- According to Phys.org, with old shared cache techniques, the directory's memory grows in linear proportion to the number of cores. This technique means it increases according to the log of the number of cores.
- This isn't a huge deal for quad-core chips, but when you get up to a 128-core chip, it will only require a third of the memory as a chip built without this technique.
- Essentially, the researchers found a simple way to keep track of what each core is doing and put them into global logical-time ordering. Although the researchers are dubious that big chip makers like Intel will adopt this new technique immediately, it does hold significant potential for massively multicore chips.
Now multicore chips can share memory more efficiently
9. 10. 15 by Kif Leswing