What Is Unified Memory?
Unified Memory Architecture (UMA) is a system design in which the CPU and GPU share one coherent memory pool. Both processors access the same physical and virtual memory space. The design removes the need for separate host and accelerator memory, so data does not have to be copied between two memory banks before the GPU can use it.
How Unified Memory Works
In a UMA system, a single address space serves the CPU, the GPU and, in some designs, other accelerators. The memory is coherent, which means each processor sees the same data. Access is transparent to software, so applications read and write shared data without explicit transfer steps. Some implementations use cache coherence to keep data consistent between processors.
Unified Memory Compared With Separate Memory
| Feature | Unified Memory (UMA) | Separate Host and GPU Memory |
|---|---|---|
| Memory pool | One shared pool | Separate banks |
| Data transfer | Not required between CPU and GPU | Explicit copies required |
| Memory coherence | Coherent across processors | Managed by software |
Use Cases
Unified memory suits workloads where the CPU and GPU work on the same data, such as graphics processing, AI inference, image and video handling, and data analysis. Systems that run these tasks benefit from one memory pool shared across processors.
Thinvent Products With This Architecture
The Thinvent range does not include a unified memory design. Thinvent products use separate system memory (DDR4 RAM) and storage (SSD or eMMC), with Intel Core, Intel Celeron, AMD Ryzen and ARM Cortex processors. The Thinvent Industrial PC IPC5, Aero Mini PC, IPC3, Treo Mini PC and Micro 5R Thin Client are available with 2 GB to 32 GB of RAM, depending on the model, for buyers who need a standard, upgradable memory layout.
