IBM has presented a technology that will make it possible to integrate three times as much memory in a processor and at the same time double the performance. It has simply decided to trash the SRAM memory that is usually used for cache in a processor. SRAM is very fast and easy to manufacture, but takes up a lot of space, which we’ve seen for a long time with both AMD’s and Intel’s processors where the large SRAM cache memories completely dominate the transistor count.
IBM has swapped the SRAM memory for DRAM, the technology used by regular memory modules. DRAM needs considerably less space, but has been considered too slow to be used as cache. IBM now says that it has managed to speed up the DRAM technology to such a degree that the difference between it and SRAM is more or less indistinguishable.
“IBM said it has been able to speed up the DRAM to the point where it’s nearly as fast as SRAM, and that the result is a type of memory known as embedded DRAM, or eDRAM, that helps boost the performance of chips with multiple core calculating engines and is particularly suited for enabling the movement of graphics in gaming and other multimedia applications. DRAM will also continue to be used off the chip.”
IBM claims that it will launch server processors using the new cache technology next year, but whether other processor manufacturers such as AMD and Intel is planning to do the same remains to be seen.
