FSB Hole or Max OC Reached?

[Gophermofur]

Yesterday I was overclocking my e4500 but ran into some difficulties when I tried to push it to 285 x 11.

Well, I finally got my Ultra-120 Extreme yesterday and temps dropped from 67 degrees to 53 degrees under load (Initially using stock Intel cooler). I figured there was still more room to overclock. The settings I was running at was 275 x 11 (3.025Ghz) and I can't remember the Vcore off the top of my head.

When i tried pushing the cpu to 285 x 11 I ran into some stability issues. Constant restarts, failures booting into windows. So I proceeded to increase the vcore by one notch at a time until it became stable enough to boot/bench. I had to increase it 3 notches/levels to get it to boot properly and not crash in windows. It took another 2 notches to get it to run orthos for 21 minutes before failing. I suppose I will have to increase it more to get it stable.

Questions for the experts:
1. That seems like a LOT of extra vcore to bump up the chip 110Mhz.
2. Could the problem not be related to vcore?
3. I've heard about the DS3 having FSB holes, but I thought they are usually found at a much higher FSB. Any chance this is a possibility?
4. Suggestions?

Btw, temps sat at about 57 on the last orthos test (load) before erroring out.

Thanks,
Mo

 

[Noubourne]

Generally the closer you are to the max chip OC, the more voltage it will require to be stable, and voltage requirement for each increase to clock speed starts to increase on a near exponential ratio in my experience.

You start off getting all kinds of Mhz for free, and then you'll start needing voltage to get it stable at a new clock. The higher the clock, the more volts needed to make it stable.

Sounds like you might have hit your chip max. Maybe someone experienced w/ your board can help you know if you're even in an FSB hole range. I'm not sure you are.

Of course CPU volts is only one factor. If you're pushing your RAM and NB harder, they too might benefit from a voltage bump.

 

[SerpentRoyal]

P965 is not a great overclocking board. Use two sticks of RAM. Set Vdimm to 2.1 with 5-5-5-15-2T timing. Bump Vcore to 1.46 (default should be 1.32). Retest.

 

[Gophermofur]

I will try your suggestion Serpent. Just as clarification, I'm running RAM at 5-5-5-15, with 1.9 Volts. The tests I am running are Small FFT's.

I hear max Vcore is 1.55. Is that a safe bet assuming temps are kept below 60 degrees?

Thanks,
Mo

 

[SerpentRoyal]

I would limit Vcore to 1.48 or 1.5. No need to go higher on air since heat will cause the CPU to lockup.

Be sure to use 1:1 memory divider to limit your memory speed to about 300MHz. This should eliminate your RAM as a potential failure mode.

The average overclock for E4x00 is around 3.2GHz.

 

[Gophermofur]

Yeah, the ram is running at 1:1 for sure. If what Noubourne is saying is correct regarding vcore needing to grow exponentially for additional overclocks, then it makes sense that I hit my max. At this point the extra 5 degrees may not be worth the 100Mhz. In any case, I'll see what I can squeeze out of it and then report back. Thanks for the suggestions.

 

[SerpentRoyal]

In general, a 0.04 bump in Vcore should net an additional 90MHz core speed. When the CPU is within 100-120MHz of the maximum stable overclock, you'll see a much smaller gain in core speed.

My chip is good up to about 3.58GHz, but the graph of Core Speed vs Vcore is only linear up to about 3.46GHz.

Also note that there is very little real-wold difference between C2D @ 3.0GHz and 3.6GHz. The best value right now is the E2160. E4300/4400 strikes a balance between price and performance. E6xx0 is over-priced. I wouldn't touch any 65nm Quad. Too much heat and no killer app to take advantage of this technology. Wait till late 08 for 45nm Quad or when your 3.0GHz C2D becomes the bottleneck with your applications.

 

[Gophermofur]

In the end, I decided the extra voltage just wasn't worth the few MHz difference. I think I might have been able to get the CPU stable at 3.2 (it was running past 6 hours) but the voltage was quite high and the temps were around 66. By comparison, I managed to get it stable (sort of) at 3.025 with a max temp of 56 degrees under load.

When I run Orthos with Small FFT test it runs fine for over 24 hours. When I run Memtest, it runs fine for 24 hours. But Blend test in Orthos seems to error out after 8 hours... any ideas as to why? Is this stable enough to run on a daily basis?

 

[SerpentRoyal]

I only run Orthos for one hour. If I encounter problem down the road, then I back-off the FSB by 2 or 3MHz. No need to torture your CPU under Orthos for one day!

 

[Navid]

Temporarily, drop your CPU multiplier to 10.
If it is the CPU that is the bottleneck, that should allow you to raise the FSB now that the multiplier is reduced.
If Orthos still fails, you know that it is something else like the FSB.