And tomorrow we might not be together…
A little more than two days passed before my E1-stepping Core i5-3570K suffered a severe loss in memory stability. I spent the rest of the third day trying to figure out why the system was crashing all of the sudden.
It seamed that as soon as Windows finished its updates, the system was no longer stable. I reloaded everything on the Mushkin SSD twice before getting it stable with none of Microsoft's automatic updates. Day four rolled along and I started having trouble with the graphics cards. Uninstalling one of the two boards helped a little, and reloading the graphics driver once again appeared to fix all my stability problems. Until it didn’t.
I pulled the CPU and put it into a known-stable system, only to discover that there were no stable settings for it. Though I’ve never seen a bad CPU right out-of-the-box, this one appeared to be my first. So, I replaced it.
And then the USB 3.0 controller died. Could a bad motherboard have killed our precious processor? Bad boards happen. I had started this build with ASRock’s Z77 Extreme4, so I reached into my pile of Z77 Extreme4s and pulled out a replacement. That one worked fine for about one day, and then started throwing memory errors. The first motherboard was surely bad, but could the CPU’s damage have been caused by a bad setting?
ASRock defaults DDR3-2133 memory to 1.5665 V. Add in the 1% over-voltage we’ve been seeing nearly every board vendor sneak in over the past several product cycles, and we come up to around 1.68 V. I wasn't going to risk any more hardware to pull an exact reading. Instead, I switched to the low-voltage RAM from our previous build, along with its not-fried CPU.
Our memory errors went away, temporarily. But the second CPU was suffering more issues with its memory controller. It had been used with those potentially-problematic voltage settings, so the chances appeared good that its memory controller was just a little less damaged than its predecessor.
A fresh board, a fresh CPU, and fresh RAM later, and all was well. Well, for about one more day. Very minor stability issues started to occur on the third motherboard sample. Then another thought struck: could the CPU cooler be warping boards?
The motherboard was already lying on the bench, so I removed the CPU cooler, loosened its brackets, and then set the CPU cooler atop the CPU. Stability returned. Even though I followed all of the italicized must-dos on the previous page, the Z77 Extreme4 appeared to be flexing too far under the force of Noctua’s installation bracket.
With no Z77 Extreme4s left to continue this fiasco, I moved on to Gigabyte’s similarly-thick Z77X-UD4H. Though the board didn’t feel any stiffer, it also didn’t throw errors after taking on the load of Noctua’s installation kit.
- The Magic Of Anticipation
- CPU, CPU Cooler, And Memory
- Motherboard, Graphics, And Power
- Case, SSD, Hard Drive, And Optical Drive
- The Initial Installation: My First Attempt
- Ten Days, Ten Solutions?
- Starting Over, This Time With Success
- Overclocking
- Test Settings And Benchmarks
- Results: 3DMark And PCMark
- Results: Battlefield 3 And Far Cry 3
- Results: F1 2012 And Skyrim
- Results: Non-Gaming Applications
- Power, Heat, And Efficiency
- The Less-Obvious Benefits Of Spending More
That sort of thing happens once in a blue moon. Don't let it bother you.
I guess Tom's tale of woe summarizes why Intel recommends against higher than 1.575 volts on the memory controller of Ivy/Sandy:
http://www.intel.com/support/processors/sb/CS-029913.htm#4
Intel® recommends using memory that adheres to the Jedec memory specification for DDR3 memory which is 1.5 volts, plus or minus 5%. Anything more than this voltage can damage the processor or significantly reduce the processor life span.
In any case, the performance benefits of overclocking memory on a Sandy/Ivy platform seem so miniscule that it's scarcely even worth considering. Buy memory capable of an appropriate speed @ 1.5V, and leave it be.
(I know Tom mentions Intel's position on memory voltage on the last page of the article, but I wanted to re-emphasize it because I've seen literally hundreds of people dismiss Intel's statement on various hardware forums. When sites like Tom's Hardware push limits, even for questionable performance gains, we all benefit -- but when someone who's on a budget and might not know any better pushes limits on his own, hard-earned hardware, the results might be tragic. Tom's experiments with this stuff so we don't have to.)
