How many "cores" is your mobile phone's GPU| Love to play computer games - 上海419贵族宝贝-上海后花园1314-上海品茶网-阿拉爱上海

In recent years, the mobile platform SoC has developed rapidly. The CPU has developed to four cores as the mainstream, and even the new "4+4" architecture has emerged. In addition to the increase in the number and frequency of CPU cores, GPU We can't ignore the promotion of "12 core" and "16 core" GPUs. In fact, how should we judge the core number of these GPUs?

If you understand the manufacturer's propaganda literally, ordinary consumers are likely to think that the "16 core" GPU will be stronger than the "12 core" GPU, and the "12 core" GPU must be much better than the three core and four core GPUs of Apple and Samsung. Is that really the case?

Just as PC platforms compete for the number of stream processors, mobile platforms begin to compete for the number of cores. However, due to the different definitions of cores by different manufacturers, many people are often confused. Now let's briefly analyze the concept of "core" by several common manufacturers:

1. ImgTec's PowerVR SGX 5XT series regards a relatively complete GPU unit as a core.

2. Nvidia Tegra Series GPU will be a ALU As a core (the number of ALUs is based on a relatively unified MAD ALU, and the role and definition of scalar ALUs may be different from each other, so they are not included temporarily).

3. Qualcomm's Adreno The series does not pay attention to the core publicity, so it is temporarily ignored.

4. ARM's Mali-T6xx Series will be a Unified Shader Unit As a core.

5. Vivante's GC The series will be one SIMD Shader pipeline As a core.

Here is a table of the current high-end GPUs of each company for comparison and understanding

The above dark blue font is a reference from Content of the post (Author: ）

The table can be understood as the "core number" of different GPUs compared to this standard under the standards of different manufacturers.

Attachment:

ALU: Arithmetic logic solver. Since the pixel coloring unit and vertex coloring unit are both 4 channels, the pixel coloring unit and vertex coloring unit of the GPU are designed as ALU with four operations.

The unified shader unit can perform both Vertex calculation and Pixel calculation.

SIMD Shader (Vec4 Shader): In order to improve efficiency, the Shader is designed as the SIMD of Vec4, which can package four data, and then use one instruction for the same processing. Of course, if the number of data is less than four, the computing power will be wasted.

Here is a brief explanation of the table. The GPU "core number" we usually understand is actually a complete GPU unit. It can be seen from the table that only ImgTec products use this standard. The SGX554MP4 in the table is the GPU on the iPad 4. It has four complete GPU units, which is the "four core" GPU we usually understand. Similar cases include the Mali-400MP4 (quad core) of the Galaxy S3 and the SGX544MP3 (tri core) of the Galaxy S4. In this table, except Mali-T678, which is dual core, the most powerful Tegra4, Adreno320 and GC8000 are actually single core products in our conventional understanding.

From this table, we can see many very "slapping" problems:

Compared with the NVIDIA standard, not to mention the "12 cores" of Tegra3, even the Tegra4 with 6 times ALU is inferior to the 128 ALUs of the SGX554MP4 and Mali-T678 in terms of the number of competing ALUs. The latter two can be used as the "128 cores" GPU of the NVIDIA standard.

Compared with Vivante's standard, GC8000's "16 cores" are not eye-catching in comparison, because the SGX554MP4 and Mali-T678 still have 32 SIMD Shaders (Vec4 Shaders), that is, "32 cores".

We cannot ignore the parameters of the GPU, but these "cores" that are often dozens of times can easily mislead consumers, and we should pay more attention to the computing performance of the GPU itself, rather than deliberately pursuing the number of cores.

In addition, it can be understood in combination with the following figure

Extended reading:

Full interpretation of refusing to fool mobile GPU (I)

Full interpretation of refusing to fool mobile GPU (II)

Learn more about new cool devices, stay tuned