- Tram Ho
Last June, Samsung and AMD announced a strategic partnership to bring AMD’s “Next Gen” GPU architecture to mobile devices. Most recently, AMD also published a white paper on their latest RDNA microarchitecture. This document shows how AMD’s high-end RX-5700 graphics card can become the future for low-power devices.
Considered the successor to AMD’s GCN (Graphics Core Next) microarchitecture, RDNA not only changes the number of small cores to memory and connections within it, it also includes the scripts and hardware used by AMD to build the latest GPUs for personal computers, handheld game consoles and other markets.
According to AMD’s white paper, GPUs built on RDNA architecture will spread across a wide range of devices, including notebooks, smartphones and some of the world’s largest supercomputers.
Will AMD’s GPUs meet the requirements of smartphones?
Although it is difficult to tell the performance of AMD GPUs through the white paper technical descriptions, we can see that RDNA offers the optimal fit for use on mobile devices. According to AMD’s white paper, the new GPUs will have L1 cache caching, shared with DCU (Dual Compute Units), which helps reduce energy consumption by reducing the number of read and write times. on memory.
The L2 cache cache is also configured with the ability to provide levels from 64KB to 512KB depending on the performance, power of the application and the silicon area it targets.
AMD’s architecture architecture roadmap.
AMD’s mobile GPU architecture will shift from supporting 64 work items to GCN architectures to 32 optimized work items with RDNA. In other words, workloads can be calculated on 32 parallel calculations simultaneously on each core.
AMD said that the benefit of this parallel computing is to distribute workloads for more cores, improve efficiency and energy efficiency. This is especially suitable for devices with limited bandwidth like mobile phones, when moving large amounts of data will consume a lot of energy.
This shows that AMD is paying close attention to memory and power consumption – the two most important areas for any GPU that succeeds on smartphones.
Take advantage of Radeon’s artificial intelligence tasks
AMD’s GCN architecture, the precursor of RDNA, also has a special advantage in machine learning or artificial intelligence workloads. As we know, AI is playing an increasingly important role in smartphone processors and will continue to become popular in the next 5 years.
RDNA supports up to 8 4-bit parallel operations and FMA calculations for machine learning tasks.
With the new architecture, but RDNA retains high-performance machine components, with the ability to support parallel 64-bit, 32-bit, 16-bit, 8-bit, or even 8-bit arithmetic operations. The Vector ALU of RDNA has twice the width of the previous generation, enabling faster processing and executing FMA multiplication calculations (Fused Multiply-Accumulate) with lower energy consumption than the others. previous generation.
FMA operation is a common calculation in machine learning applications to the extent that ARM’s Mali-G77 GPU must have its own internal hardware block to handle those calculations.
In addition, RDNA also introduced asynchronous Compute Tunneling (Asynchronous Computation) to manage shader workloads. AMD claims this “allows for workloads of computing and graphics processing to co-exist seamlessly on GPUs”. In other words, RDNA handles machine learning and parallel graphics workloads much more efficiently, minimizing the need for a separate AI processor.
RDNA is designed to be more flexible
In addition to the above advantages, AMD’s white paper also mentions a host of other improvements made to this new microarchitecture. But the most interesting thing is probably the Engine Shader and RDNA’s new Shader Array array.
Block diagram of the Radeon RX 5700 XT GPU, one of the first GPUs to use RDNA architecture.
AMD’s white paper said: ” To increase performance from low to high, other GPUs can increase the number of shader arrays and change the level of resource balance within each of these .” So this will depend on what your target platform is, as well as the number of DCU processors, L1 and L2 cache sizes, and even the number of changes in the backend renderer.
Nvidia and ARM also have ways to do the same on their CUDA and Mali GPUs when increasing or decreasing the number of processing cores depending on the performance and power targets needed. But RDNA is different from the above approaches. It offers flexibility in performance tuning and therefore power consumption in each Shader Array array. Instead of just adjusting the number of computations as rivals, the new GPU can adjust both the number of shader arrays and the Render Backend, as well as the cache volume.
This will provide a more flexible platform, with an optimized design for better expansion or contraction than previous AMD products. Even so, these are just theoretical factors, so what kind of performance can be achieved on a limited platform like smartphones is still a factor to consider.
Ever the cooperation GPU between Samsung and AMD launched
According to Samsung’s announcement in the most recent earnings report, we are still “about two years away” until the time when the company launched a new GPU based on RDNA architecture. This suggests that it may appear in 2021. At that time, it is possible that this GPU will have new tweaks and changes compared to the current RX 5700, especially when AMD will be even more optimized. energy consumption.
However, with details in the white paper on RDNA, we have an initial look at AMD’s plan to bring its famous GPU architecture to low-power devices and smartphones. The key point here is a more efficient energy-use architecture, more optimized computational workloads, and a high degree of flexibility in design to suit a variety of applications.
Refer to Android Authority
Source : Trí Thức Trẻ