AMD EPYC 9655 vs Intel Xeon 6774P Comparison

Where Each One Wins

The benchmark record is unambiguous: the AMD EPYC 9655 wins all 14 recorded head-to-head tests against the Intel Xeon 6774P. There is no workload category in the database where the Intel part takes a lead. That said, the magnitude of the AMD advantage varies sharply, which reveals where each chip is relatively stronger or weaker in a comparative sense.

The EPYC 9655's largest margins appear in integer-heavy and encryption workloads. The PassMark integer math test shows a 92% delta, and data encryption shows an 83.1% delta. These are not modest edges; they are near-doubling results. For environments running cryptographic operations, compression pipelines, or integer-heavy database queries, the AMD part is the clear choice.

The Xeon 6774P, despite losing every test, shows its most competitive result in the PassMark extended instructions test, where the AMD EPYC 9655 leads by only 14.7%. This suggests that for certain vectorized or specialized instruction workloads, the Intel architecture closes the gap considerably. The next smallest delta is 26.3% in single-thread PassMark, which means the Intel chip's per-core performance is not as far behind in lightly threaded scenarios.

For multithreaded throughput, the EPYC 9655 dominates: Cinebench R23 multi-core shows a 38.4% lead, PassMark multithread shows 38.5%, and PassMark floating point math shows 42.5%. The pattern is consistent across rendering, physics simulation, and general floating-point work. The Xeon 6774P, with fewer cores and a lower boost clock, simply cannot match the aggregate compute throughput of the AMD part.

The Intel Xeon 6774P does offer a lower launch MSRP, which can be stated once as a fact: its launch MSRP is $6760. The AMD EPYC 9655 has a launch MSRP of $11852. No cost-benefit analysis is provided here, only the recorded figures.

In summary, the AMD EPYC 9655 wins every recorded benchmark, with the largest advantages in integer math, encryption, and physics workloads. The Intel Xeon 6774P is closest in extended instructions and single-thread performance, but it never takes a win.

FAQ

Q: Which CPU has more cores?

A: The AMD EPYC 9655 has 96 cores and 192 threads, while the Intel Xeon 6774P has 64 cores and 128 threads.

Q: What is the biggest single benchmark margin between the two?

A: The largest delta is in PassMark integer math, where the AMD EPYC 9655 scores 1,139,161 versus 593,440 for the Intel Xeon 6774P, a 92% advantage.

Q: Is the Intel Xeon 6774P competitive in any single-thread test?

A: In PassMark single-thread, the AMD EPYC 9655 scores 3,847 and the Intel Xeon 6774P scores 3,047, a 26.3% delta. The Intel part is closest in the extended instructions test, where the delta is 14.7%.

Q: How do the two compare in memory bandwidth?

A: The AMD EPYC 9655 supports twelve-channel DDR5 with 576.0 GB/s bandwidth. The Intel Xeon 6774P supports eight-channel DDR5 with 409.6 GB/s bandwidth.

Q: What process nodes do the two CPUs use?

A: The AMD EPYC 9655 is built on TSMC's 4 nm process, while the Intel Xeon 6774P uses Intel's 5 nm process.

Q: What is the percentile ranking for each CPU?

A: The AMD EPYC 9655 ranks in the 100th percentile against all CPUs in the database, and the Intel Xeon 6774P ranks in the 99th percentile.

Head-to-Head Benchmarks

The database records 14 direct comparisons, and the AMD EPYC 9655 wins all of them. The most decisive result is PassMark integer math: 1,139,161 versus 593,440, a 92% delta. That is the single largest gap in the entire dataset, indicating a fundamental throughput advantage in integer operations. The AMD part also leads PassMark data encryption by 83.1%, scoring 210,555 against 115,025. Encryption workloads often rely on large integer arithmetic and memory access patterns, so the two results are consistent.

The Cinebench family of tests shows a uniform 38.4% delta across R15, R20, and R23 multi-core. In R23 multi-core, the AMD EPYC 9655 scores 132,672 and the Intel Xeon 6774P scores 95,836. This consistency across three Cinebench generations suggests the advantage scales with core count and sustained all-core boost behavior, not with any single instruction set feature.

PassMark physics shows a 61.9% delta: 25,947 versus 16,023. This is a strongly threaded workload, and the AMD part's 96 cores versus 64 cores directly contributes to the margin. Random string sorting shows a 67.6% delta, with scores of 439,682 versus 262,417, a workload that stresses memory latency and cache capacity. The AMD EPYC 9655 has 384 MB of shared L3 cache versus 336 MB on the Intel part, which likely helps in this test.

