Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
CPU Intel P4 840 D P4 820 D P4 630 P4 640 P4 650 P4 660 P4 670 AMD Athlon64 3500+ 3700+ 3800+ 4000+ X2-3800+ X2-4200+ X2-4400+ X2-4600+ X2-4800+ 1-Way Opteron Opteron 144 Opteron 146 Opteron 148 Opteron 150 Opteron 152 2-Way Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Opteron 248 Opteron 250 Opteron 252 2-Way Dual Core Opteron Opteron 270 Opteron 275 nVidia GF 7800GT GF 6800GT GF 6600GT ATI R X850 XT PE R X850 XT R X800 XT PE R X800 XT R X800 XL Memory Corsair Crucial Kingston Mushkin OCZ What are you shopping for?

Monkey Wrenches For Performance

First, Adobe software, which was primarily used for the benchmarking, has had historically problems with the allocation of virtual memory address space in a Windows environment. One known bug is the well documented launch failure of Photoshop 4.0 in system memory configurations above 1 GB total memory. There are more examples, like conflicts between MS Word and Photoshop in older versions that have finally been weeded out but suffice it to say that there are still lingering software issues with the translation of the virtual memory address space into physical addresses.

Second, many OEM boards are known to insinuate a throttle in higher system memory configurations in order to overcome the increased load on the address and command bus that will lead to clock skew and the associated more sluggish behavior of the system memory. The easy workaround for the manufacturers is to increase latencies on the chipset level, including but not limited to turning off PAT or else simply turning off the "Dynamic Mode". The benefit is that the systems will still run, even if lower quality DRAM is used. This is one of the main differences between boards sold in the OEM market and those used in the enthusiast community aftermarket.

To the best of our knowledge the first report looking into the effects of high system memory density configurations was posted on 3DExtreme several months ago, shortly after the launch of the Canterwood chipset.

Third, as specified in chapters 5.2.7 and 5.2.7.1 of the Intel 82875 Memory Controller Hub data sheet, the MCH provides Memory Thermal Management that "selectively reduces reads and writes to the DRAM when the access rate crosses the allowed thermal threshold.

Read and write thermal management operate independently and have their own 64-bit register to control operation. Memory reads typically causes power dissipation in the DRAM chips while memory writes typically causes power dissipation in the MCH."

Both can be monitored via separate mechanisms and trigger throttling either via software forced thermal management or else a counter.

In other words, there are three different possible issues, each of which could potentially be held accountable for throwing a monkey wrench into the performance gears which are:

• Software problems
• The use of an OEM system with "safe" BIOS settings
• Thermal management of the MCH

It is interesting to note here that the memory used for the high system memory configuration were Corsair modules that, in addition to the memory modules have activity-LEDs built in. Those LEDs and their associated circuitries by themselves increase the power draw of the modules in a load-dependent fashion. In other words, the more power is consumed by the memory modules as a function of activity, the higher will be the additional power overhead. Therefore, in fact, we have four different parameters that could account for the observations.

Of course, it is easy to throw out speculations and hypotheses, they would not be valid, though, if there were no data to back up the claims. Since the particular software used is not available as free download, we'll have to leave it at that and resort to standard benchmarks to see whether there are any dramatic performance hits associated with the move to 2 GB system memory. We will have some results with Photoshop towards the end of this article as well that yielded some rather surprising results.

next page:    => Test Setup, Sisoft Sandra =>