Help me find the faulty chip (EDO_ERR Part 1/3)
Вставка
- Опубліковано 30 лип 2024
- Have you ever encountered memory errors? When a memory module fails, the behavior of a computer system may be unpredictable. A best case scenario would be a blue screen (if you are using Windows) and a memory test telling you that you should check your system memory thoroughly. Modern memory modules are quite complex with large BGA memory chips, but with older technologies like EDO SIMM memory modules, we can replace memory chips without the requirement for specialized equipment - and maybe learn something along the way!
In todays video, I have a non-working 32MB EDO memory module. I tried to figure out which chip is failing, but I am afraid I will need your help! I am sorry if this video is quite technical, but maybe someone will be able to solve this puzzle! And if not, you can try your luck and guess which one of the 16 chips is faulty!
You can support me on Patreon:
/ bitsundbolts
00:00 Intro
01:09 Faulty module
03:30 Memory chip
06:49 Memory module
07:45 Motherboard
09:52 Narrowing down
11:17 Picking a side
12:42 My guess
13:15 Your help is needed! - Розваги
What about to mask SIMM pads of data bus on faulty module, small piece of tape on corresponding Dx pins. That could verify the suspect.
Wow! Very interesting idea! I will definitely try this! I think you just added an interim video my friend! Kapton Tape - here I come!!!
@@bitsundbolts waiting for your results, I am curious too 😉
This may work, but putting tape on the CAS, RAS, and DRB lines might give better results. This is because can choose which side of the memory module and capacity it may access. The datasheet for 82439HX shows this on pages 27 and 28. On my boards, the SIMM is hard to install the modules, so I do not know if there is enough room to use tape.
Thermal cameras can help sometimes. Especially if it fails after a period of time.
Or, if the part is not getting hotter that the rest, it might be an idea to cool down/warm up one chip at a time and see if the error goes away/comes back.
IPA works as well I believe. It is non-conductive and will evaporate quicker on chips that are too hot.
I actually did use a thermal camera and checked the chips. I couldn't find anything noteworthy - so, I did not include it in the video.
@@bitsundbolts Long shot. Maybe the errors start when the chip gets up to operating temp. Possibly use a can of compressed air on each chip one by one and see if the errors suddenly stop while running memtest86?
I absolutely love that idea! Your suggestion goes on the list of physical methods to find the faulty chip! Thanks!
There was a utility like CheckIt (or maybe it was that ?) that had a page to identify the faulty memory chips on an XT machine. You had to input the chip size, the memory configuration (how many chips, have parity or not, how many rows were populated, etc) and it tested and located the bad memory chip. It worked also for addon ISA memory boards. I tweaked the configuration of that program on a 286 with random freezes and 30 pin memory SIMMs and it located the bad chip. (I got a bunch of bad memory sticks and all had faulty one chip so I managed to frankenstein some to use them)
That’s really cool. I never seen that app back in the day but wish I knew about it. I always had to rely on running long tests doing that pluck-and-chuck until I at least singled out the bad stick. Memories…
Pull up OE from board, check with memtest, write bad addresses, and repeat for every chip. You will have map region->chip location
You might want to check all the solder work - I’ve dealt with “faulty” chips that turned out to be a cold solder joint with similar behavior
There are so many great suggestions. If I can narrow down a chip, I will inspect it with a microscope. It could be a cold solder joint.
I would say, your guess is correct, but without knowing the exact wiring from the Chipset to the modules it's impossible to say for sure.
Data-Lines don't have to be connected to the pins with the same name. If it is more convenient to route pin D10 on the Chipset to D3 on the RAM-Module, the PCB-Designers will do so to improve signal integrity or reduce board complexity.
Only way to know for sure is to construct a test-bench that can check the entire module on its own at full speed.
Yes, this is what bothered me and I did not expect this topic to be this complex. But when I realized this, I was too deep into the material. I have seen some EDO module testers, but they are too expensive for this single purpose. It will be a fun project to remove the chips and create 30-pin SIMMS...
