SK Hynix Previews HBM4E Memory at Computex, Cramming 48GB Into a 12-Hi Stack and Pushing Bandwidth to a Record 4 TB/s

Jun 3, 2026 at 06:05am EDT
A close-up of an HBM4E chip next to the text 'HBM4E 48GB' and '32Gb, up to 10 AI at Scale.'

The HBM race has seen major acceleration across DRAM makers, and SK Hynix is showcasing its new HBM4E solution, which increases density & bandwidth.

HBM4E Drives A 33% Die Density Uplift, and a 38% Boost in Memory Bandwidth, Setting Eyes at 4 TB/s for Next-Gen AI Datacenter Chips

NVIDIA and AMD are going to utilize HBM4 to power their latest GPUs for AI datacenters this year. Both the Rubin and the MI400 series pack some big specifications, and the HBM solutions being used on them are one of the biggest innovations.

Related Story Intel Crescent Island “Xe3P” GPU Scales To 480 GB of “Cost-Optimized” LPDDR5X Memory, Beating NVIDIA Rubin & AMD MI450X With Highest Capacity

The move to HBM4 delivers a significant improvement in bandwidth, density, and efficiency, but even with all its advanced features, the memory isn't enough, and there's demand for even more capacities coupled with higher bandwidth.

So at Computex 2026, SK Hynix is previewing its upcoming HBM4E memory die. The new solution comes with a 32Gb density, up 33% versus HBM4. With these densities, you can achieve the same 48 GB capacity as HBM4 does with 16-Hi stacks using just 12-Hi stacks. And it doesn't stop there, HBM4E is fast, with up to 16 Gbps pin speeds, a 37% boost over HBM4, which pushes the memory bandwidth to a record 4 TB/s.

With HBM4E, the main highlights are the ones we mentioned above: memory bandwidth and die density. SK Hynix is already leading the charge in producing HBM4E by showcasing samples, and we will see the first use of this memory on NVIDIA's Rubin Ultra GPUs, which arrive next year. The successor to Rubin will feature a denser design and pack multiple GPUs and HBM4E chiplets into a single package that delivers a record-setting leap in AI performance.

Spec / FeatureHBM4E (48GB, 12-Hi)HBM4 (Peak Spec, 48GB, 16-Hi)HBM3E (36GB, 12-Hi)
Capacity48GB48GB36GB
Stack Height12-Hi16-Hi12-Hi
Die Density1.5× improvement+33% per core die
Bandwidth (per pin)3.2 Gbps (+50%)Up to 16 Gbps1.2 Gbps (+20%)
Memory Bandwidth1.5 TB/sHigher (AI workloads)1.2 TB/s
I/O Channelsx1024x2048x1024
Voltage1.2V1.2V1.2V
Efficiency1.5× density gain+40% power efficiency+10% power efficiency
Process NodeTSMC N3E

AI-N B Uses HBM-Like TSVs To Stack NAND Together

SK Hynix is also working on a next-gen stacked NAND solution, which will feature a HBM-like design by stacking multiple NAND dies together using TSVs (Through-Silicon-Vias). This technology aims to offer HBM-Like Throughput with SSD-Like Capacities. The technology is similar to what HBF and Z-Angle are doing, and it can help address the memory supply-demand gap that is currently crunching away at the tech segment.

SK Hynix's 1st 96 GB LPCAMM2, V9 QLC, and V9 TLC Products Showcased Too

Besides HBM DRAM, SK Hynix also showcased its latest client-side products, such as a 96 GB LPCAMM2 module based on its 1cnm process tech. This LPCAMM2 module achieves up to 9.6 Gbps transfer rates & is based on the LPDDR5X standard. The modules should be arriving later this year for "AI PC" platforms.

The company is also working on a broad range of NAND technologies, such as its V9 NAND, which comes in QLC and TLC flavors, offering up to 2 TB of storage capacity in a cSSD form factor. These SSDs are ideal for small form factor designs, achieving high-power efficiency in a DRAM-less fashion.

About the author: A Software Engineer by training and a PC enthusiast by passion, Hassan Mujtaba serves as Wccftech's Senior Editor for hardware section. With years of experience in the industry, he specializes in deep-dive technical analysis of next-generation CPU and GPU architectures, motherboards, and cooling solutions. His work involves not only breaking news on upcoming technologies but also extensive hands-on reviews and benchmarking.

Follow Wccftech on Google to get more of our news coverage in your feeds.