From WikiChip
Difference between revisions of "arm holdings/microarchitectures/cortex-x2"
(cortex-x2) |
(add tables) |
||
| (One intermediate revision by one other user not shown) | |||
| Line 1: | Line 1: | ||
| − | {{armh title|Cortex-X2|arch}} | + | {{armh title|Cortex-X2 (Matterhorn-ELP)|arch}} |
{{microarchitecture | {{microarchitecture | ||
| − | |atype=CPU | + | | atype = CPU |
| − | |name=Cortex-X2 | + | | name = Cortex-X2 (Matterhorn-ELP) |
| − | |designer=ARM Holdings | + | | codename = Cortex-X2 |
| − | |manufacturer=TSMC | + | | core name = '''Cortex-X2''' |
| − | |successor=Cortex-X3 | + | | designer = ARM Holdings |
| − | |successor link=arm holdings/microarchitectures/cortex-x3 | + | | manufacturer = TSMC |
| − | |contemporary=Cortex-A710 | + | | introduction = 2021 |
| − | |contemporary link=arm holdings/microarchitectures/cortex-a710 | + | | process = 10 nm |
| + | | process 2 = 7 nm | ||
| + | | process 3 = 5 nm | ||
| + | | cores = 1 | ||
| + | | cores 2 = 2 | ||
| + | | cores 3 = 4 | ||
| + | | cores 4 = 6 | ||
| + | | cores 5 = 8 | ||
| + | | cores 6 = 10 | ||
| + | | cores 7 = 12 | ||
| + | | type = Superscalar | ||
| + | | type 2 = Pipelined | ||
| + | | oooe = Yes | ||
| + | | speculative = Yes | ||
| + | | renaming = Yes | ||
| + | | stages = 288 | ||
| + | | decode = 5-way | ||
| + | | isa = ARMv9.0-A | ||
| + | | feature = Hardware virtualization | ||
| + | | extension = FPU | ||
| + | | extension 2 = NEON | ||
| + | | l1i = 64 KiB | ||
| + | | l1i per = core | ||
| + | | l1i desc = 4-way set associative | ||
| + | | l1d = 64 KiB | ||
| + | | l1d per = core | ||
| + | | l1d desc = 4-way set associative | ||
| + | | l2 = 1 MiB | ||
| + | | l2 per = core | ||
| + | | l2 desc = 8-way set associative | ||
| + | | l3 = 16 MiB | ||
| + | | l3 per = cluster | ||
| + | | l3 desc = 16-way set associative | ||
| + | | predecessor = '''Cortex-X1''' (Hera) | ||
| + | | predecessor link = arm holdings/microarchitectures/cortex-x1 | ||
| + | | successor = '''Cortex-X3''' (Makalu-ELP) | ||
| + | | successor link = arm holdings/microarchitectures/cortex-x3 | ||
| + | | contemporary = '''Cortex-A710''' (Matterhorn) | ||
| + | | contemporary link = arm holdings/microarchitectures/cortex-a710 | ||
| + | | contemporary 2 = '''Cortex-A510''' (Klein) | ||
| + | | contemporary 2 link = arm holdings/microarchitectures/cortex-a510 | ||
}} | }} | ||
| − | '''Cortex-X2''' is the successor to the {{\\|Cortex-X1}}, a performance-enhanced version of the {{\\|Cortex-A710}}, low-power high-performance [[ARM]] [[microarchitecture]] designed by [[Arm]] for the mobile market. | + | |
