AMD EPYC 9965 vs Intel Xeon 6960P Comparison

Head-to-Head Benchmarks

The benchmark data paints a lopsided picture, with the AMD EPYC 9965 taking 10 of the 14 head-to-head tests. The most dramatic margins come in compute-heavy workloads. In PassMark integer math, the EPYC 9965 scores 1,926,069 against the Xeon 6960P’s 727,750, a 164.7% advantage. Floating-point math is similarly one-sided: 1,153,453 versus 527,473, a 118.7% win for AMD. Data encryption shows a 115.1% delta (348,449 vs 162,013), and data compression is 103% higher (5,679,990 vs 2,797,724). Extended instruction throughput favors AMD by 98.2% (383,298 vs 193,404), while random string sorting lands at 70.3% (633,030 vs 371,795).

Cinebench results are consistent but narrower. Across R15, R20, and R23 multi-core tests, the EPYC 9965 wins each by 23.1%. The scores are 13,775 vs 11,194 in R15, 57,397 vs 46,645 in R20, and 136,661 vs 111,060 in R23. PassMark multithread shows a 22.9% edge for AMD (160,542 vs 130,659). These are substantial margins, but they are a fraction of the 100%+ deltas seen in the PassMark math tests.

The Intel Xeon 6960P does claim four wins, and they are worth scrutinizing. PassMark physics shows the Xeon ahead by 25% (24,937 vs 18,707). The find prime numbers test goes to Intel by 18.6% (1,484 vs 1,208). Single-thread performance is close, with Intel winning by 3.4% in both passmark_single_thread and passmark_singlethread (3,287 vs 3,176). These wins are real, but they are narrower than AMD’s largest victories and occur in a minority of tests.

The overall average benchmark score tells the same story: the EPYC 9965 averages 620,487, while the Xeon 6960P averages 365,194. That is a 70% gap in average score. The EPYC 9965 sits at the 100th percentile of all CPUs, as does the Xeon 6960P, but the raw averages show how far apart they are in raw throughput.

Architecture Differences

The two processors take fundamentally different design paths. The AMD EPYC 9965 uses Zen 5 architecture on a 3 nm process from TSMC, with the Turin codename. It packs 192 cores and 384 threads. The Intel Xeon 6960P uses Granite Rapids architecture on Intel’s 5 nm process, with a 3x 598 mm² die configuration. It has 72 cores and 144 threads. That is a 120-core, 240-thread advantage for AMD — a 2.67x core count difference.

Cache layouts reflect the different philosophies. The EPYC 9965 provides 80 KB of L1 per core, 1 MB of L2 per core, and 384 MB of shared L3. The Xeon 6960P has larger per-core caches: 112 KB L1 and 2 MB L2, plus a larger 432 MB L3. So Intel gives each core more private cache, but AMD offers more total cores and a slightly smaller but still massive L3 pool.

Memory and I/O also diverge. Both support DDR5 with twelve-channel memory buses, but the Xeon 6960P has higher memory bandwidth at 614.4 GB/s versus 576.0 GB/s for the EPYC 9965. Both support ECC memory. PCIe connectivity favors AMD: 128 Gen 5 lanes versus 96 Gen 5 lanes for Intel. Neither has integrated graphics. Both have a 500-watt TDP, which is notable given the core count difference — AMD delivers 120 more cores within the same power envelope.

The process node difference is significant: 3 nm for AMD versus 5 nm for Intel. The codenames are Turin (AMD) and Granite Rapids (Intel). The AMD part numbers as EPYC 9005 series, while Intel’s is Xeon 6 Granite Rapids-AP. Both are active production parts. The release dates are close, with Intel launching 2024-09-23 and AMD following on 2024-10-09.

The Verdict

The data is unambiguous for multi-threaded throughput. The AMD EPYC 9965 outperforms the Intel Xeon 6960P in nearly every aggregate workload, often by double-digit or triple-digit percentages. If your workload scales with core count, the EPYC 9965 is the stronger choice. Its 192 cores and 384 threads deliver 164.7% higher integer math, 118.7% higher floating-point math, and 103% higher data compression scores. The Cinebench multi-core results, all 23.1% in AMD’s favor, confirm this across three different render workloads.

The Intel Xeon 6960P is not without merit. It wins physics processing by 25% and prime number finding by 18.6%. It also edges out the EPYC 9965 in single-thread performance by 3.4%. If your application is heavily dependent on single-core speed or specific physics computations, the Xeon has a measurable advantage. However, those wins are isolated and do not offset the massive multi-core deficit.

Consider the context from the nearest rivals. The EPYC 9965’s average score of 620,487 is 18.5% ahead of the AMD EPYC 9845 and 22.7% ahead of the AMD EPYC 9755. The Xeon 6960P’s average of 365,194 is essentially tied with the AMD EPYC 9754 (0.2% delta) and slightly behind the AMD EPYC 9655 (-2.2%). In other words, the Xeon 6960P is competing with AMD’s previous-generation parts, while the EPYC 9965 sits clearly above all of them.

For a builder choosing between these two today, the EPYC 9965 is the data-backed pick for any workload that can use more than 72 cores. The Xeon 6960P makes sense only if you specifically need its physics or single-thread advantages, or if you require its higher memory bandwidth of 614.4 GB/s. The 500-watt TDP on both means power draw is not a differentiator, making the EPYC’s core advantage even more compelling.

Specification Differences

• Cores: AMD EPYC 9965 has 192; Intel Xeon 6960P has 72.
• Threads: AMD has 384; Intel has 144.
• Base clock: AMD 2.25 GHz; Intel 2.70 GHz.
• Boost clock: AMD 3.70 GHz; Intel 3.90 GHz.
• Process node: AMD 3 nm (TSMC); Intel 5 nm (Intel foundry).
• Die size: Intel uses 3x 598 mm²; AMD does not list a die size.
• L1 cache: AMD 80 KB per core; Intel 112 KB per core.
• L2 cache: AMD 1 MB per core; Intel 2 MB per core.
• L3 cache: AMD 384 MB shared; Intel 432 MB shared.
• Memory bandwidth: AMD 576.0 GB/s; Intel 614.4 GB/s.
• PCIe lanes: AMD 128 Gen 5 lanes; Intel 96 Gen 5 lanes.
• Socket: AMD Socket SP5; Intel Socket 7529.
• Codename: AMD Turin; Intel Granite Rapids.
• Release date: AMD 2024-10-09; Intel 2024-09-23.
• Launch MSRP: AMD $14813; Intel $9625.

FAQ

Q: Which CPU is faster in multi-core Cinebench tests?

A: The AMD EPYC 9965 wins all three Cinebench multi-core tests by exactly 23.1%. Scores are 13,775 vs 11,194 in R15, 57,397 vs 46,645 in R20, and 136,661 vs 111,060 in R23.

Q: Does the Intel Xeon 6960P win any significant benchmarks?

A: Yes, it wins four tests. PassMark physics shows a 25% advantage (24,937 vs 18,707), find prime numbers is 18.6% higher (1,484 vs 1,208), and single-thread tests are 3.4% higher (3,287 vs 3,176).

Q: How do the core counts compare?

A: The AMD EPYC 9965 has 192 cores and 384 threads, while the Intel Xeon 6960P has 72 cores and 144 threads. AMD has 120 more cores and 240 more threads.

Q: Which CPU has more cache?

A: The Intel Xeon 6960P has a larger L3 cache at 432 MB versus 384 MB for AMD. Intel also gives each core more L1 (112 KB vs 80 KB) and L2 (2 MB vs 1 MB) cache.

Q: What are the memory bandwidth differences?

A: The Intel Xeon 6960P has 614.4 GB/s memory bandwidth, which is higher than the AMD EPYC 9965’s 576.0 GB/s. Both use twelve-channel DDR5 with ECC support.

Q: How do the average benchmark scores compare to other CPUs?

A: The AMD EPYC 9965 averages 620,487, which is 18.5% ahead of the AMD EPYC 9845 and 22.7% ahead of the AMD EPYC 9755. The Intel Xeon 6960P averages 365,194, which is within 0.2% of the AMD EPYC 9754 and 2.2% behind the AMD EPYC 9655.

Where Each One Wins

The AMD EPYC 9965 is the clear winner for throughput-oriented workloads. In integer math, it more than doubles the Xeon’s score (1,926,069 vs 727,750). Floating-point math shows a similar story at 1,153,453 vs 527,473. Data compression and encryption both favor AMD by over 100%. Extended instruction workloads are 98.2% in AMD’s favor. For render workloads, Cinebench multi-core tests consistently give AMD a 23.1% edge. Random string sorting is 70.3% faster on AMD. The PassMark multithread score is 22.9% higher. If your server runs databases, scientific computing, virtualization, or any workload that can spread across 384 threads, the EPYC 9965 is the data-backed choice.

The Intel Xeon 6960P wins in a narrower set of scenarios. PassMark physics is its strongest win at 25% ahead, which suggests an edge in certain simulation or physics-based workloads. The find prime numbers test shows an 18.6% advantage, indicating stronger performance in specific algorithmic tasks. Single-thread performance is 3.4% higher, which could matter for lightly threaded applications or latency-sensitive tasks. The Xeon also offers more memory bandwidth (614.4 GB/s vs 576.0 GB/s) and a larger L3 cache (432 MB vs 384 MB), which could help in memory-bound workloads despite the lower core count.

The practical split is simple. If you need maximum parallel compute and have software that scales, the EPYC 9965 wins by margins that range from 22.9% to 164.7%. If you need the best possible single-thread performance, physics simulation, or prime number processing, the Xeon 6960P has the edge — but those wins are measured in single-digit or moderate percentages, not the massive deltas AMD achieves in multi-threaded tests. The Xeon’s higher memory bandwidth and per-core cache are the only other advantages, and they do not translate into benchmark wins outside its four tests.