Do you have a question? Post it now! No Registration Necessary. Now with pictures!
November 9, 2006, 8:06 pm
rate this thread
Re: AMD vs Intel
Charts are here:
Not all processors are in the chart. To make up relevant numbers for
the missing processors, use ratios with respect to processors of
a similar architecture.
T5500 = 1.66 GHz/FSB667/2MB L2
E6400 = 2.13 GHz/FSB1066/2MB L2
First off, Intel's objective for the Core 2 Duo family, was
to use the same core for a number of products. Thus the
internal architecture is the same. To a first order approximation,
performance is only dependent on core speed. If you go to
the Tomshardware site, take numbers from the charts for E6400,
scale the results by 1.66/2.13 or about 0.78 times. You can refine
your analysis a bit, by looking at the trend for all the Core 2
Duo processors in the charts, as some benchmarks are really tied
to the motherboard, rather than to the processor. So you want
a bench that scales well with processor speed, so you can compare
just the core performance.
Does FSB matter ? Does my DDR2 RAM choice matter ? In an absolute
sense, yes. Probably a couple percent here and a couple percent
there. Maybe certain applications really get a kick from the FSB.
Maybe others won't. You can probably spend a couple days staring at
those charts, looking for inspiration.
OK, let's try an experiment. I'll start with this chart:
Sisoft Sandra 2007 Arithmetic ALU
The first thing we check for, is scaling. Is the benchmark scalable
with CPU speed ? We pick two processors (Conroe) with same FSB and cache.
We'll use the E6700 and the E6600. 2.666/2.4 = 1.11 24617/22038 = 1.117
Since the clock speed and the benchmark have almost exactly the same ratio,
the benchmark scales well with CPU. Therefore, we won't feel too guilty,
extrapolating the results using this chart.
E6700 2.666GHz/FSB1066/L2 4MB 24617
E6600 2.4GHz/FSB1066/L2 4MB 22038
So now that we proved the benchmark scales linearly, we take the
benchmark for the E6400 and multiply by 0.78 to get the speed of the
T5500. 19562 * 0.78 = 15258. Now we compare this to the Windsor
3800+ AM2 processor with DDR2-800 memory - it gets 14572. The
T5500 wins this one by 15258/14572 = 1.047 . Maybe we can assign
the T5500 a bit of a penalty, as after all, its FSB is actually
lower than the E6400 we used for the extrapolation.
In other words, the processors are nearly identical.
Of course, you can refine the analysis any way you choose, as there
are plenty of charts to use as raw data. Each benchmark is measuring
*something*, and your mission is to figure out what that might be :-)