Nvidia GM200 “Titan II”, AMD Radeon Bermuda XT and Fiji XT in performance test

Speculation about the next generation of graphics cards is hot. Nvidia has already given a taste of what’s to come with the Geforce GTX 970 and 980, but from AMD only loose rumors are available. The Chinese forum Chiphell is now publishing tests on what are said to be upcoming graphics cards from both parties in the top tier.

The first diagram shows the mythical graphics processors AMD Fiji XT and a scaled-down version of the Nvidia GM200, which are believed to have the marketing names Radeon R9 380X and Geforce GTX 980 Ti respectively. The performance index, which is based on 20 game titles, shows that Nvidia’s combatant is slightly faster than AMD’s, while it pulls a little more, which indicates similar energy efficiency.

For the specifications, the scaled-down GM200 is said to have 21 SMM clusters for a total of 2,688 CUDA cores – up from 2,048 for the Geforce GTX 980. For the Fiji XT, previous rumors claim that it is a whole 4,096 stream processors, which are accompanied by stacked memory and a massive memory bus of 4,096 bits.

More interesting are the bars in the higher resolution 3,840 x 2,160 pixels, where what is said to be the next generation of top-class graphics processors are housed. Here, the full-scale GM200 is noticeable, which is believed to have 24 SMM clusters for 3,072 CUDA cores, and the less talked about Bermuda XT from AMD takes first place with a graphics processor.

The details of the Bermuda XT are shrouded in obscurity, but what is clear is that it will have a reference design with combined air and water cooling. This is a direct consequence of high power consumption, which according to the post should remain at an “acceptable level” (compared to Radeon R9 290X? / Editor’s note).

Another and concluding interesting detail is that the next generation of graphics processors are said to be manufactured on a 20-nanometer technology from Globalfoundries. Whether this applies to both AMD and Nvidia is not clear, and given the energy efficiency of the Maxwell architecture, it is not excluded that Nvidia will remain on aging and at the same time price-effective 28 nanometers at TSMC.

The alleged leak is unusually extensive, not least as no less than four as yet unreleased graphics cards from both graphics card giants are represented. The information should therefore be taken with a good pinch of salt.