AMD EPYC 9655 vs AMD EPYC 9754 Comparison

The AMD EPYC 9655 and AMD EPYC 9754 are both high-core-count server processors on the AMD Socket SP5 platform, but benchmark data shows they are built for different priorities. The EPYC 9655 wins 14 of 17 head-to-head tests, while the EPYC 9754 takes 3. The average benchmark score for the EPYC 9655 is 373,484, compared to 364,371 for the EPYC 9754 — a 2.5% gap in favor of the newer part. However, the EPYC 9754 holds specific wins in data compression, encryption, and extended instruction workloads, indicating that raw core count still matters in certain throughput scenarios.

The Verdict

The data points to a clear split. For workloads dominated by single-thread performance, per-core efficiency, or physics and prime-number calculations, the AMD EPYC 9655 is the definitive choice. Its Cinebench R23 single-core score of 18,730 versus 11,850 for the EPYC 9754 (a 58.1% advantage) is decisive. Similarly, in PassMark physics, the EPYC 9655 scores 25,958 against 8,793 — a 195.2% lead. Any environment where clock speed and core architecture drive latency-sensitive or branch-heavy tasks will favor the EPYC 9655.

The AMD EPYC 9754, by contrast, is the pick for high-throughput, parallel data manipulation. Its wins in PassMark data compression (3,558,043 vs 3,271,019, an 8.1% edge), data encryption (231,891 vs 210,541, a 9.2% edge), and extended instructions (224,322 vs 203,302, a 9.4% edge) show that its 128 cores can outwork the 96-core EPYC 9655 when the task is memory-bandwidth-bound or highly parallel with minimal inter-thread dependency. For database compression, certain encryption pipelines, or vectorized instruction streams, the EPYC 9754 is the safer bet.

For mixed or general-purpose server fleets, the EPYC 9655’s 14-3 win record and higher average score make it the more versatile option. The EPYC 9754 is a specialist, excelling only in specific niches. Neither part is unlocked for overclocking, and both target the same server/workstation segment, so the decision rests entirely on workload profile.

Architecture Differences

The two processors come from different Zen generations. The AMD EPYC 9655 uses Zen 5 architecture with the codename Turin, built on a 4 nm process at TSMC. The AMD EPYC 9754 uses Zen 4c architecture with the codename Bergamo, on a 5 nm process, also at TSMC. This node difference is fundamental: the 4 nm process enables higher clock speeds and better per-core efficiency, while the 5 nm process on Bergamo prioritizes density.

Transistor counts diverge sharply. The EPYC 9655 packs 99,780 million transistors across 12 chiplets, each 70.6 mm² in size. The EPYC 9754 has 71,000 million transistors across 8 chiplets, each 73 mm². Despite fewer chiplets, the EPYC 9754 achieves more cores (128 vs 96) because Zen 4c is a dense, cache-reduced design. The EPYC 9655’s larger transistor budget supports a bigger L3 cache: 384 MB shared versus 256 MB shared. Per-core L1 cache is also larger on the EPYC 9655 (80 KB vs 64 KB), while L2 is identical at 1 MB per core.

Clock speeds reflect the architectural priorities. The EPYC 9655 has a 2.60 GHz base clock and 4.50 GHz boost clock. The EPYC 9754 has a 2.25 GHz base and 3.10 GHz boost. This 1.40 GHz boost deficit on the EPYC 9754 directly explains its single-thread performance gap. Thermal design power differs too: the EPYC 9655 is rated at 400 W, while the EPYC 9754 is 360 W, meaning the higher-clocked part draws more power despite having fewer cores.

Memory bandwidth is another differentiator. The EPYC 9655 supports 576.0 GB/s, while the EPYC 9754 is capped at 460.8 GB/s. Both have twelve-channel DDR5 memory buses and support ECC memory. PCIe capabilities are identical: Gen 5 with 128 lanes (CPU only). The EPYC 9655 was released later, and its part number is 100-000000674, while the EPYC 9754 uses 100-000001234.

Where Each One Wins

The EPYC 9655 dominates in compute-bound and latency-sensitive workloads. Every Cinebench test — R15, R20, and R23, both single-core and multi-core — goes to the EPYC 9655 with a consistent deltaPct of 58% or 58.1%. In PassMark, it wins floating-point math (662,949 vs 588,187, a 12.7% lead), integer math (1,139,221 vs 1,026,896, a 10.9% lead), multithread (156,085 vs 98,752, a 58.1% lead), physics (25,958 vs 8,793, a 195.2% lead), and random string sorting (440,608 vs 306,481, a 43.8% lead). The single-thread PassMark score is also 65.2% higher (3,845 vs 2,328). The find-prime-numbers test shows an extreme 164.7% advantage (1,599 vs 604), indicating a massive per-core throughput edge.

The EPYC 9754 wins only three tests, but they are not trivial. Data compression at 3,558,043 versus 3,271,019 is an 8.1% win, data encryption at 231,891 versus 210,541 is a 9.2% win, and extended instructions at 224,322 versus 203,302 is a 9.4% win. These are all parallel, memory-heavy operations where the 128-core count can compensate for lower clocks. The EPYC 9754’s 256 threads versus 192 threads gives it a scheduling advantage when the workload can saturate all cores without needing high single-thread speed.

