DRAM
In plain wordsThe main working memory of computers and AI chips: dense and fairly cheap, but slower than on-chip memory and needing constant refreshing to keep its data.
The dense, comparatively cheap memory that supplies the gigabytes of main memory in computers and accelerators. Each bit is stored as a small charge on a microscopic capacitor behind one access transistor; because that charge seeps away, the chip must rewrite (refresh) every row on a fixed cycle, typically every 64 milliseconds. One transistor per bit, against about six for SRAMSRAM, abbr.: The fastest kind of computer memory, built directly into processor chips. It is quick but takes up a lot of space, so chips hold only a little of it., is what makes it so much denser and cheaper per bit, and it comes in several families: DDR for servers and PCs, LPDDR for phones and some AI systems (such as Nvidia's Grace CPU), GDDR for graphics cards, and stacked high-bandwidth memoryhigh-bandwidth memory, n.: Very fast memory made by stacking memory chips on top of each other and placing them right next to the processor, so AI chips can be fed data quickly enough. for data-center accelerators. DRAM capacity, bandwidthmemory bandwidth, n.: How fast data can be moved between a chip's memory and its calculators. For chatbots, this often decides how quickly words appear, more than raw calculating speed. and price shape which models can be trained or served and at what cost, and three manufacturers — Samsung, SK hynix and Micron — dominate supply.
Origin
Robert H. Dennard of IBM invented the one-transistor DRAM cell in 1966 (US patent granted 1968); Intel's 1103 (1970) was the first commercially successful DRAM chip. Dennard later gave his name to Dennard scalingDennard scaling, n.: The rule that as transistors shrank they also used proportionally less power, so chips got faster without getting hotter. It stopped working in the mid-2000s, ending the era of ever-faster single processors..
Related
- SRAMSRAM, abbr.: The fastest kind of computer memory, built directly into processor chips. It is quick but takes up a lot of space, so chips hold only a little of it.
- high-bandwidth memoryhigh-bandwidth memory, n.: Very fast memory made by stacking memory chips on top of each other and placing them right next to the processor, so AI chips can be fed data quickly enough.
- VRAMVRAM, abbr.: The memory on a graphics card. For running AI at home it is the key number: the model has to fit in it, so more memory means bigger models.
- memory wallmemory wall, n.: The growing gap between how fast chips can calculate and how fast they can fetch data from memory. Processors increasingly sit idle, waiting for numbers to arrive.
- memory bandwidthmemory bandwidth, n.: How fast data can be moved between a chip's memory and its calculators. For chatbots, this often decides how quickly words appear, more than raw calculating speed.
- SRAMSRAM, abbr.: The fastest kind of computer memory, built directly into processor chips. It is quick but takes up a lot of space, so chips hold only a little of it.
Mentioned in
high-bandwidth memoryhigh-bandwidth memory, n.: Very fast memory made by stacking memory chips on top of each other and placing them right next to the processor, so AI chips can be fed data quickly enough., VRAMVRAM, abbr.: The memory on a graphics card. For running AI at home it is the key number: the model has to fit in it, so more memory means bigger models., Dennard scalingDennard scaling, n.: The rule that as transistors shrank they also used proportionally less power, so chips got faster without getting hotter. It stopped working in the mid-2000s, ending the era of ever-faster single processors., SRAMSRAM, abbr.: The fastest kind of computer memory, built directly into processor chips. It is quick but takes up a lot of space, so chips hold only a little of it.