Radeon RX 480
The basis of the Radeon RX 480 model is the Polaris 10 graphics processor, which has a fourth-generation GCN architecture, which is similar in many details to previously released AMD solutions.
The Polaris 10 GPU belongs to the fourth generation of the Graphics Core Next architecture, the most advanced at the moment. The basic building block of the architecture is the Compute Unit (CU), from which all AMD GPUs are assembled.
The full Polaris 10 GPU includes one Graphics Command Processor, four Asynchronous Compute Engines (ACE), two Hardware Schedulers (HWS), 36 Compute Units (CU), four geometry processors, 144 texture processors. TMU module (including four LSUs per TMU) and 32 ROPs.
The Polaris 10 crystal, containing 5,7 billion transistors, has an area of 232 mm2, which allows us to estimate a 65% increase in chip density compared to Fiji, the densest GPU AMD, manufactured using 28 nm technology (8,9 billion transistors, 596 mm2). For comparison, NVIDIA increased the density by 16% when switching to 72 nm technology (GM200 - 8 billion transistors and 601 mm2, GP104 - 7,2 billion transistors, 314 mm2).
AMD increased performance per watt by 1,7 times in the best-case scenario compared to GCN of previous iterations (for comparison, AMD took the Radeon R9 290) due to one “brute force” - the 14 nm FinFET process technology and the use of power gating and clock gating (disables from power supply and resetting the frequency of idle computing units). Additional power management functions increased the stated result to 2,8.
The memory subsystem of AMD's new GPU includes eight 32-bit GDDR5 memory controllers, providing a total 256-bit memory bus, and a 2 MB LXNUMX cache.
Radeon RX 480 Specifications
|
|
||||||
|
Chip
|
||||||
|
Frequencies
|
||||||
|
Memory
|
||||||
|
Interface and TDP
|
Clock speeds GPU in the RX 480 references, limited to 1266 MHz. As RAM, the RX 480 uses GDDR5 chips with a bandwidth of 8 Gbps per contact - this is the maximum speed available today for this type of memory.
The GCN architecture initially had the most complete set of capabilities brought by the DirectX 12 API, and in the second generation (Hawaii and Bonaire chips, which are equipped with the Radeon HD 7790, R9 260/260X and R9 360) it satisfies the conditions of feature level 12_0.
GCN version 1.3 of all functions not covered in previous iterations introduced support for half-precision calculations (FP16).