@@bitsundbolts I think maybe you could build a chip tester later once you get past this problem. It would be so awesome to have a DIY memory tester.
Regardless of PCB design the data pins have the same input & output pin !
But the address pins should be 100 % correct & must follow the chip numbering .
Part of the problem, though, is that it may not be the memory itself, but the *power supply*. I recently had a problem with trying to identify an unstable system, until I tried taking out the power supply and swapping in a new one. The memory problems and instability immediately went away.
Could be bad ripple
Interesting. I am actually using an EVGA Supernova 850 G3 - so, a really new power supply. Additionally to the faulty module, I have an identical working one. I still believe that one chip is faulty because of the consistency of how MemTest86+ reports the errors.
@@bitsundbolts My reply was more aimed at T-Squareds. I agree with you on your one :)
@@bitsundbolts Remember that PC motherboards before 2003ish primarily used the +5v rail for everything. Power supplies starting in the mid 2000s started putting most of their capacity on the +12v rail as motherboards switched to that for CPU power delivery. While you could have an 850W power supply, the percentage of that available on the +5v rail is only a tiny percentage of that. So you could be overloading the +5v rail, or getting very close to its current limit, and have issues related to excessive voltage ripple.
Group regulated supplies are even worse, if you have a heavily imbalanced load, where one rail is loaded up far more than the others, it can throw voltage regulation off on all of the other rails.
I would try to disable each chip one at a time. If there are isolating resistors on the chip select line you can simply ground the pin or you might have to cut a trace and ground the pin. If the memory error doesn't change from the unmodified scan then you disabled the "correct" chip. If you do this to a known good chip you can make a map of which chip gives which address error.
Wow fantastic video! It is tough to figure out, the Pentium is only able to address memory in 64 bit chucks so it is only exposing A31-A3 and then A2, A1 and A0 are demultiplexed out as BE7-BE0 (Byte Enable) which allows it to read/write which of the 8 bytes in that 64 block it wants. That is why it is always even numbered even when you swap that sticks, because address 0x0008 to the CPU is actually 0x0001 to the memory, 0x0010 to the CPU is 0x0002 to the memory, etc.
I think the odd cas connection are caused by simplification of the board layout. It is no important on which chip the data bit is stored since you can't hot swap the RAM to another board of another manufacturer and it shall work. This can also cause that the error when the simms are switched are not at the same position. It is not necessary for the chipset to know on which chip the bits are stored, it stores them where it is connected to. That's some pcb layout magic or some call it laziness 😄
Yes, and this is what makes it incredibly difficult to narrow down the chip. There are so many variables to consider. Looks like the solution may be in less sophisticated method: Temperature and taping off contacts :)
Very nice analysis. I was also wondering if there was a way to figure out what chip is faulty on a memory module. In my case I have a DDR3 which I'd like to rescue but the system does not see it altogether so I cannot test it. I look forward to the outcome!
Your memtest error also reminds me of a friend's laptop which kept corrupting Windows. Several journeys to the service centre for warranty repairs never identified the issue but Windows kept getting corrupted. Turned out one module was faulty when COLD! Enough to corrupt the OS at boot and enough to test fine when the technician was running their tests after a couple of minutes. I spotted that by running memtest first thing first after turning it on.
Wow - memory errors can be such a pain! The module I have here seems to behave the opposite way! Once it gets warm, errors appear.
Is this one of the DDR3 modules from your last video with the X58 chipset? I wonder if it is possible to rescue a seemingly dead module - not that easy to diagnose when its not showing up :) I would be probably going crazy and compare several pins against ground and see what values I get from a multimeter. Then compare it with working modules and see if there is something obvious.
You're a champion for even trying this. Yeah I think the bits that are good vs bad are key here and I think you're on the right track. I'm rooting for ya! :D
The CAS connections from the memory controller look al strange going to the resistors but when they leave the resistors they still have the original name.
Its just visually in the wrong order, the signal names are still correct and logically the same, those resistors are just form EMI filtering and are wired strange for layout purpose only.
wow, this is amazing work, hard work.
