[citation][nom]blazorthon[/nom]Actually, several processors use DRAM as a cache. It is MUCH faster than the DRAM used in system memory, but it is still DRAM. It's called eDRAM, embedded DRAM. I think that the Xenon and Cell processors use it, among others. It is used for higher capacity caches (it is about five times denser than usual SRAM, or something like that) than SRAM, but somewhat slower. It makes a great L3 cache with SRAM L1 and L2 caches.eDRAM isn't as dense as system DRAM, but it is still the same tech, but using a different process. System DRAM uses a different, higher performance process than CPUs and CPU cache usually do.[/citation]
Also, the main reason for DRAM's latency is because it is far from the CPU, not because it is actually THAT slow. It is a lot lower even with regular DRAM instead of eDRAM if you just put the chips next to the processor, but we can't get huge capacities cheaply that way and it would make upgrading system DRAM a lot more difficult. DRAM isn't as fast as SRAM, but it is very fast when used in situations as ideal as how SRAM is used. Remember, regardless of how fast SRAM is, it is right in the processor instead of several inches away and not even on the same board as the processor. Look at GDDR5, RLDRAM, and XDR2 memory. these are all very fast memory technologies that utilize DRAM and none off them are even still new. Heck, XDR2 is probably about as old as GDDR3 and I'm not sure about the age of current RLDRAM.
Give DRAM a superior interface such as XDR2 and RLDRAM and a cooler like what CPUs have and it can be VERY fast. Remember too, system memory isn't being given much power. We have those overclocking records with DDR3 that get minute timings at very high frequencies, I bet they have latency that is many times lower than usual system memory.
Don't count system memory as a basing point for DRAM's performance, it is not intended for high performance so it's not given circumstances that would allow high performance very often.
Facebook
Twitter
Instagram