ARM Cortex-A720

High-performance CPU core design From Wikipedia, the free encyclopedia

The ARM Cortex-A720 is a CPU core model from Arm[1][2] unveiled in 2023.[3] It serves as a successor to the ARM Cortex-A715.[4]

Launched2023
Designed byARM Ltd.
L1 cache64/128 KiB
(32/64 KiB I-cache with parity,
32/64 KiB D-cache) per core
L2 cache128–512 KiB per core
Quick facts General information, Launched ...
ARM Cortex-A720
General information
Launched2023
Designed byARM Ltd.
Cache
L1 cache64/128 KiB
(32/64 KiB I-cache with parity,
32/64 KiB D-cache) per core
L2 cache128–512 KiB per core
L3 cache512 KiB – 32 MiB (optional)
Architecture and classification
MicroarchitectureARM Cortex-A720
Instruction setARMv9.2-A
Products, models, variants
Product code name
  • "Hunter"
Variant
History
PredecessorARM Cortex-A715
SuccessorARM Cortex-A725
Close

Cortex-A700 CPU cores series focus on balanced performance and efficiency, and the CPU core can be paired with other cores in its family such as the high performance ARM Cortex-X4 or/and high efficiency ARM Cortex-A520[5] in a CPU cluster. It can be used as either "big" or "LITTLE".[6]

Architecture changes in comparison with ARM Cortex-A715

  • Update to ARMv9.2
  • 15% peak performance improvement over the Cortex-A715
  • Can down to same size as Cortex-A78 with 10% performance improvement
  • Area optimize configuration for no area cost vs Cortex-A78
  • Down L2 cache hit latency to 9 cycles (from 10 cycles)
  • Down mispredict latency to 11 cycles (from 12 cycles)[6]
  • x2 L2 bandwidth
  • DSU-120
    • Up to 14 cores (up from 12 cores)
    • Up to 32 MiB of shared L3 cache (increased from 16 MiB)

Architecture comparison

"big" core
More information μArch, Cortex-A77 ...
μArch Cortex-A77 Cortex-A78 Cortex-A710 Cortex-A715 Cortex-A720 Cortex-A725
Codename Deimos Hercules Matterhorn Makalu Hunter Chaberton
Peak clock speed 2.6 GHz ~3.0 GHz -
Architecture ARMv8.2-A ARMv9.0-A ARMv9.2-A
AArch - 32-bit and 64-bit 64-bit
Max In-flight 160 160 ? 192+[7] - -
L0 (Mops entries) - 1536[8] 0[9] -
L1 (I + D) (KiB) 64 + 64 KiB 32/64 + 32/64 KiB 64 + 64 KiB
L2 Cache (KiB) 256–512 KiB 128–512 KiB 0.25–1 MiB[10]
L3 Cache (MiB) 0–4 MiB 0–8 MiB 0–16 MiB 0–32 MiB[11]
Decode width 4-way 5-way
Dispatch 6 Mops/cycle 5 Mops/cycle[12] ? -
Close

Usage

See also

References

Related Articles

Wikiwand AI