AMD EPYC 7313P vs Intel Xeon Phi 7290 Comparison

The data presents a striking contrast between two very different server processors: the Intel Xeon Phi 7290, a 72-core Knights Landing part from 2016, and the AMD EPYC 7313P, a 16-core Zen 3 Milan chip from 2021. While their average benchmark scores are nearly identical—the Xeon Phi scores 53,469 versus the EPYC’s 53,206, a negligible 0.5% difference—the distribution of wins tells a far more interesting story. The EPYC 7313P dominates 14 of the 17 head-to-head tests, often by massive margins, while the Xeon Phi secures only three victories, all in specific data-processing workloads. This is not a close contest in most respects; it is a generational mismatch where architectural efficiency trumps raw core count.

Head-to-Head Benchmarks

The most lopsided results come from single-threaded performance, where the EPYC 7313P utterly overwhelms the Xeon Phi. In Cinebench R15 single-core, the EPYC scores 497 against the Xeon Phi’s 215, a 56.7% advantage. The gap widens in Passmark single-thread tests, where the EPYC’s 2,634 score is 81.6% higher than the Xeon Phi’s 485. This pattern repeats across all Cinebench versions: R20 single-core shows 2,072 versus 898 (56.7% delta), and R23 single-core shows 4,934 versus 2,140 (56.6% delta). The Xeon Phi’s 1.50 GHz base and 1.70 GHz boost clocks simply cannot compete with the EPYC’s 3.00 GHz base and 3.70 GHz boost, especially given the Zen 3 architecture’s superior instructions-per-clock.

Multi-core performance tells a similar story, despite the Xeon Phi’s massive 72-core/288-thread configuration versus the EPYC’s 16 cores and 32 threads. In Cinebench R23 multi-core, the EPYC scores 34,952 to the Xeon Phi’s 15,163, a 56.6% margin. The same delta appears in R15 (3,522 vs 1,528) and R20 (14,679 vs 6,368). Passmark’s multithread test shows the EPYC at 41,121 versus 17,839, again a 56.6% difference. The Xeon Phi’s 288 threads generate enormous parallelism, but each thread is so slow that total throughput falls far short. The EPYC’s 32 threads, running at more than double the clock speed with far better IPC, deliver more than twice the aggregate performance.

The Xeon Phi does manage wins in three specific Passmark subtests. Its data compression score of 564,535 beats the EPYC’s 528,167 by 6.9%. Extended instructions show a more decisive 26.6% victory, with 41,517 against 32,784. Random string sorting also goes to the Xeon Phi, 68,593 versus 62,596, a 9.6% edge. These are narrow or moderate wins in highly specialized workloads, suggesting the Knights Landing’s wide vector units and many-core layout have niches where they still excel.

Other Passmark tests heavily favor the EPYC. Data encryption shows a 65% gap, with 35,727 versus 12,505. Floating-point math goes to the EPYC at 82,260 against 47,417, a 42.4% margin. Prime number finding is even more one-sided: 346 versus 114, a 67.1% delta. Integer math favors the EPYC by 12.8% (145,558 vs 126,922), and physics simulation shows a 46.6% gap (4,229 vs 2,257). The overall average benchmark scores are within 0.5% of each other, but this masks the reality that the EPYC wins most tests decisively while the Xeon Phi’s wins are modest by comparison.

The Verdict

The AMD EPYC 7313P is the clear winner for virtually every general-purpose server workload. It wins 14 of 17 head-to-head tests, often by margins exceeding 50%. The Xeon Phi 7290’s only advantages are in data compression, extended instructions, and random string sorting, where its 72 cores and 288 threads can be fully utilized without being bottlenecked by per-thread performance. For any task that involves traditional computing, database operations, encryption, or physics simulation, the EPYC is vastly superior, frequently doubling or tripling the Xeon Phi’s output.

The data strongly indicates that the EPYC 7313P is the more capable processor for modern data centers. Its 16-core design with 32 threads, running at 3.00 GHz base and 3.70 GHz boost, delivers better performance per core and per thread than the Xeon Phi’s 72 cores at 1.50/1.70 GHz. The EPYC also has a significant architectural advantage: it uses a 7 nm process from TSMC with 16,600 million transistors, while the Xeon Phi uses Intel’s 14 nm node with 8,000 million transistors. The EPYC’s 128 MB of shared L3 cache, versus the Xeon Phi’s per-core L2 cache only, further explains its dominance in most benchmarks.

The Xeon Phi’s 94th percentile ranking among all CPUs is matched by the EPYC’s 94th percentile, but this is misleading. The average benchmark score of 53,469 for the Xeon Phi versus 53,206 for the EPYC is within the noise, but the distribution of performance is entirely different. The Xeon Phi achieves its average through sheer core count, while the EPYC achieves it through efficiency. In practice, the EPYC will feel faster in almost every real-world scenario.

