Ted Baker iPhone Cases: OPPO R9 spring fashion show

In recent years, rapid development of mobile platform SoC, quad-core CPU has been developed to mainstream, and even a new "4+4" structure. In addition to increasing the number and the frequency of the CPU core, GPU also cannot be ignored, we often see "12 stones", "16 stones" GPU propaganda, that fact and, how should we judge the GPU core at all?

If literally manufacturers propaganda, consumers are likely to believe that "16-core" GPU Council is stronger than "12 stones" of the GPU, and "12 stones" GPU than Apple, Samsung's three-core, Quad-Core GPU than many, that really is the case, please?

As the PC platform competition number of stream processors, mobile platform popular competition the number of cores, but due to the vendors for the definition of the core is not unified, so often confuses a lot of people will, briefly explain common firms for the "core" concept:

1.ImgTec a relatively complete set of GPU PowerVR SGX 5XT series unit as a core.

2. nvidia Tegra GPU series will be a core ALU as a (number with unified MAD ALU ALU units, each for the scalar ALU and definitions may have different therefore temporarily excluded).

3. Qualcomm Adreno series does not focus on the core of propaganda, so long ignored.

4.ARM Mali-T6xx series will be a unified Shader unit as a core.

5.Vivante GC series a SIMD Shader pipeline as a core.

Below the current high-end GPU with a table for comparison to understand

Dark blue font above is reference content from Baidu GPU card (author: crayon small new hip hop)

Form can be understood as different manufacturers under different GPU in comparison to the standard "core".

Attached:

ALU: arithmetic logic unit, for the vertex shader and pixel shader units are 4 channels, so the GPU's pixel shader and vertex shader units are designed to have four times the computational capabilities of ALU.

Unified calculation both Vertex Shader units, capable of Pixel calculations.

SIMD Shader (Vec4 Shader): in order to improve efficiency, Shader designed Vec4 SIMD, you can package the four data, and then use the same processing instruction. Of course if the data is less than four, computing power is wasted.

Here a small explanation of the table, we usually understand GPU "core" GPU units actually should be complete, as can be seen from the table, only ImgTec products using that standard, SGX554MP4 in a table that is on the iPad 4 GPU, it has four integrated GPU units, that is, we usually understand the "quad-core" GPU. Similar situations also Galaxy S3 Mali-400MP4 (quad-core) and Galaxy S4 SGX544MP3 (three cores). In this table, other than the Mali-T678 is a dual-core, including billed as the most Tegra4, Adreno320 and GC8000 is just our general understanding of nuclear products.

From this table we can see that a lot of very "face" issue:

Compared to NVIDIA's standards, not to mention the Tegra3 "12 stones", even 6 times ALU Tegra4, in competition number ALU as well as 128 as good as SGX554MP4 and Mali-T678, which can be used as a standard NVIDIA "128 cores" GPU.

Standards relative to Vivante, GC8000 "16 stones" in contrast is not eye-catching, since SGX554MP4 and Mali-T678 are still far ahead of SIMD 32 Shader (Vec4 Shader), namely "32 cores".

We cannot ignore the GPU parameter, but the dozens of "core" is easy to mislead consumers, and that we should focus on the GPU itself, performance, and should not try to pursue the core numbers.

May also be combined with the following illustration to understand Michael Kors Case