"Zen 3”: AMD 2nd Generation7nm x86-64 Microprocessor Core

1. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
1 of 32
“Zen 3”: AMD 2nd Generation
7nm x86-64 Microprocessor Core
T. Burd1, W. Li1, J. Pistole1, S. Venkataraman1, M. McCabe1, T. Johnson1, J. Vinh1, T. Yiu1, M. Wasio1,
H. Wong1, D. Lieu1, J. White2, B. Munger2, J. Lindner2, J. Olson2, S. Bakke2, J. Sniderman2,
C. Henrion3, R. Schreiber2, E. Busta3, B. Johnson3, T. Jackson3, A. Miller3, R. Miller3, M. Pickett3,
A. Horiuchi3, J. Dvorak3, S. Balagangadharan4, S. Ammikkallingal4, P. Kumar4
1AMD, Santa Clara, CA, 2 AMD, Boxborough, MA, 3 AMD Fort Collins, CO, 4 AMD Bangalore, India

2. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
2 of 32
Author Introduction
• B.S., M.S., Ph.D. degrees in Electrical Engineering &
Computer Science from U.C. Berkeley in 1992, 1994, 2001,
respectively
• Consultant with multiple startups in Silicon Valley prior to
joining AMD, Santa Clara, CA in 2005
• Senior Fellow Design Engineer at AMD – Physical Design
Architect for a next-gen Zen CPU core
• Authored/co-authored 28 conference and journal
publications, the book, Energy Efficient Microprocessor
Design, and an inventor on five U.S. patents
Tom Burd

3. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
3 of 32
Outline
• Market Segments and Design Goals
• Architecture
• Core Complex
• L3 Cache
• Technology
• Frequency and Power
• Client and Server Products
• Performance Measurements
• Conclusion

4. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
4 of 32
“Zen 3” Market Segments
AMD RYZEN™ 5000
SERIES MOBILE
PROCESSORS
3RD GEN AMD EPYC™
SERVER PROCESSORS
AMD RYZEN™ 5000
SERIES DESKTOP
PROCESSORS
• Single CPU core across laptop, desktop, and server
• AMD 2nd Generation TSMC 7nm FinFET CPU
• Need to balance performance and power efficiency

5. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
5 of 32
“Zen 3” Design Goals
• “Zen 3”  first “ground-up” re-design of AMD “Zen” family of CPUs
• Goal: deliver performance uplift over “Zen 2”
– Driven primarily by IPC uplift via new microarchitecture
• Same TSMC 7nm FinFET technology as “Zen 2”
– Additional design effort to push frequency in same process
• Achieved +19%1 IPC (ave.), up to +6% frequency, up to +20%2 perf/watt
1. R5K-003, 2. R5K-007 (See endnotes)
AMD chiplet strategy [Naffziger, ISSCC 2020] allows for independent
optimization of CPU and non-CPU IP in product deployment​

6. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
6 of 32
“Zen 3” Architecture
32K I-Cache (8-way)
Decode
Branch Prediction
Op-Cache
Op Queue
Dispatch
Integer Rename
Scheduler Scheduler Scheduler Scheduler Scheduler Scheduler
Floating Point Rename
ALU
BR
AGU ALU
AGU
ST
ALU
ST
AGU ALU BR
F2I
ST
MUL
MAC
ADD
MUL
MAC
ADD
F2I
ST
32K D-Cache (8-way) 512K L2 (I+D) Cache (8-way)
Load/Store Queues
Integer Register File Floating Point Register File
6 Macro Ops/Cycle Dispatched
4 Instructions/Cycle 8 Macro Ops/Cycle
3 Loads/Cycle
2 Stores/Cycle
Major changes from “Zen 2”
• L1 BTB: 5121024 entries
• Improved branch pred. bandwidth
• Int issue width: 710
• Reorder buffer: 224256 entries
• FP issue width: 46
• FMAC latency: 54 cycles
• LD/ST bandwidth: 2/13/2
• TLB table walkers: 46
[Evers, Hot Chips 2021]

7. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
7 of 32
Architectural Performance Uplift
"Zen 2" "Zen 3"
+19%
Geomean of 25 Workloads1
(Fixed 4GHz frequency, 8 cores)
“Zen 3” Performance
Contributors
Cache Prefetching
Execution Engine
Branch Predictor
Micro-op Cache
Front End
Load/Store
“Zen 2” “Zen 3”
1. R5K-003 (See endnotes)

8. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
8 of 32
“Zen 3” Core Complex (CCX)
• Increased 48 cores per CCX
• Same max. 4MB of L3 per core
• Single CCX per chiplet
• 2x L3 cache directly accessible
per core

9. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
9 of 32
“Zen 3” CCX Configs
Performance-Desktop/Workstation/Server
8 Cores, 32MB L3
APU
8 Core, 16MB L3
Value
4 Core, 8MB L3
Core Count
8 16 32 64
48
24
TDP
100
200
300
Workstation
Server
Desktop
APU
• Multiple CCXs can be
placed to create up to
64-core products
• “Zen 3” CPU core can
span from approx. 15-
320W TDP

10. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
10 of 32
L3 Cache Design
Core
Core
Core
Core
Core
Core
Core
Core
L3
Cache
• New bi-directional ring bus
• 32 bytes/cycle in each direction
• High bandwidth / low latency
• Core/L3 interface: 32 bytes/cycle in
each direction
• L3 cache supports up to 64 misses
per core and 192 misses to memory
• Switched HC  HD bitcell
– Arrays achieved -14% area, -24% leakage
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle
32B/cycle

11. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
11 of 32
L3 Slice Implementation
L
X
I
L
3
G
L3S
SW0
SW1
RPT
0
RPT
1
S6
L3S
SW0
SW1
RPT
0
RPT
1
S4
L3S
SW0
SW1
RPT
0
RPT
1
S0
L3S
SW0
SW1
RPT
0
RPT
1
S2
L3S
SW0
SW1
RPT
0
RPT
1
S7
L3S
SW0
SW1
RPT
0
RPT
1
S5
L3S
SW0
SW1
RPT
0
RPT
1
S1
L3S
SW0
SW1
RPT
0
RPT
1
S3
SW1
RPT1
RPT0
SW0
• Single ring-bus interface physical
design (L3S) stepped out 8 times,
each with two switches (SW0, SW1),
and two repeaters (RPT0, RPT1)
• Each slice configured to be a corner
or straight connection
• Wire count similar to prior 4 slice
crossbar of “Zen 2” with ~2000 wires
in both vertical and horizontal
directions
• Wires routes optimized to fly over L3
data macros, avoiding routing
channels
Clockwise 32B bus
Counter-clockwise 32B bus

12. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
12 of 32
AMD 3D V-Cache Ready
• Two columns of TSVs on left/right
side of the L3 cache
• AMD 3D V-Cache extends L3
Cache capacity by 64MB (3x)
• Total inter-die bandwidth: >2 TB/s
• All control and routing to the cores
is implemented on the base die, so
AMD 3D V-Cache can be
completely focused on density
Please attend Paper 26.4 for more details
TSV
Columns
32 B/Cycle
Bi-Directional Bus

13. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
13 of 32
Core Functional Units
• 512kB private L2 cache (8-way)
• 32kB instruction cache (8-way)
• 32kB data cache (8-way)
• Comprised of ~30 sub-blocks
• Chip pervasive logic (CPL)
– Clock/test unit

14. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
14 of 32
Technology
“Zen 2” “Zen 3”
Tech TSMC N7
Cores/CCX
4 cores,
8 threads
8 cores,
16 threads
Area/CCX 31 mm2 68 mm2
L2/core 512kB 512kB
L3/CCX 16MB 32MB
CPP 57 nm
Fin Pitch 30 nm
1x Metal Pitch 57 nm
Stdcell Track Library 6 track
Cu Metal Layers 13 (3 1x, 6 2x, 2 3x, 2 RDL)
Needed to achieve “Zen 3” program goals with same technology as “Zen 2”

15. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
15 of 32
Frequency Design Goal
To drive frequency uplift
• Structured logic placement
• Judicious cell selection
• Wire engineering
• Targeted use of low-Vt gates
• Sub-blocks (~30) kept small to
allow multiple iterations per
week
High Voltage
STA Target
Mid Voltage
STA Target
Low Voltage
STA Target
(Peak 1T Perf)
(APU, Server)
(APU, Desktop,
Server)

16. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
16 of 32
Shift in VT Mix to Drive Frequency
LowVT
LowVTLL
MidVT
MidVTLL
HighVT HighVT
LowVT
LowVTLL
MidVT
MidVTLL
“Zen 2” “Zen 3”
Frequency uplift outweighs increase in leakage power, driving higher perf/watt
27%
12%
37%
11%
13%
36%
9%
31%
12%
12%

17. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
17 of 32
Measured Silicon Frequency

18. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
18 of 32
Core Power Breakdown
Need Perf. Increase @ ISO Power
• IPC increase drives Cac increase
• Cac Target: ΔCac/ΔIPC < 1
• Cac Achieved: ~15%/~19% (Cac/IPC)
To Drive Cac Reduction
• Improved Clock Gating
• Multiple power-savings features.
• Dedicated Power Team

19. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
19 of 32
Measured Core Power Efficiency

20. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
20 of 32
Simultaneous Tape-outs
Base IPs
32MB L3
(w/ TSVs)
16MB L3
(no TSVs)
“Zen 3”
Core
CCX IPs
8 Cores + 32MB L3
8 Cores + 16MB L3
Dies
Desktop/
Server
Chiplet
(CCD)
Monolithic
APU

21. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
21 of 32
CPU Compute Die (CCD) Chiplet
System management unit (SMU)
• Microcontroller
• Power management
• Clocks and reset
• Fuses
• Thermal monitor and control
Infinity Fabric1 On-Package (IFOP) links
• 1.6 GT/s (10 bits at 1.6Ghz)
• 39 RX lanes (2 clock lanes, 1 clock gating lane)
• 31 TX lanes (1 clock gating lane)
• 4 lanes for control traffic (2 clock lanes) 81 mm2 4.15B Transistors (7nm)
1. AMD’s commercially available Infinity Fabric™ Technology

22. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
22 of 32
CCD Support for AMD 3D V-Cache
• TSVs are integrated into a single CCD design
• One CCD design saves NRE cost on mask set and
die qualification
Impact on PPA:
• Area: ~ +4% (of CCD)
• Power: negligible
• Timing: negligible
• Latency: none
Can add AMD 3D V-Cache to create new products for
workloads that benefit from the additional cache.​

23. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
23 of 32
1 CCD  Multiple Server Products
Server IO Die (12nm)
8.4B Transistors, 416 mm2
2-8 CCDs
3rd Gen AMD EPYCTM
Market Segments CCDs Cores Total L3
Cloud / Max-Throughput Compute 8 64 256 MB
High-Performance Compute 8 32 256 MB
High-Frequency / Large Memory Compute 4-8 8-24 128-256 MB
Value / Max IO Compute 2-4 16-32 64-128 MB
4x
DDR
4x
DDR
4 x16 PCIe/IFIP
4 x16 PCIe/IFIP
IFOP IFOP
IFOP
IFOP

24. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
24 of 32
1 CCD  Multiple Client Products
Client IO Die (12nm)
2.1B Transistors, 125 mm2
1-2 CCDs 4th Gen AMD RyzenTM
Processor
Market Segments CCDs Cores Total L3
Max Compute 2 16 64 MB
Best-Balanced Compute 2 12 64 MB
Mainstream / Value Compute 1 6-8 32 MB
2x
DDR
PCIe
IFOP

25. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
25 of 32
APU Monolithic Chip
Key chip features
• 8 core, 16MB L3 CCX
• 8 compute-unit (“Vega”) graphics
• 2 memory controllers
– DDR4 up to 3200 MT/s
– LPDDR4x up to 4266 MT/s
• Multimedia (MM) engines
– 2nd Gen Video Codec1, 3rd Gen Audio ACP
• 2nd Gen display controller (DC)
• I/O controllers
– PCIE® Gen4, USB-C, USB-3.1, USB-2.0, NVMe SATA
• System management unit (SMU)
• Fusion controller hub (FCH)
180 mm2 10.7B Transistors (7nm)
8 Core
16 MB L3
CCX
“Vega”
Graphics
DDR4
/
LPDDR4x
USB
PCIe
MM
DC
I/O
SMU
MM
SMU
1. GD-176 (See endnotes)
FCH
DCE

26. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
26 of 32
Per Core On-Chip Power Regulation
Each CPU core has a digital low drop-out
(DLDO) regulator
AMD Ryzen™ 4000 Series (“Zen 2”)
– All CPU VCORE tied 1:1 with VGFX regardless
of load
AMD Ryzen™ 5000 Series (“Zen 3”)
– Fine-grained power management to run each
core at its optimal frequency AND voltage
– Maximizes the performance efficiency when
running heterogenous workloads
– Projected to provide up to 18% increase in
SOC battery life1
CPU 0: 1.6GHz 1.1V
CPU 2: 2.4GHz 1.1V
CPU 4: 2.4GHz 1.1V
CPU 6: 2.4GHz 1.1V
CPU 1: 2.4GHz 1.1V
CPU 3: 3.9GHz 1.1V
CPU 5: 1.8GHz 1.1V
CPU 7: 1.6GHz 1.1V
CPU 0: 1.6GHz, 0.6V
CPU 2: 2.4GHz, 0.8V
CPU 4: 2.4GHz, 0.8V
CPU 6: 2.4GHz, 0.8V
Graphics: 2 GHz, 1.1V
CPU 1: 2.4GHz, 0.8V
CPU 3: 3.9GHz, 0.9V
CPU 5: 1.8GHz, 0.7V
CPU 7: 1.6GHz, 0.6V
Graphics: 2 GHz, 1.1V
Ryzen4000SeriesMobile
Ryzen5000SeriesMobile
1. CZM-31 (See endnotes)

27. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
27 of 32
Server Performance
Benchmark Config “Zen 2” “Zen 3” Uplift
SPECint®2017
64 Cores [7763 vs. 7H12]1 717 854 +19%
32 Cores [75F3 vs. 7532]2 444 596 +34%
SPECfp®2017
64 Cores [7763 vs. 7H12]3 543 651 +20%
32 Cores [75F3 vs. 7532]2 434 546 +26%
SPECjbb®2017 64 Cores[7763 vs. 7H12]4 249k 314k +26%
1. MLN-088B, 2. MLN-120A, 3. MLN-086B, 4. MLN-092B, 5. MLN-061A (See endnotes)
0
100
200
300
400
500
600
700
800
900
2018 2019 2020 2021
SPECint®2017_int_base
Score
SPECint®2017 Improvement by Year 1,5
“Zen” “Zen 2” “Zen 2”
“Zen 3”
7601
(32 Cores)
7742
(64 Cores)
7H12
(64 Cores)
7763
(64 Cores)
282
701 717
854

28. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
28 of 32
Client Performance
Benchmark Segment “Zen 2” “Zen 3” Uplift
Cinebench R20
Desktop [5950X vs. 3900XT]1,5 546 (4.7 GHz) 640 (4.9 GHz) +17%
Mobile [5800U vs. 5600U]2 474 (4.4Ghz) 551 (4.6 GHz) +16%
Benchmark Config “Zen 2” “Zen 3” Uplift
Cinebench R20 8 Cores [5800U vs. 4800U]2 3218 3655 +14%
PCMark 10 8 Cores [5800U vs. 4800U]3 5081 6074 +20%
PCMark Apps 8 Cores [5800U vs. 4800U]3 8663 10663 +23%
Single-
Thread
Multi-
Thread
1. R5K-004, 2. CZM-48, 3. CZM-38, 4. R5K-009, 5. R5K-025 (See endnotes)
Benchmark Game Config Uplift
CS:GOTM (DirectX® 9)
12 Cores
[3900XT
vs.
5900X]4
+46%
PUBGTM (DirectX® 11) +33%
DOTATM (Vulkan®) +24%
F1TM 2019 (DirectX® 12) +24%
BattlefieldTM V (DirectX® 12) +5%
Benchmark Game Config Uplift
League of LegendsTM (DirectX® 11)
12 Cores
[3900XT
vs.
5900X]4
+50%
Shadow of the Tomb RaiderTM (DirectX® 12) +28%
Far CryTM New Dawn (DirectX® 11) +22%
Ashes of the SingularityTM (Vulkan®) +19%
Total WarTM : Three Kingdoms (DirectX® 11) +6%
Major Gaming Uplifts with “Zen 3”: +26% on average

29. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
29 of 32
Conclusion
• “Zen 3” first major redesign of the AMD “Zen”
microarchitecture
• Scales from approx. 15W mobile to 320W server
• Delivers +19%1 average IPC (vs. “Zen 2”)
• Delivers up to +6% frequency (vs. “Zen 2”)
• Improves power efficiency up to +20%2 (vs. “Zen 2”)
• Achieves max frequency up to 4.9GHz3
1. R5K-003, 2. GD-150, 3. R5K-007 (See endnotes)

30. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
30 of 32
Acknowledgements
• We would like to thank our talented AMD
design team across Austin, Bangalore,
Boston, Fort Collins, and Santa Clara
who contributed to “Zen 3”
• Come check out our demo, 2.7,
in Demo Session #2 (on Wed)
• Attend paper 26.4 for more details on the
new AMD 3D V-Cache technology

31. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
31 of 32
End Notes
CZM-31: Testing by AMD Performance Labs as of 12/08/2020 using an AMD Ryzen 7 5800U processor on an AMD Reference Platform configured with a 53WHr battery, WLAN enabled and Bluetooth off, using 1080p
video playback (result: up to 21.4 hours) and the MobileMark 2018 benchmark test (result: up to 17.5 hours).
CZM-38: Testing by AMD Performance Labs as of 12/30/2020 using PCMark 10, PCMark Apps (APP Performance Overall), Word, Excel, PowerPoint, Edge to measure the productivity performance of a Ryzen 7 4800U
vs. Core i7-1165G7; a Ryzen 7 5800U vs. Core i7-1165G7 and 4800U; a Ryzen 5 5600U vs Core i5-113G57; and a Ryzen 3 5400U vs. Core i3-1115G4. Results may vary. PC Mark is a registered trademark of Futuremark
Corporation.
CZM-48: Testing by AMD engineering using the Cinebench R20 nT benchmark, measuring multithreaded performance of a Ryzen 7 5800U processor engineering sample vs Ryzen 7 4800U, Ryzen 7 5700U processor engineering sample vs Ryzen 7
4700U, Ryzen 5 5600U processor engineering sample vs Ryzen 5 4600U, Ryzen 5 5500U processor engineering sample vs Ryzen 5 4500U, Ryzen 3 5400U processor engineering sample vs Ryzen 3 4300U. Performance may vary.
GD-150: Max boost for AMD Ryzen processors is the maximum frequency achievable by a single core on the processor running a bursty single-threaded workload. Max boost will vary based on several factors, including, but not limited to: thermal paste;
system cooling; motherboard design and BIOS; the latest AMD chipset driver; and the latest OS updates.
GD-176: Video codec acceleration (including at least the HEVC (H.265), H.264, VP9, and AV1 codecs) is subject to and not operable without inclusion/installation of compatible media players.
MLN-061A: As of 03/16/2021, the Intel log trendline from top SPECrate®2017_int_base published scores to date for 2P 1st and 2nd Gen Intel based Xeon SP (LGA socketed) servers for each of 2017, 2018, 2019, 2020, and 2021. The AMD log trendline
from top SPECrate®2017_int_base published score to date, for 2P Intel based AMD EPYC servers for each of 2017, 2018, 2019, 2020, and 2021.
MLN-086B: SPECrate®2017_fp_base comparison based on best performing systems published at www.spec.org as of 07/06/2021. Configurations: 2x AMD EPYC 7763 (651 SPECrate®2017_fp_base, http://spec.org/cpu2017/results/res2021q1/cpu2017-
20210219-24944.html, $15780 1Ku price total, 560W total TDP) versus 2x Intel Xeon Platinum 8380 (489 SPECrate®2017_fp_base, http://spec.org/cpu2017/results/res2021q2/cpu2017-20210521-26361.html, $16198 1Ku price total, 540W total TDP) for
1.33x the performance at 1.37x the score per total CPU $; 0.83x the performance/Core; 1.28x the performance/Watt. Top EPYC 7002 Series result: 2x AMD EPYC 7H12 (543 SPECrate®2017_fp_base, http://spec.org/cpu2017/results/res2020q3/cpu2017-
20200707-23397.html). AMD 1Ku pricing and Intel ARK.intel.com specifications and pricing as of 4/6/21. SPEC®, SPEC CPU®, and SPECrate® are registered trademarks of the Standard Performance Evaluation Corporation. see www.spec.org for more
information.
MLN-088B: SPECrate®2017_int_base comparison based on best performing systems published at www.spec.org as of 07/06/2021. Configurations: 2x AMD EPYC 7763 (854 SPECrate®2017_int_base, http://spec.org/cpu2017/results/res2021q3/cpu2017-
20210622-27664.html, $15780 1Ku price total, 560W total TDP) versus 2x Intel Xeon Platinum 8380 (602 SPECrate®2017_int_base, http://spec.org/cpu2017/results/res2021q2/cpu2017-20210521-26364.html, $16198 1Ku price total, 540W total TDP) for