For server workloads like large-scale log compression, SSL/TLS termination, or SIMD-heavy media processing, the EPYC 9754’s wins matter. For general virtualization, databases, or any mixed workload, the EPYC 9655’s 14 wins make it the more reliable performer.

FAQ

Q: Which processor has the higher single-thread performance?

A: The AMD EPYC 9655. Its PassMark single-thread score is 3,845 versus 2,328 for the EPYC 9754, a 65.2% advantage. Cinebench R23 single-core shows a similar gap: 18,730 vs 11,850 (58.1%).

Q: Does the EPYC 9754 ever beat the EPYC 9655?

A: Yes, in three specific tests: PassMark data compression (3,558,043 vs 3,271,019), data encryption (231,891 vs 210,541), and extended instructions (224,322 vs 203,302). These are all 8-9% wins for the EPYC 9754.

Q: What explains the EPYC 9655’s large win in physics benchmarks?

A: The EPYC 9655 scores 25,958 in PassMark physics versus 8,793 for the EPYC 9754, a 195.2% delta. This stems from its higher boost clock (4.50 GHz vs 3.10 GHz) and Zen 5 architecture on a 4 nm node, which improves per-core efficiency in branch-heavy calculations.

Q: How do core counts compare?

A: The EPYC 9754 has 128 cores and 256 threads, while the EPYC 9655 has 96 cores and 192 threads. The EPYC 9754 has 32 more cores, but this does not translate into a win in most benchmarks due to its lower clock speeds.

Q: Are there any differences in memory support?

A: Both support DDR5 with a twelve-channel memory bus and ECC memory. The EPYC 9655 has higher memory bandwidth at 576.0 GB/s, while the EPYC 9754 is at 460.8 GB/s.

Q: Which processor should be chosen for a compression-heavy workload?

A: The EPYC 9754, based on its 8.1% lead in PassMark data compression (3,558,043 vs 3,271,019). Its 128-core design appears better suited for parallel data throughput.

Head-to-Head Benchmarks

The largest win for the EPYC 9655 is in PassMark physics, with a deltaPct of 195.2% (25,958 vs 8,793). This is more than double the score of the EPYC 9754, indicating a fundamental per-core advantage in physics simulation. The find-prime-numbers test shows a 164.7% delta (1,599 vs 604), suggesting the EPYC 9655’s integer pipeline is far more efficient. The single-thread test (PassMark) shows a 65.2% lead (3,845 vs 2,328), which compounds across all Cinebench single-core tests at 58% or 58.1%.

In multi-core Cinebench tests, the EPYC 9655 maintains a consistent 58.1% lead: R15 at 13,373 vs 8,460, R20 at 55,722 vs 35,254, and R23 at 132,672 vs 83,939. The PassMark multithread test also shows a 58.1% delta (156,085 vs 98,752). These consistent percentages suggest the per-core clock and IPC advantage translates linearly to multi-threaded rendering workloads.

Random string sorting favors the EPYC 9655 by 43.8% (440,608 vs 306,481), and floating-point math by 12.7% (662,949 vs 588,187). Integer math shows a 10.9% lead (1,139,221 vs 1,026,896). The EPYC 9754’s wins are smaller in magnitude: data compression at -8.1%, data encryption at -9.2%, and extended instructions at -9.4%. Notably, the EPYC 9754 never exceeds a 10% lead, while the EPYC 9655 often exceeds 50% or 100% in its wins.

Specification Differences

The core count is the most obvious difference: 96 cores and 192 threads for the EPYC 9655, versus 128 cores and 256 threads for the EPYC 9754. Clock speeds favor the EPYC 9655, with a 2.60 GHz base and 4.50 GHz boost, against 2.25 GHz base and 3.10 GHz boost for the EPYC 9754. Thermal design power is 400 W for the EPYC 9655 and 360 W for the EPYC 9754.

Cache configurations differ. The EPYC 9655 has 80 KB of L1 per core, 1 MB of L2 per core, and 384 MB of shared L3. The EPYC 9754 has 64 KB of L1 per core, 1 MB of L2 per core, and 256 MB of shared L3. This 128 MB L3 difference is substantial for working sets that fit in cache.

Process technology sets them apart: 4 nm for the EPYC 9655 versus 5 nm for the EPYC 9754, both from TSMC. Transistor counts are 99,780 million for the EPYC 9655 and 71,000 million for the EPYC 9754. Die configuration is 12 chiplets of 70.6 mm² each versus 8 chiplets of 73 mm² each. Memory bandwidth is 576.0 GB/s for the EPYC 9655 and 460.8 GB/s for the EPYC 9754, both on twelve-channel DDR5 with ECC support. PCIe is identical: Gen 5 with 128 lanes (CPU only). The EPYC 9655 has a launch MSRP of $11852, while the EPYC 9754 has a launch MSRP of $11900. Neither has integrated graphics, and both are locked multipliers. The architectures are Zen 5 (Turin) versus Zen 4c (Bergamo), with the EPYC 9655’s release date coming after the EPYC 9754’s.