AMD EPYC 9655 vs Intel Xeon 6781P Comparison

The AMD EPYC 9655 and Intel Xeon 6781P represent two distinct approaches to high-core-count server computing. The data shows a decisive overall victory for the AMD part, which wins 13 of 14 head-to-head benchmarks, but the Intel processor secures a notable single win that highlights a specific strength. Both processors are active, server-class parts with ECC memory support, but their architectural philosophies and resulting performance profiles differ substantially.

Head-to-Head Benchmarks

The AMD EPYC 9655 dominates the multi-threaded workload landscape. In the Cinebench series, the AMD chip scores 13,373 points in R15 multi-core, 55,722 in R20, and 132,672 in R23. The Intel Xeon 6781P trails significantly with 10,105, 42,106, and 100,254 points respectively. This translates to a consistent 32.3% advantage for AMD across all three Cinebench multi-core tests, demonstrating that the performance gap remains stable regardless of the rendering workload's specific demands.

The PassMark multithread test reinforces this pattern. The EPYC 9655 scores 156,110 compared to the Xeon's 117,946, a 32.4% difference. Integer math shows an even more pronounced separation, with AMD at 1,139,161 versus Intel's 584,834 — a staggering 94.8% lead. This near-doubling of integer throughput suggests the AMD architecture handles arithmetic-heavy server tasks with exceptional efficiency. Floating-point math also favors AMD, though by a smaller margin: 662,958 versus 507,406, a 30.7% difference.

Some of the largest deltas appear in specialized workloads. Data encryption sees AMD score 210,555 against Intel's 119,623, a 76% advantage. Random string sorting shows AMD at 439,682 versus 268,573, a 63.7% lead. Physics calculations favor AMD at 25,947 versus 17,753, a 46.2% gap. Even data compression, a common server task, shows AMD ahead at 3,271,896 versus 2,441,690, a 34% edge.

Single-threaded performance follows the same trend. The EPYC 9655 scores 3,847 in PassMark single-thread and 18,730 in Cinebench R23 single-core, while the Xeon 6781P scores 3,152 and does not have a listed R23 single-core result. The PassMark delta is 22% in AMD's favor. Extended instruction workloads are the closest contest, with AMD at 203,285 and Intel at 199,048 — a narrow 2.1% difference that nearly eliminates the gap.

The Intel Xeon 6781P claims exactly one victory: PassMark find prime numbers, where it scores 1,687 against AMD's 1,598, a 5.3% advantage. This isolated win in a prime-number-finding test suggests a specific algorithmic strength, but it stands alone against the broader trend of AMD superiority.

Architecture Differences

The two processors employ fundamentally different silicon designs. The AMD EPYC 9655 is built on the Zen 5 architecture, codenamed Turin, part of the EPYC 9005 series. It uses a 4 nm process from TSMC, packing 99,780 million transistors across 12 chiplets, each measuring 70.6 mm². The Intel Xeon 6781P uses the Granite Rapids architecture, part of the Xeon 6 line, on a 5 nm process from Intel, with a die size of 2x 598 mm². The process node advantage of 4 nm versus 5 nm likely contributes to AMD's efficiency and performance gains.

Core counts differ substantially. The EPYC 9655 provides 96 cores and 192 threads, while the Xeon 6781P offers 80 cores and 160 threads. This 16-core difference explains part of AMD's multi-core dominance. Clock speeds also favor AMD: the EPYC 9655 has a 2.60 GHz base clock and 4.50 GHz boost clock, versus the Xeon's 2.00 GHz base and 3.80 GHz boost. The higher clocks, combined with more cores, create a compound advantage for AMD in threaded workloads.

Cache hierarchies show different design philosophies. AMD allocates 80 KB of L1 and 1 MB of L2 per core, with a shared 384 MB L3 cache. Intel provides 112 KB of L1 and 2 MB of L2 per core, with a smaller 336 MB shared L3. While Intel's per-core L1 and L2 caches are larger, AMD's total L3 cache is 48 MB larger, which may benefit workloads with large shared data sets.

Memory subsystems also diverge. Both support DDR5 and ECC, but AMD uses a twelve-channel memory bus with 576.0 GB/s bandwidth, while Intel uses an eight-channel bus with 409.6 GB/s. The 166.4 GB/s bandwidth advantage for AMD is substantial and directly impacts memory-intensive server workloads. PCIe connectivity is comparable: AMD provides Gen 5 with 128 lanes, Intel offers Gen 5 with 136 lanes, giving Intel a slight edge in raw lane count.

