In-Memory Vs Near Memory Computing

Are you up for looking what is in-memory computing and near memory computing and where these technologies used now days? You are on the right spot to know the answer of this question.

In-/near memory is a new computing paradigm that has come to improve the performance of conventional memory systems.

It is a change in the architecture to improve the way in which the processing units access memory with lower latency to alleviate one of the most important power bottlenecks today.

See Also: What is STT RAM Memory

This paradigm will be used more and more in the new equipment that is arriving, and that is why it deserves that we dedicate this article to it to make this new technology known.

In-Memory vs Near Memory Computing

Here let’s check below In-memory vs near memory computing, what these are and where these used now days.

What is In-Memory Computing

In-memory computing refers to the execution of computational calculations entirely in the memory of the computer (for example, in RAM).

This procedure is made possible by the use of specialized system software running on computers in a cluster.

When computers pool their RAM resources, calculations are executed on all of them, taking advantage of the collective amounts of RAM from all computers.

In-memory computing is essentially the same as in-memory processing and in-memory computing, with the possible exception of a small nuance, as it refers specifically to computations.

If you are wondering how in-memory computing works, here we will see it. By eliminating slow data accesses and relying solely on data stored in RAM, In-memory computing increases the overall performance of calculations.

To increase overall computation performance and eliminate latency when accessing hard drives or SSDs, software on one or more computers manages computation and data in memory.

Multiple computers contribute to the computation by dividing it into smaller tasks, which are distributed to each computer to run in parallel.

In-memory computing is frequently performed using in-memory data networks (IMDG). One example is Hazelcast IMDG, which allows users to perform sophisticated calculations on large data sets across a cluster of hardware servers while maintaining high speed.

What is Near-Memory Computing

While processor innovation has outpaced memory in terms of latency and power consumption, known as the memory wall or memory technology.

Previously, system architects tried to bridge this gap by using memory hierarchies to mitigate some of the drawbacks of DRAM.

However, the limited pin count of memory packages is unable to meet the bandwidth demands of multi-core processors.

In addition, the performance of dark silicon computers has reached a plateau. This is due to the demise of Dennard scaling, the slowdown of Moore’s law, and the demise of dark silicon computers.

See Also: NVENC vs NVDEC technology

The contemporary memory hierarchy includes various levels of cache, main memory, and storage. In near-memory computing, data is brought into the cache from storage and processed there.

With near memory computing, data-centric techniques are used to process data close to where they live. Instead, near-memory computing seeks to process data as close to its location as possible.

In addition to being data-centric, near-memory computing couples compute units close to the data to minimize costly data transfers, such as 3D packets, by stacking compute and memory chips.

Near-in-memory processing is made possible by stacking logic and memory on a single silicon rail (TSV) that minimizes memory access latency, power consumption, and bandwidth.

Conclusion

In short, In-memory and near-memory computing aim to eliminate the current barriers of the current memory hierarchy, since the bottleneck between memory and (increasingly faster) processing units is one of the burdens of the current computational paradigm. Therefore, with these changes in the architecture, great gains could be obtained.

This  combined with other technologies such as FPGAs or DSAs, will make computing much more efficient, with less consumption, more performance, and more intelligent thanks to the incorporation of AI technologies.