A64 Dual Channel error = bad RAM, CPU or M/B?

I run an Athlon64 3200+ CPU on an MSI K8N Neo2 nforce3 motherboard with 1024mb of corsair TwinX 4400c25pt DDR550 RAM.

I do a lot of video editing on my system and handling large files sometimes causes CRC errors in WinRAR. I have a large 8gb file that almost always causes errors when decompressing unless I use only 1 stick of RAM, however the file always works fine on an older nforce2 athlon 3200+ dual channel system.

I have been waiting for a new bios from MSI which officially supports the E3 (venice) stepping of the Athlon64 in case that might help but now that its been released (v1.90) there has been no difference in stability.

I have tried running memtest86+ and Prime95 memory and CPU stress tests with new installations of windows 2000 pro and Windows XP pro as well as hard drive tests, changing all the IDE & SATA cables and changing the power supply from a 550W to a 500W one from another PC. I have found that if I use only 1 stick of DDR ram, so the athlon64 uses only a single channel memory interface, the errors disappear.

When it comes to overclocking there is some odd behaviour too. Memtest takes these sticks up to 293Mhz and 297Mhz each at 2.5-4-4-10 @ 2.8v before errors creep in. Dual channel works up to about 281Mhz. Using the 166Mhz divider the HT bus goes up to about 315Mhz. With 1 stick of RAM in single channel, the CPU is stable in Prime95 at 10x264Mhz (2640Mhz) or 9x293Mhz (2637Mhz) with 1.5v. Dual channel seems stable at 10x251Mhz (2510MHz) but, strangely, only up to 9x266 (2397Mhz) in Prime95 but the WinRAR problem always occurs until I wind down to 2230Mhz (10x223Mhz) where there's a better, but not 100%, chance of it succeeding in extracting the large file.

I have been struggling with this issue for a few weeks now and I'm starting to suspect that the onboard memory controller on this particular CPU may not be working correctly... or is it the Motherboard? ...Or the RAM? The socket 939 Athlon64 should work perfectly well with the dual channel certified corsair memory but i have been reduced to only using 512mb of the 1024mb I purchased to ensure error-free operation.

Anyone have any thoughts on what the problem could be?

to the Forum!
What you have described in terms of the ram and overclocking is not unusual. One stick always scales better than 2 in DC. What is unusual is the WinRAR problem. I have some ideas, but you may have tried them already as it looks like you have been at this a while. Have you tried to run single channel with the ram in slots 1 and 3? Have you tried running at 2t? I know you lose some performance with both of these, but not nearly as much as you lose running the whole system slower. Have you tried giving the ram 2.85 volts? My tccd runs best at that voltage, but it is not of the quality as your Corsair. Good luck with it.

Crank up that Vcore and run SPI 1M test at settings that is giving you the problem.

1. It sounds like you need more voltage for your CPU. Not sure what you have it now, but give it a try and report back. Don't go over 1.60 volts as this is my limit for 24/7 use at OC speeds with air cool set-up.
2. I suspect you will need more Vdimm too. You can only go so far with 2.8 volts. My BH-5 is taking 3.6 volts and loving it @ high speeds

Thank you for the welcome and for the quick reply rseven. I've been peeking around these forums for a good while now and I was almost ready to bet you'd be the first to respond, cheers!

The RAM sticks are fairly matched and I'm happy they both go to over 290Mhz at C2.5 & 1T each. Yesterday I tried updating to the latest beta bios, 1.A3mod but the results are much like the other bios' I have tried from 1.80 onwards. You are right I have tried a wide range of configurations and trawled through loads of forums in trying to figure out what's going on.

Since the RAM is double-sided the Neo2 defaults to a 166Mhz divider with the sticks in slots 1 & 3 and requires 2T timing. 1T timing causes a crash. Here are some results with the RAM at 2.5-4-3-10-13(trc)-16(trfc)-2T @ 2.85v and the A64 @ 1.45v+5% = ~1.5v with a 3x HT multiplier with WinRAR:

