Israeli researchers develop software to allow processing in memory, bypassing CPU

Jerusalem, Nov 7 (IANS) Israeli researchers have developed a software package that enables computers to perform processing directly in memory, bypassing the central processing unit (CPU), the Israel Institute of Technology said in a statement on Thursday. This innovation addresses the time- and energy-intensive data transfers between memory and the CPU, which have become a bottleneck in modern computing.By handling some computations in memory, the approach reduces reliance on the CPU, leading to significant time and energy savings.The researchers aim to solve the "memory wall" problem, where increasing processor speeds and memory storage capacity outpace data transfer speeds.Traditional computer programmes rely on separate hardware for memory operations and processing, with data being transferred from memory to the CPU for computation.To support in-memory computing, the researchers developed a platform called PyPIM, combining Python programming language with digital processing-in-memory (PIM) technology, Xinhua news agency reported.PyPIM uses new instructions that enable operations to be performed directly in memory.The platform allows developers to write software for PIM computers using familiar programming languages like Python.The researchers also created a simulation tool to help developers estimate performance improvements.The study demonstrated that using PyPIM for various mathematical and algorithmic tasks resulted in significantly faster processing with minimal code changes.--IANSas/

HamroGlobalMedia

Nov 7, 2024 - 15:41

Jerusalem, Nov 7 (IANS) Israeli researchers have developed a software package that enables computers to perform processing directly in memory, bypassing the central processing unit (CPU), the Israel Institute of Technology said in a statement on Thursday.

This innovation addresses the time- and energy-intensive data transfers between memory and the CPU, which have become a bottleneck in modern computing.

By handling some computations in memory, the approach reduces reliance on the CPU, leading to significant time and energy savings.

The researchers aim to solve the "memory wall" problem, where increasing processor speeds and memory storage capacity outpace data transfer speeds.

Traditional computer programmes rely on separate hardware for memory operations and processing, with data being transferred from memory to the CPU for computation.

To support in-memory computing, the researchers developed a platform called PyPIM, combining Python programming language with digital processing-in-memory (PIM) technology, Xinhua news agency reported.

PyPIM uses new instructions that enable operations to be performed directly in memory.

The platform allows developers to write software for PIM computers using familiar programming languages like Python.

The researchers also created a simulation tool to help developers estimate performance improvements.

The study demonstrated that using PyPIM for various mathematical and algorithmic tasks resulted in significantly faster processing with minimal code changes.

--IANS

as/