"Video Cards: 2 x 7870 LE - PowerColor PCS+ AX7870 Myst Edition"
After all, it is much more than a standard 7870.
I had no idea it was the LE until I got to the third page.
I think I'm a bit afraid to build computers now. If either of those situations would have happened to my $600 build, I would have cried and given up...
That sort of thing happens once in a blue moon. Don't let it bother you.
I guess Tom's tale of woe summarizes why Intel recommends against higher than 1.575 volts on the memory controller of Ivy/Sandy:
http://www.intel.com/support/processors/sb/CS-029913.htm#4
Intel® recommends using memory that adheres to the Jedec memory specification for DDR3 memory which is 1.5 volts, plus or minus 5%. Anything more than this voltage can damage the processor or significantly reduce the processor life span.
In any case, the performance benefits of overclocking memory on a Sandy/Ivy platform seem so miniscule that it's scarcely even worth considering. Buy memory capable of an appropriate speed @ 1.5V, and leave it be.
(I know Tom mentions Intel's position on memory voltage on the last page of the article, but I wanted to re-emphasize it because I've seen literally hundreds of people dismiss Intel's statement on various hardware forums. When sites like Tom's Hardware push limits, even for questionable performance gains, we all benefit -- but when someone who's on a budget and might not know any better pushes limits on his own, hard-earned hardware, the results might be tragic. Tom's experiments with this stuff so we don't have to.)
Heh, by that I didn't mean to dismiss your hardship. Sorry to hear it, definitely!
And thanks for all the hard work. Excellent article.
I have a serious question for you. Does Tom's know about the overclocker's secret when it comes to RAM? I've been amazed that you guys don't use it in your enthusiast builds, ever. It's pretty much the most overclockable ram ever seen, is low profile, and only costs $50 for 8GB.
I don't want to spoil the name and tell everyone, but, well... it's the only ddr 3 ram out there that uses a 22nm process.
really appreciate the hard work and the excellent, interesting article.
i liked the gigabyte board choice, better than asrock extreme4.
one question - do the recent events related to asrock extreme4 change your recommendations, especially with heavy air coolers? how about other motherboards (incl. asrock) that seem to offer better features in exchange for pcb strength?
imo mid/mini/tower cases should include some kind of standardized, customizable suspension support(from the case ceiling) for heavy air coolers. the suspension could be made from wire or metal/plastic (resizable) rods etc. or make more cases like cooler master haf xb lan box. aio coolers like nzxt kraken x40 may be an alternative.
My system is an I-7 2600K CPU and so maybe with the larger mounting surface of the 3570 CPU there may be other considerations. But if that was true, we would read about this problem in the Newegg reviews and we simply don't see the problem that you mention. You may want to look at your installation method.
You know what would be better still? Through bolts to the motherboard tray, like we see on SSI-CEB boards.
But Intel tried that with BTX, and nobody bought it. I personally blame Intel for not making BTX an extension of ATX, since cross-compatibility could have helped the cooler support mechanism and cooling tunnel survive market resistance.
I think that this is one of the reasons that Intel decided to go with CLCs for Sandybridge-e CPUs. When I had to RMA a CPU (i5 2500K) last year the tech at Intel was aghast at my not using their standard cooler. When I told him that I knew that Intel was using CLCs for their SB-e line and couldn't object, he laughed and approved the RMA.
I agree with Crashman about the cooler OEMs supplying a stiffening motherboard tray, that distributes the torque, if they are using oversized heatsinks.
I was actually more interested in this article than in the normal evaluations of systems because of your story of tribulations with the motherboard and CPU. I'm glad to know that you pursued the problem to its end because I would have had problems affording 4 mobos and 3 CPUs.
And I am sticking to CLCs in my gaming builds and recommendations in the fora. A CLC may be a bit more expensive and a little less efficient that an air cooler, but I have had yet to see one crack a motherboard and the cascading problems you experienced.