I had some memory stick in the past that were defective and using a memory tester was the best choice but it was very expensive so i did nothing.
After that , some time later, did some more research and found out that there are memory sticks on aliexpress with little sockets in place of each chip where you can place your individual chips and test them.
They exist for all sort of Ram memory, like ddr2, ddr3, ddr4 etc. Still expensive.
So another idea is to just buy one such chip socket , solder it in place of a memory chip on a good stick of ram.
Then, on the bad stick you desolder one chip at a time and test it in the testing memory stick by placing it in that chip socket ..
Chances are that you will find the bad chip before desoldering all chips from the bad stick.
Great video. Very good explanation
12:27
I've also had problems with RAM. Here are my conclusions
1. too high clock speed of the bus, it works flawlessly at 33 MHz, there are errors at 66 MHz. I had this case with 200MHz DRAMs and had to downclock to 100MHz because Windows XP kept restarting in random actions, after downclocking I didn't have to restart the system for several years. From what I remember (it was 15 years ago) errors were always on one line. It looked similar to what you see in your video.
2. Motherboard doesn't like some kind of memory. On one computer the memory works properly, on the other there are errors with the same bus timing (I had such DRAM memory - it was returned to the seller). Motherboards came from different manufacturers.
3. Memories from two different manufacturers do not like each other even though the memory specification allows them to be combined.
4. The motherboard gives too little current and part of the line gets less current and write errors occur (different voltages are used at different timings - check if the power supply after a few hours of operation maintains the voltage parameters for different voltages 3V, 5V, 12V).
Awesome video! Thank you for explaining this!
Glad you enjoyed it!
10:34 as an electronics engineer i can tell you that this connections are placed in this way purelly to optimize pcb layout. When these lines was drawn the first time they were straight for shure. When you do pcb layout, very often you are switching between schematic and PCB doing this short of stuff to avoid huge amount of vias on critical traces and just to make your data lines easier way. This is edoram era, see ddr1 era zig-zaggy traces they are called matched impedance pairs.
Curiously impedance will of course start to grow very slightly out of sync once you start adding different amount of meander to related traces, so this is not an impedance matching technique; really it's propagation delay matching or phase matching.
They're very carefully routed in order to maintain timing. Very small changes in the length of the trace can impact the timing of the signal.
Also, they're likely ordered specifically to minimize the impact of noise from adjacent lines. See row hammer.
I remember there was a method of attack that would freeze the ram chips with liquid N2 or liquid compressed air, then the bits would lock in their state and retrieval of bitlocker key was possible. Maybe freezing individual chips one by one can give you a hint for the addressing scheme since frozen ram chip should begin failing right away in memtestx86 and if in frozen state the address matches failing address from normal warm state then you got your self a suspect. Just make sure not to flood the board with water condensation 😅
Interesting. I didn't know you can freeze bits :) But yeah, looks like temperature may be a solution to find the faulty chip! Thanks!
at 10:39 you asked why some signals are crossing. I can explain. That is because of the layout consideration. At the physical BGA or package of the chip, sometimes the signals need to be adjusted when doing the layout and feed back to the schematic guy to change the wiring order.
While I can’t answer your question, I thank you for teaching me a lot about memory handling and connections! 👍
No worries! I am happy you watched my video!
The way i'd check it - at least if the custom 30-pin SIMM PCBs are 3-chip modules:
- Make one module with known good memory chips.
- Make the other module with sockets (if you can get proper ones for the chip type) and have it mounted in a socket that offers the space on, say, a 486.
- Preferably have the "tester" stick run with 2 known good chips and swap out the 3rd.
- Run memtest or another test suite to your heart's content.
That could solve the dilemma, even if you have an MMU or chipset in the equation as most motherboards have, but as the 30-pin SIMMs have less chips on them, it'd remove quite a bit of variables in any case.
It's likely to be a longer process considering memtest86 is comparably slow, but barring a dedicated memory/chip tester, it's the best I can come up with.
