Typical running time for SPEC CPU 2017 full-system simulation

[TiredTumblrina]

Hi,

I'm working on a solution to measure the performance of various system designs, and have been running into issues with the benchmark taking 3+ hours just to get to the end of reading the configuration.

The simulation is running on a i7-8700k with Fedora Linux. The python file looks like this (excluding import and requires statements)

cache_hierarchy = MESITwoLevelCacheHierarchy(
    l1d_size="32kB",
    l1d_assoc=4,
    l1i_size="64kB",
    l1i_assoc=4,
    l2_size="2048kB",
    l2_assoc=8,
    num_l2_banks=2,
)

# Memory: Dual Channel DDR4 2400 DRAM device.
# The X86 board only supports 3 GB of main memory.

memory = ChanneledMemory(DDR4_2400_8x8, 2, 64, size="3GB")

class RsysgenProcessor(SwitchableProcessor):
    def __init__(self):
        start_core = [BaseCPUCore(X86KvmCPU(cpu_id=x), ISA.X86) for x in range(0, 2)]
        end_cores = [BaseCPUCore(X86TimingSimpleCPU(cpu_id=x), ISA.X86) for x in range(0, 2)]
        cores = { "start": start_core, "end": end_cores }
        super().__init__(cores, "start")

    def switch(self):
        self.switch_to_processor("end")

processor = RsysgenProcessor()

# Here we setup the board. The X86Board allows for Full-System X86 simulations
board = X86Board(
    clk_freq="3GHz",
    processor=processor,
    memory=memory,
    cache_hierarchy=cache_hierarchy,
)

command = (
    "m5 exit;"
    + "cd /usr/share/specbench/;"
    + "source shrc;"
    + "echo \"Starting benchmark.\";" 
    + f"runcpu -c gem5bench --threads=2 --noreportable 502.gcc_r 602.gcc_s;"
    + "m5 exit;"
)

board.set_kernel_disk_workload(
    kernel=obtain_resource(resource_id="x86-linux-kernel-5.4.49"),
    disk_image=DiskImageResource(local_path="System_Workloads_gem5/spec2017/bench.img", root_partition="1"),
    readfile_contents=command,
)

board.mem_mode = "atomic_noncaching"

def handle_exit():
    print("Done booting Linux.")
    m5.stats.reset()
    board.mem_mode = 'timing'
    processor.switch()
    yield False  # E.g., continue the simulation.
    print("Done running benchmark.")
    m5.stats.dump()
    yield True  # Stop the simulation. We're done.

# functions to handle different exit events during the simuation
simulator = Simulator(
    board=board,
    on_exit_event={
        ExitEvent.EXIT: handle_exit(),        
    },
)

m5.stats.reset()

# We start the simulation
simulator.run()

My question is (short of buying a more performant PC) are there any ways to speed up the process, maybe using a class implementing AbstractClassicCacheHierarchy instead of using the MESI caches? Or is this time typical for such a simulation?