AMD EPYC 7573X vs AMD EPYC 7702 Comparison

AMD
AMD

AMD EPYC 7573X

CORE STATE Milan-X
CORE SPECS 32 Cores / 64 Threads
CLOCK SPEED 2.8 Base / 3.6 GHz Turbo
CACHE 768 MB (shared)
MAX TDP 280W
ARCHITECTURE Zen 3
nm
PROCESS 7 nm
LAUNCH DATE 2022
VS
AMD
AMD

EPYC 7702

CORE STATE Rome
CORE SPECS 64 Cores / 128 Threads
CLOCK SPEED 2000 Base / 3.35 GHz Turbo
CACHE 256 MB (shared)
MAX TDP 200W
ARCHITECTURE Zen 2
nm
PROCESS 7 nm
LAUNCH DATE 2019

PERFORMANCE BENCHMARKS

cinebench_cinebench_r15_multicore
5,948
5,900
cinebench_cinebench_r15_singlecore
839
832
cinebench_cinebench_r20_multicore
24,787
24,586
cinebench_cinebench_r20_singlecore
3,499
3,470
cinebench_cinebench_r23_multicore
59,017
58,539
cinebench_cinebench_r23_singlecore
8,331
8,264

Analysis: AMD EPYC 7573X vs AMD EPYC 7702

The AMD EPYC 7573X and AMD EPYC 7702 are both server-class processors built for the same socket, yet they represent two distinct approaches to scaling performance. The 7573X, part of the Milan-X generation, leverages 32 cores with a massive shared cache, while the 7702, from the older Rome generation, doubles the core count to 64. Benchmark data from the Cinebench suite reveals a remarkably consistent pattern: the newer 7573X edges out the 7702 across every single test, though by a narrow margin.

Head-to-Head Benchmarks

The head-to-head results are strikingly uniform. Across six Cinebench tests—spanning R15, R20, and R23, each with single-core and multi-core variants—the AMD EPYC 7573X wins every contest. The winning margin is consistently 0.8% in every single case. This uniformity is notable because it suggests a systemic architectural advantage rather than a workload-specific one.

In Cinebench R15 multi-core, the 7573X scores 5948 against the 7702’s 5900. The single-core R15 test shows 839 versus 832. Moving to R20, the multi-core result is 24787 for the 7573X and 24586 for the 7702, while single-core lands at 3499 against 3470. The R23 tests continue the pattern: multi-core scores 59017 versus 58539, and single-core 8331 versus 8264.

The 0.8% delta is consistent across both single-threaded and multi-threaded workloads. This is a crucial observation. In single-core tests, the 7573X’s higher boost clock of 3.60 GHz versus the 7702’s 3.35 GHz likely contributes to the edge. In multi-core tests, the 7573X compensates for having half the cores (32 vs 64) through a combination of higher base clock (2.80 GHz vs 2000.00 MHz) and a vastly larger L3 cache.

For context, the average benchmark score for the 7573X is 17070, placing it in the 71st percentile of all CPUs. The 7702’s average is 16932, in the 70th percentile. The delta between the two averages is less than 1%, confirming the head-to-head results. The 7573X’s nearest rivals include the Intel Core i5-11400 and Intel Core Ultra 7 164U, both with average scores within a fraction of a point (17067 and 17074, respectively). The 7702’s nearest rivals include the AMD EPYC 7742 and Intel Core i7-1260P, with scores of 16998 and 17007.

Where Each One Wins

The data shows a clean sweep for the AMD EPYC 7573X, but the interpretation requires nuance. The 7573X wins all six benchmark comparisons. However, the magnitude of these wins—0.8%—is small enough to be considered within run-to-run variance for many workloads. The practical takeaway is not that the 7573X is decisively faster, but that it achieves parity or a slight edge despite having half the physical cores.

The 7573X’s advantage is most pronounced in single-core performance, where it wins with scores of 839 (R15), 3499 (R20), and 8331 (R23). These results align with its higher boost clock and the architectural improvements of Zen 3 over Zen 2. For workloads that are lightly threaded or latency-sensitive, the 7573X is the stronger choice.

In multi-core scenarios, the 7573X still wins, but the margin is identical (0.8%). This is remarkable given the 7702’s 64 cores versus 32. The 7573X’s 768 MB of shared L3 cache—three times the 7702’s 256 MB—likely plays a significant role. For cache-resident datasets, the 7573X can keep far more working data on-die, reducing memory traffic. The 7702, by contrast, relies on raw core count, which may benefit workloads with massive thread-level parallelism that can saturate memory bandwidth.

The 7702’s strongest case is in power efficiency. Its TDP is 200 watts versus the 7573X’s 280 watts. For a server operator constrained by power budgets, the 7702 offers 64 cores at a lower thermal envelope, even if benchmark scores are marginally lower. The 7702’s base clock of 2000.00 MHz is notably lower, but the higher core count can compensate in throughput-oriented tasks that scale perfectly.

FAQ

Q: Which processor has a higher average benchmark score?

A: The AMD EPYC 7573X has an average benchmark score of 17070, while the AMD EPYC 7702 scores 16932. The 7573X also holds a higher percentile rank at 71 compared to the 7702’s 70.

Q: How significant is the performance difference in multi-core tests?

A: The 7573X wins every multi-core test by 0.8%. In Cinebench R23 multi-core, it scores 59017 versus 58539. This is a small but consistent margin, suggesting the 7573X’s larger cache and higher clock speeds offset the 7702’s double core count.

Q: Does the EPYC 7702 have any advantage in core count?

A: Yes, the 7702 has 64 cores and 128 threads, double the 7573X’s 32 cores and 64 threads. However, this does not translate to a benchmark win in the tested workloads, as the 7573X still edges ahead in all Cinebench tests.