The Retro Chip Tester also has a 30-pin SIMM adapter board - it's cheaper than the "big" RAM testers out there, but it could be of some use combined with a "socketed SIMM" board asd I described above - though there's still no guarantee that it'll catch the error as Memtest only catches it after some runtime...
On the Commodore 64, it is usual to stack a working memory chip on top of the suspected broken one. I wonder if this would work here, too. Perhaps by soldering a memory chip onto the rear of a socket, and pushing the socket onto the suspected one?
I do have memory chips in the mail that should arrive in a few weeks (shipment from China sometimes takes a long time!). I may give it a try.
The GRUB2 can make the PC run with faulty memory, with just adding badram command to its config according to memtest86+ output (which have a special output mode for this badram arguments).
Kernels that support this includes Linux, GNU Mach, the kernel of FreeBSD and Multiboot kernels in general.
I have used this feature with Linux and it really works.
I believe you can piggyback ram chips over the original and if the original is bad the one piggybacked on top will bypass it. I wonder if you could make a piggyback board similar to the one you made for the voodoo card to add ram. Use a socket to clip onto the ram chip with a good one on the other side of the board. Test one chip at a time and when the ram passes the test fully the one with the piggybacked chip is the bad one.
Haha! I love it! I just need to get a socket that fits, but very interesting idea indeed! But I wonder how the data pins would react. I think there would still be memory errors if your idea actually works, but a lot less - depending on the test that is running. E.g. if the memory chip is stuck on high, but correct value would be low - the working module would return low, but the faulty chip would still send high and the data would be wrong. The other way around, however, would be fixed by the working chip.
Judging by the memtest output, the error is more or less stable and unlikely to be caused by the cold solder joints. The chips in question seem to be the ones that use output pins DQ9-DQ12 on the SIMM - correlated to the Err-Bits from the memtest.
Cold solders (or taping the DQ pins on the SIMM as some suggested) would likely cause the SIMM to crap out across the entire address space, however in this case, the errors start at 56 Mb, meaning the last quarter of the chip is unstable, with D2 (bit 3) totally knocked out and D3 (bit 4) flapping.
In regards to the endianness, I might have guessed the wrong pins. However, taping one of them and seeing if the error stays the same across the entire address space or another Err-Bit appears would help to sort this out. I've long since let go of any SIMMs or machines capable of taking them.
On an unrelated note, I'm not sure if the EDO chips would work properly in a 386 system. Some of them have the FPM compatibility mode, some don't.
Thanks for your detailed answer. I think you're really close, if not absolutely correct. Taping contacts does indeed not work. Since the module has only 32 data lines (shared between both sides of the module), removing one bit would turn the module into a 31-bit module which doesn't work at all. I'm working hard on the next video - hopefully it will be release this week.
Regarding FPM support, yes, I agree. I paired the EDO module with an FPM module and the system boots. The BIOS summary screen as well as the tool SpeedSys report FPM mode - which makes me believe that those EDO chips will work on the 386 - but we will find out!
It could be possible to incrementally move chips from one module to the other one and see when the error gets transfered.
Other than that it might be possible to test each chip individually with an Arduino or something similar.
0:17 from Windows Vista onwards there was the utility "mdsched" on board of Windows for testing RAM. Not sure if it was capable of finding RAM errors reliably though. Had two faulty RAM modules in my life so far. Both were second hand. One from a colleague and one from an online dealer. I wonder if they did not care for ESD safety, or why the modules produced errors. 🤔
To avoid such hazzle I prefer RAM with Ecc functionality. 😅 Unfortunately CPUs that support that feature are rare.
I wouldn't be too concerned about ESD, there is ESD protection internally, you'd have to give it quite a jolt. It's not uncommon to find a latent defect in RAM, where it develops progressing fault. Especially if a ceramic capacitor on the PCB is cracked. So that is more likely to be an issue, mechanical mishandling. Or maybe it was just a faulty module and they're getting rid of them because it got ever more wonky.
