Any thoughts on the impact of .065 and the eventual impact it will have on AMD's ability to throw 2MBcach/core on your average single core and 4MB on DC? AMD has made many revisions to the K8 but we have never seen how it performs with any more than 1MB L2 because such a chip has never seen the light of day. After upgrading to a claw over a NC I can tell you that I'd never go back to the NC and i'm thinking that the same could be said when doubling the cach again. Just pondering thought's of a suprise FX62 with 4meg's of cache is all.
I suspect that the place where you would see the most gain would be synthetic benchmarks. How many real world progs would fit in 2 megs, but not in one?
Since Intel is going to larger cache, the number may increase. Sometime later, it may be worthwhile.
This is really a good question... large caches, in effect, hide memory latency by decreasing the number of times needed to go to main memory. With an IMC, I don't think a huge benefit can be realized by increasing to larger cache.
The K8 are running on low freqfency, so the L2 vs DDR-400 ratio is low. In case of cache-miss, the K8 is not waiting so much before the needed data is transfered from the RAM and being available, meanwhile doing other jobs. Thats why the low latencies are very important for the K8, less latency, the faster the data will be available.
Anyway for gaming, mostly cache capacity is not doing much, same as for the multimedia and other streaming-data software. And for this kind of software K8 rocks.
More L2 will be needed if they raise the K8 freq, and I belive that will happen with the 65nm K8s. Or maybe if they reorganize the architecture and make it more issue CPU with more execution units. Than maybe the L3 Z-RAM(possibile) will also boost the overall performance.