2.4C and P4P800 and stability - long post warning

[GotThaJuice]

EDIT: I've accidently posted this in the CPU section instead of INTEL CPU's section. I can't seem to delete this post (no permission). Could a moderatore please move this thread?

Greetings!
I've bought a new system a few months ago. Naturally, I've tried to OC it, since I've been doing that ever since my Celeron 333A.
I've been experiencing stability issues over 250FSB over the past two months. I've almost gone crazy while trying to find out what's wrong, but now I think that I've finally come to the bottom of things.


A few words about my system
For testing purposes, I've removed all but the very essential components of my system.
I'll try to keep this post as short and to the point as possible, so please bare with me.

System specs (stuff that is currently inside my case):
1) Enermax 460W PSU
2) ASUS P4P800 (unmodded)
3) P4 2.4C Q323A123 SL6WF with 1.525 vcore
4) 1 stick of Corsair 256MB VS PC3200 RAM
5) ABIT Geforce 4 Ti4200
6) Seagate 120GB 7200RPM HDD


Cooling for CPU & NB:
CPU heatsink: Spark 7's copper HS with AS-5 thermal compound
CPU fan: custom 80x80mm
NB heatsink: thoroughly cleaned stock HS with AS-5
NB fan: Spark 7's stock 70x70 fan @ 6300RPM - this fan also helps cool the CPU heatsing, as it is situated directly below the CPU fan

Ambient temperature: 20°C or lower
My case is completely open during these tests.



Overclocking and issues...
I use 2×PRIME95 with in-place FFTs for stability testing. To me, a system is rock stable if it can do at least 12 hours of 2×PRIME, F@H, ...

250FSB -> ROCK STABLE:
- vcore: default
- vdimm: 1.85
- vAGP: default
- divider: 5:4
- timings: 2,3,3,6


As I have mentioned before, I've been experiencing stability issues about 250FSB. I've discovered, that after I remove / disable a PCI card from the case, the stability increases (Skystar 2 DVB-S PCI card, onboard NIC, onboard SOUND).
The thing that has kept me puzzled the most is that:
a) increasing the vcore even to 1.65v does NOT change anything
b) fiddling with various settings doesn't help
c) fiddling with RAM dividers & timings doesn't make a difference at all
Read on!


260FSB -> Hmmm?
- CPU temp does not exceed 43°C
- System temp does not exceed 30°C
- vcore: 1.6v for testing purposes
- vdimm: 1.85v
- vAGP: various (see below)
- divider: 3:2 for testing purposes (I also tried 5:4 to no avail)
- timings: 2.5,3,3,6 for testing purposes

260FSB -> SYSTEM BEHAVIOUR:
- IF onboard NIC&SOUND are ENABLED, system will totally lock-up after 30 minutes of prime.
- IF onboard NIC&SOUND are DISABLED, the system will reboot itself after 60 minutes of prime.
- IF onboard NIC&SOUND are DISABLED AND vAGP is 1.80v, the system will reboot itself after 60 minutes of prime.
- I've noticed that if the onboard NIC&SOUND are ENABLED, increased vAGP helps the stability a bit (it adds a little time
before the system locks up or resets itself).

I think that my mobo is having severe issues. It can't handle mediocre FSB with stock chipset voltage and if any PCI cards are installed / enabled.

I have pretty much dismissed the possibility that the limit of my CPU has been reached, because:
a) 277FSB @ 1.575v: It boots into windows and performs all the tasks normally, except longer intervals (above 10 minutes) of prime and other similar programs that stress the system.
b) It reaches the WELCOME screen of WindowsXP at 300FSB with default voltage! After that, it crashes horribly.
In a) and b), my system was fully loaded (pci cards, 4 sticks of ram, cdrom, ...).



So since I've gone halfway mental because my system has been giving me gray hair for the past two months, I need you guys to either confirm or dismiss my theory that I HAVE A CRAPPY MOTHERBOARD.

If not, then what do you think that the problem is? Am I doing something wrong? What are the things that I haven't yet tried in order to solve this problem?

If so, the only solution I see is that I buy a new one.
Please don't hesistate to post suggestions or experiences with various motherboards. I'd definetly go for i875.
How are IC7 and P4C800 like? How do they fare with high FSBs? Do they cause any problems with enabled onboard features, etc.? Do all the i865/i875 motherboards experience vcore fluctuations (idle vs. load) until you do a vdroop mod or are ASUS mobos the only ones?



Sorry for the long post, but it's hard to squeeze two months worth of testing into one post . I'd really appreciate it if you'd try to answer most of my questions. Sorry for any typos, I'm trying my best .

