AMD EPYC 7663 vs Intel Xeon 6740P Comparison

Where Each One Wins

The benchmark split between the Intel Xeon 6740P and the AMD EPYC 7663 is decisively lopsided. Out of 16 head-to-head comparisons, the Intel part claims 13 victories, while the AMD processor takes only 3. That margin alone frames the competitive picture: the Xeon 6740P is the broader performer, but the EPYC 7663 holds specific niches where its architecture is clearly superior.

The Xeon 6740P dominates every Cinebench iteration, both single-core and multi-core. Its margins are remarkably consistent, hovering around 7.3% to 7.4% across R15, R20, and R23. This uniformity suggests a fundamental per-clock advantage rather than a workload-specific quirk. In PassMark's suite, the Intel chip also wins the majority of tests, including extended instructions (58.7% ahead), floating-point math (17.7% ahead), prime number finding (21.6% ahead), and physics simulation (21% ahead). The single-thread score of 2975 versus 2606 is a 14.2% advantage, reinforcing that the Xeon's core design is simply faster per thread.

The AMD EPYC 7663's three wins are concentrated in specific data-processing tasks. Data encryption is its largest victory, scoring 111725 against Intel's 82028 — a 26.6% edge. Integer math also goes to AMD, 468096 versus 388500, a 17% margin. Random string sorting is the third win, though modest at 5.1% (184367 vs 175015). These are not trivial workloads; encryption and integer-heavy operations are common in database and virtualization environments. But the pattern is clear: AMD wins where the data is dense and repetitive, while Intel wins almost everywhere else.

Architecture Differences

The two processors come from fundamentally different design philosophies. The Intel Xeon 6740P is built on Granite Rapids architecture, fabricated on Intel's 5 nm process, with a die size of 2x 598 mm². It packs 48 cores and 96 threads, running at a base clock of 2.10 GHz and a boost clock of 3.80 GHz. The cache hierarchy is generous: 112 KB of L1 per core, 2 MB of L2 per core, and a massive 288 MB of shared L3. Memory support is DDR5 over an eight-channel bus, yielding 409.6 GB/s of bandwidth. PCIe connectivity is Gen 5 with 88 lanes from the CPU. The TDP is 270 watts.

The AMD EPYC 7663 is a Zen 3 Milan part, built by TSMC on a 7 nm process. Its die is composed of 8x 81 mm² chiplets, totaling 33,200 million transistors. It offers 56 cores and 112 threads — more than the Intel part — but at lower clocks: 2000 MHz base and 3.50 GHz boost. Cache is smaller per core: 64 KB L1, 512 KB L2, and 256 MB shared L3. Memory support is DDR4 over an eight-channel bus, delivering 204.8 GB/s of bandwidth, exactly half of Intel's figure. PCIe is Gen 4 with 128 lanes. The TDP is 240 watts, 30 watts lower than the Intel chip.

The architectural divergence is stark. Intel uses a monolithic (or dual-die) approach with high per-core clocks and enormous L3 capacity, while AMD uses a chiplet design with more cores but lower clocks and half the memory bandwidth. The process node difference (5 nm vs 7 nm) partly explains Intel's clock advantage. The memory type difference (DDR5 vs DDR4) explains the 2x bandwidth gap, which likely contributes to Intel's wins in bandwidth-sensitive tests like extended instructions and floating-point math.

Notably, the AMD part has more cores (56 vs 48) and more threads (112 vs 96), yet loses most multi-threaded benchmarks. This is a direct consequence of clock speed and IPC differences. The Xeon's 3.80 GHz boost versus EPYC's 3.50 GHz boost, combined with a newer architecture, overcomes the 8-core deficit. The 288 MB L3 cache on Intel also dwarfs AMD's 256 MB, which helps in cache-resident workloads.

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 7663 has 56 cores and 112 threads, while the Intel Xeon 6740P has 48 cores and 96 threads. Despite this, the Intel part wins most multi-threaded benchmarks.

Q: What is the memory bandwidth difference?

A: The Intel Xeon 6740P supports DDR5 with 409.6 GB/s of bandwidth, exactly double the AMD EPYC 7663's DDR4 bandwidth of 204.8 GB/s. Both use eight-channel memory buses.

Q: How do they compare in single-threaded performance?

A: The Intel Xeon 6740P scores 2975 in PassMark single-thread, 14.2% higher than the AMD EPYC 7663's 2606. In Cinebench R23 single-core, Intel leads 4438 to 4137, a 7.3% margin.

Q: Which processor wins in data encryption?

A: The AMD EPYC 7663 wins decisively in PassMark data encryption, scoring 111725 versus Intel's 82028 — a 26.6% advantage. This is the largest single-test margin for AMD.

Q: What is the TDP difference?

A: The Intel Xeon 6740P has a TDP of 270 watts, while the AMD EPYC 7663 has a TDP of 240 watts. The AMD part consumes 30 watts less.

Q: Are both processors currently in production?

A: Yes, both the Intel Xeon 6740P and the AMD EPYC 7663 have an Active production status. The Intel part was released in 2025-02-23, while the AMD part was released in 2021-03-14.

