In a few days, AMD will introduce the Cayman GPU, a member of their Northern Islands Architecture. Cayman will be known as the Radeon HD 6900 Series, and we unveil the top performing single-GPU part. Thanks to our sources, we managed to get the final specifications of the parts that should be unveiled on December 15th, 2010
First and foremost, the GPU of AMD Radeon HD 6970 is known under a codename Cayman XT, standard codename for the top perforning part. Chip consists out of 24 SIMD arrays, or 384 processors for a neck-to-neck comparison with nVidia GeForce architecture. In AMD terms, 1536 cores. While this is a drop from 1600 cores inside the Radeon HD 5870, AMD is claiming that improved efficiency of the architecture will yield in higher performance.
However, raw computing power does not concur with AMD’s marketing statements. At the shipping speed of 880MHz for the GPU, Cayman XT outputs 2.7 TFLOPS in Single-Precision and 675 GFLOPS in Dual Precision. While the HD 6970 is 20 GFLOPS weaker than the HD 5870 in Single-Precision, new VLIW4 architecture brings 131 GFLOPS more in Dual-Precision mode [HD 5870: 544 GFLOPS IEEE754-DP].
Given that the clock difference is only 30MHz between HD 5870 and HD 6970, we don’t believe that clock-per-clock comparison will bring significant difference in computing performance.
The AMD Radeon HD 6970 packs 32 ROP units and is able to output 32 pixels per clock, while 96 Texture Memory Units should bring performance advantage when compared to nVidia’s approach. Memory controller remained the same 256-bit from Evergreen architecture, but the clock of GDDR5 memory was increased to 1,375MHz in Quad-Data Rate mode, i.e. 5.5 billion transfers per second. Memory bandwidth? 176.0GB/s lies at your command.
Given that HD6970 presentation put a lot of accent on Eyefinity and multi-display gaming, we were not surprised to see that AMD put at least 2GB of ultra-fast GDDR5 memory on all 6900 Series cards, bringing a clear advantage over their green competitor from Santa Clara – as GTX 580 as the top part features only 1.5GB of GDDR5 memory.
Things will get quite interesting when it comes to power consumption, as AMD is introducing the PowerTune technology, which enables dynamic overclocking and downclocking of the graphics card in order to achieve smooth gaming performance. This technology also altered the way how GPU power consumption is measured, bringing the maximum TDP up to 250W. AMD claims that TGU or “Typical Gaming Power” will not exceed 190W but the company placed an 8-pin and a 6-pin connector on the board, meaning you have up to 300W to play with. Our sources claim they already saw GPU overclocked to 1GHz+ speeds, meaning that Cayman can really stretch its legs.
Idle power is now estimated at 20 Watts, which is an improvement over previous-gen parts.
The worrysome part about the AMDs performance slides is that they were comparing the part to GeForce GTX 480, part that is practically EOL’ed by nVidia, as those GF100 dies are much more valuable as Quadro and Tesla parts. No comparisons to the GeForce GTX 570 nor GTX 580 were given.
However, when the sea of reviews hits the Internet, pay close attention to AMD’s weakest spot: Multi-GPU efficiency. According to the slide below, weak GPU scaling should be a matter of the past now.
Original Author: Theo Valich
This news article is part of our Archive, if you are looking for up-to date articles we would recommend a visit to our computer news section on the Bright Side of News frontpage. Additionally, we take great pride in our Home Office section, as well as the best VPN one, so be sure to check them out as well.