Help - unstable A64 system

Java · Post by **Java** » Sat Apr 09, 2005 7:10 pm

Sorry for the long post...

Ok. Its been a very frustrating week trying to get a new system up and stable. I've been through 2 motherboards and 3 sets of memory and I think the problem is with the power supply (S12 380).

System
A64 3200+ (90nm)
Abit AV8 (just swapped from Asus A8V)
1 Gig (512mb x 2) Mushkin Blue Line DDR 3200 (swapped from Kingston Hyper X)
Maxtor SATA 250g
ATI 9800 (non pro) AGP
Hercules Game Theater 7.1
SilverStone TJ06
S12 380
XP Pro

The build originally went fine and then I started have stability problems, which forced a windows reinstall. Eventually, I could only get one 512mb stick of memory to post. Two sticks would do nothing. I just swapped out the motherboard and memory. Again, it ran fine at first then went unstable forcing another windows install. Now I'm getting numerous fatal errors during the install.

Now that I've ruled out the motherboard and memory, I think its the power supply. I think the cause it in that I couldn't hook the ATX-12v (P4 plug) directly from the motherboard, so I used an adaptor from the Peripheral Power Connector. First, I've search all through the forum and couldn't find anyone else that ran into my problem. The bottom line is that the TJ06's upside down motherboard config with the short cables of the S12 (it has the shortest cables I've seen) make it impossible to connect to the motherboard. I got an extention for the 20 pin connector, but couldn't find an extention for the P4 connector.

I'm guessing that the peripheral power connector isn't right for the ATX-12v. A local PC shop said the peripheral connector would be fine, but if I follow correct, the ATX12v needs two +12vdc lines and the periperal power connector has one +12vdc and one +5vdc.

Any thoughts? I'm thinking about bumping up to the S12 600 in hopes that I may need a big PS and it has longer cables. My only other option is to get a different power supply with longer cables, but I really like the S12's quiet operations. Any thoughts?

Thanks much for help on this as well as the multitude of help I got lurking on this forum.

slipknottin · Post by **slipknottin** » Sat Apr 09, 2005 7:21 pm

Why not pull the PS out of the case, and mount it up higher so the cables all reach without extensions. Install windows, check stability, etc. and make sure that the PS is really the reason.

frankgehry · Post by **frankgehry** » Sat Apr 09, 2005 7:33 pm

J,

Is there any way you can set up the psu outside the case and closer to the motherboard so you can hook everything up without the adapter cable. Someone may be able to help with the adapter cable, but it might be a good idea keep things simple using the fewest number of components to get things going. - FG

Post by **Rusty075** » Sat Apr 09, 2005 7:56 pm

The fact that it will start, and that you're getting install errors and not reboots makes seem less like the PSU at all to me. If your mobo wasn't getting the proper power, I would expect it to not post at all, and certanly not to make it through the Windows install the first time.

Try the advice above first, but if that doesn't work:

Try swapping HDD's.

And try installing with nothing but the mobo, ram, vga and hdd.

Post by **MikeC** » Sat Apr 09, 2005 11:11 pm

JUst went through a few frustrating days of a system build with an A64 Winchester / nForce4 board. A64 Winchester core memory interaction is very finicky. One of the things the new Venice core A64 specifically aims to fix is memory compatibility.

According to X-bit Labs:

Every time a new Athlon 64 processor core come out AMD engineers continue tuning up their integrated memory controller. Note that this is done not only for the sake of increasing the processor performance, but mostly in order to extend the compatibility of this memory controller with the different DIMM modules and their configurations. The previous 90nm Athlon 64 processor core aka Winchester had certain performance limitations when four DDR400 SDRAM modules were used. If there were four single-side DDR400 SDRAM DIMMs installed in a system with a Winchester based Athlon 64 CPU, they could only work with the 2T timing, which caused a few percent drop below the usual performance level. If there were four double-side memory modules installed, DDR400 SDRAM would not work at all and its working frequency would automatically drop down to 333MHz.

