Sapphire Radeon HD 6870 Review - PAGE 3

The cooling solution on the Radeon HD 6800 series graphics card is similar to that found on the 5800 series, with a few small exceptions. The bottom layer of the cooler has a large black metal base that makes contact with the VRM, memory chips and GPU on the HD 6870. The GPU cooler has a large copper base that relies on thermal paste to transfer heat, while the memory and VRM use sticky thermal pads to make contact with the cooler. The thermal pads can be reused multiple times, so it is not necessary to replace these after the heatsink cover has been removed. The thermal paste on the base of the GPU heatsink however must be reapplied each time the heatsink itself is removed from the card for whatever reason.

The upper layer of the cooler consists of the fan and heatsink. The heatsink uses a three heatpipe design to transfer heat from the core into the finned array. This allows heat to be separated and dissapated. With the blower fan placed directly behind the heatsink, the hot air is blown out the back of the card, which removes it from the case. Like with previous AMD cards, the cooling fan can get quite loud when the fan speed is set to high.

The HD 6870 uses the new Barts GPU, which like the previous generation AMD graphics cards is built on a 40nm die. However, the new architecture does come with some improvements, most noticeable of these being the 25% smaller die that still packs most of what was found in the AMD Cypress core. This gives the HD 6870 1120 stream processors, 56 texture units and 32 ROPs. Improving on the performance of the HD 5700 series, the HD 6870 comes with 1GB of GDDR5 memory that comes clicked at 1050MHz, a 256-bit memory bus. The GPU clock itself is set to 900MHz, which in previous tests has proven to be quite high given the limited overclocking headroom on reference HD 6870s.

AMD has included some new technologies on the Barts GPU. These consist of Advanced Parallel Processing, HD3D and Morphological AA. Advanced Parallel processing is essentially a new name for ATI Stream technology. This is the process of adding acceleration on both a hardware and software level with parallel processing between both the GPU and CPU. There are many applications that support this technology, and these stand to benefit from greatly enhanced speed. With the new HD 6800 series AMD has updated this technology to include support for DirectCompute 11, and OpenCL; they've even added a UVD3 engine on the Barts die itself. This will give the core improved support for tessellation, and improved Divx and Blu-Ray 3D encoding. This is all part of the new EyeSpeed technology. AMD is making a big push with this technology and have teamed up with a host of partners to ensure EyeSpeed is widely supported.

Another new feature is AMD's HD3D technology. 3D computing is not new to the market, and Nvidia has in fact employed 3D support in their own products for some time now. AMD's approach to the technology is new, however. Whereas Nvidia uses a close sourced method to deliver Stereoscopic 3D, AMD is using an open-source method that allows many manufactures to create their own means of support. This means that there will not be a set, specific standard between compatible 3D displays, glasses and software.

Morphological Anti-Aliasing is an AA method that uses a post process filtering technique with Accelerated DirectCompute. This approach can deliver full screen Anti-Aliasing, faster super sampling and is compatible with DX9 through DX11.

Overclocking:

The Radeon HD 6870 from Sapphire is so far the best overclocking HD 6870 to enter our labs. When we were reviewing the reference cards that came to us directly from AMD, the 6870 was only capable of remaining stable at speeds of 970/1140MHz, not an overly impressive overclock considering the default GPU clock is already set to 900MHz. The Sapphire HD 6870 on the other hand was capable of reaching a GPU clock speed in excess of 1000MHz. The highest clock I found to be stable was 1010MHz, but there was one game (Batman Arkham Asylum) that still crashed at that clock speed setting. In the end I was forced to reduce the speed to 1000MHz. It was nice to reach 1000MHz, but this is still only a 100MHz overclock. The memory on the Sapphire card also yielded better results than the reference card, maxing out at 1200MHz (4800MHz effective).