to manufacturing chips: Hyper-NA EUV lithography.
Experts debate thermal challenges in photonic AI chips
Industry leaders say AI chips using photonics—light-based data movement—could speed up data processing and cut energy use, but new heat…