AMD engineers promised to eliminate this problem in the new Venice core and they did keep their word. Athlon 64 processors based on Venice can work with four single-side DDR400 SDRAM modules without any limitations, and if there are double-side DDR400 SDRAM DIMMs installed, they can work at 400MHz with 2T timing.

W/ my DFI nForce4 board, what it took was higher than normal RAM voltage settings to get things stable. For a while I was swapping every component trying to isolate the problem. I suggest you pay attention to the memory settings. Maybe just start by upping the voltage one notch & see what happens. Of course, you might have to reinistall windows again due to corruptions.

Straker · Post by **Straker** » Sun Apr 10, 2005 12:01 am

what Mike said.
i blew CAN$350 on a gig of OCZ EL pc3200 platinum rev2 just to make sure i wouldn't run into any annoying memory-related problems, there's plenty of even faster/more expensive memory but this is still the undisputed best for A64s. plenty of people have run into problems with memory working on one system and not another even on Winchesters made a month or two apart, since AMD is basically free to tweak their controllers with every tiny revision.

how did you rule out the memory being a problem? between not booting with both sticks and install problems (huge installs are usually pretty RAM-intensive and often break if there's any data corruption etc partway through), that sounds just as much like memory as anything else... have you tried prime95 or anything? you might want to try really loosening your memory timings/HTT and overvolting it a bit first, to see if it gets any better.

Webfire · Post by **Webfire** » Sun Apr 10, 2005 3:13 am

Well like the others said, it looks like a memory problem.
Try Memtest86 for a couple of hours if the memory produces any errors.
From what I've heard A64 has often problems with memory.

frostedflakes · Post by **frostedflakes** » Sun Apr 10, 2005 8:32 am

I also believe this may be memory related. What I would suggest is to move the memory around and try it in different slots. If that doesn't work, tweak voltage, timings, etc. and try to find a stable setting.

Java · Post by **Java** » Sun Apr 10, 2005 3:46 pm

JUst went through a few frustrating days of a system build with an A64 Winchester / nForce4 board. A64 Winchester core memory interaction is very finicky. One of the things the new Venice core A64 specifically aims to fix is memory compatibility

Thanks all for the advice and support.

On one side, I'd glad its not the PSU. The S12 is very quiet and one of my main goals is to build a quiet rig. On the other side, I think it's time for me to look toward "The Dark Side". I've been a diehard AMD fan for many years, but I think its time to switch to Intel. I've had it with the finicky systems. I just don't have time anymore to become a chipset/bios/driver/tweak/etc expert just to run a PC. Yes, AMD has more "bang for the buck", but it also has more "hassle for the savings".

Thanks again for help on this problem and the overall good advice of this forum.

frankgehry · Post by **frankgehry** » Sun Apr 10, 2005 5:54 pm

J,

I think you comment silenced the house, but intels can be quiet too. - FG

Post by **MikeC** » Sun Apr 10, 2005 11:38 pm

My mem instability problems continued, and after more messing with the incredible range of options in the DFI Lanparty NF4 Ultra D motherboard, I searched the web for solutions.

I had not realized how much flak there is out there about NF4 + A64 + memory issues. After much cruising numerous forums, I came upon the one solution that appears to have really worked (keeping fingers crossed while Prime95 runs): Increase the Vcore of the Winchester A64-3500+ from the default 1.4V to 1.5V.

This tip came from this thread: www.houseofhelp.com/forums It is interesting enough to cite the comment by sycr0 on this page:

Well guys, after many hours on end of testing, I finally am memtest stable for 4+ hours.

I ended up getting a hold of some OCZ memory to test in my system and I still got the same errors as before with my 3200XL's. Turns out the winchester model i have is known for it's weak memory controller and requires to be fed alot of voltage. I've been gradually easing it all the way up to 1.6v core which seems to be the magic point.

I just want to thank wired and ram guy for all your effort in helping me reach this resolution. I plan on taking it up with AMD since the supposed operating voltage for my cpu is 1.4v, which doesn't appear to be the case! However, I doubt if anything good will come out of it.

Thanks alot guys!
__________________
DFI LANPARTY UT nF4 ULTRA-D
AMD Athlon 64 3500+ Winchester Core (stock cooling w/ AS5)
Corsair TWINX1024-3200XL DDR400 2 x 512MB
BFG GeForce 6800 GT OC PCI-Express
Seagate Barracuda 7200.7 200GB SATA
Seagate Barracuda 120GB IDE
Plextor PX-716A
OCZ PowerStream 520W
ANTEC Black Performance Series II SX635BII
WinXP Professional SP1

The odd thing is that I have another of the same DFI motherboard model and a Winchester A64-3500+ sample from several months ago which I've run with single & dual mem of several types without paying one iota of attn to memory settings. Same BIOS, same PSU type, similar HDD, same basic OS & minimal utilities installed. That one set up w/o a single glitch. The A64-3500+ in the new rig is brand-new, store bought a week ago.

So is the difference in the CPU? Did AMD tinker with the mem controller in production?

Post by **MikeC** » Sun Apr 10, 2005 11:48 pm

BTW, Java --

I missed this in your original post:

Now that I've ruled out the motherboard and memory, I think its the power supply. I think the cause it in that I couldn't hook the ATX-12v (P4 plug) directly from the motherboard, so I used an adaptor from the Peripheral Power Connector

The fact that you're using only half the capacity of your PSU could be part of the problem. All ATX12V v2.0 PSUs have twop 12V lines. One for the 2x12V (AUX12V), which powers the CPU exclusively in A64 and P4 boards, and another 12V line for the rest. You need that 2x12V line plugged in!! Get someone to solder extensions to the cables if you have to. Defintely follow up on someone else's advise earlier about moving the PSU temporarily to reach that connector.

Classh · Post by **Classh** » Mon Apr 11, 2005 2:32 am

MikeC wrote:JUst went through a few frustrating days of a system build with an A64 Winchester / nForce4 board. A64 Winchester core memory interaction is very finicky. One of the things the new Venice core A64 specifically aims to fix is memory compatibility.

Hi,
I also have DFI nf4 mobo, and have went through a few weeks of very annoying and frustrating system crashes etc about 2-3 times a day. My mem kits are KHX3200 (DC) and the only way I could get my system stable was increasing mem voltage up to 2.9V (!!) and disabling cool and quiet.

Java: Maybe you should also try turning off that CnQ, though it's a known problem almost merely among DFI boards. Overclocking or undervolting is not even an option. (or maybe overclocking, but that's not going to make my computer any more silent)

However, now I'm very happy with my powerful and still very very quiet computer.

PS: My first post here, english and thoughts might be poor

Java · Post by **Java** » Mon Apr 11, 2005 6:07 pm

The fact that you're using only half the capacity of your PSU could be part of the problem. All ATX12V v2.0 PSUs have twop 12V lines. One for the 2x12V (AUX12V), which powers the CPU exclusively in A64 and P4 boards, and another 12V line for the rest. You need that 2x12V line plugged in!! Get someone to solder extensions to the cables if you have to. Defintely follow up on someone else's advise earlier about moving the PSU temporarily to reach that connector.

Thanks Mike. I should have followed up with all on that. I took the advice and moved the PSU so I could get a direct connect with the 2x12v, but again, I had major fatal errors.

I also have DFI nf4 mobo, and have went through a few weeks of very annoying and frustrating system crashes etc about 2-3 times a day. My mem kits are KHX3200 (DC) and the only way I could get my system stable was increasing mem voltage up to 2.9V (!!) and disabling cool and quiet.

Java: Maybe you should also try turning off that CnQ, though it's a known problem almost merely among DFI boards. Overclocking or undervolting is not even an option. (or maybe overclocking, but that's not going to make my computer any more silent)

Thanks Clashh. I tried all that, too. Turned CnQ on/off. Bumped up ddr voltage up to 2.8v, vcore up to 1.5v, and still "no go".

Again, I'm at a point where I'm tired of having to "roll the dice" with each memory or mb try and spend endless hours search this forum, anandtech, viaarena, etc hearing countless stories of problems with potential solutions. If I was overclocking, then I could accept that, but I just want basic/standard performance.

