Experiences with A1 XE and Memory DIMMs

I'm not posing a question or asking for help here, just talking about recent experiences. They may have no relevance to others' experiences or they may strike a chord. But it might help someone to hear what I've found out about my system.

Firstly, I have never been able to run two DIMMs at the same time in my XE. The symptoms have usually been that U-Boot seems to run OK but as soon as the OS (Linux or OS4) takes over and enables both DIMMs, the hardware starts giving errors, hangs or generally won't run reliably. [U-Boot only enables one of the DIMMs if it is 256 MB or more] I know this is not the same that other people experience.

I have had until recently a single 256 MB Legend DIMM, but have always wanted more, like about 512. But I can't use 2x256 MB and all the 512 MB DIMMs I've tried before have not got past the U-Boot stage.

Recently I decided to put some time into the problem, and to document the limitations if possible. I bought a pair of new 512 MB DIMMs so that I had four altogether:
1x512 MB Legend (ECC, registered);
2x512 MB Infineon (plain, non-registered);
1x512 MB Samsung (plain, non-registered).

Advice to eager experimenters: DON'T TOUCH the settings unless you know what you are doing. It is easy to make a mistake and cause a disaster.

My old 256 MB DIMM was operating at 133 MHz FSB, and my G4 CPU was operating at 800 MHz as it has been for several years. Some years ago I put a decent heat sink on it and ran it for a few days at 1066 MHz without failures or crashes, but put it back to 800 for beta-testing and it has remained there ever since.

To cut a long story short, I tried all the 512 MB DIMMs again and found that all would function properly if I dropped the FSB (and memory speed) from 133 to 100 MHz (you do this by changing J13 from 1-2 to 2-3). I could even run two at once, giving 1 gig of memory (not that I need that much). I had to alter the CPU clock multiplier switch from x6 to x8 to get my 800 MHz back, but that was no problem.

However, when I tried to increase the CPU speed above 800 MHz, I had no success at all, it just would not boot. U-Boot would run OK, Kickstart would load, but it would not get past the "now launching Kickstart" display and the WB would not appear. Even going back to my old trusty 256 MB DIMM gave the same result - if the FSB is 100 MHz, it can't run with a multiplier of more than x8. If the FSB is 133 MHz, it will run with a multiplier of x8.5 (1066 MHz).

I tried reverting to Update4 (just to be sure that it was not a "new memory system" problem, but same result, so it's not something new.

I tried going back to the old 256 MB DIMM and making sure that I could still get up to 1000 MHz, and found that the CPU would not even work at its rated 933 MHz. The core voltage (previously set to 1.85 volts) had fallen to 1.80 volts with age. I cleaned around the regulator and worked the switches on/off a few times, but eventually I had to set the switch pack to a nominal 1.89 volts in order to get the right 1.85. I'm not happy about that setting, it might suddenly start giving out 1.89, but it's not a great overload if it does. Anyway, I could run the CPU at 1066 MHz again with that core voltage, so that has not changed over the years.

Leaving the core voltage at 1.85, I rechecked the 512 MB DIMMs, but the limit of x8 at 100 MHz FSB seems to remain.

So it seems that (for my XE at least):
1) I can use two memory DIMMs if I reduce the memory speed to 100 MHz;
2) I can use 512 MB DIMMs by reducing to 100 MHz;
3) When the memory speed is set to 100 MHz, my G4 will not run with multiplier settings greater than x8 (800 MHz);
4) One of my 512 MB DIMMs (the Samsung) will run at 133 MHz FSB and 800 MHz CPU speed, so I have left it in that state.

Again, if anyone feels like trying similar tests, be very careful how you go about it. It is easy to damage DIMM sockets (they are designed for only a few insertions in their lifetime, not hundreds). It is also easy to make mistakes in DIP switch settings, mistakes that might damage your CPU. If you are going to check core voltage, you need an accurate meter and very steady hands so that you don't short anything. Don't touch things unless you know what you are doing. Always switch off the machine at the power input, not just the start/stop button. You want all voltage to the motherboard killed while handling DIMMs. But don't pull the plug out of the wall, you want to leave the computer earthed while you are near it.

I would be interested if anyone else has made any such tests recently, and with what results.

@tonyw

A wild guess: maybe some initialization of articia is need to allow it to resolve the connected ram and the needed timings?
PC bioses have lot of customization settings related to the memory bank setting and theoretically it should be possible to use two identical sticks of 1/2 GB simultaneously. Like it was said that built-in audio on A1 is broken until the correct register were set up. The problem is that there's no support from manufacturers/designers of articia

Jack

@tonyw

Have you ever tried running any 2 x ECC of any size? I seem to recall that you can expect trouble if running mixed, also seem to recall that the ECC should be placed in a specific slot (can't remember which). I have 2 x 512 ECC and never had a memory related problem.

ace

@acefnq

Hi Ace, no, I only have the one left. I bought two but sold one to a mate and it works well in his G3 XE.

I can't help wondering if all those people out there with two modules running OK received their XEs with the FSB set to 100 MHz, and have left it there. It might explain why some people have a lot of trouble with DIMMs and some people have none.