GotThaJuice

 

[puji]

As described and discussed in n many threads, the main issue the p4 has is at a certain fsb is the divider, memory and the mother board. Only certain types of memory and motherboards will work with a divider over 250 fsb. Also certain setting of the memory for write and read can severely affect the stability.
I have an AI 7 and the kingston 3500 I used in the beginning did not let me go over 250 stable. Reading up on this board I discovered many people had that problem and switched to BH5 type memory. I did the same and since then I am only limited by my CPU and heat. At my mother board there is also a GAT (read and write) setting and if I do not disable the last two setting is has the same effect: the system is unstable over 250 with the divider set 5:4.
By the way one stick was much more stable then two. I would advice,
either to get good ADATA memory which goes easy over 280 1:1 or try to hunt up some BH5 like the older kingston 3000 /3200.

 

[GotThaJuice]

I see. Well, that certainly sheds some light on the whole situation. And it actually makes sense .
Maybe converting to a 2.8C/2.8E is another solution, since it's unlikely that you'd need more than 250FSB on air.

 

[batboy]

Do you have the AGP/PCI locked to 66/33?

 

[aNTiChRisT]

Just something i stumbled across, duno if its relevant... just out of memory it was http://support.microsoft.com/default.aspx?scid=KB;en-us;q155053

~t0m

 

[Sophisticated]

update your Bios to the latest???? if not do so....about the 2 instances of prime....you do not need 2 instances of prime...WHY??? because "VIRTUALLY" hyperthreading gives u to processors but in reality you have 1....HT is basically for performance increase...not for stabilit testing....Prime is more for the reality side....so try 1 instance of prime.....about your vdimm... why is it 1.85 or can the P4P go that low...i thought it stops like at 2.55vdimm????

 

[GotThaJuice]

batboy: Yes, my PCI/AGP is locked at 33/66. Spread Spectrum is also disabled. I am using the latest bios (I think it's 1.016). Just thought I'd mention that too.

Sophisticated:
Using latest bios. Ooops, I meant 2.85v... Sorry about that .
If I use only 1 instance of prime, my CPU load is only 50%. That's not much of a stability test IMHO.

aNTiChRisT:
You're psychic, right? I've noticed a black screen with a blinking indicator at the top for about 2 seconds before Windows loading logo appears. I'll get this fixed . I'm not sure that this is relevant to my instability problems though - my system has been behaving the same way for the past two months and that "black screen" has only been present since yesterday. I guess that all those numerous crashes followed by reboots can't be too got for the HDD

 

[Sophisticated]

its at 50% VIRTUALLY because of hyperthreading like i mentioned.....take hyperthreading off if it makes u happy and that will put your CPU at 100% Virtually and in reality since no hyperthreading

 

[GotThaJuice]

So why is my load temp almost 10 degrees lower when I only use 1 instance of prime, instead of 2? Is that virtual too?

 

[Csybe]

its at 50% VIRTUALLY because of hyperthreading like i mentioned.....take hyperthreading off if it makes u happy and that will put your CPU at 100% Virtually and in reality since no hyperthreading

 

[aNTiChRisT]

Heh, i just remembered this thread, and saw it through random tutorial clicking. Np, hope ya get it fixed

~t0m

 

[Falkentyne]

So why is my load temp almost 10 degrees lower when I only use 1 instance of prime, instead of 2? Is that virtual too?

Because, both HT cores are being fully used. The chip is doing more work by multithreading, so thus you are going to get more heat.

You have to know how the core is designed and what HT is really all about. For one thing, hyperthreading itself allows better thermal load. This is why you will have idle temps as much as 6-10C lower than with HT disabled. (Test it for yourself).
Under load, 1 prime is more typical of what you will use your system for. Unless you want to test multithreaded programs, 2 primes isn't going to be realistic. You are *STILL* getting 100% CPU load with 1 prime-- it's just the 2nd hyperthreading core that's being unused. Put it this way: Do you actually think for one iota, that your framerates would double, (or even increase by 25%) if both cores were being used?

I don't remember the full details (and I'm not a chip guru), but as it is now (single threaded), 1 fully loaded HT core will run at the chip's main speed (example: 3.2 ghz). If both cores are fully loaded, you will be running at 2x1.6 ghz (still 3.2 ghz, but with the work evenly split).

Now, multithreaded applications will use both HT cores by themselves, running at 3.2 ghz total, while using one of the cores to do a task more effeciently, such as rendering water, foilage, etc, without slowing down the game considerably.

So, clearly, the chip is doing more work when both HT cores are loaded, thus more heat. But it's not working any _faster_. Dual priming is only testing simulated multithreaded performance.

Even hyperthreading isn't fullproof.
A badly written application can STILL improperly hog the CPU, slow things down far more than it should (though nowhere near as much as if you were using a non-HT processor) and raise temps higher than even prime95 can do, just by opening the application and doing nothing. (Eudora 4.2 is guilty of this), as well as some other apps that were written during the win95/98 days.