Single Channel (1 & 3): 10x262Mhz @ 166Mhz divider = 2625Mhz @ 217Mhz ram ... PASS
Single Channel (1 & 3): 9x291Mhz @ 166Mhz divider = 2620Mhz @ 241Mhz ram ... PASS
Dual Channel (1 & 2): 10x262Mhz @ 200Mhz divider = 2625Mhz @ 262Mhz ram ... FAIL
Dual Channel (1 & 2): 9x245Mhz @ 200Mhz divider = 2207Mhz @ 245Mhz ram ... FAIL
Dual Channel (1 & 2): 10x230Mhz @ 200Mhz divider = 2300Mhz @ 230Mhz ram ... PASS

Thanks for your suggestions Sniper.nkc and for letting me know what your max vcore is for air-cooling. I tried upping the voltage to 1.45v+10% which is about 1.6v and running SPI 1M tests with the mem up at 2.85v. Those tests passed but I'm still getting this CRC problem with winRAR extractions until I clock right down or go single channel @ 2T - pretty odd!

What wierd behaviour. I think the RAM is fine so it could be some issue with either the CPU or the motherboard. Whaat do you think?

ColdKiss said:

Thank you for the welcome and for the quick reply rseven. I've been peeking around these forums for a good while now and I was almost ready to bet you'd be the first to respond, cheers!

The RAM sticks are fairly matched and I'm happy they both go to over 290Mhz at C2.5 & 1T each. Yesterday I tried updating to the latest beta bios, 1.A3mod but the results are much like the other bios' I have tried from 1.80 onwards. You are right I have tried a wide range of configurations and trawled through loads of forums in trying to figure out what's going on.

Since the RAM is double-sided the Neo2 defaults to a 166Mhz divider with the sticks in slots 1 & 3 and requires 2T timing. 1T timing causes a crash. Here are some results with the RAM at 2.5-4-3-10-13(trc)-16(trfc)-2T @ 2.85v and the A64 @ 1.45v+5% = ~1.5v with a 3x HT multiplier with WinRAR:

Single Channel (1 & 3): 10x262Mhz @ 166Mhz divider = 2625Mhz @ 217Mhz ram ... PASS
Single Channel (1 & 3): 9x291Mhz @ 166Mhz divider = 2620Mhz @ 241Mhz ram ... PASS
Dual Channel (1 & 2): 10x262Mhz @ 200Mhz divider = 2625Mhz @ 262Mhz ram ... FAIL
Dual Channel (1 & 2): 9x245Mhz @ 200Mhz divider = 2207Mhz @ 245Mhz ram ... FAIL
Dual Channel (1 & 2): 10x230Mhz @ 200Mhz divider = 2300Mhz @ 230Mhz ram ... PASS

Thanks for your suggestions Sniper.nkc and for letting me know what your max vcore is for air-cooling. I tried upping the voltage to 1.45v+10% which is about 1.6v and running SPI 1M tests with the mem up at 2.85v. Those tests passed but I'm still getting this CRC problem with winRAR extractions until I clock right down or go single channel @ 2T - pretty odd!

What wierd behaviour. I think the RAM is fine so it could be some issue with either the CPU or the motherboard. Whaat do you think?

Click to expand...

Looks like all is fine except for Winrar decompression. You may have to settle with 2T. I am at 2T and I can get 2.7 Ghz. No way I can acheive that with 1T. What is your best OC at 2T?

ColdKiss said:

Thank you for the welcome and for the quick reply rseven. I've been peeking around these forums for a good while now and I was almost ready to bet you'd be the first to respond, cheers!

Click to expand...

Thanks, ColdKiss. I'm a little late getting back to you tonight as I'm building a DFI based system for my bro.
I read your response and I'm a bit confused. Okay, I'm always confused, but I'm a bit more confused. The memory should run at 1t doublesided or not. My ram is double sided and has always run at 1t. If you were running 4x256 single sided it would run at 2t or 4x512 it would be 2t at 333MHz. This makes me think something is not right from the get go. Is the CPU or the mobo? Man, that is really hard to determine. The only way you can relly be sure is to try another cpu or mobo, ehich isn't very practical. How long have you had the system? The reason I ask is could the board be rma'd? If I was goig to bet I would guess it's the board as MSI makes more defective boards than AMD makes defective cpu's.
BTW in looking at your results it seemed you did better in single channel. I take it you still ahad the same WinRAR problem? The pass/fail results were from memtest?
I have to tell you I'm presently stumped, but I will mull it over and see if I can think of anything else to try. I'm kind of fried by building a new rig, that didn't go so easy. Whatever, faults MSI has, their manual is far better than MSI's. I mean not even close.
Good luck.

