How CPU and GPU Work Together

A CPU (central processing unit) works together with a GPU (graphics processing unit) to increase the throughput of data and the number of concurrent calculations within an application. GPUs were originally designed to create images for computer graphics and video game consoles, but since the early 2010’s, GPUs can also be used to accelerate calculations involving massive amounts of data.

A CPU can never be fully replaced by a GPU: a GPU complements CPU architecture by allowing repetitive calculations within an application to be run in parallel while the main program continues to run on the CPU. The CPU can be thought of as the taskmaster of the entire system, coordinating a wide range of general-purpose computing tasks, with the GPU performing a narrower range of more specialized tasks (usually mathematical). Using the power of parallelism, a GPU can complete more work in the same amount of time as compared to a CPU.

Diagram depicts the difference between the control/logic unit of CPU and GPU
Image from Nvidia


Difference Between CPU and GPU

The main difference between CPU and GPU architecture is that a CPU is designed to handle a wide-range of tasks quickly (as measured by CPU clock speed), but are limited in the concurrency of tasks that can be running. A GPU is designed to quickly render high-resolution images and video concurrently.

Because GPUs can perform parallel operations on multiple sets of data, they are also commonly used for non-graphical tasks such as machine learning and scientific computation. Designed with thousands of processor cores running simultaneously, GPUs enable massive parallelism where each core is focused on making efficient calculations.

CPU vs GPU Processing

While GPUs can process data several orders of magnitude faster than a CPU due to massive parallelism, GPUs are not as versatile as CPUs. CPUs have large and broad instruction sets, managing every input and output of a computer, which a GPU cannot do. In a server environment, there might be 24 to 48 very fast CPU cores. Adding 4 to 8 GPUs to this same server can provide as many as 40,000 additional cores. While individual CPU cores are faster (as measured by CPU clock speed) and smarter than individual GPU cores (as measured by available instruction sets), the sheer number of GPU cores and the massive amount of parallelism that they offer more than make up the single-core clock speed difference and limited instruction sets.

GPUs are best suited for repetitive and highly-parallel computing tasks. Beyond video rendering, GPUs excel in machine learning, financial simulations and risk modeling, and many other types of scientific computations. While in years past, GPUs were used for mining cryptocurrencies such as Bitcoin or Ethereum, GPUs are generally no longer utilized at scale, giving way to specialized hardware such as Field-Programmable Grid Arrays (FPGA) and then Application Specific Integrated Circuits (ASIC).

Examples of CPU to GPU Computing

CPU and GPU rendering video
— The graphics card helps transcode video from one graphics format to another faster than relying on a CPU.

Accelerating data — A GPU has advanced calculation ability that accelerates the amount of data a CPU can process in a given amount of time. When there are specialized programs that require complex mathematical calculations, such as deep learning or machine learning, those calculations can be offloaded by the GPU. This frees up time and resources for the CPU to complete other tasks more efficiently.

Cryptocurrency mining — Obtaining virtual currencies like Bitcoin includes using a computer as a relay for processing transactions. While a CPU can handle this task, a GPU on a graphics card can help the computer generate currency much faster.

Does HEAVY.AI Support CPU and GPU?

Yes. The GPU Open Analytics Initiative (GOAI) and its first project, the GPU Data Frame (GDF, now cudf), was the first industry-wide step toward an open ecosystem for end-to-end GPU computing. Now known as the RAPIDS project, the principal goal is to enable efficient intra-GPU communication between different processes running on GPUs.

As cudf adoption grows within the data science ecosystem, users will be able to transfer a process running on the GPU seamlessly to another process without copying the data to the CPU. By removing intermediate data serializations between GPU data science tools, processing times decrease dramatically. Even more, since cudf leverages inter-process communication (IPC) functionality in the Nvidia CUDA programming API, the processes can pass a handle to the data instead of copying the data itself, providing transfers virtually without overhead. The net result is that the GPU becomes a first class compute citizen and processes can inter-communicate just as easily as processes running on the CPU.