Radeon RX Vega 56 is a graphics card from AMD, The core frequency of the graphics card is 1153 MHz. I think AMD may increase this core frequency, but AMD may not do this for stability reasons, The number of transistors is one of the important parameters to measure the performance of a graphics card. The number of transistors in this graphics card is 12.5 billion, which belongs to the upper middle level in the same level of graphics, Radeon RX Vega 56 needs extra power, and the power supply interface is 8+8 Pin, GCN 5.0-based core architecture will bring a higher performance experience, 2048 bit bit of memory interface, allowing it to instantly through put data, Thanks to14 nm's process technology, the heat generated by the graphics card can be further reduced, Radeon RX Vega 56 has 256 texture units (this parameter value is one of the important parameters that determine the performance and grade of the chip. At the same frequency, the one has more texture units will have better performance), It has a good single-precision floating-point computing power of 10.5 TFLOPS, 495 mm² chip area can accommodate more transistors, The number of transistors is one of the important parameters to measure the performance of a graphics card. The number of transistors in this graphics card is 12.5 billion, which belongs to the upper middle level in the same level of graphics, The graphics processor used by the GPU is Vega 10 XL, Radeon RX Vega 56 has only 8 GB HBM2 video memory, but because of its good data processing speed, the actual working ability is still good, The maximum power consumption of Radeon RX Vega 56 is 210 W (only the heavy computing task can reach the maximum power consumption, generally the power is less than 210 W), The number of stream processing units is 3584. Since the stream processing unit directly affects the processing power of the graphics card, the processing power of Radeon RX Vega 56 is good, Its 1469 MHz Boost frequency is supposed to balance the heat and performance, In terms of memory frequency, it seems that AMD is not generous, only equipped with low frequency memory, resulting in its memory frequency is only 1.6 Gbps.

The Radeon RX Vega 56 uses a 14nm FinFET "VEGA 10" core. The Vega 10 XT chip based on the new VEGA architecture has 3584 stream processors with 56 operating units, a total of 144 TMUs, and 32 ROPs. A 1024-bit dual-channel memory controller consists of a total of 2048-bit memory control unit with a size of 8GB.



The VEGA 10 chip has the following features:

1. Based on the 14nm LPP FinFET process.

2. The 64 sets of NCU units are divided into 4 Compute Engines, and each Compute Engines contains 16 sets of NCUs. NCU is the largest underlying innovation since the introduction of the GCN architecture. It incorporates Rapid Packed Math (RPM) and can handle 512 8-bit, 256 16-bit or 128 32-bit operations per clock cycle. Especially a pair of FP16 floating point operations can be handled in a single FP32 ALU. If the two FP16 operations are compatible with each other, they can be packaged together for processing as FP32, thereby doubling the peak throughput. At the same time, the register can also split an FP32 into two FP16s for reverse operation. The NCU unit is optimized for high frequency and high IPC. It can perform calculation and graphics processing at the same time, and can change the width of SIMD unit according to different loads. Tasks that required multiple computing units to complete in the past, will be done with only one unit, without a waste of GPU performance.

3. Four sets of Geometry Engines corresponding to Compute Engines and four sets of DSBRs (drawing flow binners). DSBR can eliminate unnecessary processing and data transmission on the GPU, improve performance and reduce power consumption. AMD claims that DSBR can achieve 10% frame rate improvement in the game, while saving up to 33% of the memory bandwidth without increasing power consumption. .



4. The total number of ACE asynchronous computing units is 4 groups, and there are also 2 sets of HWS designed for asynchronous operations. HWS is used to better schedule hardware resources and improve the efficiency of asynchronous operations.

5. The secondary cache reaches 4MB, improving the power efficiency while reducing the memory bandwidth requirements, and saving energy in the memory operation.

6. Using the new generation HBM2 memory from Micron, the unit bandwidth is twice that of the first generation HBM, reaching 2Gbps/pin; the single Die capacity is up to 4 times that of a generation, and the occupied area is reduced by 50% compared with GDDR5 memory.

7. In order to better utilize the advantages of HBM2 memory, in VEGA, AMD has joined the High-Bandwidth Cache Controller (HBCC), which is used to manage HBM2, system memory, non-volatile memory and network storage. With a maximum virtual addressing space of up to 512TB, HBCC can accommodate and manage very large data sets and fine-tune the transfer of data. HBCC is regarded as the biggest innovation in the VEGA architecture. Simply put, the entire system memory can be used as video memory. It is equivalent to a graphics card that can have terabytes of high-speed memory. In other words, it implements a certain level of integrated memory pool. AMD calls this part "HBCC Memory Area" (HMS).

8．VEGA 10 is AMD's first GPU with "Infinity Fabric" technology, which connects the GPU with the main logic modules such as memory controller, PCI-E controller, display engine and video acceleration module. This modular design allows the GPU to be transformed according to the customer's different needs, and is also applied to the AMD "Zen" architecture CPU.