AMD EPYC 9654 vs Intel Core Ultra 5 135H Comparison

The AMD EPYC 9654 and Intel Core Ultra 5 135H occupy opposite ends of the computing spectrum, and the benchmark data reflects this chasm. The EPYC 9654 is a 96-core, 192-thread server behemoth with a 360W TDP, while the Core Ultra 5 135H is a 14-core, 18-thread mobile processor with a 28W TDP. Across all six shared Cinebench tests, the EPYC 9654 wins outright, with deltas ranging from 445.1% to 756.2%. However, both processors share an identical 81st percentile ranking among all CPUs, and their average benchmark scores are remarkably close—29409 for the EPYC versus 29093 for the Ultra 5. This contradiction—massive per-test wins yet similar overall averages—suggests the Ultra 5's additional PassMark tests (not shared with the EPYC) pull its average up, while the EPYC's sheer multi-core might is concentrated in Cinebench workloads. The data implies a clear divide: the EPYC 9654 is for throughput-heavy server tasks, while the Core Ultra 5 135H, with its integrated Arc Xe-LPG graphics and mobile market segment, targets efficiency and portability.

The Verdict

The data dictates a simple choice based on workload and platform. For any multi-core rendering, scientific computing, or virtualization scenario where the Cinebench R23 multicore score of 101675 versus 11875.5 matters, the AMD EPYC 9654 is the only logical pick—it leads by 756.2% in that test. Conversely, the Intel Core Ultra 5 135H is the sensible option for mobile or space-constrained systems where its 28W TDP and integrated graphics are essential, despite being 744.4% behind in Cinebench R23 single-core. The EPYC 9654’s 6-0 sweep in head-to-head benchmarks leaves no ambiguity in raw compute, but the Ultra 5’s existence is justified by its market segment and efficiency profile. Neither processor is a compromise; they are purpose-built tools. The EPYC 9654, with its launch MSRP of $11805, is a server-workstation investment, while the Core Ultra 5 135H, at a $342 launch MSRP, targets premium mobile computing. If the question is raw performance per benchmark, the EPYC wins every time; if the question is portability and power envelope, the Intel part has no rival in this comparison.

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 9654 has 96 cores and 192 threads, while the Intel Core Ultra 5 135H has 14 cores and 18 threads. This core disparity drives the EPYC’s massive multi-core benchmark advantage.

Q: How big is the multi-core performance gap in Cinebench R23?

A: The EPYC 9654 scores 101675 in Cinebench R23 multi-core, which is 756.2% higher than the Core Ultra 5 135H’s 11875.5. This is the largest delta across all shared tests.

Q: Is the Intel Core Ultra 5 135H competitive in single-core performance?

A: No. The EPYC 9654 leads in Cinebench R23 single-core with 14354 versus 1700, a 744.4% difference. The Intel part’s higher boost clock of 4.60 GHz does not overcome the EPYC’s architecture advantage in this metric.

Q: Do both processors have integrated graphics?

A: No. The Intel Core Ultra 5 135H includes Arc Xe-LPG 96EU integrated graphics, while the AMD EPYC 9654 has no integrated graphics. This makes the Intel part suitable for systems without a discrete GPU.

Q: What memory channels do they support?

A: The EPYC 9654 uses twelve-channel memory with 460.8 GB/s bandwidth, while the Core Ultra 5 135H uses dual-channel memory with 89.6 GB/s bandwidth. The EPYC’s memory bandwidth is over five times higher.

Q: How do their average benchmark scores compare?

A: The EPYC 9654 has an average score of 29409, and the Core Ultra 5 135H averages 29093—a difference of just 1.1%. This near-parity is misleading because the EPYC lacks PassMark scores, while the Intel part has 11 additional PassMark results.

Architecture Differences

The AMD EPYC 9654 is built on the Zen 4 architecture, codenamed Genoa, using a 5nm process from TSMC. It packs 78,840 million transistors across a die size of 12x 72 mm². In contrast, the Intel Core Ultra 5 135H uses the Meteor Lake architecture on a 7nm Intel process, with no transistor or die size data provided. The EPYC’s cache hierarchy is radically different: 64 KB L1 per core, 1 MB L2 per core, and 384 MB shared L3. The Intel part offers 112 KB L1 per core, 2 MB L2 per core, and only 18 MB shared L3. This 366 MB L3 difference is critical for data-heavy server workloads. The EPYC supports ECC memory, while the Intel part does not. PCIe lane counts also diverge sharply: the EPYC provides 128 Gen 5 lanes, versus 8 Gen 5 lanes for the Ultra 5. These architectural choices reflect the EPYC’s server pedigree versus the Ultra 5’s mobile focus, with the latter integrating graphics and the former prioritizing memory bandwidth and expansion.

Specification Differences

The two processors differ in nearly every specification field. The EPYC 9654 has 96 cores and 192 threads; the Ultra 5 135H has 14 cores and 18 threads. Base clocks are 2.40 GHz for the EPYC and 3.60 GHz for the Intel part, while boost clocks are 3.70 GHz and 4.60 GHz, respectively. TDP is the starkest contrast: 360W for the EPYC versus 28W for the Ultra 5. The EPYC uses AMD Socket SP5, while the Intel part uses Intel BGA 2049. Process nodes differ (5nm TSMC versus 7nm Intel). Cache totals are 384 MB L3 shared for the EPYC versus 18 MB L3 shared for the Intel part. Memory support is DDR5 for both, but the EPYC uses twelve-channel memory with 460.8 GB/s bandwidth, whereas the Ultra 5 uses dual-channel with 89.6 GB/s. ECC memory is supported only on the EPYC. PCIe lanes are 128 Gen 5 for the EPYC versus 8 Gen 5 for the Ultra 5. Integrated graphics exist only on the Intel part (Arc Xe-LPG 96EU). Market segments are Server/Workstation versus Mobile. Release dates are 2022-11-09 for the EPYC and 2023-12-13 for the Ultra 5. Launch MSRP is $11805 for the EPYC and $342 for the Ultra 5. Neither processor has an unlocked multiplier. The EPYC’s part number is 100-100000789, while the Ultra 5’s is SRMZ0.

Head-to-Head Benchmarks

The six Cinebench tests show a complete dominance by the AMD EPYC 9654. In Cinebench R15 multi-core, the EPYC scores 10248 versus 1718, a 496.5% win. Single-core R15 shows 1446 versus 244, a 492.6% delta. Cinebench R20 multi-core yields 42703 for the EPYC and 7834 for the Intel part, a 445.1% difference. R20 single-core is 6028 versus 1105, a 445.5% gap. The largest margins appear in Cinebench R23: multi-core at 101675 versus 11875.5 is a 756.2% blowout, and single-core at 14354 versus 1700 is a 744.4% difference. The EPYC wins all six tests, with zero wins for the Core Ultra 5 135H. The data shows that regardless of test generation (R15, R20, R23) or thread count (single or multi), the EPYC’s Zen 4 architecture and 96 cores provide an insurmountable advantage. The Intel part’s higher boost clock of 4.60 GHz does not translate into any single-core win, suggesting the EPYC’s per-core efficiency at 3.70 GHz is superior in these specific workloads.

Where Each One Wins

The AMD EPYC 9654 wins every shared benchmark, making it the definitive choice for multi-threaded server workloads. Its Cinebench R23 multi-core score of 101675 indicates exceptional performance in rendering, simulation, and compilation tasks that scale across 192 threads. The 460.8 GB/s memory bandwidth and 384 MB L3 cache further support data-intensive applications. The EPYC’s 128 PCIe Gen 5 lanes enable massive expansion for GPUs and accelerators. This processor is built for a server rack, not a laptop.

The Intel Core Ultra 5 135H wins no shared benchmarks, but its data suggests wins outside the Cinebench suite. Its 28W TDP makes it viable for thin-and-light laptops where power efficiency is paramount. The integrated Arc Xe-LPG 96EU graphics provide a display output without needing a discrete GPU. The 89.6 GB/s memory bandwidth, while lower, is sufficient for mobile tasks. The higher base clock of 3.60 GHz and boost of 4.60 GHz could benefit bursty, short-duration workloads, though the Cinebench data does not confirm this. The Ultra 5’s PassMark scores—such as 251651 in data compression and 73509 in integer math—offer a broader performance profile, though these are not directly comparable to the EPYC. In this head-to-head, the Intel part’s wins are in platform suitability, not raw compute. The data shows the EPYC for throughput, the Ultra 5 for mobility.