1.42x the performance at 1.46x the score per total CPU $; 0.89x the performance/Core; 1.37x the performance/Watt. AMD 1Ku pricing and Intel ARK.intel.com specifications and pricing as of 4/6/21. SPEC®, SPEC CPU®, and SPECrate® are registered
trademarks of the Standard Performance Evaluation Corporation. see www.spec.org for more information.
MLN-092B: SPECjbb® 2015-MultiJVM Critical comparison based on best performing 2P systems published at www.spec.org as of 10/26/2021, 2x AMD EPYC™ 7763 scored 313,824 SPECjbb® 2015-MultiJVM Critical-jOPS (339,338 max-jOPS,
https://www.spec.org/jbb2015/results/res2021q3/jbb2015-20210701-00688.html ) which has 47% higher critical server-side Java® operations than the top “Ice Lake” 2x Intel® Xeon® Platinum 8380 that scored 213,195 critical-jOPS (269,094 max-jOPS,
https://www.spec.org/jbb2015/results/res2021q3/jbb2015-20210810-00701.html ). 2x AMD EPYC 7H12 scored 248,942 critical-jOPS (315,663 max-jOPS, http://spec.org/jbb2015/results/res2020q2/jbb2015-20200423-00550.html ). SPEC® and SPECjbb®
are registered trademarks of the Standard Performance Evaluation Corporation. See www.spec.org for more information.
MLN-120A: SPECrate®2017_int_base and SPECrate®2017_fp_base comparison based on best performing 32-core processor-based systems published at www.spec.org as of 10/26/2021. Integer Configurations: 2x AMD EPYC 75F3 (596
SPECrate®2017_int_base, http://spec.org/cpu2017/results/res2021q2/cpu2017-20210409-25541.html , $9720 1Ku price total, 560W total TDP) versus 2x Intel Xeon Platinum 8362 (526 SPECrate®2017_int_base,
http://spec.org/cpu2017/results/res2021q3/cpu2017-20210802-28469.html , $10896 1Ku price total, 530W total TDP) for 1.13x the performance at 1.27x the score per total CPU $; 1.13x the performance/Core; 1.07x the performance/Watt. Floating-Point
Configurations: 2x AMD EPYC 75F3 (546 SPECrate®2017_fp_base, http://spec.org/cpu2017/results/res2021q2/cpu2017-20210409-25543.html , $9720 1Ku price total, 560W total TDP) versus 2x Intel Xeon Platinum 8362 (465 SPECrate®2017_fp_base,
http://spec.org/cpu2017/results/res2021q3/cpu2017-20210802-28467.html , $10896 1Ku price total, 530W total TDP) for 1.17x the performance at 1.316x the score per total CPU $; 1.17x the performance/Core; 1.11x the performance/Watt. AMD 1Ku
pricing and Intel ARK.intel.com specifications and pricing as of 4/6/21. SPEC®, SPEC CPU®, and SPECrate® are registered trademarks of the Standard Performance Evaluation Corporation. see www.spec.org for more information.
R5K-003: Testing by AMD performance labs as of 09/01/2020. IPC evaluated with a selection of 25 workloads running at a locked 4GHz frequency on 8-core "Zen 2" Ryzen 7 3800XT and "Zen 3" Ryzen 7 5800X desktop processors configured with
Windows® 10, NVIDIA GeForce RTX 2080 Ti (451.77), Samsung 860 Pro SSD, and 2x8GB DDR4-3600. Results may vary.
R5K-004: Testing by AMD performance labs as of 09/01/2020 with a Ryzen 5950X processor vs a Core i9-10900K configured with NVIDIA GeForce GTX 2080 Ti graphics, Samsung 860 Pro SSD, 2X8 DDR4-3600, Windows 10 and a Noctua NH-D15s
cooler. Single-core performance evaluated with Cinebench R20 1T benchmark. Results may vary.
R5K-007: Testing by AMD Performance Labs as of 09/01/2020 using Cinebench R20 nT versus system wall power during full load CPU test using a Core i9--10900K, Ryzen 9 3900XT, Ryzen 9 5900X, Ryzen 9 3950X, and a Ryzen 9 5950X configured
with: 2x8GB DDR4-3600, GeForce RTX 2080 Ti, Samsung 860 Pro SSD, Noctua NH-D15s cooler, and an open-air test bench with no additional power draw sources. Results may vary.
R5K-009: Testing by AMD performance labs as of 09/01/2020 measuring gaming performance of a Ryzen 9 5900X desktop processor vs. a Ryzen 9 3900XT in 11 popular titles at 1920x1080, the High image quality preset, and the newest graphics API
available for each title (e.g. DirectX® 12 or Vulkan™ or DirectX® 11). Results may vary.
R5K-025: Testing by AMD Performance Labs as of September 23, 2020, using Ryzen(TM) 5000 Series processors: 5600X, 5800X, 5900X, 5950X and 3rd Gen Ryzen(TM) processors: 3600X, 3800XT, 3900XT and 3950X, configured with DDR4-3600C16
and NVIDIA GeForce RTX 2080 Ti, in Cinebench R20 1T. Results may vary.

