AMD, one of the most famous multinational semiconductor company, has RDNA 3 as its recent project. RDNA 3 will be great graphics architecture succeeding the current RDNA 2. It seems to be a great turn for AMD to be the first to utilize the design “ Multi chip module.”
MCM-Multi chip module
Take a break to get to know what an MCM design is in reality. MCM, or “Multi chip module,” refers to utilizing several small chips working together as a single piece to form a “Super chip.”
Advanced Micro Devices Inc. is using the same design since the launch of Zen and still maintaining with Zen+, Zen 2, and Zen 3, an architecture, the latter that keeps the chiplet or CCD unit as the smallest base, including the essential changes to take its internal composition closer to the conventional monolithic designs.
Why would AMD take the jump to MCM design with RDNA 3?
While talking about the merits and demerits of Intel and AMD processors, the problems are near in a precise manner, and the reality is that can be taken slowly to a conclusion.
An MCM design lets a huge chip divide into several tiny chips. This makes the transfer to the wafer straight-forward, as there are very few possibilities of failures.
Here is an example to understand it in a better way. Imagine a GPU like the one Radeon RX 6900 XT has that adds 5,120 shaders. Usually, GPU uses a monolithic core design, which means all the shaders are built on a single chip. It can be possible to adapt an MCM design as it will be sufficient for GPU division into two chips with the same no. of 2,560 shaders. These interconnected shaders would surely work as a single unit.
It’s obvious and easy to get a chip with 2,560 totally effective shaders than a chip having 5,120 totally effective shaders. By taking the reference of this example, we have precisely described the significance and value of the MCM designs.
Navi 31 & 10240 shaders
NVIDIA got the achievement of overcoming the challenge of 10,000 shaders including a monolithic core design wasn’t a problem, besides this, the RTX 3090 has 10,496 shaders, but we can still see silicon- level view here as NVIDIA has acquired an easily adaptable architecture to an MCM design.
It’s casual to approach a condition where monolithic core designs are progressively hard to maintain by raising the no. of transistors and making manufacturing processes smaller. These things play the role of a proper limiting factor to overcome with MCM designs easily. AMD has made the declaration with Zen architecture for its feasibility. It has also revealed the limits of such designs with Zen 3.
By taking a leap to an MCM design, AMD could make the best quality RDNA 3-based GPU with 10,240 shaders and 160 computing units. AMD needs to start looking for something different from raw power.
There no talk without reason. The ray-tracing performance of the RX 6000 has been proving after looking at the first tests in Quake 2 RTX under Vulkan. For now, AMD has no competition with DLSS 2.0 technology from NVIDIA.
Comments