It’s a common knowledge that computers run on rams, any application run on a pc is temporarily stored in the ram. So why can’t we just dedicate 32Gbs of ssd storage to ram and have our pcs run on it? That way one could have a 256Gbs of ram easily, without the need of the ram sticks and for so much cheaper.
In: 0
A top of the line NVMe SSD does something like 4GB/s with a read latency (time between command given to start a read operation and data getting back) of ~300 microseconds. My (out of date DDR4 of middling performance) RAM has a read performance of 41GB/s and a read latency of 80.6 nanoseconds (0.86 microseconds, 300+ times less). Remember that 1 nanosecond = one cycle at 1GHz, 5 cycles at 5GHz (the speed of a brand new high end CPU). Even at 80ns that means reading from RAM equals your CPU not being able to do anything for hundreds of cycles. At hundreds of microseconds, reading from an SSD would punt your performance back to the 90s era.
The L3 cache of my CPU, a few megabyte memory directly on the CPU for stuff that even RAM would be slow for, has a read speed of **390GB/s**. L2 cache, 710GB/s. There is an even smaller and faster L1 cache but I only have the trial version of AIDA64 and it blanks out that stat – it does show that it has a latency of 1.1ns, 80 times faster than my RAM.
Do you get why SSDs make lousy operative memory now?
Latest Answers