The genre of mobile GPU rendering principles--IMR, TBR, and TBDR

The genre of mobile GPU rendering principles--IMR, TBR, and Tbdr

The mobile GPU can only be considered as a small child, although children can be more advantageous than adults on some occasions (such as acrobatics, contortion, etc.), but there are innate differences in power, mainly in theoretical performance and bandwidth.

Compared with the desktop GPU 256bit or even 384bit of bit width, 1.2-1.5ghz high-frequency memory, mobile GPU not only to share memory bandwidth with the CPU, but also generally use a dual 32bit bit width, lpddr2-800 or about 1066 of the memory system, The total bandwidth is generally within 10gb/s, the sad Tegra 3 uses a single channel memory mode, with ddr3l bandwidth but 6.4gb/s.

The GPU's most powerful ipad 4 bandwidth is still 17gb/s (image from AnandTech)

The highest memory bandwidth in the mobile processor is the ipad 3/4, because they use the retina screen, 2048x1536 high-resolution GPU bandwidth requirements higher, but even these two products, 17gb/s bandwidth and PC graphics on the 200gb/s more than the bandwidth is more than pediatrics.

Without high bandwidth there is no high-volume texture data, and there is no quality. Although bandwidth is not the only factor restricting the development of mobile GPUs, under current constraints, mobile GPU manufacturers are concerned about how to improve GPU performance and image quality in the smallest possible bandwidth requirements, the previous introduction of texture compression is a method, there is a different way to use the rendering, mainly IMR, TBR and Tbdr and so on.

"Immediate rendering Mode"--IMR

IMR (Immediate Mode Rendering) is literally-the submission of each rendering request will begin immediately, which is a simple and rough idea, the pros and cons are very obvious, if not for performance concerns, this way will be very convenient, However, the rendering of IMR is non-discriminatory, and the masking process is still rendered by the processor, which results in more meaningless read and write operations, wasting a lot of performance and bandwidth.

In short, IMR this rendering method on the mobile GPU evaluation can only be "negative points, rolling!" ”。

The clever "map rendering"--TBR

IMR silly approach is not advisable, then a clever way to--TBR (tile Based Rendering, map rendering), it will need to render the screen into a block (tile), each block of coordinates by the intermediate buffer in the form of a list in the system memory. The benefit of this rendering method is that the relative IMR reduces unnecessary rendering tasks, with the drawback that the masking debris remains small and requires an intermediate buffer.

TBR rendering divides the game screen into different chunks

Re-evolved rendering style--TBDR

TBR is much smarter than IMR, but there are still a lot of flaws, tbdr (Tile Based Deferred Rendering, map delay rendering), it is similar to the TBR principle, but using deferred rendering (Deferred Rendering), Combined with perfect pixels, the HSR (Hidden surface removal, hidden surface elimination) further reduces the need for rendering and reduces bandwidth requirements. In fact, these changes are somewhat similar to the rendering on a PC.

One key to tbdr rendering is deferred rendering

Several other manufacturers are using TBR technology, TBDR is mainly used by imagination, which is one of their biggest chips.

Support for TBDR is also mentioned in Microsoft's DX11.1 upgrade, as the Windows 8 system is also optimized for tablet and touch optimization, and it is also necessary to optimize the technology that is commonly used in the TBDR mobile platform.

The genre of mobile GPU rendering principles--IMR, TBR, and TBDR