If you are seeing issues with the command timing, it sounds like a bump in the signal which is somthing that happens in connectors and wiring that slows down the signals.
Have you tried resocketing the CPU just for S&Gs. While you have the proc out, make sure no foreign objects are in the socket such as hair, packing materials etc. Same goes for the dimm slots.
Next to run Dual requires more power and will start to propagate more issues. Get/borrow a voltmeter/DVM and check the voltages while prime is running. If the 12V is dipping below 11.75 then that there lies your problem. Check the 3.3V also as some power for the interface circuits draw from there, less than 3.2 is a problem.

"If all else fails, fix a Long Island Tea and call it a night!".

AlabamaCajun said:

"If all else fails, fix a Long Island Tea and call it a night!".

Click to expand...

Hey, I'll drink to that! Cheers!

Has anyone ever come across a problem with the A64 on-chip memory controller? If this was an old-style sysytem I'd be suspecting the northbridge as being dodgy.

Sniper.nkc, I haven't really tried for a max OC just yet as this single/dual channel memory issue has been confounding me since I got this board a couple of months ago. However, it does 2640Mhz stable @ 1.5v (single channel) and I ran at 2520Mhz (dual channel) for a wee while after passing a day of Prime95 and before I discovered the joys of WinRAR. Maybe there is a little more headroom available so I'll check out more 2T timings and let you know. What's your best OC for 1T if 2.7Ghz is your best for 2T? Have you tried to see what difference in OC you get in single and dual channel configurations? The OC difference I get is over 300Mhz!

Damn, rseven, you've been up concentrating on building a new PC and you still have the will to pop in here and reply? I'll drink to that, especially since you must be cross-eyed by now after your wrestles with DFI documentation...
There does seem to be something very wrong in this here PC box because it's got me pretty damned confused too. Try as I might, the RAM on this board will only run 2T timings if in slots 1 & 3 but will do 1T or 2T if in slots 1 & 2 albeit with an unstable OC despite resetting the bios. I'm inclined to think its the CPU that is wonky, but that is probably influenced by it being much easier for me to unplug than the Neo2 is to extract. Single channel is much more stable and OCs far better than dual channel does. Even in dual channel I get a better OC with higher multis than with higher memory fsbs even though there is still lots of headroom available. Strange stuff... something is wrong with the dual channel memory interface. The results in my previous post were showing WinRAR extraction tests as they fail when SuperPi and Prime95 pass in windows. Should I RMA this CPU? I only hope they don't replace it with some half-dead one after all my patient waiting for venice availability.

Thanks for pointing me towards possible connector issues AlabamaCajun. I re-seated everything tonight, but still the same dull results. Do you think its likely that there is a problem with motherboard traces or connectors that could be reflected by the timing issues I am encountering? I used my trusty Fluke multimeter to measure the load voltages at +12.02v, +4.97v and 3.31v.
Hmm, now I'm more inclined to think its a motherboard problem...

Aaargh! So confusing.. I think I'll join you guys out on the veranda for some of that cool Long Island iced tea...!

ColdKiss said:

Has anyone ever come across a problem with the A64 on-chip memory controller?

Sniper.nkc, What's your best OC for 1T if 2.7Ghz is your best for 2T? Have you tried to see what difference in OC you get in single and dual channel configurations? The OC difference I get is over 300Mhz!

Click to expand...

At 1T command, the best I can do in Dual Channel 230 X 10 @ 2-2-2-6. I have stayed with Dual Channel because I get a performance boost. My ram is designed for tight timings, so I am limited to have far I can relax the settings. One stick of ram went further than 2 sticks and that is obvious and understood.

Is there a way I can do a large file decompression? I have winrar too and I want to check how it does with a large file. What can I use to simulate what you are doing?

Well, I must thank all you guys for your tips - I am making some progress!

Although the TCCD RAM passes memtest+ at up to 284Mhz in dual-channel at 1T; I find if I run it with a 166Mhz divider I can get a fairly high OC of 10x261Mhz or 9x290Mhz (2610Mhz) as opposed to running it in sync which, as you can see, has led to various problems.

