Vector and Scalar Processing

Modern graphics core architectures feature multiple processing units which are either vector or scalar based. Both are supported by different versions of PowerVR architecture – Series 5 supporting vector, and Series 6, Series 7 and Series 8 supporting scalar.

• Scalar processing units operate on a single value per processing unit.

• Vector processing units work on multiple values per processing unit.

Vector

Vector processing can be very efficient, as the execution unit can work on multiple values at the same time rather than just one. For colour and vertex manipulation, this type of architecture is extremely efficient. Traditional rendering operations are, therefore, well suited to this architecture as calculations often operate on three or four elements at once.

The main drawback of vector architectures is that if scalar values or vectors smaller than the processor expects are used, the additional processing element width is wasted. The most common vector width is four, which means that a shader or kernel mainly operating on three component vectors will operate these instructions with 75% efficiency. Having a shader that works on only one scalar at a time may take this number down to as low as 25%. This wastes energy and performance as parts of the processor are not doing any work. It is possible to optimise for this by vectorising code, but this introduces additional programmer burden.

Scalar

Scalar processors tend to be more flexible in terms of the operations that can be performed per hardware cycle, as there is no need to fill the additional processing width with data. Whilst vector architectures could potentially work on more values in the same silicon area, the actual number of useful results per clock will usually be higher in scalar architectures for non-vectorised code. Scalar architectures tend to be better suited to general purpose processing and more advanced rendering techniques.