Arctic Islands GPU Will Double Current Generation’s Performance Per Watt!

Arctic Islands GPU: Due to the shift to a lower node, it appears that 2016’s GPUs will be a significant improvement over the current generation.

At AMD’s earnings call, Lisa Su said that the next generation of AMD graphic cards will double the performance per watt offered by the current generation (Fiji) lineup, whereas previous reports said that Nvidia’s Pascal has 17 Billion transistors (more than twice as much as the Maxwell flasghip).

Arctic Islands GPU
Arctic Islands GPU

Arctic Islands processors will ‘Twice the performance per watt’ of Fiji

Although the huge performance increase has been hinted at previously, this is one of the rare occasions when the corporation has made the increased official. On the most recent earnings call for the 2015 fiscal year’s third quarter results, Lisa Su said:

According to AMD, the flagship of the Arctic Islands is known by the codename “Greenland,” and if this is accurate (and there’s no reason to believe it isn’t), then we should expect performance to grow by a factor of 100 compared to Fiji.

But what does that amount to, numerically speaking? Our current understanding of the FinFET procedure allows us to provide a rough estimate. In contrast to AMD, whose participation in TSMC’s 16nm FinFET+ node has not yet been confirmed, Nvidia is already there.

More reliable claims have surfaced, however, suggesting that AMD has moved on to the 16FF+ node rather than the 14nm FinFET.

Assuming this to be the case, we can expect AMD’s upcoming procedure to follow the same guidelines as Nvidia’s next. Here’s the official rundown of TSMC’s 16FF+ node:

The Fiji GPU’s die size is 596mm2 and it contains 8.9 billion transistors and 4096 stream processors. To get this kind of speed boost, the Greenland (Arctic Islands) processor will likely need to have roughly 18 billion transistors.

Assuming an initial die size of 500mm2, you can put as many as 15 billion transistors on it. Considering the performance boosts from architectural upgrades, the 500mm assumption can see an increase of 66%+ (in reality, anything from 70% to 80%).

The accuracy may not be 100%, but it’s good enough for our purposes. The actual number of cores would be impossible to guess without knowing the size of the stream processor, which we obviously don’t have.

A reasonable estimate puts the figure around 8192 for a 600 mm die and 6784 for a 500 mm die (minus a few multiples of 128). However, the GM200 only has 8 billion transistors, therefore doubling the density would result in roughly 16 billion transistors for a 600mm chip according to TSMC specifications, presuming that the transistor characteristics would remain unchanged.

Assuming a 500mm2 device for the first 16nm FinFET+ run yields a transistor count of about 14 billion. Nvidia’s next-generation Pascal GPUs may have a new transistor-to-core ratio and design if the story of 17 billion transistors being brought to the table by Nvidia turns out to be genuine (improbable).

While TSMC’s 20nm back-end technology lends credence to rumors of a 600mm2 die, I have trouble believing that these businesses will debut with such enormous dies.

I think the likelihood of an initial 500mm2 die is far higher than the possibility that competition will compel them to increase to that size.

In any case, the first true, 4K @60fps monsters will arrive in the DirectX 12 battlefield in 2016, and the competition will be intense. Being a PC gamer right now is unquestionably a positive experience.

Please keep visiting for updates. Keep our site bookmarked so you can quickly return to check for new content. Like Elden, Ring Gamer Transforms Their Avatar