🇨🇳Stalecollected in 6m

Huang Hosts SK Hynix HBM4 Engineers Birthday Dinner

Huang Hosts SK Hynix HBM4 Engineers Birthday Dinner
PostLinkedIn
🇨🇳Read original on cnBeta (Full RSS)

💡NVIDIA CEO woos HBM4 team—vital supply chain signal for AI infrastructure builds.

⚡ 30-Second TL;DR

What Changed

Event held on Feb 14 at 99 Chicken in Santa Clara

Why It Matters

This personal gesture highlights NVIDIA's reliance on SK Hynix for HBM supply, critical for AI GPU performance in Blackwell and future platforms. It may ease concerns over memory shortages in AI data centers.

What To Do Next

Track SK Hynix Q1 earnings for HBM4 yield updates impacting NVIDIA GPU supply.

Who should care:Enterprise & Security Teams

Key Points

  • Event held on Feb 14 at 99 Chicken in Santa Clara
  • 30 core SK Hynix engineers focused on DRAM/HBM4
  • Jensen Huang personally served drinks to guests
  • Signals strong NVIDIA-SK Hynix collaboration

🧠 Deep Insight

Background and context from public sources — not the original article. 6 sources cited.

🔑 Enhanced Key Takeaways

  • SK Hynix secured 70% of NVIDIA's HBM4 orders for the Vera Rubin platform, utilizing 12-layer and 16-layer stacks.[1]
  • HBM4 features a 2048-bit I/O interface doubling HBM3E's bandwidth, custom logic die via TSMC 12nm/5nm processes, and 30-micrometer thinned dies for 16-high stacks.[1]
  • SK Hynix began HBM development in 2009, co-developed first TSV HBM with AMD in 2014, and released HBM3 in 2021 with mass production in 2022.[3][4]

🛠️ Technical Deep Dive

  • HBM4 doubles bus width to 2048-bit I/O from HBM3E's 1024-bit, enabling >2.0 TB/s per stack at lower clocks for improved power efficiency.[1]
  • Replaces DRAM base die with TSMC 12nm/5nm logic die integrating memory controllers and PHY, reducing latency by ~20%.[1]
  • 16-high stacks achieved by thinning DRAM dies to 30 micrometers while maintaining 775-micrometer height limit.[1]
  • Power usage 40% more efficient than prior gen, >10 Gbps speed, capable of processing over 400 Full HD 5GB movies per second per stack.[3]

🔮 Future ImplicationsAI analysis grounded in cited sources

SK Hynix HBM4 will enable real-time training of trillion-parameter AI models
HBM4's high throughput and efficiency address energy costs prohibitive in current hardware for Vera Rubin platform.[1]
NVIDIA-SK Hynix 'one team' ties will lock in long-term HBM supply contracts
Custom HBM4 development from design stage creates strategic dependencies unlike standardized memory.[2][6]

Timeline

2009
SK Hynix begins HBM development ahead of industry.
2014
Co-develops first TSV HBM with AMD.
2021-10
Develops world's first HBM3.
2022-06
Begins HBM3 mass production for NVIDIA.
2025-03
Delivers HBM4 samples to NVIDIA.
2026-01
Secures 70% of NVIDIA HBM4 orders for Vera Rubin.
📰

Weekly AI Recap

Read this week's curated digest of top AI events →

👉Related Updates

AI-curated news aggregator. All content rights belong to original publishers.
Original source: cnBeta (Full RSS)

This is a summary, not the original. Read the source, or get the weekly briefing.

Weekly AI briefing

One email a week. Unsubscribe anytime.