Stability Testing Walkthrough

Stability testing is a long, drawn-out process that takes a considerable amount time, and Haswell doesn't make it any less. Stability testing on Haswell is in fact much harder than ever before because of the chip's added complexity, especially if you aren't testing each part individually. The following is my own personal stability testing routine—what I use to test motherboards and memory in all of my reviews. My routine may not prove stable for all users but should serve as a good base to getting the most you can out of Haswell.

One of the biggest changes Haswell stability testing brings with it is the behavior Haswell exhibits under certain workloads. If you use an application with AVX-based instructions, Haswell CPUs will boost voltage by as much as 0.100 V when using normal offset voltage values, which can significantly increase power consumption and heat output at both the board's CPU and VRM. A quick search on any forum, including our own, will unearth reports of users testing with applications we have used for years to ensure stability, but there is far more "argument" as to what works properly now because of Haswell's behavior with AVX's instruction sets. There are a few tools that offer AVX-based stability tests, with IntelBurnTest and AIDA64 proving to be favorites of many. Prime95 and OCCT still have many fans too. AIDA64 is all I require for AVX-load testing, but it does require customized testing options for "FPU" only testing. If you do not remove the other tests, you'll get the expected voltage boost many users and guides reported without the full AVX workload, or the possible heat. Shortly after Haswell's launch, [H]ardOCP's reviewer found AIDA64 to have some shortcomings when used for stability testing, and, yes, AIDA64 might not be the best choice if their reviewer did not adjust those settings. I've seen quite a few posts in the past eight weeks questioning the usefulness of AIDA64 stability testing for Haswell, but that is not a question I find myself asking. For me, AIDA64 is a powerful and valuable tool that must be included in Haswell stability testing.


The other tests and applications I run are fairly basic, spanning a wide gamut of workloads most users are likely to use. Each test, other than 3DMark, has a completely different workload which then focuses on pushing a different part of the system. All products must pass all of these tests and at least 8 hours of AIDA64's stability testing with only the "FPU" option selected. I will run through all of the tests in the screenshot, and only THEN do I start my AIDA64 testing. Only if all of those pass do I enable the other CPU- and memory-based test options in AIDA64 and let it run for 24 hours. I can usually do my benchmarks for reviews at this point using the same tests; after discarding all my saved stability testing results. Once that is complete, I save the BIOS settings to a profile in the BIOS itself, and then reset the BIOS to re-run all the tests for my "stock" numbers in reviews. At that point, I re-enable the OC profile and test again.

And that is all. It's not that complex, and, yes, some of the testing I do requires purchased software, and if any test fails, the point of failure is usually a great indicator of where problems may be. SuperPi32M is great for memory and memory controller testing, while wPrime and Cinebench are more CPU-focused. Games obviously test VGA and PCIe stability, and some CPU stability, while Handbrake and AIDA64 are good for encoding and AVX. My final test is not shown here—I spend some time playing DICE's Battlefield 3. I need to get a couple 1000+ ticket rounds in on one of the larger maps before I feel a system is fully stable, and the last test is obviously my most favorite. My tests also include many applications you have to pay for, but their value in testing greatly surpasses the small expense involved. Using trustworthy tools is something I cannot emphasize enough since their quality will play a big role in the results you generate. Then you have to compare the test results of those tools to see if that overclock pays off or not, so let's take a look at what gains my overclock provided!