The gem5 simulator is a modular platform for computer-system architecture research, encompassing system-level architecture as well as processor microarchitecture.
gem5 is an open source computer architecture simulator used in academia and in industry. gem5 has been under development for the past 15 years initially at the University of Michigan as the m5 project and at the University of Wisconsin as the GEMS project. Since the merger of m5 and GEMS in 2011, gem5 has been cited by over 2900 publications. gem5 is used by many industrial research labs including ARM Research, AMD Research, Google, Micron, Metempsy, HP, Samsung, and others.
gem5 provides four interpretation-based CPU models: a simple one-CPI CPU; a detailed model of an in-order CPU, and a detailed model of an out-of-order CPU. These CPU models use a common high-level ISA description. In addition, gem5 features a KVM-based CPU that uses virtualisation to accelerate simulation.
gem5 features a detailed, event-driven memory system including caches, crossbars, snoop filters, and a fast and accurate DRAM controller model, for capturing the impact of current and emerging memories, e.g. LPDDR3/4/5, DDR3/4, GDDR5, HBM1/2/3, HMC, WideIO1/2. The components can be arranged flexibly, e.g., to model complex multi-level non-uniform cache hierarchies with heterogeneous memories.
gem5 decouples ISA semantics from its CPU models, enabling effective support of multiple ISAs. Currently gem5 supports the Alpha, ARM, SPARC, MIPS, POWER, RISC-V and x86 ISAs. However, all guest platforms aren't supported on all host platforms (most notably Alpha requires little-endian hardware).
The CPU models and caches can be combined in arbitrary topologies, creating homogeneous, and heterogeneous multi-core systems. A MOESI snooping cache coherence protocol keeps the caches coherent.
In application-only (non-full-system) mode, gem5 can execute a variety of architecture/OS binaries with Linux emulation.
Multiple systems can be instantiated within a single simulation process. In conjunction with full-system modeling, this feature allows simulation of entire client-server networks.
gem5’s objects are arranged in OS-visible power and clock domains, enabling a range of experiments in power- and energy-efficiency. With out-of-the-box support for OS-controller Dynamic Voltage and Frequency (DVFS) scaling, gem5 provides a complete platform for research in future energy-efficient systems. However, the existing DVFS documentation is out of date. You can find this page at the old wiki.
CPU model that plays back elastic traces, which are dependency and timing annotated traces generated by a probe attached to the out-of-order CPU model. The focus of the Trace CPU model is to achieve memory-system (cache-hierarchy, interconnects and main memory) performance exploration in a fast and reasonably accurate way instead of using the detailed CPU model.
gem5 can be included in a SystemC simulation, effectively running as a thread inside the SystemC event kernel, and keeping the events and timelines synchronized between the two worlds. This functionality enables the gem5 components to interoperate with a wide range of System on Chip (SoC) component models, such as interconnects, devices and accelerators. A wrapper for SystemC Transaction Level Modelling (TLM) is provided.
gem5 comes with an integrated NoMali GPU model that is compatible with the Linux and Android GPU driver stack, and thus removes the need for software rendering. The NoMali GPU does not produce any output, but ensures that CPU-centric experiments produce representative results.
The gem5 simulator is released under a Berkeley-style open source license. Roughly speaking, you are free to use our code however you wish, as long as you leave our copyright on it. For more details, see the LICENSE file included in the source download. Note that the portions of gem5 derived from other sources are also subject to the licensing restrictions of the original sources.
The gem5 simulator has been developed with generous support from several sources, including the National Science Foundation, AMD, ARM, Hewlett-Packard, IBM, Intel, MIPS, and Sun. Individuals working on gem5 have also been supported by fellowships from Intel, Lucent, and the Alfred P. Sloan Foundation.
Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation (NSF) or any other sponsor.