| + | '''Cortex-X2''' ''(Matterhorn-ELP)'' is the successor to the '''{{\\|Cortex-X1}}''' ''(Hera)'', a performance-enhanced version of the '''{{\\|Cortex-A710}}''' ''({{armh|Matterhorn|l=arch}})'', low-power high-performance [[ARM]] [[microarchitecture]] designed by [[Arm]] for the mobile market. | ||
| + | |||
| + | === [[Cortex]]-X === | ||
| + | :;[[ARM]] • [[Cortex]] | ||
| + | {| class="wikitable" style="text-align: center; | ||
| + | |- | ||
| + | ! Year !! Cortex-X Core !! Cortex-A Core | ||
| + | |- | ||
| + | | [[2020]] || {{armh|Cortex-X1|l=arch}} (''{{armh|Hera|l=arch}}'') <br>{{armh|Cortex-X1C|l=arch}} (''{{armh|Hera-C|l=arch}}'') || {{armh|Cortex-A78|l=arch}} (''{{armh|Hercules|l=arch}}'') <!--<br>{{armh|Cortex-A78AE|l=arch}} (''{{armh|Hercules-AE|l=arch}}'')--> <br>{{armh|Cortex-A78C|l=arch}} (''{{armh|Hera Prime|l=arch}}'') | ||
| + | |- | ||
| + | | [[2021]] || {{armh|Cortex-X2|l=arch}} <br>(''{{armh|Matterhorn-ELP|l=arch}}'') || {{armh|Cortex-A710|l=arch}} (''{{armh|Matterhorn|l=arch}}'') <br>{{armh|Cortex-A510|l=arch}} (''{{armh|Klein|l=arch}}'') | ||
| + | |- | ||
| + | | [[2022]] || {{armh|Cortex-X3|l=arch}} (''{{armh|Makalu-ELP|l=arch}}'') || {{armh|Cortex-A715|l=arch}} (''{{armh|Makalu|l=arch}}'') | ||
| + | |- | ||
| + | | [[2023]] || {{armh|Cortex-X4|l=arch}} (''{{armh|Hunter-ELP|l=arch}}'') || {{armh|Cortex-A720|l=arch}} (''{{armh|Hunter|l=arch}}'') <br>{{armh|Cortex-A520|l=arch}} (''{{armh|Hayes|l=arch}}'') | ||
| + | |- | ||
| + | | [[2024]] || <s>{{armh|Cortex-X5|l=arch}} (''{{armh|Chaberton-ELP|l=arch}}'')</s> <br>{{armh|Cortex-X925|l=arch}} (''{{armh|Blackhawk|l=arch}}'') || {{armh|Cortex-A720AE|l=arch}} (''{{armh|Hunter-AE|l=arch}}'') <br>{{armh|Cortex-A725|l=arch}} (''{{armh|Chaberton|l=arch}}'') | ||
| + | |- | ||
| + | | [[2025]] || {{armh|Cortex-X930|l=arch}} (''{{armh|Travis|l=arch}}'') || {{armh|Cortex-A730|l=arch}} (''{{armh|Gelas|l=arch}}'') <br>{{armh|Cortex-A530|l=arch}} (''{{armh|Nevis|l=arch}}'') | ||
| + | |- | ||
| + | |} | ||
| + | |||
| + | == Architecture == | ||
| + | |||
| + | === Key changes from {{\\|Cortex-X1}} === | ||
| + | The processor implements the following changes: <ref>{{cite book |date=2021-05-25 |title=Arm Launches Its New Flagship Performance Armv9 Core: Cortex-X2 |url=https://fuse.wikichip.org/news/5269/arm-launches-its-new-flagship-performance-armv9-core-cortex-x2/ |website=WikiChip Fuse |last=WikiChip Fuse}}</ref> | ||
| + | * Instruction set ARMv9.0 | ||
| + | * 10 cycle pipeline down from 11, created by reducing dispatch stage from 2 cycles to 1 | ||
