Unified Memory Architecture
As a Unified Memory Architecture (UMA) is called computer architecture in which various components (eg. As the CPU and the GPU ) together on a common memory , instead of working to own dedicated memory.
x86 computer systems
The more general term for x86 systems is Memory Mapped Graphics (MMG). In the early days of the personal computer, this was the usual way of displaying graphics. Another name for the principle is shared memory .
The elimination of a dedicated memory enables more cost-effective production of the system. The shared access to the main memory results in a loss of performance for the system: When the graphics processor (GPU) is accessing the memory, it is not available to the CPU and the CPU has to wait idle. In 3D mode, this data transfer rate required by the GPU increases many times over due to complex texture calculations. In addition, the memory area used by the GPU (mostly between 128 and 256 MB) cannot be used permanently by the CPU and thus decimates the available main memory. Graphics performance is also lower because access to main memory is slower than access to dedicated graphics memory.
SGI / MIPS systems
In the mid-1990s, SGI brought workstations onto the market in which the CPU and graphics processor shared a high-performance memory. The memory was organized 256 bits wide, which resulted in 3.2 GByte / s at 100 MHz clock frequency. The chipset used was called Cobalt, the main advantage was a flexible division between main memory and graphics memory and the elimination of the PCI bus as a bottleneck.