Seriously, I’ve always been asked why I put faster DDR3-1866 RAM with my AMD recommendations. That’s usually down to two things: higher bandwidth, room for tweaking and more headroom for a higher CPU overclock. For most people its better to just leave RAM timings at stock speeds, but a few enthusiasts will try to drop the speed of those kits in return for tighter timings and a lower voltage. Some people will go to extreme lengths to make their system “their own”.
So in the case of AMD, is it better to have faster RAM? I’ve always maintained that APU processors would benefit more from it as the integrated GPU shares RAM from the system. Builds with an APU and slower DDR3-1333 have shown markedly lower scores due to a bottleneck in speed and bandwidth. Since AMD has chosen to integrated the northbridge to the processor and squeeze in a working graphics core, things get strained when you’re playing a game and both the CPU and GPU tax the RAM. But what happens when you scale up above DDR3-1600? AMD itself recommends DDR3-1866 for the best performance, but does it really improve things? Tom’s Hardware went about testing that and I’ve posted some of the results below.
For the most part in the past, AMD-Optimised memory has been a joke. Its only with AMD’s recent joint deal with Patriot that there’s actually RAM kits out there that are “optimised” for use with Llano or in a Bulldozer build. In previous years I actually held off recommending kits that were advertised as optimised for AMD systems because I believed it was marketing rubbish used to get older kits and chips off the shelves of the manufacturers. These days things are much better and we don’t see kits advertised for AMD rigs that use 1.65v just to boot up. 1.5v is the JEDEC requirement for DDR3 RAM to be bootable and with Intel’s new Nehalem architecture it wasn’t a good idea to run RAM with a voltage higher than 1.65v as that would screw with the new memory controller.
In addition, Intel’s Nehalem shipped with a DDR3-1333 capable memory controller and if you wanted any higher speeds you’d have to overclock both the processor and board and manipulate the memory ratio to get your desired speed. You had to work within 1.5v limits as Intel wouldn’t guarantee that the chip would work with higher-voltage 1.65v RAM modules. It actually released a statement warning that modules that went above 1.65v could negatively impact the life of the chip. AMD’s IMC in the Athlon and Phenom processors could handle both the higher voltages and higher RAM speeds and that’s where the older, more power hungry and less efficient kits were pushed to.
Today that’s not the case. Sandy Bridge and Ivy Bridge both sport DDR3-1600 memory controllers and most kits out on the market are rated for 1.5v. We no longer have EPP profiles limited to Nvidia-only chipsets, with XMP replacing that poorly improvised overclocking system. We have kits that go all the way to DDR3-2800Mhz with huge heat spreaders and ultra-low voltages for maximum efficiency. And now we’re seeing that higher performance RAM benefits us less and less as time goes on. So if you’re considering an APU system, let the graphs below save you some money.
I won’t bother with the benchmarks at DDR3-1600, since most modules run at the same speeds at that point. Tom’s underclocked a number of different RAM modules to the standard speeds and found that there’s no difference in choosing a more expensive kit at DDR3-1600 versus choosing one of the more affordable modules from companies like Transcend. DDR3-1866 brings about a 1GB/s bandwidth improvement, but in games and applications the difference is negligible. At most, in games you’ll see a 3-5fps benefit and its a difference you won’t notice anyway. The application benchmark shows an even smaller difference, with two of the higher-performance modules actually performing slower than the reference DDR3-1600 kit.
I’m jumping straight to the DDR3-2400 runs because its exactly the same story. Bandwidth improvements are now just under 4GB/s, but the difference in games and benchmarks is still as close as the taste of red and blue smarties (i.e. there’s no discernable difference). There’s only a four frame lead in DiRT3, a two frame improvement in Metro 2033 and four second drop in the WinRAR benchmark.
Is there a case for higher-end RAM for system builders? Unless you’re planning super-high overclocks that you’ll be running day in and day out, then perhaps there’s a case for it. Bragging rights is always a good reason, but at the cost to your wallet and the benefits can’t always be proved when your LAN buddies come over to gawk. More RAM is better than higher-speed RAM in most scenarios, so if you’re in the market there’s only two things you need to look out for; 1) DDR3-1600 4GB kits at a good price and preferably; 2) modules that can have their latencies lowered in future. Lowering latencies allows your RAM to run at its optimum settings and may help in smoothings things over for your system, if not speeding it up.
Source: Tom’s Hardware
Discuss this in the forums: Linky