Q: What is the difference in power consumption?

A: The 7573X has a TDP of 280 watts, while the 7702 has a lower TDP of 200 watts. This makes the 7702 a more power-efficient option for core-heavy deployments, despite its lower clock speeds.

Q: Are both processors compatible with the same server platforms?

A: Yes, both use the AMD Socket SP3. They also share identical memory support: DDR4, eight-channel memory bus, and 204.8 GB/s memory bandwidth. Both support ECC memory and PCIe Gen 4.

Q: Which processor has a higher boost clock?

A: The 7573X boosts to 3.60 GHz, while the 7702 boosts to 3.35 GHz. The 7573X also has a higher base clock at 2.80 GHz versus 2000.00 MHz.

Specification Differences

The specification sheets reveal several key divergences beyond the core count. The 7573X is a 32-core, 64-thread processor, while the 7702 is a 64-core, 128-thread processor. The 7573X operates at a base clock of 2.80 GHz and boosts to 3.60 GHz. The 7702 has a base clock of 2000.00 MHz and a boost clock of 3.35 GHz. The 7573X carries a TDP of 280 watts, whereas the 7702 is rated at 200 watts.

The most dramatic difference is in cache hierarchy. The 7573X features 768 MB of shared L3 cache, while the 7702 has 256 MB. Both have 512 KB of L2 per core, but the L1 cache differs: the 7573X has 64 KB per core, and the 7702 has 96 KB per core. The transistor counts also differ significantly: the 7573X has 33,200 million transistors spread across 8x 81 mm² dies, while the 7702 has 3,800 million transistors on a 74 mm² die.

The release dates are separated by nearly three years: the 7702 launched on 2019-08-06, and the 7573X on 2022-03-21. The 7573X has a launch MSRP of $5590, while no launch MSRP is listed for the 7702. Both processors share the same socket, memory support, and PCIe Gen 4 interface, though the 7573X specifies 128 lanes (CPU only) for PCIe.

Architecture Differences

The architectural gap between these processors is generational. The 7573X is based on Zen 3, with the codename Milan-X, and belongs to the EPYC 7003 series. The 7702 is based on Zen 2, with the codename Rome, and belongs to the EPYC 7002 series. Both are built on a 7 nm process at TSMC, but the design philosophies differ.

The 7573X’s Zen 3 architecture brings a redesigned core layout that improves instructions per clock (IPC) over Zen 2. This is evident in the single-core benchmark wins, where the 7573X outperforms the 7702 despite a narrower core-count advantage. The 7573X’s 768 MB L3 cache is a defining feature of the Milan-X line, designed to accelerate workloads with large working sets that would otherwise hit memory. The 7702’s 256 MB L3 is smaller, reflecting the older design’s focus on core density over cache capacity.

The transistor count disparity—33,200 million for the 7573X versus 3,800 million for the 7702—is partially explained by the cache. The 7573X uses 8x 81 mm² dies, a chiplet-based approach that integrates the massive L3 stack. The 7702’s single 74 mm² die is much simpler. The 7702 compensates with more cores and a lower TDP, making it a power-conscious choice for scale-out workloads. The 7573X, with its higher TDP and newer architecture, targets performance-per-thread and cache-sensitive applications.

Both processors support eight-channel DDR4 memory with 204.8 GB/s bandwidth, so memory throughput is not a differentiator. The 7573X’s advantage lies in how efficiently it uses that bandwidth, thanks to the larger cache. In summary, the 7573X is a newer, cache-rich processor that trades core count for architectural efficiency, while the 7702 is a core-dense workhorse from the previous generation. The benchmark data shows the 7573X’s approach yields a slight but consistent performance lead across all tested Cinebench scenarios.

DETAILED SPECIFICATIONS

SPECIFICATION
EPYC 7573X
EPYC 7702
Core Specs
Cores
32
64 +100.0%
Threads
64
128 +100.0%
Base Clock (GHz)
2.8
2,000 +71328.6%
Boost Clock (GHz)
3.6
3.35 -6.9%
Frequency (GHz)
2.8
2,000 +71328.6%
Turbo Clock (GHz)
3.6
3.35 -6.9%
Multiplier
28
20 -28.6%
SMP CPUs
1
2 +100.0%
Cache
L1 Cache
64 KB (per core)
96 KB (per core)
L2 Cache
512 KB (per core)
512 KB (per core)
L3 Cache
768 MB (shared)
256 MB (shared)
Power
TDP (W)
280
200 -28.6%
Configurable TDP
225W
Architecture
Architecture
Zen 3
Zen 2
Codename
Milan-X
Rome
Generation
EPYC (Zen 3 (Milan))
EPYC (Zen 2 (Rome))
Process Size
7 nm
7 nm
Transistors
33,200 million
3,800 million
Die Size
8x 81 mm²
74 mm²
Foundry
TSMC
TSMC
Memory
Memory Support
DDR4
DDR4
Memory Bus
Eight-channel
Eight-channel
Memory Bandwidth
204.8 GB/s
204.8 GB/s
ECC Memory
Yes
Yes
Platform
Socket
AMD Socket SP3
AMD Socket SP3
PCIe
Gen 4, 128 Lanes(CPU only)
Gen 4
AMD Multi-Die
CCDs
8
Cores per CCD
4
IO Process Size
12 nm
Other
Market
Server/Workstation
Server/Workstation
Production Status
Active
Active
Launch Price
$5590
Part Number
100-000000506​WOF100-000000506​
100-000000038
Package
FCLGA-4094
FCLGA-4094
View EPYC 7573X Details View EPYC 7702 Details