CPU core over-voltage and Sempron overclocking?

[BlueWeasel]

I finally made the jump over the weekend from a budget XP mobile to a budget Sempron 3100+.

Sempron 3100+ w/ TT Venus 7 HSF
Epox 8KDA NF3
2x512MB GEIL PC3200 Value RAM

I picked up the newest E6 revision of the 3100+ Sempron (90nm, 256k L2, 1.40v stock voltage) and an Epox 8KDA3I NF3 board off the FS/FT trade forums.

I spend the better part of today following Zebo's A64 OC'ing guide trying to find out the maximum HT/FSB/CPU/memory values for the MB and chip, and here's what I've determined the maximum values to be:

HTT/FSB = 320 @ 3x
CPU = 2.55 to 2.65Ghz (see below)
RAM - 215 @ 2.5-4-3-7-1T

I was suprised and thrilled to see the board be able to hit 320Mhz. However, for some reason, the Vcore seems to be overvolted by about 0.1v. When running at the default voltage, the BIOS reports the Vcore to be 1.48v. Bumping the voltage up in the BIOS to +0.05v gives a reading of about 1.57v, and bumping it up further to +0.10v gives a reading around 1.67v.

Given a ~10% maximum voltage increase for overclocking, I'm a little worried to go above 1.57v. However, the load temps never go above 50C even when the Vcore is reported as 1.67v.

Using the max HT/FSB of 320, the best OC I've gotten so far is 2.56Ghz (8x320) at 1.57v. I lowed the multiplier down to 8x, but the RAM is running at 213Mhz, which is just below the max value of 215 Mhz.

Would it be worth it to run the Vcore at the 1.67v reported voltage for a higher clock speed (say an additional 100Mhz) at the expense of a significantly lower FSB and RAM bandwidth?

 

[furballi]

Double check by downloading and running CPUz. I'd limit Vcore to 1.55 max. Lowering the RAM speed to 133 from 166 will cost you about 50MHz core speed.

Have two sticks of Corsair Value Select 2 x 512MB ($50AR at Fry's). These modules will easily run at 230MHz at 2.5-3-3-5-1T. Perhaps you will also get similar performance.

I have the older Sempron 3100 130nm Paris. The default voltage for this chip is 1.5V, so the safe upper limit is 1.65V. At 1.6Vcore, the CPU is stable in Prime95 at 2.51GHz with 166 memory divider (231MHz RAM speed). 1M digits Super Pi time is 36.8 sec.

Currently running 2.44GHz at 38 seconds with Cool n Quiet ON.

I'd recommend that you keep the 9x multiplier and stay at 166 divider if possible. Run 1M digits SP and post your time at various settings. Also run Prim95 in ALL three modes for at least 15 minutes to quickly check for stability. You may discover that the maximum stable speed may be 100MHz below the boot speed.

The sweet spot for these chips is around 2.4 to 2.5GHz.

ECS NF3 is a very good board. BIOS will only overclock to 250MHz. However, you can use Clockgen to overclock in Windows.

Contrary to spec, this is not a hot running chip. 26 to 32C ide, and 46C max under load at 73F room temperature. Am using the Zalman 7700Cu.

http://shop3.outpost.com/product/4705559

 

[BlueWeasel]

CPU-z gives me the same voltage as in the BIOS. In order to keep the 9x multiplier and 166 divider, the chip would need 1.67v (what the BIOS is reporting, but only a +0.10v bump in voltage).

It really makes me think the BIOS voltage being reported is wrong. I've seen chips a little bit over-volted, but nothing like this.

 

[furballi]

I'd stick with 1.57V.

 

[BlueWeasel]

Originally posted by: furballi
I'd stick with 1.57V.

In the end, I believe that's where I'm going to stop. I gave the 2 sticks of Corsair a try, but I just couldn't push them as far as I needed in order to go back up to the 9x multiplier and 166 memory divider...even with 2.8v.

I don't think that running with the +0.10v bump in voltage would hurt much, especially since my temps at that voltage stay right around 50-51C. But I'd only gain 50Mhz or so of clock speed, while my HTT/FSB would drop about 25-30Mhz and my memory bandwidth would be slightly below 200Mhz (it's 213 right now).

 

[furballi]

My Paris Sempron is a hotter core, but I still run at 48C with 1.6V (73F ambient). Once you've reached the point of saturation, additional voltage will generate more heat, with not much more core speed (<50MHz).

Try both settings and test with 1M digits Super Pi version 1.4 or 1.5. Please post those numbers.

 

[mindwreck]

1.6v seems to be the point of return for semprons, any higher and it just creates more heat.

 

[PingSpike]

Originally posted by: BlueWeasel

Originally posted by: furballi
I'd stick with 1.57V.

In the end, I believe that's where I'm going to stop. I gave the 2 sticks of Corsair a try, but I just couldn't push them as far as I needed in order to go back up to the 9x multiplier and 166 memory divider...even with 2.8v.

I don't think that running with the +0.10v bump in voltage would hurt much, especially since my temps at that voltage stay right around 50-51C. But I'd only gain 50Mhz or so of clock speed, while my HTT/FSB would drop about 25-30Mhz and my memory bandwidth would be slightly below 200Mhz (it's 213 right now).

0.10v in voltage for 50mhz of clockspeed is a bad trade, even if you didn't have a drop memory speed. Once you reach the point of diminished returns...its tempting to keep pushing through it but is it really worth it?