Specification Differences

| Field | Intel Xeon 6740P | AMD EPYC 7663 |

|---|---|---|

| Cores | 48 | 56 |

| Threads | 96 | 112 |

| Base Clock | 2.10 GHz | 2000 MHz |

| Boost Clock | 3.80 GHz | 3.50 GHz |

| TDP | 270 W | 240 W |

| Socket | Intel Socket 4710 | AMD Socket SP3 |

| Architecture | Granite Rapids | Zen 3 |

| Codename | Granite Rapids | Milan |

| Generation | Xeon 6 (Granite Rapids-SP) | EPYC (Zen 3 (Milan)) |

| Process Node | 5 nm | 7 nm |

| Foundry | Intel | TSMC |

| Die Size | 2x 598 mm² | 8x 81 mm² |

| Transistors | Not listed | 33,200 million |

| L1 Cache | 112 KB (per core) | 64 KB (per core) |

| L2 Cache | 2 MB (per core) | 512 KB (per core) |

| L3 Cache | 288 MB (shared) | 256 MB (shared) |

| Memory Support | DDR5 | DDR4 |

| Memory Bandwidth | 409.6 GB/s | 204.8 GB/s |

| PCIe | Gen 5, 88 Lanes | Gen 4, 128 Lanes |

| Launch MSRP | $4650 | $6366 |

Head-to-Head Benchmarks

The largest single-test victory for the Intel Xeon 6740P comes in PassMark extended instructions, where it scores 116992 against the AMD EPYC 7663's 73719. That is a 58.7% margin, the most lopsided result in the entire comparison. This test measures vector and SIMD throughput, and the combination of DDR5 bandwidth and newer architecture clearly pays off.

The second-biggest Intel win is in PassMark find prime numbers, 822 versus 676, a 21.6% gap. This is an integer-heavy workload that stresses branch prediction and core efficiency. The physics test also goes to Intel by 21% (9705 vs 8020), suggesting better floating-point throughput. Floating-point math itself is 17.7% in Intel's favor (296102 vs 251535).

In Cinebench, the margins are tighter but consistent. R15 multi-core gives Intel 7544 versus 7032, a 7.3% win. R20 multi-core is 31437 versus 29304, also 7.3%. R23 multi-core is 74851 versus 69773, again 7.3%. Single-core tests show the same pattern: R15 at 7.4%, R20 at 7.3%, and R23 single-core at 9850 for AMD versus an implied higher score for Intel (the head-to-head only lists deltaPct for R23 multi-core, but the single-core scores are in the benchmark arrays: Intel 4438, AMD 4137).

PassMark multithread gives Intel an 88061 to 82087 win, a 7.3% margin that mirrors Cinebench. Data compression goes to Intel by 6.2% (1537129 vs 1447020). Single-thread PassMark is 14.2% in Intel's favor (2975 vs 2606).

The AMD EPYC 7663's biggest win is data encryption at 26.6% (111725 vs 82028). This is notable because encryption often relies on dedicated instructions and memory access patterns that favor AMD's chiplet layout. Integer math is AMD's second win at 17% (468096 vs 388500), which is surprising given Intel's core advantage, but suggests AMD's SMT implementation handles integer-heavy threads well. Random string sorting is AMD's third win at 5.1% (184367 vs 175015), a modest edge in a memory-latency-sensitive test.

The Verdict

The data paints a clear picture. The Intel Xeon 6740P wins 13 of 16 head-to-head benchmarks, including every Cinebench test and the majority of PassMark workloads. Its advantages are broad and consistent: 7.3% in multi-threaded rendering, 14.2% in single-thread throughput, and a massive 58.7% in extended instructions. For general-purpose server workloads, database queries, scientific computing, and any task that benefits from high memory bandwidth (409.6 GB/s versus 204.8 GB/s), the Xeon 6740P is the superior choice. Its 288 MB L3 cache and 3.80 GHz boost clock provide tangible performance across nearly every measured category.

The AMD EPYC 7663, despite having 8 more cores and 16 more threads, only wins in three specific tests: data encryption, integer math, and random string sorting. These are real workloads, and the 26.6% encryption advantage is substantial for security-focused applications. The 17% integer math win is also relevant for certain data-processing pipelines. However, these wins are narrow and do not compensate for the overall performance gap. The EPYC's lower TDP (240 vs 270 watts) is a point in its favor for power-constrained deployments, but the benchmark data does not suggest it closes the performance gap.

The choice is straightforward. If the workload involves encryption-heavy operations, integer-dense processing, or random string manipulation, the AMD EPYC 7663 has demonstrable strengths. For virtually everything else — rendering, floating-point math, physics, compression, general multi-threading, and single-threaded response — the Intel Xeon 6740P is the data-backed winner. The Intel part also carries a lower launch MSRP ($4650 versus $6366), though that should not be the primary decision factor given the performance differential. The Xeon 6740P is the more capable processor in the majority of scenarios, and the benchmark evidence supports that conclusion strongly.