32. 2.7: “Zen 3”: AMD 2nd Generation 7nm x86-64 Microprocessor Core
© 2022 IEEE International Solid-State Circuits Conference
32 of 32
Disclaimer
The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions,
and typographical errors. The information contained herein is subject to change and may be rendered inaccurate for many reasons,
including but not limited to product and roadmap changes, component and motherboard version changes, new model and/or
product releases, product differences between differing manufacturers, software changes, BIOS flashes, firmware upgrades, or the
like. Any computer system has risks of security vulnerabilities that cannot be completely prevented or mitigated. AMD assumes no
obligation to update or otherwise correct or revise this information. However, AMD reserves the right to revise this information and
to make changes from time to time to the content hereof without obligation of AMD to notify any person of such revisions or
changes.
THIS INFORMATION IS PROVIDED ‘AS IS.” AMD MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO
THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY INACCURACIES, ERRORS, OR OMISSIONS THAT
MAY APPEAR IN THIS INFORMATION. AMD SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF NON-
INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL AMD BE LIABLE
TO ANY PERSON FOR ANY RELIANCE, DIRECT, INDIRECT, SPECIAL, OR OTHER CONSEQUENTIAL DAMAGES ARISING
FROM THE USE OF ANY INFORMATION CONTAINED HEREIN, EVEN IF AMD IS EXPRESSLY ADVISED OF THE POSSIBILITY
OF SUCH DAMAGES.
© 2022 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, EPYC, Infinity Fabric, Ryzen, and
combinations thereof are trademarks of Advanced Micro Devices, Inc. Other product names used in this publication are
for identification purposes only and may be trademarks of their respective companies. PCIe is a registered trademark of
PCI-SIG.