4GB of DDR400 not working... Any ideas?

toothpick

Distinguished
Jun 29, 2006
9
0
18,510
I am working on a customers PC and for once am completely stumped on a problem. It is a Dell 4600 P4 3.0GHz with an Intel 865 chipset. It has 4 DIMM slots and 4GB max memory. It has 4 x 1GB DDR400 Micron (I think) chips in it running in dual channel.

Problem is the memory fails Goldmemory when all 4 chips are installed. It passes every time with 1, 2 (dual channel) or 3 chips in. But when you stick the 4th chip in it freezes every time in random places during the test. Also it only shows 3710MB is there. I know it may not show the whole amount but with 2GB installed it shows 2047MB so thats only about 1MB less than actual. So why is it nearly 400MB off with 4 chips installed?

I have checked each chip individually and they all pass, no matter what DIMM slot they are in. Sisandra shows they are all identical chips manufacturered on the same day. All of the chips are fine, they just dont work when all 4 are there. Today I even tried a different, high end motherboard and got the same results.

This customer is very insistant on having 4GB (not sure why...) and spent over $1000 just on the memory so far so I need to work this out somehow. Any help would be greatly appreciated.
 

maury73

Distinguished
Mar 8, 2006
361
0
18,780
You'll never see 4GB of RAM, simply because the 32bit memory limit is 4GB of total memory, including ROM BIOS and memory mapped I/O.
No matter the CPU can address 48 or 56 bit, BIOSes are still limited to 32bit addressing modes and Windows XP is a 32bit OS.
Unless you get a 64bit OS you'll never be able to use all 4GB of RAM.

The freezeing issue could be a signal integrity problem: try lowering the memory clock to 166MHz (DDR333), since very little boards and very lucky DIMMs can run at DDR400 with 4 sticks connected.
 

h_alone

Distinguished
Jun 16, 2006
5
0
18,510
If your system is getting unstable when yiou have the 4 DIMMs, Check how many ranks is each of the DIMMs. Best is if each ram has only one rank. The chipset supports upto certain number of ranks. if you get 2 ranks DIMMs, it cuts the number of DIMMs you can use in the system.
Normally the DIMMs with lower rank are more expensive. They are used mainly for servers to alow higher RAM configuration
 

angry_ducky

Distinguished
Mar 3, 2006
3,056
0
20,790
You can't address all 4gb of RAM without a 64 bit OS, such as Windows XP x64, certain Linux distros and the forthcoming Windows Vista.

And then you're screwed because that P4 probably isn't a 64-bit CPU.

I can't see how anyone could need 4GB of RAM, let alone spend $1000 for it. This kind of stupidity is common in Dell users, I hear.
 

toothpick

Distinguished
Jun 29, 2006
9
0
18,510
Thanks for all of your help. I agree having 4GB of RAM is completely pointless unless you are running a server and I have told this customer that but he insists. He even paid for me to get a whole new case/mb in order to get the 4GB working right. I tried to talk him into letting me build an AMD 64Bit system as that is all I build/use. But he wants Intel.

I took Maury's advise and lowered the chips to 333Mhz and left the test running all night. It passed 6 times without any problems. I guess I can leave it like that but he is not going to like me slowing down his RAM, even though I know this will not make a noticable difference in speed. According to Goldmemory it benchmarks just as fast when set to 333Mhz.

Not seeing the entire 4GB is not a big deal. I just want to make sure it passes all tests because I have been getting random memory related blue screen errors and that is unacceptable.

If anybody has any ideas on how to get it working while running at 400Mhz let me know. I considered raising the voltage a little. You think that would help? Otherwise I will just leave it at 333Mhz and tell him thats the best I can do.

BTW what is the max memory supported by 64 Bit? Before now I was unaware that 32/64bit deturmined the maximum memory. Other than this guy, nobody I know has ever had over 2GB of RAM so I guess I never needed to know.

Thanks again for the information guys!
 

maury73

Distinguished
Mar 8, 2006
361
0
18,780
Hi there,

With 4 DIMMs the problem is the longer PCB tracks that cause signal integrity decay: long tracks raise the parasite capacitance and above a certain lenght the signals switch so slowly that they're impossible to sample at 400MHz.
The signal tracks on regular mobos are terminated and their lenght tuned for best operation with 2 DIMMs and it's simply impossible to tune them for both options (2 & 4), only with registered DIMMs or very high voltages it's really possible.

I personally saw 4 DIMMs running at DDR400 stable with registered DIMMs on server mobos only. The only desktop system I saw running like that is an A8N mobo with 4x512MB Kingston ValueRAM with voltages raised by 200mV (very dangerous!) and slower timings.

Try the best compromise: high speed with high latency or low speed with low latency.
Anyway don't raise voltages abose 100mV over stock recommended setting.
 

h_alone

Distinguished
Jun 16, 2006
5
0
18,510
It is not just the track length. Each motherboard have a list of supported ram configurations. It is a matter of loading capacitance on the chipset (or uP for AMD). The loading capacitance is track length + input capacitance of the ram modules (ranks).
Since the motherboard designers know their track lengths, they quote you the maximum number of ranks you can have on each channel. If you exceed that it may work when the system is extra cold and break when the system heats up.
Raising the voltage is just an overclocking and you have to be careful not to fry your rams. It will also affect the life time of your modules. Lowering the speed should be OK. However it doesn't guarantee that the system will remain stable when it gets quite hot.
 

maury73

Distinguished
Mar 8, 2006
361
0
18,780
The problem with high temps is due to the RDSon characteristic: MOS RDSon increases with temp resulting in higher output impedance and higher propagation times.
Input load capacitance is not a so big issue: as an example Samsung DDR400 chips have an input capacitance of 0.5pF @2.6V+0.1V per pin. So if you have double sided DIMMs you will have a total of 8x2x2x0.5pF = 16pF per pin on each channel (not considering the termination resistors that make the buffers "see" a lower cap) and any SSTL-2 output buffer must drive at least 50pF with SR=25 and Z=50 with the Weak Strenght setting (and DIMMs use Full Strenght).
 

oenomel

Distinguished
Jan 23, 2006
160
6
18,685
I'm not a memory guru..but I thought I read somewhere that changing the mem from 1T to 2T helped in this situation.....

Of course yer still slowing it down.......just not as obvious...

Anyone else hear this? Or am I wrong?
 

Nitro350Z

Distinguished
Apr 19, 2006
416
0
18,780
hmm... Thats intresting, I always thought 1Tb was the limit for 64bit, I guess I know better now.

Anyways, couldn't you put more than 4gb of memory in a 32bit system if the memory was fully buffered and registered? FBDIMMs is what I'm thinking of, but they are really expensive and intended for servers. I'm just wonderin.
 

mesarectifier

Distinguished
Mar 26, 2006
2,257
0
19,780
No. In the case of an Opteron/A64 system, they could run >4gb RAM, but the 32bit OS doesn't permit it. I'm not an expert so I can't tell you what, but something about 32bit code cannot address more than 3.7something gigabytes of RAM.

You could have 16gb in one of those systems and still only see under 4. As soon as you install a 64bit OS, all the RAM will appear.
 

1Tanker

Splendid
Apr 28, 2006
4,645
1
22,780
I'm not a memory guru..but I thought I read somewhere that changing the mem from 1T to 2T helped in this situation.....

Of course yer still slowing it down.......just not as obvious...

Anyone else hear this? Or am I wrong?
Yes, but that's on Athlon64 mobo's. Usually Intel mobo's can handle 4 sticks of RAM better. ie: 4x512MB would be fine, but with 4x1GB you run into OS problems.
 

Corasik

Distinguished
Jan 24, 2006
92
0
18,630
No point anyway with 32bit windows. XP is limited to 2Gig of memory for applications, that can be 'hacked' up to 3gig but the applications have to be compiled to recognise that mode anyway, which excludes 99.9% of software.