|
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's P4 820D and 670 More Power to Duallies | |
|
(Review by MS September 3, 2005) |
| P4-670 |
Synthetic Benchmarks
Faster (left) vs. Dual Core (right)
EIST
The P4 820 which is already running at the lowest multiplier supported by the Prescott core design can naturally not be further clocked down under idle conditions. The P4 670 on the other hand should drop its clock speed from 3800 to 2800 MHz (19x => 14 x multiplier) if idle signals are asserted. We used PC-Wizard 2005 to monitor the clock speed. In the case of the 855 chipset and the Pentium M, this utility shows the changes in multiplier and bus speed very reliably, in the case of the 955XE chipset and the P4 670, all we saw was a fluctuation between 3805 and 3785 MHz, consistent with Spread Spectrum Modulation but not with any throttling. The power measurements we performed (see later in this review) also coroborated our idea that no clock reduction happened on the test platform at hand.
Synthetic CPU tests are somewhat irrelevant, on the other hand, they still give some sort of idea about the general horse power of a given processor. In most cross-platform comparisons, those numbers are meaningless altogether but for reference within the same platform, the numbers are at least worth taking a quick glance at:
SiSoft Sandra CPU Multimedia Benchmark
The SiSoft Sandra CPU Multimedia Benchmark clearly takes advantage of both SMP and HyperThreading. Note that under those conditions the Athlon64 FX57 - probably the fastest gaming processor out there - comes in dead last. Note further that overall, the P4 670 and the 820 perform head to head.
Memory Bandwidth
The SiSoft Sandra CPU memory bandwidth benchmark shows that centering the selling argument for the new Intel chipsets on the higher potential memory bandwidth of DDR2 is rather absurd. Granted, the memory devices themselves can achieve higher bandwidth, on the other hand, similar as in the case of the ill-fated Intel 820 "Camino" chipset, the processor host bus is the limiting bottleneck which is exacerbated by the overhead incurred by arbitrating between two cores. In other words, in the final analysis, the total bus utilization is only in the order of slightly over 50% (of 8400 MB/s on the level of the memory bus proper).
Memory Latencies
We are only comparing the 820 D (solid blocks) to the 840 EE (transparent blocks) processor here and what is interesting is that apparently it is not only the cache that is running at clock speed and, by extension, slower but also the main memory accesses are burdened with higher latencies. Since nothing else was changed, the source of these higher latencies needs to be looked for in the slower clock speed of the CPU-internal memory management unit and/or the bus arbitration that is handled on the level of the CPU as well.
|
Pentium4 820 D (dual core) |
next page: => WorldBench5 (I) =>
All advice and educational articles on LostCircuits are free, but if you feel you can, please make a small donation to us!