|
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 |
LOSTCIRCUITS
|
|
| Intel Pentium4 600 Series | |
|
(Review by MS February 21, 2005) |
| Intel P4 630+ At: |
Memory Latencies
Cachemem 2.65 measures the latencies encountered by accessing different size blocks across the memory subsystem (including cache) using increasing stride lenght to generate offsets of the starting address. Increasing strides will eventually hit page boundaries during transfers and, thus, incur higher latencies. Increasing block sizes will either fit into the different levels of cache or else generate the need to access the main memory. In this case, we used 3D plots of the stride x block size matrix for a direct comparison of the older Galatin-based Extreme Edition with the new 600 series at 800 MHz and 1066 MHz each.
For illustration, we have plotted the different cache and memory levels of the Galatin core: red: L1 cache; yellow: L2, turquoise: L3; pink: main memory. By defining different block sizes, the use of one or the other part of the memory system can be forced. In this case, we are showing actual processor cycles as metric.
Latencies in [ns]. ExtremeEdition vs. 660: in general, the ExtremeEdition (solid columns) has lower latencies (better) than the P4 660 (transparent blocks). Keep in mind that the chipset and memory latency settings are identical, meaning that what the differences we are looking at are the latencies of the memory management unit within the processor.
Same as above at 1066 MHz bus. Note that the latency differences within the main memory block are fairly constant, which excludes the memory itself from being the culprit.
In the graphs above, it is somewhat difficult to appreciate that there is more to that story as well, namely, the older ExtremeEdition (blue and turquise) runs faster at longer strides but the Prescott MMU (pink and yellow) appears to be optimized for strides of about 128 Bytes length where it accesses the main memory within 38 ns as opposed to the 67 cycles needed by the older core. This could play an important role in some of the applications where the Prescott core shows muscle.
Let's see how the memory profiles translate into real world performance.
|
Intel P4 Northwood 2.4 (Clearance Sales?) |
next page: => WorldBench5 =>
All advice and educational articles on LostCircuits are free, but if you feel you can, please make a small donation to us!