Originally Posted by
BlindFreddie
The memory modules concerned are G-Skill "F3-12800CL7D-2GBHZ - DDR3-1600 CL7-7-7-18 1.9V", from my Asus P5K3 Delux board.
I also have 2 x 1GB Kingmax "FLGD45F-B8MF7 MAEH DDR3-1600" that run CL7. It's not printed on the labels but I think they need 1.9V for CL7 too. Labels date them as 2008 and 2009.
Ket: "1. Any settings defaulted to for memory in the UEFI are directly read from the speed data stored on your memory modules. If 1.9v was set, that is the fault of the memory manufacturer incorrectly setting parameters in the modules speed data, not asrocks."
I haven't recorded all of the timings in the SPDs, but as I said, there are (much slower) JEDEC timings with 1.5V set. It's only the XMP profile that sets 1.9V.
Every board that I've set up other than the Z68 Pro3-M has defaulted to JEDEC timings, and required the user to explicitly select the XMP.
If Asrock had done that, the damage would not have occurred.
They should also realise that there are such DDR3 modules out there and put a guard against applying more than 1.65V (say) without explicit user permission.
Too much "form over function"!
I haven't taken the risk of setting timings & 1.5V Vdimm manually with the current OCZ 1.65V sticks, swapping in the G-Skills and booting the modules again, as I don't know whether the manual settings would be preserved across memory module changes.
Even if it worked it would be too risky to leave that way in case the CMOS got cleared somehow.
Ket: "2. "Soaring" voltage to 1.24v from 1.23v? That is a very, very minimal amount of fluctuation you can expect to see on any board."
Yep! (Do you detect just a little hint of sarcasm maybe?) 1.23V is also a very low max voltage to avoid fast degradation of an SB CPU. The CPU core seems to have been damaged as well as the IMC. And Vcore has never been north of 1.3V, and that was only applied for a couple of min during initial setup experiments.
Agree re. VCCSA and VCCIO, tho higher VCCSA might get you a higher mem divider.
"... my 2500k has recently died ..."
I'm not overly surprised, and expect it was a combination of high Vcore, high temps, and time.
You might have got away with 1.36/1.4V, but 90 deg C with that would have been OTT. Remember that unlike metals, where electrical resistance increases with temperature, in semiconductors resistance decreases with temp, thus increasing the leakage currents and power, raising the temps even higher! In early transistor circuits, such thermal runaway was a major problem.
1.36V seems a lot for SB at only 4.5GHz. Maybe you had a slow leaky chip.
Back in the Q9650 days leeghoofd posted that he was backing off Vcore on all his machines to 1.25V, which I thought was rather conservative. But he was right, and it's still not a bad limit for current machines that are 100% loaded 24/7.
Your cooling must have been suboptimal too. Pushing a good HS like a TRUE with decent fan to 90C would require well over 250W I think.
Splave: "Hmmm, I've had the same chip running 32m about 5 hours a day 5 days a week. Anywhere from 1.92 to 2.1v on my voltage hungry samsungs. Still as strong as ever... "
Well, my experience was as I described, and I assume it was caused by Vdimm = 1.9V.
I'd advise getting some less hungry memory than your samsungs. DDR3 is cheap now. Have you found much performance increase from running your memory faster anyway?
"I'd hope your setting your own voltages, if your caring enough to run faster than standard mem."
I already had the mem sitting idle, and I didn't get time to set my own voltages before the damage occurred.
I haven't tried to RMA the 2600K as it's not Intel's fault, and putting 1.9V on the IMC would have voided the warranty, so it would be unethical too.
Bookmarks