SK hynix has now given out HBM4 units to NVIDIA for test runs of its new "Rubin" AI GPUs. This move puts the South Korean group in front of Micron and Samsung in the fight for high-speed memory deals.
HBM4 is key for high-level computing, as it mixes memory and working parts in one unit. Korean news hub DealSite notes SK hynix has begun to share some HBM4 with NVIDIA. This lets NVIDIA push ahead with its Rubin GPU plans, aiming to hit the market by the end of 2025.
In the HBM4 match
- Micron Seems to lag as it has low HBM4 make rates and faces bumps in setting up its work lines.
- Samsung Though Samsung is doing well with HBM4, its make rates aren't out yet. But, Samsung just got HBM3E deals from AMD, which may give it a leg up in getting future HBM4 work from NVIDIA.
Even with SK hynix's head start, NVIDIA plans to get HBM4 from different places, like it does with HBM3, maybe counting on Micron and Samsung too.
The handout of HBM4 kits shows NVIDIA's Rubin GPU plan, likely called the R100, is on track to show up by the end of 2025. This date fits with NVIDIA's six-month product cycle. The Rubin GPUs might have
- TSMC N3 tech,
- Eight HBM4 blocks,
- A 3nm "Grace" CPU part,
- A big push on saving power.
Getting HBM4 from SK hynix boosts NVIDIA's reach to fill needs for its next AI units. SK hynix's spot as a top choice is set to grab a big slice of HBM4 orders, growing its influence in the memory area.
