Intel Core Ultra 5 235HX vs Intel Core Ultra 9 386H Comparison

FAQ

Q: Which processor has the higher overall average benchmark score?

A: The Intel Core Ultra 5 235HX records an average benchmark score of 52073, while the Intel Core Ultra 9 386H scores 43210. The Ultra 5 sits at the 91st percentile among all CPUs, compared to the 88th percentile for the Ultra 9.

Q: How does the single-core performance compare between the two?

A: The Ultra 5 235HX wins the Cinebench R23 single-core test by 136.7%, scoring 4903 versus 2071.5. In Passmark single-thread, the Ultra 5 leads by 11%, with scores of 4683 versus 4218.

Q: Does the Core Ultra 9 386H win any benchmark at all?

A: Yes, the Ultra 9 386H wins the Passmark physics test with a score of 3028, beating the Ultra 5's 2550 by 15.8%. That is the only head-to-head win out of 17 recorded tests.

Q: What are the core and thread counts for each processor?

A: The Core Ultra 5 235HX has 14 cores and 14 threads. The Core Ultra 9 386H has 16 cores and 16 threads. Neither processor uses simultaneous multithreading.

Q: Which processor uses which manufacturing process and foundry?

A: Both are built on a 3 nm process. The Ultra 5 235HX (Arrow Lake-HX) is fabricated by TSMC, while the Ultra 9 386H (Panther Lake-H) is fabricated by Intel.

Q: How do the two compare in memory bandwidth?

A: The Ultra 5 235HX supports DDR5 with dual-channel memory and a bandwidth of 102.4 GB/s. The Ultra 9 386H supports DDR5 and LPDDR5X with dual-channel memory and a higher bandwidth of 115.2 GB/s.

Where Each One Wins

The recorded data shows a dominant split: the Core Ultra 5 235HX wins 16 of 17 head-to-head benchmarks. Its biggest advantages appear in heavily threaded workloads. In Cinebench R23 multicore, it scores 34731 versus 20547, a 69% lead. In Cinebench R20 multicore, the gap is 13.8% (14587 versus 12820). The Ultra 5 also leads in Passmark multithread by 15.4% (40849 versus 35399) and in integer math by 11.3% (97177 versus 87284).

The Ultra 5's single-core advantage is even more pronounced in certain tests. The Cinebench R23 single-core result shows a 136.7% delta, with the Ultra 5 at 4903 versus 2071.5. Cinebench R15 single-core shows a 62.8% lead (494 versus 303.5). Passmark single-thread is closer, at 11% (4683 versus 4218).

The Core Ultra 9 386H wins only one test: Passmark physics. Its score of 3028 beats the Ultra 5's 2550 by 15.8%. This suggests the Ultra 9 has a specific strength in physics simulation workloads, but it does not translate into wins elsewhere. Across data compression, encryption, extended instructions, floating point math, random string sorting, and prime number finding, the Ultra 5 leads by margins ranging from 5.9% to 21%.

For practical use, the Ultra 5 235HX is the choice for rendering, compiling, and general productivity where multicore and single-core speed both matter. The Ultra 9 386H has a narrow niche in physics-based calculations, but its lower power envelope of 25 W TDP versus 55 W may matter for systems where thermal limits are strict.

Architecture Differences

The two processors come from different Intel lineups. The Core Ultra 5 235HX belongs to the Core Ultra Series 2, uses the Arrow Lake architecture, and carries the codename Arrow Lake-HX. The Core Ultra 9 386H belongs to the Core Ultra Series 3, uses the Panther Lake architecture, and carries the codename Panther Lake.

Both are built on a 3 nm process, but the foundries differ. TSMC fabricates the Ultra 5, while Intel fabricates the Ultra 9. The Ultra 5 has a transistor count of 17,800 million and a die size of 243 mm². The Ultra 9 has no recorded transistor count or die size.

Cache layouts differ in L2 and L3. Both have 192 KB of L1 per core. The Ultra 5 has 3 MB of L2 per core and 24 MB of shared L3. The Ultra 9 has 2.5 MB of L2 per core and 18 MB of shared L3. The Ultra 5 therefore provides more cache per core and a larger shared pool.

The integrated graphics differ. The Ultra 5 uses Arc Xe-LPG Graphics with 48 execution units. The Ultra 9 uses Intel Xe3 Graphics. The Ultra 9's integrated graphics is the newer generation, but the benchmark data does not include graphics tests.

PCIe lane counts also differ. The Ultra 5 provides Gen 5 with 20 lanes (CPU only). The Ultra 9 provides Gen 5 with 12 lanes (CPU only). The Ultra 5 offers more PCIe connectivity for discrete components.

