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:

LOSTCIRCUITS	SHORTCUTS: A new bus speed Heat and Hertz Test Configurations Cachemem, Sandra Everest and Speed Derating Sysnark2004 Internet ContentCreation Office Productivity WorldBench5 -1 WorldBench5 -2 3D Rendering 3ds max Cinebench2003 Futuremarks Gaming A-Z DOOM3 FarCry Conclusions and Paradoxes Give Us Some Feedback to Help Us Improve our Reviews
Intel Pentium4 3.46 Extreme Edition / 925XE chipset
(Review by MS November 16)

Memory Performance Part II

ASIDE FROM SANDRA, there are a number of different offshots of the original DOS Stream with emphasis on various parameters including prefetch and buffering algorithms. We pretty much know what Sandra does, however, the details of the different other benchmarks are still somewhat elusive. One of the newer benchmarks out there is Lavasys' Everest that, similar to AIDA, differentiates between READS and WRITES and in this case, the results are somewhat puzzling.

Everest Memory READ performance

In case somebody did not notice, the "old" 925X platform totally trashes the 925XE chipset, as long as a Prescott processor is used as opposed to the eXtreme Edition

Intel P4 Northwood 2.4 (hard to find)

Everest Memory WRITE performance

On WRITES, things are more uniform than on reads for the simple reason that prefetch by definition can only be done on READS

One thing that needs to be pointed out is that for READS, in both Sandra and Everest, the same processor when running at a lower multiplier, that is as a P4 550 (3.4 GHz), consistently outperformed the 560 (3.6GHz) speedgrade. There are several possibilities that could cause this, from bad software algorithms that incorrectly factor in the actual clock speed to other hardware issues. One thing to keep in mind is that memory READs predominantly stress the chipset and the CPU whereas WRITES stress the memory array itself.

In a normal scenario, that is, if the benchmark is written correctly, we would expect an increase in processor clock speed to be associated with an increase in bandwidth - at least as long as there is some kind of CPU bottleneck. In order to get to the bottom of what is going on here, we sidestep this review a bit to look at the memory bandwidth benchmarks as a factor of clockspeed of the exact same processor configured to run as 530 (3.0GHz), 540 (3.2GHz), 550 (3.4GHz) or 560 (3.6GHz).

Everest Memory Bandwidth vs. CPU Clockspeed

WRITES are not affected by clock speed. READS, on the other hand increase in bandwith with the CPU speed until they hit a cutoff at about 3.4 GHz - after which the performance drops rather dramatically

Sandra Memory Bandwidth vs. CPU Clockspeed

Sandra shows the same picture. Note that this graph uses a non-zero offset for reasons of clarity

So what is going on here? The most likely explanation is that there is another throttling mechanism in place that caps memory transfers as soon as the load on the CPU or chipset reaches a nominal maximum, At this point, we are not exactly in the clear with the interpretation of our findings but we will follow up with more details and explanations shortly.

next page:     => System Benchmarks: Sysmark2004 =>