Some thoughts about memory clock speed (800, 1066, 1333, 1600, ..) and
latency timing (CL7, CL9..).
About a year ago, in 2008, many people were saying that DDR2 was giving
a bettera "bang for buck":
http://www.tomshardware.com/reviews/ram-speed-tests,1807.html
On the last ("Conclusion") page, they said:
"Our conclusion is very simple: you get the best bang for the buck if
you stick to the mainstream of the memory market, which currently is
still DDR2-800 or 1066, preferably at low latencies. DDR3-1066 and -1333
memory do not yet result in better performance, and so should only be
considered by hardcore enthusiasts, who aim for maximum overclocking
performance knowing that they will get little benefit for spending a
fortune."
But this was prior to I7.
I read that the recent trend has changed, and many recommend going
with DDR3 for a performance system. (And DDR3 prices came down,
at least until this July's price surge.)
In this recent review they compare memory latencies to memory
clock speed:
http://www.tomshardware.com/reviews/core-memory-scaling,2342-10.html
they conclude:
"It is interesting to see that low latencies typically are more
important than clock speed bumps. This is not always the case on the AMD
Phenom II X4 or Intel's Core i7 platform. Both come with memory
controllers integrated into the CPU core, while the CPU controller is
part of the (X48) chipset in the case of the Core 2 Quad on Socket 775.
Apparently, latency seems to be much more of an issue on non-integrated
memory controllers."
And they recommend:
"Enthusiasts should go for DDR3-1600 and low latencies, while others will
get the best bang for the buck at DDR3-1333 speeds and low latencies.
CL7 timings can be considered ideal, but refrain from paying a
significant premium over CL8 or CL9 memory."
See also charts comparing faster processor vs. faster memory
performance:
http://www.tomshardware.com/gallery/RAM-Core2-scaling,0201--6464jpg-.html
However, personally I still won't go for the fastests CPUs.
HTH,
Igor
--
PDML Pentax-Discuss Mail List
PDML@pdml.net
http://pdml.net/mailman/listinfo/pdml_pdml.net
to UNSUBSCRIBE from the PDML, please visit the link directly above and follow
the directions.
