apache_lives :
Makes you wonder wether Intel's dual die quad core chip's have advantages because of 2 independent L2 caches, rather then AMD's share between 4 etc?
The Phenom has three or four independent L2 caches (one per core) and shares the L3 cache.
Besides, didn't you make a lot of hot air about Vista's NUMA awareness being a major advantage for the Phenom processor over Windows XP? That turned out to make squat all difference. How do we know all these visions you're making won't simply fall flat on their face as well? 4x4 was supposed to be great too, haven't seen anything from that in 18 months, guess that still has growth room for the future? I think AMD just realised it was a scrapheap and simply cancelled all development on the 4x4 platform. True potential indeed.
The Phenom does not use NUMA as it has a single bank of RAM, hence a unified memory arrangement. Only the multiple-socket, multiple-IMC, multiple-RAM-bank Opteron 2xxx and 8xxx use NUMA.
TechnologyCoordinator :
Standard deviation is used one using statistical data to project numbers onto an entire population.
However, when you are looking at benchmark numbers, which are reproducable and absolute, you don't have to worry about standard deviations.
Bottom line, the processor got beat in that benchmark.
The numbers are not absolutely reproducible and you DO have to worry some about standard deviations between benchmark runs. A well-set-up system should have minimal deviation between runs, but that is not always the case. I wonder how many runs of each benchmark gets performed by the HW review sites and what the SD, mean, and median values are for each run. I have a hunch it's only one run of each benchmark on many HW review sites.
TechnologyCoordinator :
True.
However, you are distracting from several different points.
#1 - You can benchmark CPU performance with games at lower resolutions
#2 - At low resolutions the CPU is the bottleneck
#3 - At high resolutions the GPU is the bottleneck
#4 - On desktop machines, in virtually all normal usage situations, INCLUDING MULTITASKING, FSB is NOT a bottleneck
[TC Meltdown]
HOWEVER WE ARE TALKING ABOUT FREAKING BENCHMARKING. GOT IT? SO IF AT A LOW RESOLUTION PROCESSOR A GETS 350 FPS AND PROCESSOR B GETS 250 FPS GUESS WHAT, ONE IS BETTER. DO YOU UNDERSTAND!?!?!?!?!?!?!?
OMG AGAHAHAHAHHAHAHAHAHAHAHAHAH!?!?!?!?!?!
[/TC Meltdown]
The OP for that comment probably meant to say "you do not need a quad to get playable framerates at those resolutions," Yes, the CPU is a bottleneck of top absolute performance at that resolution, but it may not be enough of a bottleneck to hinder gameplay.
amdfangirl :
True, but how many of us use Intel Core-based processors at 133Mhz FSB?
I do, a Core 2 Duo U7500 running on a 133 MHz QDR FSB. Not many Core-based CPUs run on such a slow FSB and no desktop units do but a few mobile units do and run decently enough, apparently.[/quotemsg]
kassler :
Yes, matlab is one type of application that will run better on Intel. The reson is that it doesn’t use memory. .
Hah! It takes 300 MB memory just to *start* the application on the Windows XP x86_64 machine I use at work. Let's just say that there is a very good reason that there is an x86_64 version of MATLAB and that I've run my machine with 8 GB RAM out of RAM several times working on some data sets. On a side note, that machine runs a Q6600 and I find its multitasking ability to be perfectly acceptable. I can run four instances of MATLAB to light up all four cores and it still chugs along nicely. The execution time of four identical calculations at once is pretty much the same as three, two, or one, so I think that my 1.5 GB data set per thread calculations both hit the memory hard and show that the FSB is not a bottleneck. We DO see a big bottleneck with some Xeon servers doing OpenMPI apps, where the scaling pretty much drops off all of a sudden in a "hit the brick wall" fashion but the Q6600 doesn't show this kind of behavior with what we're doing.