Carrizo in Review: How does AMD's A10-8700P Perform?

We already mentioned that the TDP of the new Carrizo models can vary between 12 and 35 watts – it is unfortunately not known which setting HP chose for the Pavilion 17. The same chassis is also available with Intel ULV processors as well as inexpensive Beema APUs, so we expect a similar consumption of around 15 watts. This evaluation also corresponds with the consumption values later in this review.

If you consider the TDP of the two Kaveri-based predecessors FX-7600P (35 watts) and FX-7500 (19 watts) did not even have an integrated chipset, AMD really did a great job with Carrizo: On average, the A10-8700P beats the old FX-7500 by almost 20 percent, which is mainly a result of the Turbo, which works properly now. The APU runs at 3.1 GHz in the majority of single thread applications, only 100 MHz below the specified maximum clock (1.8-3.2 GHz). The clock is reduced to 2.3 GHz when you stress all four cores, but this is still higher than the nominal clock. In combination with the slightly increased per-MHz performance, the chip is fast enough to compete with a Core i3-5005U or i3-5010U. Carrizo cannot keep up with the faster Broadwell chips however – even the slightly higher clocked FX-8800P should (at a similar TDP configuration) lag behind a Core i5-5200U. Still, the advantage in respect of the single thread performance is remarkable.

While the difference in the CPU clocks between the FX-8800P and A10-8700P is limited to a couple of hundred MHz, AMD clearly reduced the GPU performance of the smaller model: Our A10 only uses 386 shader units (= 6 Compute Units) from the original 512 units of the Carrizo chip, only the FX model can use all 512 ALUs (= 8 Compute Units).

Thanks to much higher GPU clocks of up to 800 MHz (manufacturer’s information, it seems that the review unit only runs at 720 MHz), the A10-8700 can still easily beat its predecessors FX-7500 (384 shaders, up to 533 MHz) and A10-7300 (384 shaders, 533 MHz). All in all, Carrizo manages an advantage of around 15 percent; if the memory of the Pavilion is really only clocked at 667 MHz (DDR3-1333), the real difference should be higher than 20 percent.

The duel with the HD Graphics 5500 (Broadwell) is usually won by Carrizo if the game is heavily limited by the GPU. AMD’s solution is a couple of percent ahead in Tomb Raider or Evolve, whereas Dota 2 or The Sims 4 show pretty much the opposite results. We can say the same thing for both GPUs: More than minimum details are not possible in current titles – but you can expect at least medium settings and 1366x768 pixels if you only play classics like Counter-Strike: GO or Diablo III. By the way, the Radeon R6 does lose less performance compared to the HD 5500 at higher resolutions, because the limitations by the CPUs are reduced.

The performance of the A10-8700P should be even higher with faster memory, and – similar to the predecessor – it is possible to use the processor graphics in combination with a dedicated GPU in a dual-graphics configuration. We will have a detailed look at this in combination with Windows 10 in a separate article.

Update from 07/31: More gaming benchmarks added.

The frame rate in our Dota 2 scene fluctuates at around 24 fps (42.5 watts total consumption) directly after the start, until it drops to 21 fps (31.6 watts) after two minutes. We can see the reason when we look at the clocks: The A10-8700P starts at 1.8 GHz (CPU) and 530 MHz (GPU), respectively, before the values drop to 1.3 GHz (CPU) and 380 MHz (GPU).

Contrary to Dota 2, Diablo III reduces the processor performance in favor of the GPU, which is why the APU starts at 1.3 GHz for the CPU, but 600 MHz for the GPU (41.8 watts). We can once again see a drop from 47 to 39 fps due to the TDP after around 2 minutes, but this time only the GPU is affected and runs at 430 MHz (31.8 watts).

Overall, we can say that the performance is steady under sustained 3D load. We cannot determine significant drops that we know from some Core-M devices (due to the extremely low TDP).

H.264 decoding should not be a real challenge for any processor in 2015 – and it isn’t for Carrizo: There are no problems and the CPU load is not very high even with our very demanding sample video in the 4K resolution, 60 fps and almost 100 Mbps.

H.265/HEVC

Our first attempts to test Carrizo’s HEVC decoder in Windows 8.1 via DXVA2 (native) and the Media Player Classic failed – contrary to Broadwell or GM206, our Carrizo system still decoded all the videos in software. This problem was only solved after we followed a tip from AMD and used the full version of PowerDVD 15 (we want to thank CyberLink at this point for the provided test version).

The power consumption of the Carrizo system is noticeably increased during H.265 playback, despite the dedicated Fixed-Function units: The additional consumption compared to an idling Windows Desktop is around 10 watts in our first sample video (Tears of Steel 4K, 2.6 Mbps, 24 fps); exactly the same as a Core i5-5200U that we used for the comparison. Interesting: The Intel system even needed 2 watts more when we used the hybrid decoder – it seems that software decoding on the Broadwell CPU cores is more efficient than the use of the GPU.

The Core i5-5200U was not able to provide smooth playback of our second sample video (Big Buck Bunny 4K, 6.1 Mbps, 60 fps) independent of the player and decoding method. Carrizo has issues as well (GPU load of up to 70 percent), but can at least provide smooth playback – advantage for AMD. The power consumption of the overall system (around 31 watts), however, is around 20 watts higher than the idle value and reaches levels similar to 3D games or demanding CPU benchmarks.

Finally another word about Windows 10: Microsoft’s latest operating system is already prepared for HEVC/H.265 and should be able to play correspondingly decoded videos without additional software. We will compare the differences to Windows 8.1 and PowerDVD in terms of consumption and hardware utilization in another review.

Is Carrizo a success against Intel's overwhelming market share?  One thing is certain: AMD’s latest APU generation is the biggest leap in terms of power efficiency in the last couple of years. The manufacturer now once again provides the most powerful graphics solution within the 15-watt segment, even though the reviewed Pavilion 17 is probably slightly slowed down by the not perfectly configured memory. The CPU part also performs well and at least does not have to fear the current Core i3 chips. It is not enough to compete with the Core i5 models or higher, but they also have a much higher price. We really want to praise the modern H.265 decoder, which – even though not as frugal as hoped – even provides smooth playback of demanding 4K videos.

It remains to be seen how the notebook manufacturers can utilize the potential. HP’s Pavilion 17 shows good approaches like the Full HD display and the standard dual-channel memory configuration, but details like the obviously faulty BIOS (e.g. memory clock) still affect the overall impression a bit right now. We would also like to see more devices without an additional dedicated GPU, but in return with fast DDR3-1866 or DDR3-2133 modules. This should provide remarkable performance in combination with a generous TDP limit, especially for the top model FX-8800P.