CINEBENCH 11.5 OpenGL
The A8-3850 manages to creep by the Phenom II in OpenGL performance when both are using the GIGABYTE HD 6670 dedicated card. The A8-3850 and A6-3650 both took a very small blow using CrossFireX for this test. According to AMD, OpenGL should only use the GPU the primary display is connected to, which is the on-die GPU in this case. While it is only about 0.5 FPS difference, it shows the small overhead for CrossFireX is still there, even when only one card is in use.
3DMark11 Benchmark
In this test, the 6MB of L3 cache won as the Phenom II 955 holds a substantial lead. When you consider the gap between the A6 and A8 is only 20E and 5P in these tests, the 109E and 56P different between the A8 and Phenom II seems huge. These are both a 300 MHz clock speed bump, and a memory bandwidth drop when moving to the Phenom II.
Media Expresso 6.5 GPU Accelerated Conversion
As with Handbrake, the same 5 minute 720p .mov H.264 file is converted to fit an iPhone4 with the software’s built-in preset. Once again, the CrossFireX issue we saw in the OpenGL testing comes up. This too can be a driver issue, but I am more inclined to say its a software issue in this case.
In the test where the APUs use the on-die GPU and overclocked, they are out performing the dedicated HD 6670. This should not be the case given the HD 6670 has more SIMD cores, faster clock speed, and faster memory. It cannot be an issue with CPU overhead as the faster Phenom II 955 + HD 6670 only holds a 2 second lead over to the A8-3850 + HD 6670.
Older Game Engine Test with Portal 2
Portal 2 was tested at maxed out settings (8x MSAA, 16 x AF, Vsync ON, Very High Texture). AMD states when CrossFireX is enabled but cannot be used, the system should default to the faster of the two GPUs. This is simply not the case as the CF test are closer to the on-die GPU’s performance. To verify I checked the AMD Resource Monitor. It indicates only the GIGABYTE HD 6670 was at full speed, but the on-die GPU processor was doing most of the work. This could just be an issue with the Resource Monitor.
Resident Evil 5 Benchmark
In the Resident Evil 5 DX10 tests I have full use of CrossFireX. Both the A6-3650 and A8-3850 with the GIGABYTE HD 6670 in CrossFireX broke 60 FPS average. A 25% boost in performance over just the HD 6670. In the DX9 tests, CF results were almost identical to the 6550D. It was the CF tests with the 6530D which results in performance closer to the HD 6670. It almost split the difference between the on-die 6530D and the dedicated HD 6670.
Alien Vs. Predator DirectX 11 Benchmark
AvP was tested with the default of 16x AF, Tessellation on, Advanced Shadows on, and SSAO on. The Medium test was done with all the DX11 features turned off, 2x AF, shadows on medium, and textures on medium.
While these entry level AMD Radeon GPUs can play DX11 titles, they don’t play them very well. Even when I turn off all the DX11 features, I can barely get 30 FPS with CrossFireX enabled. Keep in mind AvP is heavily taxing in DX11 mode, so some DX11 games may be playable with an APU CrossFireX setup.