High Bandwidth Flash (HBF), a new memory format presented at Hot Chips 2026, offers significantly more capacity than traditional High Bandwidth Memory (HBM) at a comparable cost. However, HBF's lower bandwidth makes it unsuitable for many AI workloads where speed is critical. OXMIQ Labs demonstrated that while HBF can drastically reduce the number of GPUs needed to store large models like Kimi-K2, HBM remains superior for maximizing inference throughput and delivering a lower cost per token when bandwidth is the limiting factor. The consensus suggests HBF may serve as a specialized memory tier for large, less frequently accessed datasets rather than a direct replacement for HBM. AI
IMPACT High Bandwidth Flash offers a potential solution for accommodating extremely large AI models, but its performance limitations may restrict its adoption for high-throughput inference tasks.
RANK_REASON New memory technology presented at a research conference. [lever_c_demoted from research: ic=1 ai=0.7]
- HBM4E
- High Bandwidth Flash
- High Bandwidth Memory
- Hot Chips 2026
- Kimi-K2
- NAND flash
- OXMIQ Labs
- SanDisk
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →