AMD EPYC 7663 vs Intel Xeon 6732P Comparison

The AMD EPYC 7663 and Intel Xeon 6732P are both server-grade processors that land in the same performance tier overall, but the recorded data shows they get there in very different ways. The EPYC 7663 wins 14 of 17 head-to-head benchmarks, including dominant results in integer math and encryption, while the Xeon 6732P takes three wins that point to distinct strengths in vector-heavy and latency-sensitive workloads. Both sit in the 98th percentile versus all CPUs in the database, yet the older Zen 3 part consistently outscores the newer Granite Rapids chip in this matchup.

Head-to-Head Benchmarks

The EPYC 7663 sweeps every Cinebench test, and by nearly identical margins. In Cinebench R23 multi-core it scores 69773 against 63621, a 9.7 percent lead, and the same 9.7 percent gap appears in R20 multi-core (29304 vs 26720), R15 multi-core (7032 vs 6412), R23 single-core (9850 vs 8981), R20 single-core (4137 vs 3772), and R15 single-core (992 vs 905). That uniformity is telling: rendering throughput scales with the EPYC's 56 cores, and its 2000 MHz base with 3.50 GHz boost still delivers better per-thread results than the Xeon's 3.80 GHz base and 4.10 GHz boost in these tests. The single-core sweep suggests Milan's architecture is genuinely more efficient per clock in Cinebench's workload mix.

The biggest win anywhere in the dataset is encryption. The EPYC 7663 posts 111725 in PassMark data encryption versus 63848 for the Xeon 6732P, a 75 percent blowout. Integer math tells a similar story: 468096 vs 334340, a 40 percent gap. Random string sorting goes to AMD 184367 vs 133467, 38.1 percent ahead. These are the numbers that matter for database engines, compression pipelines, and cryptographic workloads.

The Xeon's wins are narrower but real. Extended instructions (SIMD) is its strongest result: 106697 vs 73719, a 30.9 percent lead for Intel. Floating point math also favors the Xeon, 261703 vs 251535, a 3.9 percent edge. PassMark physics is essentially a tie with a slight Intel advantage, 8109 vs 8020, 1.1 percent apart. Rounding out the AMD column, multi-thread goes to the EPYC 82087 vs 74849 (9.7 percent), data compression 1447020 vs 1339480 (8 percent), find prime numbers 676 vs 628 (7.6 percent), and single-thread 2606 vs 2506 (4 percent).

FAQ

Q: Which CPU is faster overall?

A: The EPYC 7663 wins 14 of 17 recorded benchmarks and holds a higher average benchmark score, 161973 versus 143444. Typical margins cluster around 9.7 percent, with much larger gaps in integer math (40 percent) and encryption (75 percent).

Q: Does the Xeon 6732P win anything?

A: Yes, three tests. It leads in PassMark extended instructions by 30.9 percent (106697 vs 73719), floating point math by 3.9 percent (261703 vs 251535), and physics by 1.1 percent (8109 vs 8020).

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

A: The EPYC 7663 wins every single-core test: Cinebench R23 single-core 9850 vs 8981 (9.7 percent), and PassMark single-thread 2606 vs 2506 (4 percent). This holds despite the Xeon's higher listed clock speeds.

Q: What memory does each platform support?

A: The EPYC 7663 uses DDR4 on an eight-channel bus with 204.8 GB/s of bandwidth. The Xeon 6732P uses DDR5, also eight-channel, with 409.6 GB/s of bandwidth, exactly double the EPYC's figure.

Q: How do their rivals compare in the database?

A: The EPYC 7663's average score sits 0.3 percent below the AMD EPYC 9375F (162497) and 1 percent above the AMD EPYC 9355P (160358). The Xeon 6732P is within 0.2 percent of the AMD Ryzen 9 PRO 9965X3D (143735) and the Intel Xeon 674X (143103).

Q: Do both chips support ECC memory?

A: Yes, both are validated for ECC, consistent with their server and workstation market segment.

The Verdict

The data points one direction for compute density. If the workload is rendering, encryption, integer-heavy processing, sorting, or general multi-threaded throughput, the EPYC 7663 is the stronger pick: it wins 14 of 17 tests, often by double digits, and posts the higher average score despite being the older design. Its 56 cores and 112 threads convert directly into the 9.7 percent Cinebench and PassMark multi-thread leads over the Xeon's 32 cores and 64 threads.

The Xeon 6732P earns its place on a different axis. Its 30.9 percent extended instructions win makes it the clear choice for SIMD-dependent workloads, and its 409.6 GB/s of DDR5 bandwidth doubles the EPYC's 204.8 GB/s, which matters for memory-bound applications regardless of raw compute scores. Platform I/O also favors Intel: 136 lanes of PCIe Gen 5 versus 128 lanes of Gen 4. Workloads that stream large datasets or need the newest platform features have a reason to pick the Xeon even with the benchmark deficit. Note the power envelope differs too: 350 W TDP for the Xeon against 240 W for the EPYC, so the AMD chip delivers more performance per watt of rated TDP in the recorded tests.

Specification Differences

The two chips diverge on nearly every spec line. Core counts: 56 cores and 112 threads for the EPYC 7663 against 32 cores and 64 threads for the Xeon 6732P. Clocks run the other way: the Xeon bases at 3.80 GHz boosting to 4.10 GHz, while the EPYC bases at 2000 MHz boosting to 3.50 GHz. TDP is 240 W for AMD, 350 W for Intel.

Platform details differ completely. The EPYC uses AMD Socket SP3; the Xeon uses Intel Socket 4710. Memory splits along generational lines, DDR4 with 204.8 GB/s bandwidth for AMD versus DDR5 with 409.6 GB/s for Intel, both eight-channel with ECC. PCIe is Gen 4 with 128 CPU lanes on the EPYC versus Gen 5 with 136 CPU lanes on the Xeon.

Cache hierarchy favors Intel at the per-core level: 112 KB L1 and 2 MB L2 per core, versus 64 KB L1 and 512 KB L2 per core on the EPYC. AMD counters with a larger shared L3, 256 MB against 144 MB. Release timing spans the generations: the EPYC 7663 launched with a $6366 launch MSRP, the Xeon 6732P with a $5295 launch MSRP. Neither has an unlocked multiplier, and both are active production parts aimed at servers and workstations.

Architecture Differences

The EPYC 7663 is a Zen 3 design, codenamed Milan, from the EPYC 7003 series. It is built on TSMC's 7 nm process and packs 33,200 million transistors across an 8x 81 mm² chip layout. That eight-die construction is how AMD fits 56 cores and the 256 MB of shared L3, and it underpins the chip's multi-threaded dominance in the benchmark data.

The Xeon 6732P is Granite Rapids, specifically Granite Rapids-SP within the Xeon 6 generation, fabricated on Intel's own 5 nm node. The database does not record transistor counts or die sizes for this part. The generational gap shows up most clearly in the platform rather than the compute scores: DDR5 support, Gen 5 PCIe, and doubled memory bandwidth are the tangible advantages of the newer architecture. The EPYC's counterargument is raw execution: a 75 percent encryption lead and a 40 percent integer math lead show the older design still outperforming in integer and cryptographic throughput. Per-core cache is where Intel's newer architecture pays off, 2 MB of L2 per core feeding its higher clocks and helping deliver the extended instructions win.

Where Each One Wins

Pick the EPYC 7663 for throughput-bound server workloads. The recorded data shows it ahead in rendering (all six Cinebench tests, 9.6 to 9.7 percent), encryption (75 percent), integer math (40 percent), string sorting (38.1 percent), data compression (8 percent), prime number computation (7.6 percent), overall multi-thread throughput (9.7 percent), and even single-threaded tests (4 percent). Any consolidation, virtualization, or batch processing stack that leans on integer and cryptographic throughput will favor the AMD part, and it does all this at a lower 240 W TDP rating.

Pick the Xeon 6732P where the platform and vector units do the work. Its 30.9 percent extended instructions lead suits SIMD-heavy code paths, its 3.9 percent floating point edge helps scientific and numeric workloads, and its physics result confirms competitive latency-sensitive behavior. The bigger draw is infrastructure: 409.6 GB/s of DDR5 bandwidth and 136 Gen 5 lanes make it the right fit for builds that need maximum memory throughput and the latest I/O, accepting a roughly 10 percent multi-threaded deficit against the EPYC in exchange.