@@SianaGearz thanks for the information! ☺️👍
My first thought was to desolder each individual chip and test them in some arduino-based test bed, but arduino may be fast enough to test chips “at full speed” and it may matter in this case
Also it may be that chips itself are ok but there is some problem like solder issue, trace issue, corroded contact etc
And all those issues may be temperature-dependent, so for example in slow machine such circumstances (full speed + high temperature) may never occur
I had a weird faulty ram module years ago, running linux, that would constantly corrupt my file system so I thought it was a harddrive problem, then i thought it was the onboard controller and finally i did a memtest and there it was, a consistent flaw and when i swapped out the modules the file system corruption stopped happening
Yeah, memory errors are really hard to find. You blame everything else first :)
Honestly i'm not convinced i'd be playing around with salvage RAM chips from the era! They're kind of all suspect bad. ISSI manufactures FPM and EDO DRAM ICs that you can use.
... is what i'd be saying 3 years ago. Today, good luck with that availability :(
The answer to "why the lines are crossing in the schem" when it comes to RAM is usually because they are not really crossing, it was more convenient to lay it out that way for routing reasons, and when you look at the PCB, they aren't actually crossing. Because you can just shuffle the lines around in a number of ways, the host will be none the wiser. You can even do this with ROM chips but then you need to bake the data shuffled correspondingly. This is also why i will refuse to take a guess at which chip is at fault.
nice Video 👍
What if to try to test only selected address ranges? Theoretically chip which corresponds to the selected address range should become warmer than other chips while tested and you could identify it for example with thermal camera.
one way to narrow it down is to limit memtest to some range and then scope which chip gets accessed, idk just an idea, never tried it
1. When possible test ram with Prime95. I found that memtest86 could throw error sporadically, even one in 6 passes, so very, very long testing is needed to be sure of good memory subsystem. In Prime95 the same ram throw error in something like 5 minutes!
2. Did You even look at this ram stick under microscope? Check if some solder joints are weak.
I put these modules in a memory tester and see what comes out.
Unfortunately, I do not have one.
why dont you make a mamory tester with an arduino? those rams are slow enough that you can use them with one
I am looking into this, but I will need time to hopefully figure something out. I have seen some promising projects that my help getting something together!
I've literally got a screenshot of a memtest that 'passed', with literally 3 errors on the text display itself.
It literally said "Memory test passed, No0errors found, press ECC to continue"
No joke whatsoever here, I can send you the screenshot photo if you'd like.
The issue was bad capacitors in that case.
I think the only way to find the bad chip is by trial and error. Replace one chip at a time until it works, then you know which is the bad one. Having said that, obviously that is a terrible job and would take a crazy amount of time with the memtest taking so long. I wonder if there is a better source for the memory chips then harvesting them from old parts.
first of all try to clean memory contacts with isopropyl and rubber eraser (eraser for pencil). Some memory may be not faulty, just dirty.
You can try freeze spray on one chip while is testing in memtest to see if anything change. Sometime it works.
I have a feeling you are right - it seems like the module starts failing after some time when the heat increases. Could be that by cooling it down (maybe a fan may already be enough) to make the stick work. I may try that. Thanks!
I really need to know. What is Cheese-kill?
the only help i can give you, is giving a like, a view and a comment.
Thank you for your support! That is all I need!
about needing matching memory modules on a Socket 7 board.. is this true on Socket 7 boards that use SD-RAM instead of SIMMs?
They were transitional, just like you can find 486 boards the support either or both of 4x 30 pin or 1x 72 pin per bank, there are Socket 7 boards that support pairs of 72 pin or single SDRAM DIMMS - and while a 486 with both may allow mixing, a bank filled with 30 pin and a 72 pin, the technical differences prevent mixing on Socket 7 boards with both
SD-RAM (and modern DDR RAM) sticks have a 64-bit bandwidth, while 72-pin SIMMs are 32 bits wide ans 30-pin SIMMs are 8 bits wide. Rule is (unless you count RAMBUS/RD-RAM), you have to "fill out" the memory bandwidth for the system to work properly.