| + | * Reorder buffer (ROB) increased by 30% from 224 entries to 288 | ||
| + | * dTLB increased by 20% from 40 entries to 48 | ||
| + | * Bfloat16 data type support | ||
| + | * Support for Aarch32 removed | ||
| + | * SVE2 SIMD support | ||
| + | * DSU-110 | ||
| + | ** Up to 12 cores (up from 8 cores) | ||
| + | ** Up to 16MB L3 cache (up from 8MB) | ||
| + | * CoreLink CI-700/NI-700: up to 32MB SLC | ||
| + | |||
| + | Performance claims: | ||
| + | *Comparing the {{\\|Cortex-X2}} <ref>{{cite book |title=Arm Announces Mobile Armv9 CPU Microarchitectures: Cortex-X2, Cortex-A710 & Cortex-A510 |url=https://www.anandtech.com/show/16693/arm-announces-mobile-armv9-cpu-microarchitectures-cortexx2-cortexa710-cortexa510 |website=anandtech.com}}</ref> to the {{\\|Cortex-X1}} with the same process, <br>clock speed, and 4MB of L3 cache (also known as ISO-process): | ||
| + | **16% greater integer performance / IPC | ||
| + | **100% greater ML performance | ||
| + | *30% peak performance improvement over the {{\\|Cortex-X1}} in smartphones | ||
| + | :(3.3GHz, 1MB L2, 8MB L3) | ||
| + | * 40% faster than an Intel Core i5-1135G7 at 15W (3.5GHz, 1MB L2, 16MB L3) | ||
| + | |||
| + | === Comparison === | ||
| + | |||
| + | :;"Prime" core | ||
| + | {| class="wikitable sortable" cellpadding="3px" style="border: 1px solid black; border-spacing: 0px; width: 100%; text-align:center; | ||
| + | |- | ||
| + | ![[Microarchitecture|Architecture]] | ||
| + | !{{armh|Cortex-A78|l=arch}} | ||
| + | !{{armh|Cortex-X1|l=arch}} | ||
| + | !{{armh|Cortex-X2|l=arch}} | ||
| + | !{{armh|Cortex-X3|l=arch}} | ||
| + | !{{armh|Cortex-X4|l=arch}} | ||
| + | !{{armh|Cortex-X925|l=arch}} | ||
| + | !{{armh|Cortex-X930|l=arch}} | ||
| + | |- | ||
| + | !Code name | ||
| + | |''{{armh|Hercules|l=arch}}'' | ||
| + | |''Hera'' | ||
| + | |''{{armh|Matterhorn|l=arch}}-ELP'' | ||
| + | |''{{armh|Makalu|l=arch}}-ELP'' | ||
| + | |''{{armh|Hunter-ELP|l=arch}}'' | ||
| + | |''Blackhawk'' | ||
| + | |''Travis'' | ||
| + | |- | ||
| + | !ISA | ||
| + | | colspan="2" |[[ARMv8]].2-A | ||
| + | | colspan="2" |ARMv9.0-A | ||
| + | | colspan="3" |ARMv9.2-A | ||
| + | |- | ||
| + | !Peak clock speed | ||
| + | | colspan="3" |~3.0 GHz | ||
| + | |~3.3 GHz | ||
| + | |~3.4 GHz | ||
| + | |~3.8 GHz | ||
| + | |~4.2 GHz | ||
| + | |- | ||
| + | !Max in-flight | ||
| + | |2x 160 | ||
| + | |2x 224 | ||
| + | |2x 288 | ||
| + | |2x 320 | ||
| + | |2x 384 | ||
| + | |2x 768 | ||
| + | | | ||
| + | |- | ||
| + | !L0 (Mops entries) | ||
