Raymond said:
Thank you very much for your response. Are you able to explain in
layman terms what the difference in cache size means? I saw that the
T7300 is 4MB and the T7250 is 2MB, but I don't understand what it means.
Thanks again.
Cache is temporary, very high speed memory, that keeps a copy
of recently used instructions or data. If the instructions or
data that are needed, are in the cache, then a trip to the much
slower DDR2 DRAM in the computer, is not needed. When the cache
is present, perhaps for every 10 accesses the processor wants
to make, 9 of them are answered by a cache inside the processor.
The system memory might only need to be accessed 1 time out of
the 10 total attempts. (Numbers were just made up - there used to
be an entire book in my company's library, with examples of
real numbers for caches of different size and associativity.)
In practical terms, a larger cache improves performance by a
small amount. For AMD processors, perhaps the difference between
512KB and 1MB of cache, might be 200MHz equivalent extra core speed.
In other words, if half the cache goes missing, then the processor would
have to run 200MHz faster, to produce computed results at the
same rate. You can see this, by comparing the AMD "P.R." rating,
between processors with different sized caches, but with the
same core clock speed. (
www.amdcompare.com)
I'm not sure right off hand, what a good conversion factor
between cache size, and extra clock rate needed to match it,
would be for Core2. We can try Tomshardware, to find out. I picked
three processors, that are close in terms of core frequency (second
column from left), but with different sized L2 caches.
SLA3H 1.80 GHz E2160 06 800 MHz 65 nm L2 = 1 MB LGA775
SL9TB 1.80 GHz E4300 06 800 MHz 65 nm L2 = 2 MB LGA775
SLA4U 1.86 GHz E6320 06 1066 MHz 65 nm L2 = 4 MB LGA775
In this benchmark (the first one that came up in fact), the
cache is making no difference. The results are clustered
together. That means the cache isn't making a difference.
There could be some code, a tight loop, that sits in L1, and
all three processors have enough L1 to do the job. Maybe there
isn't significant stuff stored in L2 while this benchmark is
running. Such a benchmark or activity, is "core limited". The
cache isn't helping.
http://www23.tomshardware.com/cpu_2007.html?modelx=33&model1=879&model2=883&chart=435
Winrar, on the other hand, is known to be memory intensive. Even though the
processors have (close) to the same clock rates, the one with the biggest
cache finished 37 seconds before the one with the smallest cache.
http://www23.tomshardware.com/cpu_2007.html?modelx=33&model1=879&model2=883&chart=434
Games can show similar patterns. Some have a higher dependence on
a big cache, than others. Some run no faster when on a processor
with the bigger cache.
Photoshop is pretty memory intensive, so it should also show some
effect from the cache. Tomshardware doesn't know how to run a
Photoshop benchmark, so you'll have to excuse their choice of
filters. The spread seen here, is actually much smaller than I
would have expected.
http://www23.tomshardware.com/cpu_2007.html?modelx=33&model1=879&model2=883&chart=437
Anyway, that is enough examples for now. People don't spend all day
compressing things with Winrar, so for the most part, you're not
going to miss the difference that much. It is when you have a
"grinding task", with a long run time, and a high dependence on
cache, that the difference in execution time starts to mount up.
Maybe it'll be when you are running DVDshrink perhaps. Something
that runs for hours. But for instantaneous desktop response, it
might not make that much difference. A slow disk drive might be
more of an annoyance, than the difference in processors.
Paul