That last straw was I got fed up with this and the 80 mile roundtrip drives to Fry's to return the bad memory, board, etc. I must say that while the Fry's sales and support team was useless in this exercise (other than saying "gee, we get a lot of memory returned"), they have a great return policy. Since I didn't want to make the 80 mile trek, I unloaded all that I got from them (MB, Proc, Mem, and HD). No problem. No restocking fee.

Now I'm somewhat back to where I started. I have a silverstone TJ06/S12 with my existing peripherals (video/sound card) in search of a CPU/MB. I guess I have a Tin Man in search of a heart (or is that a Scarecrow in search of a brain)

.

For the A64 fans, believe me, I wanted it to work. I love the promise of the price/performance and the cooling/quiet potential. For the few days I had stability, I had the CPU idle temp at about 32' C with the TJ06 windtunnel. It works as advertised. Very quiet and cool. My next step was to try passive cooling of the CPU with Silverstone's NT01. I understand the new P4 600 series has Intel's version of CnQ, so there is some hope of getting back to some of the A64's cooling/quiet potential. I will still try the NT01 and see where it goes.

Any P4 experts in the crowd? Any suggestions?

Thanks, again, for the support...

Post by **MikeC** » Mon Apr 11, 2005 6:54 pm

IMO, it's not an Intel vs AMD thing -- it's just supporting the technology that makes more sense for low noise & low power waste. But if it doesn't work...

I have to say the problems I had were not resolved by any setting of the CPU Vcore, mem voltage or any other switch in the endless DFI BIOS. I finally did a component-by-component swap/analysis and identified the culprit as the motherboard All the components used on the board worked perfectly on my 2nd same-model board without any tweaking.

It was a lemon board.

I pushed & shoved & threatened & almost raised my voice at this store with not so good a return policy & finally got them to give me an exchange for a new DFI board.

It booted into Windows on first try without any touching of the BIOS with all 4gb of RAM. It's purring away running memtest86 right now.

With CPUBurn, at max it, draws <140W AC. That's only marginally higher than the p4-2.8 northwood system I have as my main system these days. With the PSU in the A64 system I am using, barely 100W is delivered to the components. A comparable Intel 660 (3.6G) draws ~180W AC in a similar system, and overheats with the same HS model. (IE, starts throttling at 70C & keeps throttling at that them.)

Post by **NeilBlanchard** » Tue Apr 12, 2005 7:04 am

Hello:

I totally agree with what Mike is saying here: I have built 4 Athlon 64 machines, each with different motherboards, and different power supplies -- and ALL of them were totally stable. The most recent was also the most extreme:

Athlon 64 4000+ (Newcastle) with 4GB Corsair RAM, a Gigabyte motherboard, and a S12 500watt PS -- and it has never crashed, even with some undervolting to 1.4v.

Also:

Athlon 64 2800+, 1GB Mushkin, AOpen mobo, CleverPower 300watt PS, undervolted to 1.375v

Athlon 64 3000+, 1GB Muskin, ECS mobo, Fortron PS

Athlon 64 3200+, 1GB Corsair, Asus mobo, Seasonic PS

I have used either Corsair or Mushkin RAM; some of it the expensive stuff, but also some of the low priced stuff. I only had to "experiment" on one of these machines, with a different motherboard and different RAM -- this was mostly in order to get CN'Q working.

I also built a number of Athlon XP machines and all of them are stable, too. So my point is: you cannot assume that all AMD systems are unstable or hard to build, based on one situation.

The one thing that you can almost guarranty by moving to an Intel-based system: that it will either run hotter and/or louder than a similar AMD system.

el_ · Post by **el_** » Tue Apr 12, 2005 12:21 pm

mikec you are running the dfi with 4gb ram? did you realize the divider lock when running 4sticks? try just ram in slots 2 and 4 and you should get much better memory bandwidth. new bios 3.31 v1 comes with memtest 1.55 and should really help test for stability. goto dfi street for more info.

java the mushkin blue is great ram but you must give it tons of voltage and there are about 6 different IC, pcb and even colors for the ram. memtest it at 2-3-3 2.7volts but it will do 2-2-2-5 with 3.2volts if bh or ch flavor if you got mosels then you didn't buy from outpost/fry's.

talk to ninja at the mushkin forums tell him el sent you and he should help ya.

andyb · Post by **andyb** » Thu Apr 14, 2005 3:35 pm

Had masses of problems with an MSI RS480M2 mobo with RAM, using a 3000+ winchester CPU, got anther system order using one of them boards and 2 different types of RAM turning up.