I couldn't see any apparent difference in performance between 100 and 133 FSB. Presumably the few apps I tried manage to fit their "fast" bits into the CPU caches.

I don't think there is any difference between the two memory slots, I would be very surprised if an ECC DIMM behaved differently in one of them. ECC is not supported by the motherboard, anyway.

@tonyw

Have you seen this?:

http://amigans.net/modules/newbb/view ... ost_id=7531#forumpost7531

I have using my A1 with those two 512MB sticks inserted for a month, 133MHz FSB and 933MHz CPU (same that I had before, no changes) without a problem, uptimes up to 5 days.

Saluditos,

Ferr?n.

@Ferry

So the question remains: are your and my A1s different, or are the DIMMs different? No one has yet compared two A1s side-by-side with DIMMs that work in one but not in the other, AFAIK.

I keep thinking that there must be a marginal hardware problem in there somewhere, but I don't have expensive enough equipment to go looking (My CRO only goes to 100 MHz) and I'm not confident that you could improve matters anyway.

Investigations are not helped by the complete lack of hardware diagnostics. In order to investigate such problems, we would need a proper memory tester, CPU tester, DMA tester, interrupt tester, PCI tester, etc.

You can't run diagnostics at the same time as an operating system and expect an exhaustive test. A memory test program has to check every part of memory, not just a random block that the OS thought it could spare. A memory test program should be run from an external box that connects to the CPU socket. If you have to run the test program on the CPU itself, it should test a block of memory at a time, relocate itself to another part of memory space and retest from there, until it has tested all parts of memory. When I used to design hardware and write such diagnostics back in the 1970s, memory was small (a megabyte was unheard of) and a full memory test on 64kB of RAM would take an hour per pass (CPUs were slow then, too ).

You could re-write the GPL memtester to boot from U-Boot and run stand-alone, but it would be a big job and you'd get no thanks for it. I've thought about it in the past but the need is not great enough.

@tonyw

I think the trick with getting pairs of DIMMs to work is to be sure to use "Registered" DIMMs. It is my understanding that these have more buffering and present less of a load to the memory controller.

I loaned a pair to Atheist at our local Vancouver user group and haven't heard back if he got them working together. I will have to torture him on IRC to see what the deal is.

The registered DIMMs have some extra chips between the edge connector and the memory chips themselves. They are a bit "taller" than non-registered DIMMs.

The ECC registered DIMMs that I have here are a CT32M72S4R7E. These are Crucial 256MB ECC CL2 chips. I have the memory but no motherboard. Wanted an XE but never had the $$ when they were out.

@gregthecanuck

I've tried a couple of registered DIMMs in the past and I still have one here, but I don't see any difference between its "readiness to work" and that of a non-registered version. As usual, others may have different experiences.

@tonyw

Gday tony i also have 2 X 512 ecc registered ram moduals running in my A1
also Michaels A1 is running 2 X 512 ecc registered ram

both of us have exact same ram up and running

This is the memmory chips installed in A1

double sided MicroTechnologies memory stick.

MT18LSDT6472G-133B1?? PC133R-333-542-B1
US?????????????? S029BGZ????? 200236N
512MB, SYNCH, 133MHz, CL3, ECC
??? (BARCODE IS HERE)
2077-063

the chips on the sticks are - 0234??????? 1-1
????????????????????????????? MT?? 48LC64M4A2
????????????????????????????? TG??????? -75 B

@poweramiga

I have the 256MB CL2 version of that DIMM.

The chips are MT 48LC32M4A2-7E B.

I had loaned out a pair of IBM 512MB ECC registered DIMMs to Atheist but couldn't get a coherent answer out of him on IRC.

Is this info remotely useful to anyone?? I'm tired - off to be for me.... :)

I changed from 800 to 933MHz with a 1x 512MB Kingston stick
and just had a look at the memory speed and it is at 133, never changed that.

Edited on top sorry:
Very true that differences have been made just eg. the A500... Nothing worked in mine with "green-orange" lights after the third change from the shop, maybe the rev?DD???? nowbody knows.. Everything worked with the original "green-red" lighted with my friend. Now we are in spec's and numbers. Not so very happy situation. Maybe Eytech & the other company(s) tried to make them better as they were comming out (as they did?)

What About a new Topic about" What is You'r A1 nro...
Ok. I'll try it after a cig;)

@tonyw

Quote:

Very true, such a test would bring some light to this obscure problem with mem sticks and A1, or at least some clue...

Quote:

It is supposed that the makers did all the above, isn't it? ;?)

Quote:

I have even found different results with different versions of memtester, so go figure... :?/

Quote:

The first computer I used was a Vic-20 with 5KB, so I know the feeling ;?)

Quote:

That's why I did Real-World? tests, like many rounds of copying files to RAM: (different sizes), checksuming and deleting them, again and again, in order to find any error. Of course, I don't expect it to be a totally accurate test, but at least it has been intensive and filling as much mem as possible. As I said, no problems so far, but I keep an eye on it, just in case.

Saluditos,

Ferr?n.

@Ferry

"Real-World" tests are the Final Test that you use when everything else has been debugged. The problem is that, with so many other things happening at the same time (interrupts, DMA, etc), if something goes wrong, you have no way of isolating it to a cause.