![[ICO]](/icons/blank.gif){kind=icon}
![[PARENTDIR]](/icons/back.gif){kind=icon}
![[DIR]](/icons/folder.gif){kind=icon}
![[ ]](/icons/unknown.gif){kind=icon}
1/4 0 11 19 <- Travisdowns test
ROB XMM YMM
size GPRs regs regs year microarchitecture class
? ? ? ? 2011 Bobcat AMD low-end
72 72 37 - 2003 K8 AMD mainline
72 72 72 - 2009 Phenom II
128 92 128 64 2015 Carrizo (Excavator)
192 145 144 66 2017 Zen
224 141 147 143 2019 Zen2
? ? ? ? 2020 Zen3
32 32 32 - 2013 Silvermont Intel low-end
78 56 65 - 2016 Goldmont
95 68 77 - 2017 Goldmont+
96 96 96 - 2007 Penryn Intel mainline
128 128 128 - 2009 Lynnfield
168 131 112 112 2011 Sandy Bridge
168 116 110 110 2012 Ivy Bridge
192 142 134 134 2013 Haswell
224 151 150 150 2015 Skylake
352 247 203 203 2021 Rocket Lake
The Bobcat and Zen3 do not exhibit the step in the results that the
others exhibit (maybe the Bobcat cannot overlap the execution of two
main memory accesses enough to trigger the effect that the program is
designed for; for the Zen3 I have no theory). Apparently one of the
assumptions used in the measurement program does not hold for these
CPUs.
Combining this with stuff from other sources, we get:
ROB GPR FPR flags Year
64 33 44 2013 Jaguar AMD low-end
72 72 2005 K8 Palermo (Sempron) AMD mainline
72 72 72 2009 Phenom II
128 92 128 2015 Carrizo (Excavator)
192 168 160 2017 Zen Zen1-3
224 180 160 2019 Zen2
256 192 160 2020 Zen3
320 224 192 126 2022 Zen4
448 240 384 192 2024 Zen5
ROB GPR FPR XMM YMM ZMM Year
32 32 32 32 2013 Silvermont Intel low-end
78 2016 Goldmont
95 2017 Goldmont+
208 2020 Tremont
256 2021 Gracemont
416 2024 Skymont
126 2002 Northwood (Pentium 4) Intel mainline Netburst
40 1999 Coppermine (Pentium III) Intel mainline P6-based
96 2007 Yorkfield (Core 2 Quad)
96 2007 Penryn (45nm Core 2)
128 2009 Lynnfield (Core ix-xxx)
168 2011 Sandy Bridge (Core ix-2xxx)
168 2012 Ivy Bridge (Core ix-3xxx)
192 2013 Haswell (Core ix-4xxx)
224 2015 Skylake (Core ix-6xxx)
352 280 224 224 224 224 2019 Sunny/Willow Cove
512 288 320 320 320 220 2021 Golden Cove
576 2024 Lion Cove
ROB GPR FPR Year
192 2013 Apple A7 Cyclone
560 2019 Apple A13 Lightning
630 354 384 2020 Apple A14/M1 Firestorm, more info, more on the microarchitecture
ROB GPR FPR Year
128 128 2016 Cortex A72 ARM big
76 73 70 2016 Cortex A73
128 2018 Cortex A76 & Neoverse N1
160 2019 Cortex A77
160 2020 Cortex A78
160 147 124 2021 Cortex-A710
224 2020 Cortex-X1 (A78 sibling) ARM bigger
288 2021 Cortex-X2 (A710 sibling)
320 2022 Cortex-X3
256 2021 Neoverse V1 (A78 sibling) ARM biggest
ROB GPR FPR Year
96 96 96 2015 Samsung M1 Samsung big (Aarch64 architecture)
100 96 96 2017 Samsung M2
228 192 192 2018 Samsung M3
228 192 176 2019 Samsung M4
228 192 176 2020 Samsung M5
256 224 224 - Samsung M6 (not released)
122 2015 Qualcomm Kryo (Snapdragon 820)
128 2017 Qualcomm Centriq 2400 (Falkor)
680 416 416 2024 Qualcomm Oryon
180 2018 Cavium Thunder X2
220 256 256 2021 Marvell Thunder X3
Cortex A73 and Cortex
A75 do
not use a reorder buffer.
ROB GPR FPR Vec Year 1120 228 240 128 2024 SiFive P870 ROB size apparently is 280 merged instructions
| Name | Last modified | Size | Description |
|---|---|---|---|---|
| Parent Directory | - | ||
| Travisdowns/ | 2021-04-17 19:08 | - | |
| bobcat | 2019-07-09 14:47 | 6.0K | |
| carrizo | 2019-07-08 17:15 | 6.0K | |
| goldmont | 2019-07-08 17:22 | 6.0K | |
| goldmont+ | 2019-07-08 17:25 | 6.0K | |
| haswell | 2019-07-08 17:08 | 6.0K | |
| ivy-bridge | 2019-07-08 18:14 | 5.9K | |
| k8 | 2019-07-08 17:37 | 6.0K | |
| lynnfield | 2019-07-08 17:43 | 6.0K | |
| penryn | 2019-07-08 17:48 | 6.0K | |
| phenom2 | 2019-07-08 17:13 | 6.0K | |
| sandy-bridge | 2019-07-08 17:39 | 6.0K | |
| silvermont | 2019-07-09 14:02 | 6.0K | |
| skylake | 2019-07-08 17:50 | 6.0K | |
| zen | 2019-07-08 17:22 | 6.0K | |