The memory bandwidth on a 386DX and a 486 are 32 bits wide, thus you need at least 4x 30 pin SIMMs or 1x 72-pin SIMM.
The Pentium memory pathway (and almost anything newer) is 64 bits wide, thus 2x 72 pin SIMMs or 1x SD-RAM DIMM. (or 8x 30 pin SIMMs, but never seen a Pentium board with 30-pin memory).
There is also some finagling regarding parity and ECC memory, but those are the basics :-)
RAMBUS RIMMs are kinda on its own - it's some sort of a ring bus that needs at least two modules and "dummy" sticks, as you can't have empty sockets with that tech.
Assuming parity bits aren't included, the Pentium architecture has a 64-bit data bus, while 72-pin FPM or EDO SIMMs are 32-bits wide. This necessitates the use of pairs*. SDRAM 168-pin DIMMs are 64-bits wide, which means they are accessed individually and don't need to be installed in pairs.
*There are some Pentium chipsets that support a "half-bank" mode that allows individual 72-pin SIMMs to be installed, by only using 32-bits of the data bus. This cuts the memory bandwidth in half, significantly reducing system performance.
@@Tylonfoxx I have a Dell Precision board (socket 423) that uses RAMBUS. I can tell you the ram sticks run very hot and I mean very! However, the speed is really good even though the Willamette P4 is pretty slow. I just have the board for nostalgia but it’s fully working. The board is picky about what ram you use…you can’t mix 45ns and 60ns sticks for example and it’s important to have matching sticks by the same vendor. Oh and if not all slots are filled or you have a bad stick then the board makes a shrill constant beep. The most annoying board when it screams at you something is wrong 😂
In 1995 year a program exists for verifing chips. Its write on specific adress and verify back. Runs on dos but i have no idea what person create. Its written in basic or pascal. To many years ago but i hope the idea help.
Thanks for sharing. I will do some research and see if I can dig something up. Thanks for letting me know!
You should try a thermal camera. See if one chip is getting hotter than the others.
I did try a thermal camera. Now I'm blaming myself not including the footage in this video. I was looking for anything suspicious, but all chips had the same operating temperature. I couldn't spot anything odd and that is why I didn't bother to include the footage. Thanks for the suggestion though!
@Bits und Bolts Going through pass 2 until it errors likely hints that heat helps show the errors. One thing you can try is while it is spewing errors, touch each module for a few seconds with an ice cube in a sandwich bag. This will bring the temperature back down for only that chip. Careful not to hold it too long, as it will propagate the cooling through the board and to other chips. Touch each one, one by one. You may see the errors go away and come back as you heat and cool that specific chip.
@@speedycpu This is a good idea. You can also buy “cold in a can” to cool specific chips.
Err-Bits are bits 18 and 19. (0c00 = 0000 1100 0000 0000). According to your diagrams ~7:00 this should be U2. Could be another IC, depending on the real schematic of your module, but you can just check continuity according to pinouts.
the grouping (~11:00) is because of the way the pins are located on the BGA ICs and how they can be routed on the PCB.
Regarding the guess for the chip - did you consider little endian ordering of bytes? I also had U2 in one of my assumptions, but I settled for U4 because of the order of bytes when they are stored in memory. Unfortunately, I don't know how MemTest86+ displays those values
@@bitsundbolts yes, accorging to my knowledge intel is MSB, so you count from right to left. I don't really know if endiness is a thing here and obviously didn't take it into account. I may be wrong. I like the idea with the tape from another guy. Kapton would be ideal here as it is very thin and stable.
Yes, I liked that idea as well, but spoiler: with one data line taped, the system doesn't boot. I may try other pins, but it makes sense. If one data pin is taped, all 32-bit values miss one bit - so, you get absolutely no correct response from the memory. So, the conventional memory area is also affected and no operation is possible. At least this is how I justify the behavior of the system. But I keep trying the other suggestions too - time to get out the hot air station to heat up individual memory chips :)
@@bitsundbolts you could use 4 modules. That way, the device may boot as at least one range of memory is functional. But maybe the memory controller doesn't recognize the rest because of missing bit.