Power envelopes differ, with AMD rated at 400 W TDP and Intel at 350 W TDP. The AMD part draws more power but delivers significantly higher performance, suggesting a favorable performance-per-watt trade-off in the benchmark data. The launch MSRP for the AMD EPYC 9655 is $11852, while the Intel Xeon 6781P is $8960.

The Verdict

The benchmark data presents a clear hierarchy. The AMD EPYC 9655 holds a 100th percentile ranking among all CPUs, while the Intel Xeon 6781P sits at the 99th percentile. In head-to-head competition, AMD wins 13 of 14 tests, with an average benchmark score of 373,479 compared to Intel's 315,524. The AMD part's nearest rivals include the AMD EPYC 9535, which trails by 1.6%, and the Intel Xeon 6960P, which AMD leads by 2.3%. The Xeon 6781P's nearest rivals are all AMD parts or AMD-based Threadripper processors, with the AMD EPYC 9575F trailing by 1.2% and the AMD Ryzen Threadripper PRO 9985WX leading by 1.6%.

For organizations prioritizing raw multi-core throughput, integer math, encryption, or memory bandwidth, the EPYC 9655 is the data-supported choice. Its 32.3% lead in Cinebench multi-core and 94.8% lead in integer math are decisive. The Xeon 6781P offers no multi-core advantages and only a narrow 5.3% win in prime number finding. The AMD processor's higher TDP of 400 W versus 350 W is the only major drawback, but the performance gains justify the additional power envelope.

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 9655 has 96 cores and 192 threads. The Intel Xeon 6781P has 80 cores and 160 threads.

Q: What is the largest performance gap between the two processors?

A: The largest gap is in PassMark integer math, where the AMD EPYC 9655 scores 1,139,161 against the Intel Xeon 6781P's 584,834, a 94.8% difference.

Q: Does the Intel Xeon 6781P win any benchmarks?

A: Yes, it wins the PassMark find prime numbers test with a score of 1,687 compared to AMD's 1,598, a 5.3% advantage.

Q: What are the memory bandwidth specifications for each processor?

A: The AMD EPYC 9655 has a twelve-channel memory bus with 576.0 GB/s bandwidth. The Intel Xeon 6781P has an eight-channel bus with 409.6 GB/s bandwidth.

Q: How do the processors compare in single-threaded performance?

A: The AMD EPYC 9655 scores 3,847 in PassMark single-thread, while the Intel Xeon 6781P scores 3,152, giving AMD a 22% lead.

Q: What is the TDP difference between the two?

A: The AMD EPYC 9655 has a TDP of 400 W, while the Intel Xeon 6781P has a TDP of 350 W.

Where Each One Wins

The AMD EPYC 9655 is the clear winner for virtually all server workloads. Its 32.3% lead in Cinebench R15, R20, and R23 multi-core tests makes it the superior choice for rendering, simulation, and other compute-heavy tasks. The 94.8% advantage in integer math positions it strongly for database operations, financial modeling, and general server-side computation. Data encryption performance, with a 76% lead, makes it ideal for security-focused applications, VPN gateways, and encrypted storage systems. The 63.7% edge in random string sorting benefits log processing, data indexing, and text-heavy workloads. Physics calculations show a 46.2% advantage, supporting scientific computing and physics simulation. The 34% lead in data compression and 30.7% lead in floating-point math further cement its position for data centers handling large datasets. The 22% single-thread advantage ensures responsive performance even in lightly threaded applications.

The Intel Xeon 6781P's sole victory in prime number finding, with a 5.3% edge, indicates a narrow niche in number-theoretic computations and certain cryptographic algorithms. Its 136 PCIe Gen 5 lanes versus AMD's 128 lanes might appeal to configurations requiring maximum I/O expansion, though this advantage does not appear in the benchmark results. The lower 350 W TDP could benefit power-constrained environments, though the data shows this comes at a significant performance cost. For organizations running prime-number-heavy research or specialized number theory workloads, the Xeon 6781P offers a targeted advantage; for all other applications, the benchmark data overwhelmingly supports the AMD EPYC 9655.