The Ultra 5 has an unlocked multiplier, while the Ultra 9 is locked. Memory support differs: the Ultra 5 supports DDR5 only, while the Ultra 9 supports both DDR5 and LPDDR5X. Neither supports ECC memory.

Specification Differences

| Specification | Intel Core Ultra 5 235HX | Intel Core Ultra 9 386H |

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

| Cores | 14 | 16 |

| Threads | 14 | 16 |

| Base clock | 2.90 GHz | 2.10 GHz |

| Boost clock | 5.10 GHz | 4.90 GHz |

| TDP | 55 W | 25 W |

| Socket | Intel BGA 2114 | Intel BGA 2540 |

| Architecture | Arrow Lake | Panther Lake |

| Foundry | TSMC | Intel |

| L2 cache | 3 MB (per core) | 2.5 MB (per core) |

| L3 cache | 24 MB (shared) | 18 MB (shared) |

| Memory support | DDR5 | DDR5, LPDDR5X |

| Memory bandwidth | 102.4 GB/s | 115.2 GB/s |

| PCIe lanes | 20 (Gen 5) | 12 (Gen 5) |

| Integrated GPU | Arc Xe-LPG 48EU | Intel Xe3 Graphics |

| Multiplier | Unlocked | Locked |

| Release date | 2025-01-12 | 2026-01-04 |

| Part number | SRVFL | SA4R5Q9EH |

The Ultra 5 has a higher base clock (2.90 GHz versus 2.10 GHz) and higher boost clock (5.10 GHz versus 4.90 GHz). The Ultra 9 has more cores (16 versus 14) but a much lower TDP (25 W versus 55 W). The Ultra 5 uses an older socket (BGA 2114) compared to the Ultra 9's BGA 2540.

Head-to-Head Benchmarks

The largest single benchmark gap belongs to the Ultra 5 in Cinebench R23 single-core. It scores 4903 against 2071.5, a 136.7% advantage. This is an outlier compared to other single-core tests, where the Ultra 5 leads by 62.8% in R15 (494 versus 303.5) and 13.8% in R20 (2059 versus 1809).

In multicore rendering, the Ultra 5 leads by 8.6% in Cinebench R15 (3500 versus 3223), 13.8% in R20 (14587 versus 12820), and 69% in R23 (34731 versus 20547). The R23 result stands out as a much larger margin than the R15 and R20 results, suggesting the R23 workload amplifies the Ultra 5's advantages.

Passmark results consistently favor the Ultra 5. Data compression shows a 21% lead (426417 versus 352365). Data encryption shows 20.4% (32697 versus 27150). Extended instructions show 19.5% (34808 versus 29138). Floating point math shows 19.6% (129819 versus 108527). Integer math shows 11.3% (97177 versus 87284). Random string sorting shows 20.9% (50929 versus 42135). Prime number finding shows a narrower 5.9% lead (361 versus 341).

The Ultra 9's only win is Passmark physics, where it scores 3028 versus 2550, a 15.8% margin. In multithread, the Ultra 5 still wins by 15.4% (40849 versus 35399), despite the Ultra 9 having 16 cores against 14.

The Ultra 5 also wins Passmark single-thread by 11%, with identical scores of 4683 versus 4218 in both the single_thread and singlethread entries.

The Verdict

The data does not support a balanced recommendation. The Intel Core Ultra 5 235HX wins 16 of 17 benchmarks and holds a 91st percentile rank versus the Ultra 9's 88th. Its average benchmark score of 52073 places it among desktop-class rivals like the AMD Ryzen 9 5950X (0.2% higher) and the Intel Core i9-13900F (0.7% lower). The Ultra 9's average of 43210 aligns with the Intel Core i9-12900 (0.7% higher) and the AMD Ryzen AI Max PRO 385 (0.3% higher).

For most workloads, the Ultra 5 is the faster chip. It wins in rendering, compression, encryption, math, and single-thread tasks. Its higher clocks and larger caches give it a clear edge despite having fewer cores. The unlocked multiplier adds flexibility for systems that can handle the 55 W TDP.

The Ultra 9 386H should be selected only when the Passmark physics win matters or when the 25 W TDP is a hard requirement. Its 16 cores do not translate into a multicore win, and its lower clocks put it behind in nearly every test. The newer Panther Lake architecture and LPDDR5X support are notable, but the recorded benchmarks show no performance benefit from them.

Builders with power headroom should choose the Ultra 5 235HX. Builders constrained to a 25 W envelope and needing physics simulation performance have one specific reason to pick the Ultra 9 386H. Otherwise, the data points firmly to the Ultra 5.