| + | |1536 <ref>{{cite book |title=Arm's New Cortex-A78 and Cortex-X1 Microarchitectures: An Efficiency and Performance Divergence |url=https://www.anandtech.com/show/15813/arm-cortex-a78-cortex-x1-cpu-ip-diverging }}</ref> | ||
| + | | colspan="2" |3072 | ||
| + | |1536 | ||
| + | |0 | ||
| + | | | ||
| + | | | ||
| + | |- | ||
| + | !L1-I + L1-D | ||
| + | |32+32 KiB | ||
| + | | colspan="2" |64+64 KiB | ||
| + | | colspan="2" |64+64 KiB | ||
| + | |64+64 KiB | ||
| + | | | ||
| + | |- | ||
| + | !L2 | ||
| + | |128–512 KiB | ||
| + | | colspan="3" |0.25–1 MiB | ||
| + | |0.5–2 MiB | ||
| + | |2–3 MiB | ||
| + | | | ||
| + | |- | ||
| + | !L3 | ||
| + | | colspan="2" |0–8 MiB <ref>{{cite book |last=Schor |first=David |date=2020-05-26 |title=Arm Cortex-X1: The First From The Cortex-X Custom Program |url=https://fuse.wikichip.org/news/3543/arm-cortex-x1-the-first-from-the-cortex-x-custom-program/ |website=WikiChip Fuse }}</ref> | ||
| + | | colspan="2" |0–16 MiB | ||
| + | | colspan="2" |0–32 MiB | ||
| + | | | ||
| + | |- | ||
| + | !Decode width | ||
| + | |4 | ||
| + | | colspan="2" |5 | ||
| + | |6 | ||
| + | |10 <ref>{{cite book |date=2023-05-29 |title=Arm Cortex-X4, A720, and A520: 2024 smartphone CPUs deep dive |url=https://www.androidauthority.com/arm-cortex-x4-explained-3328008/ |website=Android Authority}}</ref> | ||
| + | |10 | ||
| + | | | ||
| + | |- | ||
| + | !Dispatch | ||
| + | |6/cycle | ||
| + | | colspan="3" |8/cycle | ||
| + | | colspan="2" |10/cycle | ||
| + | | | ||
| + | |- | ||
| + | |} | ||
| + | |||
| + | == References == | ||
Latest revision as of 19:22, 15 April 2025
| Edit Values | |
| Cortex-X2 (Matterhorn-ELP) µarch | |
| General Info | |
| Arch Type | CPU |
| Designer | ARM Holdings |
| Manufacturer | TSMC |
| Introduction | 2021 |
| Process | 10 nm, 7 nm, 5 nm |
| Core Configs | 1, 2, 4, 6, 8, 10, 12 |
| Pipeline | |
| Type | Superscalar, Pipelined |
| OoOE | Yes |
| Speculative | Yes |
| Reg Renaming | Yes |
| Stages | 288 |
| Decode | 5-way |
| Instructions | |
| ISA | ARMv9.0-A |
| Extensions | FPU, NEON |
| Cache | |
| L1I Cache | 64 KiB/core 4-way set associative |
| L1D Cache | 64 KiB/core 4-way set associative |
| L2 Cache | 1 MiB/core 8-way set associative |
| L3 Cache | 16 MiB/cluster 16-way set associative |
| Cores | |
| Core Names | Cortex-X2 |
| Succession | |
| Contemporary | |
| Cortex-A710 (Matterhorn) Cortex-A510 (Klein) | |
Cortex-X2 (Matterhorn-ELP) is the successor to the Cortex-X1 (Hera), a performance-enhanced version of the Cortex-A710 (Matterhorn), low-power high-performance ARM microarchitecture designed by Arm for the mobile market.
