Why can't I overclock??!
 

After reading about people getting 4+GHz overclocks with the "E" stepping P4 Prescott 3.0, I spent months finding the CPU (spec code SL7PM). But the best I can get out of it is 3.45GHz (prime95 stable).

Above 3.45GHz, the small FFT torture tests which stress the CPU are fine. But when I choose the large FFT torture test, it fails after about 10mins. I take that to mean that my memory or motherboard is failing.

Here's my system specs:

Abit IC7 @ 230MHz FSB
2x512MB Kingston HyperX PC4000 (500MHz DDR)
80GB Western Digital 7200rpm JB
GeForce 5700LE
450W Eye-T power supply with PFC (came with case)
Thermalright SP-94 copper heatsink
Panaflo 92mm fan for heatsink
4x Panaflo 80mm case fans

The FSB is set to 230MHz, so the memory runs at 460MHz (below the 500MHz spec of the Kingstons). The timings are set to 3,3,8. PCI/AGP clocks are locked to 33/66MHz. I checked to see that no thermal throttling is occuring (i.e. not overheating).

Anyone got any ideas why I can't get past 3.45GHz? Is it the power supply (don't have anything demanding on it apart from the CPU).

because every cpu is different urs might only be stable to 3.45 ghz otherwise it loses stability hence the drawback of overclocking someone might get something better on the same setup as you. moral of story you will have to live with it.

Generally when small FFT torture test works and blend does not, then memory is the culprit.

Try overclocking with just one stick of RAM. Then try the other. Let us know how you do.

Try running memtest86 to verify your memory operates well at the overclocked speeds. Maybe give sticks of ram extra cooling like a heatsink.

Also, if your bios supports it you may want/need to slightly increase the voltage provided to your cpu and/or memory to permit the faster operation.

You could also try swapping out a borrowed power supply unit, not just for different watt capacity but also stability of voltage rails can be an issue. Try disconnecting any high-draw components eg second hard drives, optical drives, if you have a major graphics card put a cheap one in temporarily which uses less juice. Obviously if your testing still falls over you can reconnect these.

Try massively cooling your system with (say) a 12 inch desk fan and/or putting it by an open refrigerator. This will give you a clue if thermal problems are the culprit. (in which case you would proceed to take more practical cooling enhancements).

I would have thought you could get stable operation faster than you are, but as moo pointed out, sometimes individual chips are less overclockable than others in a batch. In this case you either have to live with it or find a nice person who doesn't like overclocking their rig, and who is willing to swap their overclockable chip for your less capable one.

Some success
 

Thanks everyone. I reduced the memory timings from 3-3-3-8 to 3-4-4-8 and managed to get more stability out of it. I'm now running at 3.5GHz (a 16.5% overclock). :razz:

Strange thing now is that I can still push it further to 3.6GHz. Prime95 runs fine with no errors. But after about 15-20mins, the computer spontaneously rebooted. Temperature of the CPU was 71.5 degrees at that point. :nuke:

Any ideas about that? This time around is it the power supply? Or should I just be happy with 3.5GHz, call it quits and leave it alone? :surrender

Here's the benchmark results:

Intel(R) Pentium(R) 4 CPU 3.00GHz
CPU speed: 3499.45 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2
L1 cache size: unknown
L2 cache size: 1024 KB
L1 cache line size: unknown
L2 cache line size: 128 bytes
TLBS: 64
Prime95 version 23.5, RdtscTiming=1
Best time for 384K FFT length: 11.854 ms.
Best time for 448K FFT length: 14.454 ms.
Best time for 512K FFT length: 16.106 ms.
Best time for 640K FFT length: 19.160 ms.
Best time for 768K FFT length: 22.962 ms.
Best time for 896K FFT length: 27.739 ms.
Best time for 1024K FFT length: 30.964 ms.
Best time for 1280K FFT length: 41.066 ms.
Best time for 1536K FFT length: 49.327 ms.
Best time for 1792K FFT length: 59.371 ms.
Best time for 2048K FFT length: 66.544 ms.

You should get a better cooling device. 71.5 deg is a lot. Also check the mobo temperature. It will give you a good indication whether the chipset is heating too much. Memory overheating is also something to be considered.

maybe this will help.
[url="http://forums.extremeoverclocking.com/showthread.php?t=79266"]http://forums.extremeoverclocking.com/showthread.php?t=79266[/url]

I am running a Prescott 3.2 GHz at 10% overclock (3520 MHz) prime stable. Maybe I can go higher, but I (still) haven't had time to try.

I did however (contrary to the Northwood) get higher throughput by running two instances of Prime95. It seems that Prime95 can't fill the ridiculously long 31 stage pipeline of Prescott (and that's the reason Prime95 is slower on Prescotts, I guess). But you can make up for that by running two instances.

But you'd better time this yourself, because I'm using fast dual channel 400 MHz memory. And for Northwood, don't even consider it (unless one instance is running trial factoring).

Patrik, how do you run 2 instances of Prime95? Do you set the processor affinity?

I didn't set processor affinity, I just started one copy with
./mprime -d >> log.txt &
and the second copy with
./mprime -A1 -d >> log0001.txt &
Here I am using the Linux version, mprime. They have been running on different logical processors when I have watched them with "top" a few times. So I guess I don't have to set the affinity.

Also I think you can just run prime95.exe from two different folders. But that would be like two seperate computers (from PrimeNet's point of view) because it wouldn't share worktodo.ini etc.