DRAM
Dynamic random-access memory; the main commodity memory type whose limitations led to HBM and which has stagnated since 2012.
As defined in dwarkesh_asianometry_dylan_patel_how_the_semiconductor_indu
Where it comes up · 19
- Samsung: Semiconductors over Smartphones9 claims
- Dwarkesh Patel Podcast - Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI compute3 claims
- Intel: Cyclical Recovery or Secular Demise?2 claims
- Dwarkesh Patel Podcast - What will computer chips look like in 2045?2 claims
- Alex Sacerdote - How to Invest Through Technology Cycles - [Invest Like the Best, EP.477]2 claims
- Gavin Baker - AI, Semiconductors, and the Robotic Frontier - [Invest Like the Best, EP.385]2 claims
- Neil Movva - Making AI 10x Cheaper - [Invest Like the Best, EP.488]2 claims
- CHIPS Act: Securing Semiconductor Supply1 claim
and 11 more
Most important claims · 10 of 34Search all claims for “DRAM” →
- Unlike human brains, where storing and loading memory is cheap and math is expensive, chips find loading data very expensive and math very cheap, and math is getting cheaper faster than memory because of fundamental DRAM limits, so models should be designed to use far more compute.Etched - Building AI Hardware to Make Inference Faster and Cheaper - [Invest Like the Best, EP.480]
- Intel's early role in the PC revolution offers a template for being early in a growth cycle, as it rode the wave for two and a half to three decades despite DRAM-style cycles, because more PC buyers dampened visible cyclicality.Andrew Homan & Chris Miller - Redefining Semiconductor Progress - [Invest Like the Best, EP.396]
- DRAM margins run about 35% to 40% at the top of the cycle and 5% to 10% at the bottom, and Samsung rarely loses money in DRAM while Hynix and Micron lose significant money at the trough.Samsung: Semiconductors over Smartphones
- Hardware companies cannot control their full stack as software companies can, because they depend on HBM, DRAM, TSMC, and cross-border supply chains, and Cerebras' 2024 IPO attempt failed over CFIUS and timing, so getting lucky on timing and macro conditions is critical.Eric Vishria - A Decade of Lessons Investing in Software & Hardware - [Invest Like the Best, EP.486]
- KH Lee, the third-generation leader, recognized that memory semiconductors would grow as electronics expanded and personally financed the purchase of a memory operation that had been poorly financed, then pursued a multi-decade buildout.Samsung: Semiconductors over Smartphones
- The DRAM industry consolidated over three decades from 10 or 11 players to three because poorly financed firms could not survive industry downturns.Samsung: Semiconductors over Smartphones
- On-chip SRAM is roughly three orders of magnitude less dense than DRAM-based HBM; Blackwell carries 288 gigabytes of HBM around its logic die against about 500 megabytes of on-die SRAM, though SRAM offers much faster access.Neil Movva - Making AI 10x Cheaper - [Invest Like the Best, EP.488]
- Most of Samsung's profit comes from component businesses, not finished consumer electronics: memory generated 55% of profit in 2022, mobile phones 25%, display panels 14%, and all other consumer electronics only 3%, so consumer electronics is typically under 30% of profit.Samsung: Semiconductors over Smartphones
- A ChatGPT-based server needs about 5x the memory of a standard server, and an L5 fully autonomous vehicle needs more than 30x the DRAM and over 100x the NAND of an internal combustion vehicle, so AI and autonomy create large memory demand.Samsung: Semiconductors over Smartphones
- Over the past ten years, DRAM bit volumes have grown about 23% per year and NAND about 37% per year, with NAND growing faster because it both rides product growth and takes share from hard drives.Samsung: Semiconductors over Smartphones