gamerk316 :
And IMO, Mantle becomes moot if MSFT tackles the driver overhead in DX12, like it is said they are doing. No dev is going to do a OpenGL EX/libgcm backend for the PS4, DirectX 11.1/Native backend for the XB1, DirectX+Mantle backend for the Windows, and OpenGL/Mantle backend for Linux; they aren't going to re-write their backend on four separate occasions.
Hum! When MANTLE was announced some people in forums said that it was unneeded because would provide negligible performance gain over DX: "the overhead is minimal", they said us. My reply was that devs would not be interested in MANTLE for minimal single digit performance gains and that we would expect 30--50% gains in BF4. Benchmarks are showing that MANTLE provides big gains, as expected. Now you pretend that MANTLE is not needed because Microsoft will eliminate the overhead in DX12...
Nice to know that MANTLE is useful on putting pressure on Microsoft, but the problem is that DX12, even if it fits your expectations, will be only released for Windows when developers are interested in other platforms as well. Your argument about no dev using MANTLE was answered many times before giving to you announcements of main devs already using MANTLE or planing to use it soon.
gamerk316 :
Kaveri Memory Scaling:
http://www.hardcoreware.net/kaveri-memory-speed-performance-scaling/
So as expected, diminishing returns after DDR3-1866. At that point, GPU is a larger bottleneck the RAM speeds.
http://www.hardcoreware.net/kaveri-memory-speed-performance-scaling/
![average-performance.png](/proxy.php?image=http%3A%2F%2Fimg.hardcoreware.net%2Fwordpress%2Fwp-content%2Fuploads%2F2014%2F01%2Faverage-performance.png&hash=5c5576262441264af7d65ada19543d39)
So as expected, diminishing returns after DDR3-1866. At that point, GPU is a larger bottleneck the RAM speeds.
That graph represents the average of a 'GPU' test (1080p and medium settings) and a 'CPU' test (720p and low settings). The names are irrelevant, call them "A" and "B" if you want. Take BF4 benchmarks, in the 'GPU' test the gain from using 2400MHz instead 1600MHz is the double of the gain measured in the 'CPU' test, which is the expected trend because bandwidth is more needed for GPU than for CPU. In more GPU oriented tests we can see nice bumps in gaming performance by using 2133 and 2400 memory
![AMD-Kaveri-Memory-scaling.png](/proxy.php?image=http%3A%2F%2Fsemiaccurate.com%2Fassets%2Fuploads%2F2014%2F01%2FAMD-Kaveri-Memory-scaling.png&hash=1943ee4fcd3073b8a99e6b83f61e2ebd)
When interpreting those graphs, it must be understood that 2133MHz is clocked 33% above 1600MHz whereas 2400MHz is clocked only 13% above 2133MHz.
Here more memory scaling benchmarks with 2400 compared to 2133 and 1866
![A10-percentages.jpg](/proxy.php?image=http%3A%2F%2Fcwsmgmt.corsair.com%2Fmedia%2FBlogs%2FDustinImages%2FA10-percentages.jpg&hash=232750a7e0ce0d9574a0a91a500f24ea)
And here you can find gaming memory scaling (800/1066/1333/1600/1866/2133) under linux, which can give an idea of how SteamOS will perform. Again 2133MHz provides nice gains and it is cheaper in my country ;-)
P.S.: Thanks to 8350rocks by the post about clocks and voltages.