Yes the CPU's/Mobo's are very picky about RAM, but from experience, the are very consice, they either work perfectly or dont work (within 3 mins).

I have used many system with different RAM, they are all picky, Intel used to make the pickiest Memory Controllers of them all, then VIA got theiract together and we have all been spoiled........ until now.

Just pick your RAM carefully.

By the way, the problem is probably the lack of power getting to the componens on the board.

Andy

Java · Post by **Java** » Thu Apr 14, 2005 6:16 pm

By the way, the problem is probably the lack of power getting to the componens on the board.

Thanks, Andy.

Why do you think that and how and I diagnose? That was the orignal intent of my post (it started in PSU forum). Even after I "correct" my ATX-12v connection problem, it was still unstable. In the back of my head, I'm wondering if the psu is undersized. At 380w and being very efficient (seasonic SJ), I thought it would be plenty. I've read some post saying get huge PSU's and others saying you could never justify a 300w. How can I rule out the PSU as being undersize or faulty?

Thanks, again

el_ · Post by **el_** » Thu Apr 14, 2005 10:48 pm

two rails with 25A combined shouldn't be a problem. post your voltages especially 12volt voltage.

Post by **MikeC** » Thu Apr 14, 2005 11:02 pm

el_ wrote: divider lock when running 4sticks

All that RAM is need for this machine, which will do very intensive Photoshop with near-1GB images 7 video rendering. Do you know if tthere is a workaround to...

1) make the board recognize all 4GB instead of only 3.2?
2) make the mem run at its spec'd 400 MHz instead of 333MHz?

el_ · Post by **el_** » Thu Apr 14, 2005 11:11 pm

1. now that is weird only sees 3.2gb or your ram?

2. I think it is a problem with the on chip memory controller and might be fixe d with the E stepping or Venice Core. I don't know of any work arounds right now but maybe new bioses will address the problem.

What memory bandwidth do you get with sisoft sandra? should be about 7000

andyb · Post by **andyb** » Sun Apr 17, 2005 12:42 pm

Hiya Mike.

RE: Your 3.2GB of RAM situation.

I have been informed by a very reliable source that the 3.2GB is a windows limitation, and should be resolved with XP64, a dual Xeon running Windows Server (unknown version), had the same problem, it's not CPU, or Chipset, or memory bound.

Older Windows OS'es used to only see 2GB, and needed registry tweaks to get it to see more, 3.5GB seems to be the limit (you may be able to tweak Windows to see another 300MB), the other 500MB apparantly seems to be used for caching the hardware.!!! (strange concept, did'nt really understand that one).

The issue with the A64's running with 4 DIMM's at 333 is the Memory controller.

The current limitations are:

1, As we all know, picky about RAM (board design is also an issue).
2, 4 x Double sided DIMM's automatically locked to 333.
3, 4 x Single sided DIMM's can only run at T2 timings at 400.

Venice will change that, the new memory controller in Venice has promised to:

1, Make systems less picky about RAM (very much needed).
2, Allow 4 x Sticks of Double sided RAM run at 400, but only at T2 timings.
3, Allow 4 x sticks of Single sided RAM run at 400, at T1 timings.

Opteron systems can of course work with 4GB RAM (3.5GB OS limited) at 400 when using Registered RAM, presumably some of the Server versions of Windows can actually make use of larger quantities of RAM, it would be an idea to find out which ones.

Regards Andy

Post by **NeilBlanchard** » Mon Apr 18, 2005 5:38 pm

Hi Mike:

MikeC wrote:1) make the board recognize all 4GB instead of only 3.2?
2) make the mem run at its spec'd 400 MHz instead of 333MHz?

