There is no doubt that AMD and NVIDIA will launch their ‘latest and greatest’ graphics cards in June 2015. While the exact schedule is lacking details, it is rumored that both companies will split their unveilings between Computex Taipei 2015 (June 2-5) and E3 2015 (June 16-18).
AMD recently fired the first salvo, with a teaser trailer for its upcoming Fiji XT graphics card, high end model for the otherwise rebranded number of GPUs belonging to the ‘300’ series.
As you can see on a leaked image from TechPowerUp, this board is going to be quite smaller than the previous high-end cards, even though it is rumored to feature a cooling breakthrough; All-in-One liquid cooling solution which fits in the space previously reserved for the heatpipe/heatsink/fan combination we got used to over the past 13 years or so.
While it was rumored that Fiji Pro and Fiji XT will carry the name R9 390 and 390X, it looks like AMD decided to change the branding and give their high-end the same treatment NVIDIA uses for their highest-of-all-ends products. While the name is not known for now (Radeon FURY being one of candidates), we know the GPU will use 4GB of HBM (High Bandwidth Memory), which represents another first from the company that brought us GDDR memory (GDDR, GDDR2, GDDR3, GDDR4, GDDR5) and now industry’s first mass produced stacked DRAM. While majority of players much bigger than AMD showed us different variations of Stacked DRAM using TSV (Through-Silicon-Via), including Micron, Intel and others – there’s no denying that AMD is a company that ‘launches first’. Two years ago, AMD and Hynix made a small announcement that they will jointly develop the next generation of graphics memory, and even though neither company showed silicon until now – the results are probably much more impressive than lab experiments we witnessed so far.
Given that we will not see HBM around NVIDIA until the second generation of memory arrives in 2016, we can expect that the green people will push with GDDR5 for one more generation, and the question will be – is 4GB with a massive 4096-bit bus and 512 GB/s bandwidth better than 4/6/8/12/16/24 GB of memory connected to a 256/384/512-bit bus?
Who knows, perhaps the future will show us a Fiji XT2 with two GPUs and 8GB of memory in total? That future might be closer than some people think…