Where Each One Wins

The AMD EPYC 7313P wins in all Cinebench tests, both single-core and multi-core, across R15, R20, and R23 versions. This makes it the obvious choice for rendering, 3D modeling, and any CPU-bound creative workload. Its Passmark wins include data encryption, floating-point math, integer math, multithread, physics, prime number finding, and single-thread tests. This covers most general computing, scientific simulation, financial modeling, and high-frequency trading applications. The EPYC’s 32 threads and 3.70 GHz boost clock make it ideal for workloads that require both high clock speeds and reasonable parallelism.

The Intel Xeon Phi 7290 wins only in data compression, extended instructions, and random string sorting. These are narrow victories, with margins of 6.9%, 26.6%, and 9.6% respectively. The extended instructions win suggests that the Phi’s Knights Landing architecture, with its wide vector units, excels at SIMD-heavy code such as certain scientific calculations or media processing. The data compression and string sorting wins indicate that its 288 threads can saturate memory bandwidth when processing large datasets that are not latency-sensitive. However, these are niche applications, and even here the EPYC is competitive, losing by single digits in two of the three tests.

For a mixed workload server, the EPYC 7313P is the safer choice. It wins in the majority of tests and loses only in areas where the Xeon Phi’s unique many-core design can be fully exploited. The Xeon Phi is a specialist tool, best suited for highly parallel, vectorized workloads that can keep all 288 threads busy. It is not a general-purpose processor, and its performance in single-threaded and latency-sensitive tasks is poor.

FAQ

Q: Which processor has more cores?

A: The Intel Xeon Phi 7290 has 72 cores and 288 threads, while the AMD EPYC 7313P has 16 cores and 32 threads.

Q: Which processor is faster in single-threaded performance?

A: The AMD EPYC 7313P is significantly faster, scoring 2,634 in Passmark single-thread versus the Xeon Phi’s 485, an 81.6% advantage. Cinebench R23 single-core shows a 56.6% gap in favor of the EPYC (4,934 vs 2,140).

Q: What is the performance difference in multi-core workloads?

A: The EPYC 7313P wins every multi-core Cinebench test by 56.6%, and Passmark multithread by the same margin. For example, Cinebench R23 multi-core scores 34,952 for the EPYC versus 15,163 for the Xeon Phi.

Q: In which tests does the Xeon Phi 7290 beat the EPYC 7313P?

A: The Xeon Phi wins three tests: Passmark data compression (564,535 vs 528,167, a 6.9% edge), extended instructions (41,517 vs 32,784, a 26.6% edge), and random string sorting (68,593 vs 62,596, a 9.6% edge).

Q: How do their average benchmark scores compare?

A: The Xeon Phi 7290 has an average benchmark score of 53,469, while the EPYC 7313P scores 53,206. The Xeon Phi is 0.5% ahead, and both rank in the 94th percentile of all CPUs.

Q: What are the clock speeds of each processor?

A: The Xeon Phi 7290 has a 1.50 GHz base clock and a 1.70 GHz boost clock. The EPYC 7313P has a 3.00 GHz base clock and a 3.70 GHz boost clock.

Architecture Differences

The Intel Xeon Phi 7290 uses the Knights Landing architecture, built on Intel’s 14 nm process with 8,000 million transistors. It has 72 cores and 288 threads, with 32 KB of L1 cache and 512 KB of L2 cache per core, but no L3 cache. The EPYC 7313P uses the Zen 3 architecture (codename Milan), built on TSMC’s 7 nm process with 16,600 million transistors across four 81 mm² dies. It has 16 cores and 32 threads, with 64 KB of L1 cache, 512 KB of L2 cache per core, and a substantial 128 MB of shared L3 cache. The Xeon Phi’s lack of L3 cache is a major disadvantage, as the EPYC’s 128 MB shared cache can hold far more working data, reducing memory latency.

The socket and platform differences are significant. The Xeon Phi uses Intel Socket 3647, while the EPYC uses AMD Socket SP3. The EPYC supports eight-channel memory with a bandwidth of 204.8 GB/s, whereas the Xeon Phi supports DDR4 memory but without a specified bus width or bandwidth in the data. The EPYC also offers PCIe Gen 4 with 128 lanes (CPU only), while the Xeon Phi’s PCIe support is not specified. Both support ECC memory, and neither has integrated graphics. The EPYC has a TDP of 155 watts, lower than the Xeon Phi’s 245 watts, making it more power-efficient per unit of performance.

The release dates are far apart: the Xeon Phi launched on 2016-06-19, while the EPYC launched on 2021-03-14. This five-year gap explains the architectural differences. The EPYC is built on a newer, denser process, has a more modern core design, and benefits from years of microarchitectural improvements. The Xeon Phi’s Knights Landing was a specialized many-core architecture designed for high-performance computing, but it sacrificed single-thread performance and cache capacity to achieve its high core count. The EPYC’s Zen 3 delivers a balanced design that excels across all benchmark categories, making it the more versatile and practical choice for server deployments. The Xeon Phi’s production status is not listed, while the EPYC is marked as active, further indicating the Phi’s obsolescence.