AMD Introduces Radeon R9 Fury Series Graphics Cards With Fiji GPUs

[+FML]

Yowza! We expected one new Fiji graphics card, but we got four new Fiji-based graphics products with HBM memory from AMD's latest announcement. Next to announcing its new 300-series GPUs, AMD also introduced not just one graphics product based on the new Fiji silicon, but four! Meet the Radeon R9 Fury X, the Radeon R9 Fury, the Radeon R9 nano and Project Quantum. The Radeon R9 Fury X is the high-end solution, which packs 4096 GCN cores and 8.9 billion transistors. The Fury X is the water-cooled variant, while the Fury is only air-cooled. They feature 1.5x the performance per watt over the R9 290X, and heaps more performance.   The Radeon R9 Fury X’s liquid cooler is capable of dissipating up to 500 W, but the board only has a TDP of 275 W. According to AMD, the 6-phase VRM circuitry is even able to provide up to 400 A of current, making this card "an overclocker's dream." AMD did not reveal all that many details about the Radeon R9 Nano but did mention that it is six inches long (shorter than a Mini-ITX motherboard), packs notably more power than the R9 290X, and that it has twice the performance per watt To top off the new launch, Project Quantum takes the cake. It is AMD's own small-form factor gaming PC that features not one, but two Fiji GPUs. Again, AMD was a bit light on the details, but it did tell the audience that the bottom houses the computing power and processing, while the top half takes care of cooling. Presumably, liquid cooling is used. The Radeon water-cooled R9 Fury X will cost $649 and will be available on June 24. The air-cooled R9 Fury will sell for $549 and be on shelves a couple of weeks later on July 14. AMD did not reveal what Project Quantum and the Radeon R9 Nano would cost but said they will be available in the summer and fall, respectively. The new HBM memory from AMD takes a radically different approach to creating a memory subsystem for a graphics card, whereby rather than placing GDDR5 memory on the PCB around the GPU, it uses 3D stacked memory on a silicon interposer, which is placed on the same substrate as the GPU itself. The result is that the memory system takes up 94 percent less space, enabling the assembly of much smaller graphics cards with more power than before. HBM memory is also much faster, with lower latency (due to being closer to the GPU) and much higher bandwidth. It may run at much lower frequencies, but rather than having a 256- or 512-bit memory interface, it runs over a 4096-bit interface. (Wide and slow wins the race.) You can read more about HBM memory here in our past coverage. Of course, we’re still short on many of the details. We don’t know yet at what clock speeds the GPUs or memory run, or many other tech specs. Naturally, we can expect these details to be revealed soon, though.

Source:Tom's Hardware