Home > News content

Amd rdna2 graphics card core exploration: zen2 soul attachment!

via:驱动之家     time:2020/11/22 21:44:22     readed:53

When RX 6800 was officially released from the stock market, we had analyzed its backgroundDesign and characteristics of rDNA 2 Architecture。 Now, according to the rnda 2 kernel diagram, the foreign god has drawn the distribution of each module.

The research and development of rDNA 2 architecture is led by David Wang, senior vice president of engineering R & D of AMD radon technology division. His famous GCN architecture is his masterpiece.

Wang qishang worked for ATI in his early years, and later joined AMD with the acquisition. He once left amd for Synaptics Xinsi technology a few years ago.

In 2018, Wang qishang returned to AMD at the invitation of Su Zifeng, and made great efforts again. With the rDNA 2 architecture, AMD's graphics card returned to the same status as NVIDIA in high-end.

It was under the leadership of Wang qishang that rDNA 2 architecture took 16 months to complete the research and development, and successfully achieved the original design goals: double the performance, improve energy efficiency at least 50%, fully support dx12u and ray tracing and other technical features.


Amd has made it clear that the design of RNA 2 architecture is based on the advanced ideas of Zen CPU architecture. There are also reports that Zen development team has directly involved in the design of rDNA 2. In the layout of rDNA 2 architecture, we can see the shadow of Zen, specifically Zen 2 / 3.

RDNA 2 architecture has 80 groups of computing units, which are divided into two groups. Next to them are two groups of infinite cache with 64MB capacity each. Its function is somewhat similar to that of level 3 cache. Is it similar to the design of two groups of CCX modules and two groups of three-level cache on Zen 2 / 3?

The big core of rDNA 2 architecture, big Navi (nvai 21), is manufactured in TSMC's 7Nm process,There are 26.8 billion integrated transistors with a core area of 519 square millimetersIn contrast, the previous 7Nm Navi 10 integrated 10.3 billion transistors with a core area of 251 square millimeters.

in other words,With 1.61 times more transistors, the area of Navi 21 core is only increased by 1.07 times. The calculation scale, performance and energy efficiency of Navi 21 core are doubled, and 50% respectively.


The following is an introduction to the different modules of rDNA 2:

Red: WGP, also known as the work group processor, is composed of two computing units (Cu), with a total of 40 groups, including 5120 Alu units, which is commonly known as the stream processor.

Blue: Infinity cache unlimited cache, there are two groups, each group has 64MB capacity, the total capacity is 128MB.

Gray: Infinity fabric interconnect bus, infinite cache controller.

Cyan: secondary cache, 256Kb in each group, 16 groups in total, with a total capacity of 4MB.

Orange: instruction front end, geometry processor (in doubt).

Yellow: ROP grating units, 8 in each group, 128 in total.

Pink: raster shader, primitive unit.

Orange: gddr6 video memory controller physical layer, each group 32 bit, total bit width 256 bit.

Green: XGMI / Infinity connection controller, PCIe controller, display controller, codec controller, etc.



Reprint please indicate the source: fast technology

#AMD#Video card#framework#Zen 2#RDNA

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments