Finishing the Build, Thoughts on APUs and Mantle
The motherboards in this review were chosen for one reason: they were among the cheapest options at the time I ordered them. There's not much to say here except that the MSI A88XM-E45 motherboard, still a good deal at the $65 I paid (but now selling for $73), is not the most economical choice for an AMD board. $10 – $20 could be shaved off the total build price if a lower-priced option was selected, but I have to say this MSI board is really nice for the money!
Memory and Storage
I needed to save some money here to keep the build cost down. How about a $50 hard drive and some sticks of RAM harvested from a Dell tower? Bingo.
It doesn’t get much more “commodity” than this
The RAM I used accross all systems is generic Samsung DDR3-1600 (CL11, if you care), and I only used 4GB for the benchmarks on each system. With memory prices still quite high in the current market (my 4GB would cost about $40 at current prices) I couldn’t justify spending more given the constraints of a low-cost system. For storage I decided to go with the traditional approach. A 1TB WD Blue hard drive was going for $50 when I picked it up (now back up to $60), and it provides enough storage to hold a huge library of games.
I thought a hard drive made more sense than a low-cost SSD here, but for around the same price an 120GB SSD could be found – though it’s not always so easy to go with just a small drive when you can only have one in the system. However, with prices dropping considerably of late a 256GB SSD might be sufficient given the crazy low price of a drive like the Crucial MX100.
For power I selected the lowest cost 80 PLUS certified PSU I could find, and it turned out to be an EVGA 430-watt model for $40. For our single-GPU needs this is more than enough power, and the quality of the unit seemed pretty good. It also has a very quiet 120mm fan, and didn’t put off any of the shrill noises I’ve heard from some really cheap PSUs. Done.
Naturally, my selections in the components category will not please everyone, but it’s Ok. All’s fair in war and PC builds. Or something.
Where are the APU Results?
Before going on I think I need to say a few words about the absence of APU results in the following benchmarks. Basically you're going to need a discrete GPU. Sorry, even AMD's highest offering (the A10-7850K) doesn't present a compelling 1080p gaming experience as an APU. And it's just too expensive at $185 to work on a tight budget.
APUs also need very fast RAM to really compete, and even so when you look at benchmarks (like the ones I posted when looking at APU memory scaling) the results aren't that impressive, especially with 1600MHz DDR3. Mantle might be the X-factor here, but that would mean a tiny subset of possible game titles until something changes in the market. And speaking of Mantle…
Mantle and DirectX
I'll talk about Mantle briefly here, and it's disappointing that something which can work so well to reduce dependency on the CPU for gaming is implemented in so few games. If I can revisit this in a year and directly compare mantle powered games vs DirectX on the same titles, I would love to. I have personally experienced Mantle providing a better overall experience on a couple of AMD systems, and it's very impressive where it has been well implemented.
For the sake of consistency between platforms, I decided to test every game exactly the same way and I couldn't do this with Mantle. I scrapped the idea, even though with only one Mantle title with a built-in benchmark in my test suite (Thief) it wouldn't have been very compelling. And I know what you’re probably thinking: Hello – BF4!! Sorry. I couldn’t go down that manual run-through road with this many tests given my own time constraints.
The fact is, when correctly implemented a game (like Thief) plays more smoothly with Mantle than with DX11 (and with a higher minimum frame rate) on the same hardware. But with almost all modern games running on DirectX I ran the graphics benchmarks using DX11 to give a better example of an average FPS number from different sources, and to keep the results consistent between platforms.
So for the benchmarks to follow, each combination was tested using the same basic build, with a discrete GPU (and the same API), 4GB of RAM, and a traditional spinning hard drive for the OS and games.
So without any further delay lets see how these tests ended up!