AMD APU: How Much Progress Has Been Made?

Final Thoughts

Progress?

3 years on from Llano with 4 generations of APUs having graced the market what conclusions can we make about AMD’s APUs? Well I think its a mixed bag but a broadly positive picture. Despite AMD’s best efforts to unify the CPU and GPU, as well as their recent effort to unify memory resources with HSA, the majority of games and productivity applications still remain fairly fragmented at a hardware level. Most applications lack the support for programming environments (OpenCL, HSA) that can unify this hardware fragmentation. Simply put many applications are still largely GPU or CPU dependent. With AMD’s APUs it is clear to see that the CPU part is the weakest link: at best we’ve seen around a 10% boost since Llano in overall CPU performance with slightly reduced power consumption but the results do vary a lot by application, some have gained more than that but some have also declined since Llano. Is that enough? I don’t think so given what Intel have been doing since 2011 but on the whole the CPU performance is still “enough” thanks to AMD’s dedication to bringing quad cores to the market and at the same price as Intel’s dual cores. AMD’s answer to the CPU performance problem will always be the GPU and while that isn’t going to win them any respect with enthusiasts, for most light PC users and casual gamers AMD’s APUs do provide enough CPU power to get by comfortably with larger GPU capabilities than rival products giving them the edge.

Moving onto the all important GPU and I think this is clearly AMD’s strong footing with the APUs. No other “CPU & GPU” combination unit on the market can match AMD’s APU performance in the desktop space, Intel’s Iris Pro graphics probably can but these are BGA (e.g. not socketed and can’t be upgraded) and are not aimed at desktop users, not to mention they are significantly more expensive. AMD’s APU results that we’ve shown here only demonstrate the tip of the iceberg: we’ve shown AMD’s Kaveri APU has so much more to offer when you overclock it, add faster memory or use Hybrid Crossfire. More importantly GPU performance has improved a lot, something we wanted to validate when we started this testing. Whether the GPU performance is “enough” for the market AMD are going for is still a debatable question, AMD’s APUs are more geared at the casual and light gamers as opposed to die-hard enthusiast gamers who need the best of the best. Our tests show results from 720p gaming where Kaveri did very well but to generalise AMD’s Kaveri’s APUs are good for 1080p gaming at low-medium settings at the 30-45 FPS mark. For most enthusiasts they will crave more performance and discrete GPUs like from AMD’s R9 Series are better equipped to deliver that, but at a price & power consumption premium.

The Future

AMD’s APUs have demonstrated significant GPU progress over the four generations of the APU as well as incremental CPU performance improvements but what does the future hold for AMD’s APU? Well I think it is clear the immediate problem is memory bandwidth. The graphics units in AMD’s rely on the system memory interface to the rest of the system. Currently 1600MHz is the mainstream memory speed for the desktop market while 1866 and 2133MHz are the more “high-end” speeds. Anything above 2133MHz is categorically an “enthusiast” product. We did all our testing with 1866MHz DDR3 memory, something we think matches the APU in terms of its price-point, but even this is a hugely limiting factor as our memory scaling article revealed: AMD’s Kaveri APUs need 2400MHz memory to shine. DDR4 couldn’t be coming at a better time for AMD; if they can bring DDR4 support to the next-generation APU platform then it would overcome that memory bandwidth bottleneck by offering faster memory speeds: DDR4 starts at 2133MHz. This is important because AMD’s iGPUs on their APUs can be best thought of as DDR3 graphics cards, anyone that has read reviews of certain graphics cards will know that the difference between DDR3 and GDDR5 versions can be as much as 10-20%: memory bandwidth plays a huge role in GPU performance. However, it isn’t just all about the GPU. The future for AMD also NEEDS to include the CPU. AMD cannot cause a paradigm shift towards GPU-centric programming in the market overnight, they are certainly trying with the advent of Mantle and HSA but it’s going to require broader support across the market which will take time. You can also rest assured Intel will do everything it can to keep x86 CPU performance at the heart of the PC market so it certainly is an uphill battle, but one AMD are capable of taking on as they’ve already demonstrated so far.

What do you think of the progress of AMD’s APUs? Have they made enough performance gains since Llano? Let us know your thoughts!

Thanks to AMD for supplying us with all the APUs over the past few years that made this article possible.