Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ Search Prices:

It appears very unlikely that the extra L3 cache has anyy impact on streaming memory performance, however, as always, the proof is in the pudding.

Q.E.D.

Cachemem 2.65 MMX

With respect to the Extreme Edition of the P4, Cachemem is certainly the most interesting synthetic benchmark in that it should clearly show where memory traffic can be absorbed by the L3 cache rather than having to be routed all the way to the system memory. The cut-off block size is naturally expected to be 2 MB since that is all that would fit into the L3 cache. In other words, the main differences between the P4 3.2 and the P4 EE are expected with block sizes larger than 512kB and up to 2048 kBytes.

Access latencies, lower is better. It is no surprise that the standard P4 and the P4 EE are performing on par in the low and high end of the block size range, however, the 1024 and 2048 block transfer sizes show the EE (solid blocks) way ahead of the standard Northwood (transparent blocks).

From those numbers, it is easy to predict that all applications that use data blocks smaller than 2 MB and larger than 512 kB will see a sizeable boost in performance, prime candidates are some 3DGames with relatively uncomplicated geometry data. Office applications are expected to see a boost as well, despite the fact that most of the data will fit into the L2 cache, the differences will show in multitasking environments where the overall data flood will compete for limited space in the 512 kB L2 of the Northwood core. It is also predictable that non-recurrent data as they occur in e.g. video encoding (depending on the application) will not see much improvement for the P4EE, likewise, CAD applications with their huge amount of geometry data could exceed the size of the L3 cache, which means that they will not benefit from the extra L3.

next page:    => Winstones =>

All advice and educational articles on LostCircuits are free, but if you feel you can, please make a small donation to us!
Thank you!