One thing to remember:
1 prime alone will heat your system far more than what a standard 3D game would do, *With* a hot video card pouring extra heat into the case.....(the only thing that can usually do worse as far as heat output, is a badly written app.....)

 

[freezerx]

I think can be your ram. it's at 3200 and at 5:4 it is at 400. @ 260 it's at 416.

 

[Sophisticated]

Because, both HT cores are being fully used. The chip is doing more work by multithreading, so thus you are going to get more heat.

You have to know how the core is designed and what HT is really all about. For one thing, hyperthreading itself allows better thermal load. This is why you will have idle temps as much as 6-10C lower than with HT disabled. (Test it for yourself).
Under load, 1 prime is more typical of what you will use your system for. Unless you want to test multithreaded programs, 2 primes isn't going to be realistic. You are *STILL* getting 100% CPU load with 1 prime-- it's just the 2nd hyperthreading core that's being unused. Put it this way: Do you actually think for one iota, that your framerates would double, (or even increase by 25%) if both cores were being used?

I don't remember the full details (and I'm not a chip guru), but as it is now (single threaded), 1 fully loaded HT core will run at the chip's main speed (example: 3.2 ghz). If both cores are fully loaded, you will be running at 2x1.6 ghz (still 3.2 ghz, but with the work evenly split).

Now, multithreaded applications will use both HT cores by themselves, running at 3.2 ghz total, while using one of the cores to do a task more effeciently, such as rendering water, foilage, etc, without slowing down the game considerably.

So, clearly, the chip is doing more work when both HT cores are loaded, thus more heat. But it's not working any _faster_. Dual priming is only testing simulated multithreaded performance.

Even hyperthreading isn't fullproof.
A badly written application can STILL improperly hog the CPU, slow things down far more than it should (though nowhere near as much as if you were using a non-HT processor) and raise temps higher than even prime95 can do, just by opening the application and doing nothing. (Eudora 4.2 is guilty of this), as well as some other apps that were written during the win95/98 days.

One thing to remember:
1 prime alone will heat your system far more than what a standard 3D game would do, *With* a hot video card pouring extra heat into the case.....(the only thing that can usually do worse as far as heat output, is a badly written app.....)

I think can be your ram. it's at 3200 and at 5:4 it is at 400. @ 260 it's at 416.

Falktyne words explained it all....for freezerX's comment that might be that problem also so put your Vdimm to the max it wont hurt Anything...alot of RAM can take more then 3.0v...

 

[Sophisticated]

havent seen this anywhere so

WELCOME TO THE FORUMS!!!!!!!!

 

[GotThaJuice]

WELCOME TO THE FORUMS!!!!!!!!

Thanks

for freezerX's comment that might be that problem also so put your Vdimm to the max it wont hurt Anything...alot of RAM can take more then 3.0v...

My vdimm is already at its max [2.85], so I'd have to do a vdimm mod to get up to 3v.
Even if I set my divider to 3:2, the situation is the same.
I am using 2.5-3-3-6 and my ram can do 215MHz with those timings.
I'll run memtest at 260FSB 5:4 2.5-3-3-6 to see if there are any incompatibilities with 5:4. I've already tried running it at 255FSB 5:4 2.5-3-3-6 and it's completely error-free (memtest that is), 260FSB is the next thing on my list.
You know, I really have no idea where to point fingers anymore .

 

[Sophisticated]

see now your gettin the hang of it

 

[puji]

Thanks


My vdimm is already at its max [2.85], so I'd have to do a vdimm mod to get up to 3v.
Even if I set my divider to 3:2, the situation is the same.
I am using 2.5-3-3-6 and my ram can do 215MHz with those timings.
I'll run memtest at 260FSB 5:4 2.5-3-3-6 to see if there are any incompatibilities with 5:4. I've already tried running it at 255FSB 5:4 2.5-3-3-6 and it's completely error-free (memtest that is), 260FSB is the next thing on my list.
You know, I really have no idea where to point fingers anymore .

I have to point to the memory again. The Ch5 would not do better with the more relaxed timing (because it is a memory divider problem with certain read and write timings). I even could show that at the same speed more relaxed timings had worse effects, like not even completing the boot sequence, instead of crash after five minutes. Increasing of the vdimm had little effect too. The AI7 has 3.2 volt vdimm. Did nothing for stability except at 1:1. With the bh5 it is different, except these relaxed timing phenomena.

 

[Sophisticated]

its only so much his RAM can do man chill i can see if it wus CAS 3 or something but its 2.5 so thats still good and not that relaxed...relaxed is something like 3-4-4-8.....and if u didnt notice he got the OC problem fixed

 

[GotThaJuice]

OK, I have been runin memtest86 @ 260FSB 5:4 2.5-3-3-6 for 10 hours and 26 minutes (76 passes) - no errors.
And I have no idea, what else to try Above 250 FSB, my system just decides to lock up at a certain point of load. That's so wonderful .