Richland Graphics and Platform
The entire package has been given a good once-over, and it has allowed AMD to increase clocks in a fairly decent way. Going from 800 MHz for the GPU to 844 MHz is a nice boost. I was honestly surprised to see this part reach 4.1 GHz stock speed for the cores, I guess I was expecting 4 GHz at most. AMD has been able to achieve this, plus increase turbo clocks fairly generously, all without breaking the TDP bank. Not only are the turbo modes higher, but they stay “available” for longer under load as compared to the previous products.
If there is one issue that is a bit of a sticking point, it is the ability to run the GPU portion of the APU along with a standalone graphics card. This means that if a user installs a card that does not CrossFire with the integrated unit, they will lose all of that processing power. The GPU portion is only active for 3D applications as well as GPGPU and OpenCL when it is run either by itself or in CrossFire mode with the 6670, 6570, and 6450 series of video cards. If a user were to install anything higher, it disables the GPU portion of the APU, thereby making it unavailable for any kind of workload that it could possibly attend to. When we hear about the potential 779 GFLOPS of performance from this single chip, it is only when the GPU portion is active. This will always be the case for Richland, Trinity, and the older Llano APUs. AMD is not talking about it in depth yet, but this could change with Kaveri and its ability to share memory address space with the CPU. Make no mistake, the current APU does a very impressive job when it comes to compute. It is unfortunate that we cannot access that performance when used in conjunction with a high end video card.
One of the advantages of Richland is that it still uses the FM2 socket infrastructure. The current top end chip, the A85X, is still one of the best motherboard chipsets on the market. It is currently the only single chip solution to offer up to eight SATA 6G ports. The latest Intel Z87 offers six ports, something that AMD has included since the introduction of the SB850 southbridge some three plus years ago. It also features four USB 3.0 ports to boot.
Richland still supports a maximum of two modules with four cores. It has a grand total of 4 MB of L2 cache split between the modules. It has 20 PCI-E 2.0 lanes available to be split in multiple combinations. PCI-E 3.0 does not look to make it to AMD’s products until Kaveri and FM2+. Rumor is that PCI-E 3.0 was designed into Trinity, but the extra time, money, and manpower to certify it was just not worth it. AMD wanted Trinity out the door as soon as possible, and they also figured that for its target market, it was not worth it to pursue that course until they absolutely had to.
The list of advanced features includes up to SSE 4.2, SSE 4a, AVX, FMA 3/4, AES, and XOP. Most of these operations are still not widely supported in software, but AMD has kept up with Intel in terms of what they can support. While Haswell supports AVX2 and is a beast when it comes to this type of operation, AMD apparently has just as fast of an AES unit as the latest Intel models.
AMD has a full stable of Richland APUs that it is releasing, and I also had the chance to test the A10 6700. This is a 65 watt TDP part that is multiplier locked, so overclocking is somewhat limited. It features a core speed of 3.7 GHz with a turbo going up to 4.3 GHz. The graphics portion is also clocked up to 844 MHz, which in total is pretty impressive considering that it is just a 65 watt TDP part, and it is not all that far behind the 100 watt 6800K.
Prices for Richland products stretch from $69 for the A6-6400K up to $142 for the A10-6700 and A10-6800K.