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?

For anyone who has difficulties understanding some of the issues, click here to enter the FAQ site.

The following is more or less work in progress, that is a sequence of findings and insights that led to some interesting conclusions. March 17, 1998

One of the spookiest topics on all bbs sites is always the question which one is the best .....? IMHO, there is never a best or a worst in a generalized sense, it always depends on what the environment is like and what do YOU expect from your equipment. There may be some exceptions to the rule but these exceptions can only apply to very simple applications or hardware.

Also, and this is most important, there simply is no equality, every single piece of equipment is different, even if they come from the same production line. A case in point is the system RAM which is one of the smallest components in the setup but has more impact on the overall performance of a system than e.g. the hard disk or the modem. Lately there have been quite a few improvements such as SDRAM which, even more recently has been added an EEPROM, performing the task of enabling "serial presence detection" (SPD).

The SPD is a small Electrically Erasable Programmable Read Only Memory (EEPROM) that contains the data sheet of the module in a [HEX] format according to the JEDEC specifications. Supposedly, the information contained within the SPD is read by the BIOS which then accordingly configures the state machines of the individual DIMM slots or else the global state machine of the system, that is the access latencies for the memory modules.

The question we tried to ask was whether there are differences between different DIMMs or not. It is difficult to tell unless one pushes the system really to the absolute limit, which is what we did by setting the memory and system bus to 83 MHz.

The setup: AMD K6 200 running at 208 MHz, everything else standard as described in the LAN solutions
The program: Memspd (here is the link). What memsp(ee)d does is it runs a number of sequential tests, addressing each MB of RAM and measuring the transfer speed from 16 bytes up to 8192 kB.
The candidates: Mitsubishi SDRAM 10 ns EEPROM (32 MB)
Micron SDRAM 10 ns EEPROM (32 MB)
both from Coast to Coast Memory

The results: In this pure measuring environment, both brands gave identical results, that is at 83 MHz:

• 435.0 Mb/sec = 2.4 ns/byte for the L1 cache = 176% of the main RAM
• 310.3 " 3.4 for the L2 cache = 125% "
• 246.9 " 4.2 for the DIMMs = 100% (reference)

At 66 MHz, these values dropped to:

• 418.2 MB/sec 2.5 ns/byte = 202%
• 263.6 " 4.0 = 127%
• 207.0 " 5.1 = 100%

So, no differences? Well the one striking thing running the test was that the Mitsubishi SDRAM gave identical values for each MB of RAM that was addressed in both DIMMs (3 µsec/KB transfer up to blocks of 64KB and 4 µsec/KB from 128 to 8192KB) whereas in the case of the Micron chips every single MB gave different readings, varying from 2 µsec/KB to 5 µsec/KB (this one only at 8192).

How does this translate into real life?
Quite simple.
With the Mitsubishi DIMMs, I achieved a Norton SI index of 98.5.
With the Microns, I achieved Windows protection errors (I hate to say it).

Also, both Mitsubishi Dimms gave identical results, whereas there was a lot of variability with the Micron chip-based modules running on the test system. That still does not mean that everybody else will have the same experiences and/or that every Micron component-based module behaves this way. However, it is well known that DIMMs with Micron chips obtained from sources other than the original factory or Crucial may not be as high quality as the original modules.

The same is true for any other brand, there are established vendors that only buy high quality bins, there are many other 3rd party vendors who are bargain-hunting on the spot market and who are further using the least expensive 4-layer PCBs. Considering the amount of headaches that a faulty or marginal DIMM can cause, it is not worth trying to save a few bucks here.

March 18, 1998

The one thing that came out of all those benchmarks is that the overall performance of all 4 modules was not only similar but actually identical. This is consistent with the idea that memory modules and the chips used are passive components that can only react, whereas the memory controller in the chipset is the factor that actually dictates the mode of operation and, therefore, the performance. In other words, the brand and rating of the memory does not really matter if the memory access parameters are set to identical values in the BIOS.

March 20, 1998

I just discovered something interesting. Last night, still brooding over the numbers I got from Memspeed, I decided to look at the Mitsubishi DIMMs again. The system had been up and running for several hours and so I did a strip reboot (Ctrl + F5) to measure the transfer rates again. Big surprise, with 64 Megs of SDRAM, everything went as expected, no fluctuations at all up to MB #63 which all of a sudden gave me extremely slow readings (up to 10 µsec per KB). In the next rerun of the test, the same thing happened but the slowdown already started at MB #62. Additional repetitions of the test showed progressively deteriorating performance, that is, on the next trial, slow-downs already started at #60, moreover, #62 and 63 didn't give any recognizable readings at all anymore. Each further run victimized another MB of RAM. Interestingly, after letting the system cool off overnight, today everyting was back to normal. Upon opening the case, I discovered that the IDE cables were partially shading one of the DIMMs which, thus, would not get enough ventilation. As a result, one of the modules was getting quite warm on one end....

Conductance is inversely correlated to temperature, which means that upon warming up of the chips, it is actually possible to see a dramatic slow-down of the transfer rates even on a system level. What that means is not that it doesn't matter which brand or quality or memory is used, but that even the best memory will be subject to temperature derating of its performance.

March 22, 1998

I straightened out the cable mess in the system, the UDMA ribbon cables were cut to the shortest possible lenght, giving each IDE data cable only the minimum length necessary. This way, it was possible to completely elimintate curling of these cables around other components within the case. The net effect is a much improved tidyness and most importantly, no more heat pockets caused by lack of ventilation. On the memory side of things, mo more erratic last MBs even after several hours of heavy use.

next page:    => Memory general FAQs =>