Here's a summary of my OC experiences with the Neo2 so far:

Single-channel (slot 1 only) sync @ 200Mhz, 1T or 2T = 2640Mhz, Prime95 pass, WinRAR pass, stable
Single-channel (1 & 3) sync @ 200Mhz, 1T or 2T = crash
Single-channel (1 & 3) async @ 166Mhz, 1T = crash
Single-channel (1 & 3) async @ 166Mhz, 2T = 2640Mhz, Prime95 pass, WinRAR pass, stable
Dual-channel sync @ 200Mhz, 1T or 2T = 2520Mhz, Prime95 pass, WinRAR fail (unless 2230Mhz), general instability
Dual-channel async @ 166Mhz, 1T or 2T = 2610Mhz, Prime95 pass, WinRAR pass, stable

What do you make of these results? Do you recognise any patterns here and is there some tell-tale giveaway of a failing component?

Sniper.nkc you seem to have quite a large difference in between 1T and 2T OC results whereas I find a negligible difference using TCCD ram. I just stumbled across this WinRAR 'test' a few days ago when uncompressing some 8gb of raw DV footage. If you like, you can try it too by making an .iso file of a nice fat DVD or a ghost partition image and then use WinRAR to compress it into a number of 100mb parts then decompress the whole thing onto another hard drive. Let me know how you get on.

I'm running with the A64 @ 1.45v+5% = ~1.5v and the RAM at 2.5-3-3-10-13(trc)-16(trfc)-2(trrd)-3(twr)-1(twtr)-4(trtw)-1(twcl)-2T timings @ 2.85v.

Since playing around with timings has made some difference, what's the best place for getting info on appropriate timing and configuration parameters for A64tweaker with the Neo2? In fact, what timings do you guys run in your setups? Rseven, I notice you run your RAM async - what was your max OC at relaxed timings but in sync?

"Once more into the breach dear friends..."

Good morning ColdKiss. I looked for this thread at 2am and in the grog of the night I couldn't find it. (Gettin' old!) I looked at my manual last night, a rarity, and sure enough it says that when using 2 double sided pc400 sticks of ram, it runs them at pc333. It must be something with the cpu memory controller that dictates these parameters.
To answer your question, my ram will go as high as 276MHz runing 1:1. The timings are 3-4-4-8 1t at 2.85 vdimm. I have an OCZ ddr booster coming next week ($36 shipped from Monarch) and I will try to run it with a tad more voltage to see if I can go higher. My async performance with the lower latency is really quite good and I'm not sure there is anything to be gained here, but you know about curiosity and the cat.
I'm glad to here you are making some progress. I have been thinking about this issue and I'm not getting any profound insights. If I were you, I would be up nights with it spinning in my head, of course that would probably just be the effect of the booze....LOL BTW, what is your HT multi set at? Mine is at 3x. I doubt it makes a big difference, but some folks have set it to 2x.
Okay, as always good luck.

This is looking more like your chip his the stumble point. My P4 won't go past 255 which I blame on the NB. If you can prime DC at 166 and get 2.6G out it I would stay with that or try relaxed memory timings of 3-3-3-10 and see what that would do. Try rsevens 3-4-4-8 or 3-4-3-10. With the 10 you might get a little higher FSB of 8-12Mhz.

ColdKiss said:

Well, I must thank all you guys for your tips - I am making some progress!

Although the TCCD RAM passes memtest+ at up to 284Mhz in dual-channel at 1T; I find if I run it with a 166Mhz divider I can get a fairly high OC of 10x261Mhz or 9x290Mhz (2610Mhz) as opposed to running it in sync which, as you can see, has led to various problems.

Click to expand...

Try not to mix results/performance between memtest86 and your winrar problems. Memtest86 is for testing you ram speed and to see how high it will OC before errors. For example, my ram is rated at PC3500 @ 2-2-2. At a minimum, my sticks should pass memtest86 at 217.

