Sapphire Radeon HD 6950 & 6970 review
Author: Luka Rakamaric
Date: 04 Mar 2011

With the GTX 580, although not on a new production process, NVIDIA managed to raise the bar quite substantially over the ‘previous generation’ GTX 480. The chip architecture has been improved, the yields have been improved, and the power consumption and clocks both went in the desired direction, down and up, respectively. Today we will be reviewing the cut down version fitting into the 225W thermal envelope, as well as comparing it to the two newcomers from ATI, which are supposed to directly challenge NVIDIA’s new flagship pair.

First, we will start with NVIDIA’s GTX 570. It is based on the same chip powering the GTX 580, the GF110. This second coming of Fermi proved to be quite an improvement over the previous generation, but it was present only in the very expensive NVIDIA’s flagship. GTX570 would best be described as the mixture of GTX 480 and GTX 470, with the aforementioned improvements. From the GTX 480 it gets the number of shaders and TMUs, which is now 480 and 60 respectively, and even surpasses it in clocks. From the GTX 470 it gets everything else. 40 ROPs, 320 bit memory bus, and the1.25 GB frame buffer. Of course, it also gets the two 6 pin power connector requirement form the latter, so you know it consumes less than 225 W.

All the architectural enhancements of the GF110 are present here as well. You get twice the FP16 performance compared to the GF100 based 400 series, and the improved Z-culling system. When you look at the percentage of the increases over the GTX 470, you get a 125 MHz core clock increase parallel to the increase in shaders. With the extra TMUs, you get around 30% of theoretical performance increase. However, the memory throughput is only 13.5 percent higher, and lower than the GTX 480. So we should see a greater improvement in shader heavy games and slightly less in memory intensive scenarios. Unlike the GTX 470, the 570 looks exactly like its older brother. You get the same PCB, the same (great) vapor chamber cooler and the same plastic shield.  Also, unlike the 460 and lower models, there’s still no support for HD audio bitstreaming, which only shows that the GF110 is a tweak, rather than a major redesign of the GF100 GPU.

Let's continue with AMD’s new products. We will first take a look at the full card in the shape of Radeon HD 6970.It is not a 5970 successor, bit we believe we have covered the AMD naming strategy shift when the 6800 series came out. This is a single GPU card based on the Cayman GPU. Since AMD had more time to make a successor to the 5870, we see that it is a completely different product. You do get "only" 1536 processors (1600 in the 5870) but they are based on a different architecture. The number of texture units went up by 20% to 96, and the memory went to 5.5 GHz effective, and to a double amount at 2 GB. Despite the smaller number of processors, the chip is larger than the one used in the 5870 (2.64 billion to 2.15 billion). The core clock is also improved at 880 MHz, up by 30 from the 5870.

Let's see pictures of these two Sapphire cards:

next >>