MemTest86 General Observations.

[Jim]

This program apparently does not run under Windows. In some regards that is a good thing, because it allows you to experiment w/ aggressive Bios settings without screwing up your Windows install.

At the same time, however, it means that the testing environment is not "Real Life". The programs that I use to tweak my machine DO run under Windows. That means when I am running MemTest 86, the tweaks are not applied. The RAM therefore is less stressed while running Memtest86 than it is in normal operation.

I just ran MemTest86 on Superpuppy 2, (w/ its 768 Meg of so-so RAM), and despite using the most aggressive Bios settings that the P5A-B Rev 104 has :
Speed = 7ns (143 MHz)
SDRAM CAS Latency = 2T
SDRAM RAS Precharge = 2T
SDRAM RAS to CAS Delay = 2T
Enhanced Page Mode Count = Disabled - (That one is always trouble)
Internal Page Detection = Enabled
SDRAM Pipe Function = Enabled
SDRAM x111-2111 Mode = Enabled
IO Recovery Time = Disabled

The PC133 CL3 - PC100 CL2 "Junk RAM" in Superpuppy 2, (Which crashes my DFI w/ WP Credit settings enabled), passed MemTest86 @ 100 MHZ without a hitch.

Am now heading over to Uptown to see if I can test my Hynix @ 133 MHz on one of their machines.

[Jim]

Further on "MemTest86+ v1.65" After a week of watching paint dry I have gotten some interesting and conflicting results. The tests were conducted on a variety of boards for reasons which will become apparent.

Board # 1: Via 733 MHz w/ Onboard everything including processor @ 133 FSB w/ SDRAM timings set @ 3/3/3/6 : Test = Memtest86 v3.2

Tested : 2x CL3 PC133 256 Meg sticks of Hynix HYM71V32635HCLT8-H (wherein the "L" stands for "Low Power" and the "-H" stands for CL2 @ 100 MHz (2/2/2/5) & CL3 @ 133MHz (3/3/3/6)
Passes : 1. Total errors = 3.

Board # 2 : FIC VA_503+ Rev 1.2 w/ K6-2+ 550ACZ @ 4x120 MHz w/ SDRAM timings set variously in an effort to pass test : Test = MemTest86+ v1.65

Tested : 2x CL2 PC133 256 Meg sticks of Hynix HYM71V32635HCT8P-K wherein the "-K " stands for CL2 @ 133 MHz.
Passes : Several - all terminated by me on account of excessive errors. Also tried varying processor voltage which reduced the number of errors; but did not eliminate them; - best result @ 2.4v.

Tested : 2x CL3 PC133 256 Meg sticks of Hynix HYM71V32635HCLT8-H
Passes : Several - all terminated by me on account of excessive errors. Best results @ 2.4v.

Board # 3 : "A OPEN MX6B EZ" w/ P3 533 @ 4x133 w/ SDRAM timings set @ 2/2/2/? : Test = MemTest86+ v1.65

Tested : 8x CL2 PC133 256 Meg sticks of Hynix HYM71V32635HCT8P-K
Passes : Ranged from 1 to 15 per stick. Total Errors = 0.

Tested 4x CL2 PC133 256 Meg sticks of Hynix HYM71V32635HCT8-K
Passes : 1 per stick. Total Errors = 0.

Tested 2x Infineon CL2 PC133 (Part # unknown)
Passes : 1 per stick. One stick passed w/ 0 Errors, the other repeatedly froze the machine.

The following tests were run w/ The SDRAM timing settings @ "By Spd"

Tested 2x CL3 PC133 256 Meg sticks of Hynix HYM71V32635HCLT8-H

Passes : Several - all terminated by me on account of excessive errors.

NOTE : These two sticks of RAM were; and possibly still are the best two I possess **. The problem here is that this type of board has a RAM identification feature built into the bios which persistantly misidentifies these sticks as CL2.@ PC133. This can be overridden w/ some of the settings, but not all of them. This same RAM identification feature misidentifies the Infineon CL2 PC133 sticks as CL3 PC133, which again can be overridden w/ some of the settings; but not all of them. This results in the Infineon getting a "Soft" test, while these two sticks of Hynix get a "Hard" test.

** Note : These two sticks are the only sticks that I possess that : 1) allow enabling 4 bank interleaving w/ WPCredit; & 2) allow using the "Any bank" setting in WPCredit; - which gives a big gain in "Memory Write" speed. I think the reason for this is that they are "Low Power" SDRAM which I have been running @ "Normal Power". - All my other RAM is "Normal Power".

Tested 1x stick of CL3 PC133 256 Meg Hyundai
Passes : 2 - complete miserable failure @ 133 MHz.

Tested 1x stick of CL3 PC133 256 Meg Micron
Passes : 1 Errors = 0

Tested 1x stick of CL3 PC133 256 Meg Infineon
Passes : 2 Errors = 0.

Tested 2x sticks of CL3 PC133 128 Meg Infineon
Passes : 1 per stick. Errors = 0.

I then decided to retest one of the 8 sticks of Hynix CL2 PC133, (HYM71V32635HCT8P-K), in a Super 7 board. Since the DFI can only be overclocked through software which runs under windows, my DFIs could not be used. My FIC seems to be in the early stages of the "capacitor problem", so I examined my ASUS P5A-Bs carefully.

Board # 4 : ASUS P5A-B Rev 104, (Good Caps), w/ K6-2+ 550ACZ @ 4x120 @ 2.3 v. w/ SDRAM timings set variously in attempt to pass test.

Tested : 1 CL2 PC133 stick of Hynix HYM71V32635HCT8P-K
Passes : Several - all terminated by me on account of excessive errors, inspite of trying various settings.

CONCLUSION : MemTest86, in all of its various versions, is a poor tool for testing RAM on OVERCLOCKED Super 7s. WHY? Answer : Because Super 7s do not have sophisticated memory timing settings available in the bios. To fine tune a Super 7s memory timings, you have to use software that only runs under DOS or Windows. e.g. 1) WPCredit; e.g. 2) Tweak Bios.

Because of this, you cannot make the appropriate memory timing adjustments needed to run in the 120 - 124 MHz range. See SDRAM timing specifications required for 125 MHz in Hynix technical specifications download.

That said, it may be possible to use nk's "Bios Patcher" to alter the bios to include the necessary settings.

[Uranium235]

Interesting observations Jim. I, myself am constantly trying to understand why hardware responds in certain ways.
One thought I have regarding your tests is that your RAM may not be the only culprit for the errors you get. Memtest86+ is also testing the CPU's cache, the motherboards L2 cache and the chipset's memory controller.

Here is an article regarding memory diagnostics that you may find interesting: http://www.realworldtech.com/page.cfm?A ... 2001232443
Dean Kent got good results with the original Chris Brady version Memtest86.
I wonder if you would see similar results with Chris Brady's Memtest86 3.2 release.
http://www.memtest86.com/

I've found that the Microsoft Windows Memory Diagnostic is a good quick-and-dirty test to run for redundancy purposes.
http://www.majorgeeks.com/download3955.html

I've always had good results with Crucial PC133 RAM, my personal favorite.