cpu cores

[atc_traffic856]

What makes one core of a cpu better or different than another one?
Example Celeron and Smithsfield - is it the quality of the tranistors or the proximity of the transistors/ thickness of the wall structures??
Thank you>>

[hockey man]

First off, the Celeron is not a Core. It is the general product name for a CPU line. Smithfield is a Pentium D core. Obviously, I don't know as much as Saphalline does, but this is what I do know:

Smithfield
64 bit Support: Yes
Cooling Device: Heatsink and Fan
Hyper-Threading Support: No
L1 Cache: 12KB+16KB
L2 Cache: 2 x 1MB
Multimedia Instruction: MMX, SSE, SSE2, SSE3
Process Type: 90 nm

Celeron D -Prescott
64 bit Support: No
Cooling Device: Heatsink and Fan
Hyper-Threading Support: No
L2 Cache: 256KB
Manufacturer Warranty: 3-year warranty
Multimedia Instruction: MMX, SSE, SSE2, SSE3
Process Type: 90 nm

So, first and obvious things. 64 Bit support. Smithfield has it, Prescott does not. Niether off then use HT, however, that is dying so no worries. L1&L2 Cache- very important. More is better. That allows the CPU to have faster access to frequently used data. Instructions are ways that CPUs think, I'm not too familar with any if those. 90 nm is thinkness of the scilicon used- I think. After that Clock speed is fairly strait forward. Then you have Un measureable core things. Certian cores are just better at different tasks. That is why you have to check benchmarks.

[saphalline]

Check out our new website (www.sheaiden.com) for the raw stats on all the core revisions. I'm planning on expanding the charts to include TDP and max heat dissipation and all that, but for now it's still good info.

In terms of the transistors, those have changed over the years, mostly by getting smaller. SOI and strained silicon are the hot things nowadays, but there's no "quality" difference between the transistors in a Celeron vs a Pentium D (if a transistor doesn't work then the CPU won't work and it's thrown out before it hits the streets). The main differences between core revisions are in the design of the microarchitecture, circuit layout, and the features.

For instance, the Celeron D is kept behind the performance of the Pentium 4 600 series based on FSB speed (533 vs 800), core speed (max of 3.2GHz vs max of 3.8GHz), L2 cache size (256KB vs 2MB), and in HyperThreading (Celeron D = no, P4 600 = yes). That's about it. Otherwise, they both sport a modified version of the 90nm Prescott core revision (which is based on NetBurst 2 with a 31-stage pipeline) and have roughly the same circuit layout (with the P4 600's having a larger die size due to the extra L2 cache).

Both Intel and AMD have their little tricks for keeping the "value" CPU's behind their main "flagship" CPU's. Mostly they just deactivate stuff to keep them down. The most popular way to do this is to give less L2 cache to the value CPU's, but a slower FSB is also quite popular (in the case of AMD's K8-based Semprons they only have a single-channel onboard memory controller). And of course you never want the value CPU's to get anywhere near the raw clock speed of the flagship CPU's, because then what's the point of getting a more expensive CPU?

[atc_traffic856]

Thankyou Spl for your time, and super web site> Since I was away for awhile - 90nm is the distance between the Tranistors??
Thank you again>.

[saphalline]

90nm is the average distance across one transistor plus the distance to the next one. This is just an average, though, and is not to be taken as the absolute measure of a process. Some transistors are closer together, some are farther apart, some are smaller, some are bigger. CPU cores are not just a stark well-ordered grid - their designs are mostly science but include a healthy dose of art as well. Afterall, if it was all science, then computers would have no need for humans! :p