1) Run WinXp Pro 64?
2) Get the next generation Athlon 64? I don't know if Wincesters have this issue -- is yours a Winchester or a Newcastle?

Post by **MikeC** » Mon Apr 18, 2005 5:51 pm

It's a Winchester. I had a chat with an AMD rep today, who suggested the problem would probably not be resolved with a Venice core. I didn't quite catch the full drift, but he seemed to be suggesting that the Venice core simply allows a wider variety of RAM/timings. There is no fundamental change in the mem controller. But don't quote me on this; it was a short chat.

I also had a long chat with a Corsair mem tech who said they had 2 DFI nF4 boards running with 4 1G sticks of Corsair mem like mine. We went through >30 mins of BIOS tweaks w/o luck.

The system still crashes when trying to boot into Windows. This did not come up as an issue with the replacement DFI board right away. Maybe I didn't notice because it booted up fine with all 4 sticks the first few times. Bit like throwing dice...

Finally, the Corsair man asked me to run memtest86+ on each stick of RAM separately. First one has passed 3 times now. I suppose if they all pass, he'll suggest yet another install of windows.

Green Shoes · Post by **Green Shoes** » Mon Apr 18, 2005 8:00 pm

Almost makes you want to switch to mac, doesn't it? They can supply you with up to 8 gigs of RAM stable in your G5 setup.....provided you're willing to part with $3050 just for the ram

Post by **NeilBlanchard** » Tue Apr 19, 2005 8:58 am

Hi Mike:

Which Corsair RAM are you using? I had no problems at all with a Newcastle Athlon 4000+, on a Gigabyte GA-K8N Ultra-9, with four Corsair XMS 1GB sticks similar to these:

http://www.newegg.com/app/ViewProductDe ... 560&depa=1

They ran quite warm, though not alarmingly so; and they ran at 333mHz at the stock latencies. I could not improve on any of the speeds -- but it was totally stable.

Post by **MikeC** » Tue Apr 19, 2005 9:56 am

TWinx 2048-3200C2. Basically, 2 matched pairs of 1024s. EAch of the sticks proved to be without errors in 3-4 passes (by itself) of memtest86+.

I've decided it HAS to be a windows issue. Maybe the flaky mem settings caused some kind of data corruption somewhere that none of my dianostic tools are able to pick up. So I am installing Win XP once again. If this fails to solve the problem, it's time for a drop kick from the 2nd fl balcony.

andyb · Post by **andyb** » Tue Apr 19, 2005 3:32 pm

I had a chat to someone today, I have nothing against DFI (never used a board), I was told all DFI mobo's made over the last 2ish years have all been RAM picky, Athlon, Athlon 64, AND P4 boards. So, It may not be a memory controller, chipset, or CPU issue, just a DFI issue.

See my sig, I use to have 2 x 512MB Kingston C2.5 sticks, they are flaky in an MSI RS480M2 mobo, same CPU + PSU, and latest BIOS, it must be board layout, works OK @ 333 C3, will get back to you on 2 weeks when I have tested with GEIL ram (listed as good fo this board).

Also Mike, one last thing, if youy have a drive nuking tool, use it, and re-install, using the most stable RAM you have working with that board, image the drive, and test it with different RAM modules and combinations.

Regards Andy

PS: This may be a silly question, but how do you actually get memtest to work. Boot disk.???, what settings.???

Java · Post by **Java** » Tue Apr 19, 2005 3:41 pm

If this fails to solve the problem, it's time for a drop kick from the 2nd fl balcony.

You should get some good hang time from the 2nd floor.

Kidding aside (I know it hurts), this is why I'm thinking about Intel. I can't chew anymore time doing the endless "New MB/Memory/Reload Windows Dance".

I keep talking about getting an Intel vs actually doing it so I guess I'm waiting for someone to tackle me and say "oh it's just your X", but your problems aren't giving me hope.

Help - unstable A64 system

Help - unstable A64 system

problem

Intels can be quiet too