ColdKiss said:

Here's a summary of my OC experiences with the Neo2 so far:

Single-channel (slot 1 only) sync @ 200Mhz, 1T or 2T = 2640Mhz, Prime95 pass, WinRAR pass, stable
Single-channel (1 & 3) sync @ 200Mhz, 1T or 2T = crash
Single-channel (1 & 3) async @ 166Mhz, 1T = crash
Single-channel (1 & 3) async @ 166Mhz, 2T = 2640Mhz, Prime95 pass, WinRAR pass, stable
Dual-channel sync @ 200Mhz, 1T or 2T = 2520Mhz, Prime95 pass, WinRAR fail (unless 2230Mhz), general instability
Dual-channel async @ 166Mhz, 1T or 2T = 2610Mhz, Prime95 pass, WinRAR pass, stable

What do you make of these results? Do you recognise any patterns here and is there some tell-tale giveaway of a failing component?

Click to expand...

If you take winrar out of the equation, the information you provided is normal.

ColdKiss said:

Sniper.nkc you seem to have quite a large difference in between 1T and 2T OC results whereas I find a negligible difference using TCCD ram.

Click to expand...

My BH5 ram is not the same as TCCD. I don't have the flexability to loosen timings to get high OCs @ 1:1 ratio.

To me, winrar is just a program that will stress your CPU and memory. If you fail with winwar, it just means you have instability at that OC setting. More Vcore and Vdimm may solve it, but you hit it on the nail about the 166 Mhz divider as this does reduce the speed on your memory which translate to better stability. For me to hit 2.7 Ghz, I have to use the 166 Mhz divider, and there is no way around it with my configuration.

I think all is well with your computer, just use the 166 Mhz divider to zip through your winrar work without crashing @ high OC speed.

Hi guys, thanks for your replies. As if it one overclocking problem wasn't enough to keep me up at nights: I've now started overclocking my car as well as my comp! I've always thought it wasn't quick enough and now I've discovered that its only got a 2-litre engine instead of the 2.3-litre one its meant to have (damn previous owners!) so the last few days have been spending extracting the old one. What fun...

rseven, have you gotten your hands on that OCZ booster yet? I wonder how your TCCD will respond to more voltage as I've heard that newer TCCD modules can take higher volts but some older modules are comfortable with closer to default voltage. My HT multi is at 3x but I'll try 2x. Do you find any drawbacks with async performance? What kinds of applications, if any, suffer as a result and do you think cas latency is important for an A64?

How did you get on testing WinRAR on your setup sniper.nkc? My comp does seem to be running fine at 9x288Mhz with a 166Mhz divider even though the RAM is capable of running 1:1 and that is a relief because having a flakey computer is a nightmare. If there is a problematic component in this setup I agree with AlabamaCajun in thinking that the memory controller ont this Venice CPU is duff and can't handle high frequencies in dual-channel. Unlucky... but I'm lucky I discovered the fault after memtest, Prime95 and spi all worked fine. I feel so tempted to RMA but then who knows what kind of CPU I'd get back?

I have been playing around with the memory timings AlabamaCajun, and it does seem to make a difference. I changed trc to 13 (=tras+trp) and set ref to 166Mhz 3.9us, twtr to 1 and driver strength to weak whilst running a 200Mhz divider at 10x240Mhz and it managed to pass WinRAR once which it couldn't do before. Upon trying a whole bunch of memory timings and reading through loads of tweak guides after a particularly long car fixing session I did manage to get myself all confused with the numbers and called it a night.

Perhaps there is potential there to investigate. What are your RAM timings from A64tweaker?

ColdKiss said:

rseven, have you gotten your hands on that OCZ booster yet? I wonder how your TCCD will respond to more voltage as I've heard that newer TCCD modules can take higher volts but some older modules are comfortable with closer to default voltage. My HT multi is at 3x but I'll try 2x. Do you find any drawbacks with async performance? What kinds of applications, if any, suffer as a result and do you think cas latency is important for an A64?

Click to expand...

I got the OCZ booster yesterday and it was a bust. No increase in performance with vdimm at 2.9 or 3.0 volts. In fact, it performed worse with the booster than it did without it. It barely fit in slot 4, and it generated a lot of heat. I just requested an rma from Monarch.
As for async performance, I find the lower latency pretty much cancels out the higher running speeds. In my case though, I got the tccd thinking it would run 1:1. If I thought I was going to run async I would have gotten other, less expensive, ram that would have run with tighter timings like 2-2-2-5. It's called live and learn. BTW, when testing the booster I tried to run tighter async timings as well with no success.