A veteran graphics analyst takes AMD's first graphics card with stacked memory out for a test drive.
AMD made a breakthrough with their Graphics core Next (GCN) architecture when they brought it out in late 2013. It proved to a viable, scalable, and enduring design which has served them well. This will be the last of that generation, the last of the Fuji chips, and remarkably the test platform for their stacked 3D memory they call High-Bandwidth Memory (HBM).
With 30 percent more performance and 30 percent lower power than the previous generation Radeon R9 290X AIB, the 175W Radeon R9 Nano becomes the world’s most power efficient Mini ITX enthusiast graphics card available—period. This sets a new bar on size and performance. Its price is going to challenge a few users, but in graphics you really do get what you pay for, so we think this is a good deal if you can afford it.
The R9 Nano is the first GPU to deliver AMD’s new 3D stacked 4 GB of AMD’s new high-bandwidth memory on the main die itself. They run it at 500 MHz (1.0 GHz effective) and that gives the GPU 512 GB/s bandwidth—a scorcher.
By stacking up the memory (on the GPU) lots of board space can be saved, and more efficient cooling can be realized. The net result is a smaller, cooler, less power consuming AIB that goes really fast. The AIB has a tiny heat sink (by GPU standards) and a single, 92mm, single axial fan.
Cooling is not the only part of the equation; the new Nano gets the best performance per watt and price—check the Pmark scores. (The Pmark is JPR’s comparison of price, wattage, and performance to calculate a products comparative value.)
We first saw the AIB at E3 in June, and we’ve been waiting for it ever since. The R9 Nano is a marvel of technology improvements, operating at 175 watts and requiring only one 8-pin power plug, this killer board is targeted at 4K gaming and virtual reality. This is an enthusiast AIB, and is priced according. Selling at $649 it will be available by the second week of September.
This will be the last GPU AMD offers based on the 28nm HPX process, and they’ve squeezed every ounce of performance out of the Fiji while reducing power. The Radeon R9 Nano features 64 compute units (CU) which have 64 stream processors per CU, which works out to a total of 4096 stream processors, the same as the flagship R9 Fury X.
The AIB’s GPU has 256 texture mapping units and 64 raster operation units. Compared to the Radeon R9 Fury X with 8.6 TFlops and a 1050 MHz GPU clock, the Radeon R9 Nano hits 8.19 TFlops with a 1000 MHz engine clock. It accomplishes that partially by employing greater memory bandwidth over GDDR5 via a 4096-bit memory interface. Normally a wide bus width will cost you in power consumption, but AMD came up with some clever low power switches which they are not talking about.
How the latest card stacks up in the specs. Source: Jon Peddie Research