Cortex-X[edit]
| Year | Cortex-X Core | Cortex-A Core |
|---|---|---|
| 2020 | Cortex-X1 (Hera) Cortex-X1C (Hera-C) |
Cortex-A78 (Hercules) Cortex-A78C (Hera Prime) |
| 2021 | Cortex-X2 (Matterhorn-ELP) |
Cortex-A710 (Matterhorn) Cortex-A510 (Klein) |
| 2022 | Cortex-X3 (Makalu-ELP) | Cortex-A715 (Makalu) |
| 2023 | Cortex-X4 (Hunter-ELP) | Cortex-A720 (Hunter) Cortex-A520 (Hayes) |
| 2024 | Cortex-X925 (Blackhawk) |
Cortex-A720AE (Hunter-AE) Cortex-A725 (Chaberton) |
| 2025 | Cortex-X930 (Travis) | Cortex-A730 (Gelas) Cortex-A530 (Nevis) |
Architecture[edit]
Key changes from Cortex-X1[edit]
The processor implements the following changes: [1]
- Instruction set ARMv9.0
- 10 cycle pipeline down from 11, created by reducing dispatch stage from 2 cycles to 1
- Reorder buffer (ROB) increased by 30% from 224 entries to 288
- dTLB increased by 20% from 40 entries to 48
- Bfloat16 data type support
- Support for Aarch32 removed
- SVE2 SIMD support
- DSU-110
- Up to 12 cores (up from 8 cores)
- Up to 16MB L3 cache (up from 8MB)
- CoreLink CI-700/NI-700: up to 32MB SLC
Performance claims:
- Comparing the Cortex-X2 [2] to the Cortex-X1 with the same process,
clock speed, and 4MB of L3 cache (also known as ISO-process):- 16% greater integer performance / IPC
- 100% greater ML performance
- 30% peak performance improvement over the Cortex-X1 in smartphones
- (3.3GHz, 1MB L2, 8MB L3)
- 40% faster than an Intel Core i5-1135G7 at 15W (3.5GHz, 1MB L2, 16MB L3)
Comparison[edit]
- "Prime" core
| Architecture | Cortex-A78 | Cortex-X1 | Cortex-X2 | Cortex-X3 | Cortex-X4 | Cortex-X925 | Cortex-X930 |
|---|---|---|---|---|---|---|---|
| Code name | Hercules | Hera | Matterhorn-ELP | Makalu-ELP | Hunter-ELP | Blackhawk | Travis |
| ISA | ARMv8.2-A | ARMv9.0-A | ARMv9.2-A | ||||
| Peak clock speed | ~3.0 GHz | ~3.3 GHz | ~3.4 GHz | ~3.8 GHz | ~4.2 GHz | ||
| Max in-flight | 2x 160 | 2x 224 | 2x 288 | 2x 320 | 2x 384 | 2x 768 | |
| L0 (Mops entries) | 1536 [3] | 3072 | 1536 | 0 | |||
| L1-I + L1-D | 32+32 KiB | 64+64 KiB | 64+64 KiB | 64+64 KiB | |||
| L2 | 128–512 KiB | 0.25–1 MiB | 0.5–2 MiB | 2–3 MiB | |||
| L3 | 0–8 MiB [4] | 0–16 MiB | 0–32 MiB | ||||
| Decode width | 4 | 5 | 6 | 10 [5] | 10 | ||
| Dispatch | 6/cycle | 8/cycle | 10/cycle | ||||
References[edit]
- ↑ WikiChip Fuse (2021-05-25). Arm Launches Its New Flagship Performance Armv9 Core: Cortex-X2.
- ↑ Arm Announces Mobile Armv9 CPU Microarchitectures: Cortex-X2, Cortex-A710 & Cortex-A510.
- ↑ Arm's New Cortex-A78 and Cortex-X1 Microarchitectures: An Efficiency and Performance Divergence.
- ↑ Schor, David (2020-05-26). Arm Cortex-X1: The First From The Cortex-X Custom Program.
- ↑ (2023-05-29) Arm Cortex-X4, A720, and A520: 2024 smartphone CPUs deep dive.
Facts about "Cortex-X2 (Matterhorn-ELP) - Microarchitectures - ARM"
| codename | Cortex-X2 (Matterhorn-ELP) + |
| core count | 1 +, 2 +, 4 +, 6 +, 8 +, 10 + and 12 + |
| designer | ARM Holdings + |
| first launched | 2021 + |
| full page name | arm holdings/microarchitectures/cortex-x2 + |
| instance of | microarchitecture + |
| instruction set architecture | ARMv9.0-A + |
| manufacturer | TSMC + |
| microarchitecture type | CPU + |
| name | Cortex-X2 (Matterhorn-ELP) + |
| pipeline stages | 288 + |
| process | 10 nm (0.01 μm, 1.0e-5 mm) +, 7 nm (0.007 μm, 7.0e-6 mm) + and 5 nm (0.005 μm, 5.0e-6 mm) + |