Holy macaroni - you are right! It does boot with 2x8MB in S1+S2 and the taped module in S4 (2x32MB in S3+S4) - now we are back in business! Thanks for your great idea!
You want to remove EDO Chips for a 386 motherboard.. But 386 only supports FPM.. So EDO is useless in a 386.
If I am not mistaken, most EDO chips are also capable to run in FPM mode. I am not 100% sure, but FPM and EDO are not that different. It is just a different way how RAS and CAS signals are used.
SOME EDO chips can indeed run in FPM mode. Only differences are that EDO have somewhat faster timings on RAS/CAS, and include some burst mode features (the "Extended" part of "Extended Data Out") - most Non-EDO boards can also indeed still use EDO sticks, but at the slower FPM mode.
The most important is whether or not the EDO chips are 5V tolerant - FPM traditionally runs at 5V while EDO sometimes runs at 3.3V.
My experience and impression is that the majority of compatibility issues encountered using EDO with a chipset designed for FPM have to do with the chipset not recognizing the 5-bit presence detect code on the newer EDO module and failing to POST. Most 72-pin EDO modules and some later FPM modules use the presence detect pins to advertise capacity, timings, etc. This is accomplished with passive components on the SIMM PCB, not the RAM ICs, so transplanting the EDO ICs to a new PCB avoids the problem.
@@Tylonfoxx The chips on BuB's memory modules are EDO: chip model marking end with 5, which usually means EDO. FPM chip model marks end up with 0 in most cases.
Hey you should be able to overclock your memory kit to 3600 3800 on that 5950x, it would give you a nice boost, just saying :P
I tried this once, but I did notice some instability. I value stability over speed. Now that DDR5 is out, maybe some day I get a good deal on 128GB of faster memory. I use RAM disks quite a lot (scratch disks) - wouldn't mind to get more memory.
@@bitsundbolts you should check what those rams can do individually. Like take all out and just leave one. Overclock it till you cant do more mhz find the stable spot and see if all of them can do it. Usually you dont need to touch the voltages on a single stick, auto should do.
then once you know the actual limit then yeah put all of them back and touch the voltages till its stable. ryzen 3000 can do about 3200 on 4 sticks (mine does 3333) but ryzen 5000 usually do 3600 on 4 sticks and most 3200mhz memory can do just fine at 3600 or more.
And indeed, mine is stable as a rock. About 60 days uptime at the moment :P
Quad rank (beacouse 64Gb of memory) of Samsung 8 gigabit B-die (14-14-14 at 3200 Mt/s can only be B-die) is very hard on memory controller. 3200 or maybe 3400 is maximum that memory configuration will run fully stable.
@@laharl2k those sticks individually will probably do 4400 2:1:1 or 4400 2:1 at 1800 FCLK (1.4V 18-18-18-36-54, tRRD 4-6, tFAW 16), but together - no chance. Quad rank is very hard on memory controller, especially quad rank of Samsung 8Gigabit B die.
@@volodumurkalunyak4651
well not too far here, quad rank micro e-die here, and my 3600x is doing fine at 3200 but it can do 3333 and 3400 but i rather stay on the safe side. :P
Couldn't you just buy the chips you needed instead of cannibalizing a stick of ram?
I am not cannibalizing this stick of ram. In its current condition, it is unusable. I can either discard the entire stick - or try to salvage the working memory chips.
Since you're working on creating your own RAM sticks from these chip, I'd think desoldering all the chips and then making up a pluggable single-chip RAM stick which you can then test each chip alone is probably the quickest/best route to go, rather than studying an ambiguous diagram.
You'll probably need such a tester anyway at some point to validate RAM chips in the future.
And perhaps he could also wire up one of those IC clamps to potentially test chips right on the ram stick (assuming it would work given the connections to other chips on the ram module board)