Barcy's cache size question

[cubicleslave]

I don't post often, but I lurk a lot. There's one thing I've been wondering which hasn't been talked about (as far as I can tell). Intel's C2D Conroe contains 4MB of shared L2 cache, shared between 2 cores. But from what information I've been able to find, AMD's Barcelona will have a 2MB shared L3 cache, shared between 4 cores. This sounds rather small. Has anyone done any kind of comparative analysis of the effect of such a small cache, and it being L3 instead of L2? The closer to the CPU you can put fast memory, the better, so sharing at L3 instead of L2 is also bad, right?

 

[RichPLS]

Mostly it has to do with power consumtion thus heat generated not to mention die real estate... plus AMD just switched to a 65nm process which apparently is not as effecient as Intel's process.

 

[RichPLS]

also I read that Intel is exploring stackable memory sandwiched on the CPU with a terabyte/sec bandwidth... iirc.

 

[ajfink]

You DO know K10 has 512kb of L2 per core as well, right?

 

[RichPLS]

yes - 2MB total...

 

[benzene]

I also thought that AMD's use of an on die IMC meant that it was less reliant on cache and so didn't need as much. I could well be wrong mind, I'm sure someone will correct me if this is the case.

 

[RichPLS]

True, iow's... more cache does not sig help AMD... it does tho for Intel...

 

[qcmadness]

True, iow's... more cache does not sig help AMD... it does tho for Intel...

Exclusive cache + IMC vs Inclusive cache.

 

[Eviltwin17]

its not the size that counts, its how you use it lol


intel has always been on this more = better mindset

amd has always looked for different and better approaches than intel in the past


AMD is always looking for more efficient ways to solve problems, thus the core2duo was born because intel learned from this. Now all they have to do is get rid of that aging front side bus and theyll be golden

 

[piesquared]

Mostly it has to do with power consumtion thus heat generated not to mention die real estate... plus AMD just switched to a 65nm process which apparently is not as effecient as Intel's process.

Incorrect. AMD's 65nm tech is superb.

 

[RichPLS]

Is that why changing to 65nm from 90 there was a performance hit...

 

[piesquared]

Is that why changing to 65nm from 90 there was a performance hit...

No that is not why.

 

[Viperabyss]

Mostly it has to do with power consumtion thus heat generated not to mention die real estate... plus AMD just switched to a 65nm process which apparently is not as effecient as Intel's process.

Incorrect. AMD's 65nm tech is superb.
AMD's 65nm tech is no doubt superb. but comparing with Intel's 65nm, AMD's process technology is still inferior. there are tons of problems associate with AMD's 65nm technology.

for one all 65nm X2s have trouble booting when cooled below freezing temperature. I'll find the links later.

another is that, although arguably, AMD X2s have problem reaching above 3.0Ghz. X2 3600 Brisbane only has the maximum overclocking potential of 3.1Ghz. however you can also attribute that to architectural problem.

 

[qcmadness]

Is that why changing to 65nm from 90 there was a performance hit...

In fact a small redesign is needed.
Since K8 is based on 130nm process, for 65nm process, some small tweaks should be done. But AMD now focuses on NGA and other projects.

 

[RichPLS]

they just do not have the resources...

 

[qcmadness]

they just do not have the resources...

I would say that AMD's 65nm is not good compared to Intel. But as we know, only a few companies have 65nm fabrication technology in their fabs.