Or, you can try upping the memory voltage by .005v with the settings you had, and run Memtest again. I've had MANY occasions where a specific set of timings caused errors in Memtest, but a small bump in DRAM voltage eliminated any further errors during testing in Memtest OR in Prime95 version 26.6 Blend mode. Testing in Memtest isn't a surety of stability after making changes to memory timings. It ONLY tells you if there are any glaring instability or incompatibility issues. You need to run Prime95 version 26.6 and choose the Blend mode option, for 8 hours, or you're in real trouble of potentially having system corrupting micro-errors.
You won't get blue screens, or restarts, or any of that. What you'll get is about 3 to 6 months later, half your data is corrupted. If that's only the OS, not that big of a deal as you can always reinstall. If it's all your game files, pictures, music, etc., it becomes a very bad day scenario. Memory is FAR more likely to create system wide corruption from fiddling with DRAM speed and timings than when overclocking the CPU. Take it seriously, because it is.
If however you can pass two or more passes of all 13 (11 on free versions) Memtest86 tests and 8 hours of Blend mode, then your memory is as reasonably stable as you can ever expect to get it.
I probably might also try manually setting the primary timings, and about three of the secondary timings, and leaving the remainder of the timings individually on Auto, with DOCP off. DOCP only exists to translate XMP settings to AMP, which is the AMD equivalent. If DOCP/AMP are turned off, it can't change your settings. Or you can just leave it at what it likes like Rogue leader suggested.
The difference really isn't going to be all that significant.