Author: Robert Hyatt
Date: 06:48:21 08/09/00
Go up one level in this thread
On August 09, 2000 at 04:52:28, David Blackman wrote: >On August 08, 2000 at 21:49:12, Robert Hyatt wrote: > >>The problem is that the CPU goes faster, but memory hasn't >>improved at all in 10 years. DRAM chips still take just as long to >> dump/rewrite today as they did 10 years ago. And crafty needs a >>fair amount of memory >>bandwidth, or else a big cache, which the 21264 nicely provides... > >Are you sure memory speed hasn't improved? I was seeing latencies around 300ns >to 1000ns in both cheap PCs and mid-range workstations in 1995. Now i see 100ns >to 200ns in cheap PCs. Not as impressive as the CPU core speed increase, but >still worth having. (Of course Crays were fast, even a Cray-1 was about 120ns i >think, but i never got to use them.) Since I have been watching, DRAM speed has been in the 100ns range. Remember that I mentioned DRAM access time... not all the other nonsense between the DRAM chip and the CPU. We were buying 60-70ns DRAM in the middle 90's. The Cray was using this same speed DRAM in the early 90's. Two things have helped to offset the non-improving memory speed: cache and wider bus transfers. Some memory technologies (IE SDRAM) is no faster on random access than its predecessor, but if you read a big chunk of memory, all of the 8-byte reads (except for the first) are faster. As there is some internal buffering going on that helps for large transfers. But that raw latency is still there. Another good comparison: Cray-1 took 7-8 clocks to read from memory. The XMP took this to 14 clock cycles or so, all the way to the C90 which took this to 25 clock cycles. If you take cpu cycle time, and then memory access time, you get a flat line for memory speed. As the cpu got faster, it simply took more cycles to read from memory. ie 7*12.5 (cray 1 clock) is about 14*8, the XMP time, which is about 25*4, the C90 time...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.