Floating point math shows a 42.5% delta: 662,958 versus 465,314. Data compression shows a 41.6% delta: 3,271,896 versus 2,309,868. Find prime numbers shows a 26.4% delta: 1,598 versus 1,264. Extended instructions shows the closest result at 14.7%: 203,285 versus 177,273. Single-thread PassMark shows 26.3%: 3,847 versus 3,047.

The smallest deltas are in extended instructions and single-thread, which suggests the Intel Granite Rapids core has competitive per-core execution for certain instruction mixes. However, the AMD Zen 5 core still holds a meaningful edge in single-thread PassMark, and the gap widens substantially as thread counts scale.

Specification Differences

The two CPUs differ in nearly every major specification field. Core count: 96 versus 64. Thread count: 192 versus 128. Base clock: 2.60 GHz versus 2.50 GHz. Boost clock: 4.50 GHz versus 3.90 GHz. TDP: 400 watts versus 350 watts. Socket: AMD Socket SP5 versus Intel Socket 4710.

Memory configuration diverges significantly. The AMD EPYC 9655 uses a twelve-channel DDR5 memory bus with 576.0 GB/s bandwidth. The Intel Xeon 6774P uses an eight-channel DDR5 bus with 409.6 GB/s bandwidth. Both support ECC memory. PCIe lanes also differ: the AMD part provides Gen 5 with 128 lanes (CPU only), while the Intel part provides Gen 5 with 136 lanes (CPU only). The Intel part has more PCIe lanes, which may matter for storage or accelerator density, but the AMD part has more memory channels and bandwidth.

Process node differs: TSMC 4 nm for AMD versus Intel 5 nm for the Xeon. The die size figures are not directly comparable, as AMD lists 12x 70.6 mm² chiplets and Intel lists 2x 598 mm². The AMD part has a transistor count of 99,780 million; the Intel part does not list a transistor count in the database.

Cache hierarchy differs in per-core allocation. The AMD EPYC 9655 has 80 KB L1 per core and 1 MB L2 per core, with 384 MB shared L3. The Intel Xeon 6774P has 112 KB L1 per core and 2 MB L2 per core, with 336 MB shared L3. The Intel part has more L1 and L2 per core, but the AMD part has more total L3.

Release dates differ: the AMD EPYC 9655 was released on 2024-10-09, and the Intel Xeon 6774P on 2025-05-21. The AMD part launched earlier. The Intel part has a lower launch MSRP of $6760 versus $11852 for the AMD part.

Architecture Differences

The AMD EPYC 9655 is based on the Zen 5 architecture, codenamed Turin, and belongs to the EPYC 9005 series. It uses a chiplet design with 12 dies, each 70.6 mm², fabricated on TSMC's 4 nm process. The Intel Xeon 6774P is based on the Granite Rapids architecture, part of the Xeon 6 family, with a 2x 598 mm² die configuration fabricated on Intel's 5 nm process.

The core designs differ fundamentally. Zen 5 is a new microarchitecture for AMD, and the database shows its single-thread PassMark score of 3,847. Granite Rapids is Intel's data-center core design; its single-thread PassMark score is 3,047. The 26.3% delta in single-thread reflects both architectural IPC and the higher boost clock of the AMD part (4.50 GHz versus 3.90 GHz).

Cache architecture reflects different design philosophies. Intel gives each core 112 KB of L1 and 2 MB of L2, which is larger than AMD's 80 KB L1 and 1 MB L2. However, AMD's shared L3 pool of 384 MB exceeds Intel's 336 MB. For workloads with large shared working sets, the AMD part has more aggregate cache. For per-core local data, the Intel part has more.

Memory architecture is a major differentiator. AMD uses twelve DDR5 channels, Intel uses eight. The resulting bandwidth figures are 576.0 GB/s versus 409.6 GB/s, a 40.6% difference that aligns with the multi-core performance gaps in memory-sensitive benchmarks like random string sorting and data compression.

PCIe lane count favors Intel: 136 lanes versus 128 lanes, both Gen 5. This is a narrow advantage but relevant for systems with many NVMe drives or GPUs. Both parts have no integrated graphics and are unlocked for multiplier adjustment, meaning they run at fixed clock ratios.

The production status for both is active. The AMD part has a part number of 100-000000674; the Intel part has a part number of SRWPC. Neither part is in the workstation or desktop segment; both are classified as server or workstation parts.