Intel Core Ultra 5 235HX vs Intel Core Ultra 5 338H Comparison

Head-to-Head Benchmarks

The head-to-head data shows a dominant performance profile for the Intel Core Ultra 5 235HX, which wins 16 of the 17 recorded benchmark comparisons. The single exception is the PassMark physics test, where the Core Ultra 5 338H takes a narrow 5.5% lead (2697 vs 2550). Every other test, whether single-threaded or multi-threaded, favors the 235HX by a wide margin.

The largest gaps appear in Cinebench workloads. In Cinebench R23 single-core, the 235HX scores 4903 against 2044 for the 338H, a 139.9% advantage. The R23 multi-core result is similarly lopsided: 34731 vs 16331, a 112.7% lead. These are the two biggest deltas in the entire comparison, indicating that the 235HX holds a substantial per-thread performance edge alongside a massive parallel workload advantage.

Cinebench R15 and R20 tell the same story. In R15 multi-core, the 235HX scores 3500 against 2504, a 39.8% margin. The single-core R15 result shows 494 vs 305, a 62% lead. In R20, the multi-core score is 14587 vs 10213 (42.8% ahead), and the single-core score is 2059 vs 1441 (42.9% ahead). The consistency of these deltas across Cinebench generations suggests the 235HX does not just scale better with more cores; it also extracts more performance from each core.

PassMark results follow a similar pattern, though with some variation. The data compression test shows 426417 vs 276539, a 54.2% lead. Data encryption shows 32697 vs 21367, a 53% lead. Extended instructions score 34808 vs 23906, a 45.6% margin. Floating point math shows 129819 vs 84067, a 54.4% lead. Integer math delivers 97177 vs 64934, a 49.7% advantage. Random string sorting scores 50929 vs 34082, a 49.4% margin. The multithread benchmark shows 40849 vs 28717, a 42.2% lead.

The gaps narrow in single-threaded PassMark tests. The single-thread score is 4683 vs 4180, a 12% lead for the 235HX. Prime number finding, which is often sensitive to memory latency and core architecture, shows a smaller 18.8% delta (361 vs 304). These two results are the smallest margins outside the physics test, indicating that the 338H is more competitive in lightly threaded or latency-bound workloads, but still loses ground.

The physics test is the only recorded win for the 338H, and its 2697 score represents a 5.5% improvement over the 235HX's 2550. This is an isolated result, and the magnitude is modest compared to the double-digit and triple-digit deltas elsewhere. The overall average benchmark score reflects the disparity: the 235HX averages 52073, placing it in the 91st percentile of all CPUs, while the 338H averages 33989, placing it in the 84th percentile. The 235HX sits alongside desktop-class parts like the AMD Ryzen 9 5950X and Intel Core i9-13900F in the database's nearest rivals, while the 338H competes with mobile and server parts such as the Intel Core Ultra 7 165H and Intel Xeon 6353P.

Architecture Differences

The two processors come from different Intel product lines. The Core Ultra 5 235HX belongs to the Core Ultra Series 2, based on the Arrow Lake architecture with the codename Arrow Lake-HX, and is fabricated on a 3 nm process at TSMC. The Core Ultra 5 338H belongs to the Core Ultra Series 3, based on the Panther Lake architecture, also on a 3 nm process but fabricated at Intel. The transistor count and die size are recorded for the 235HX (17,800 million transistors on a 243 mm² die), while those figures are not available for the 338H.

Core and thread counts differ significantly. The 235HX has 14 cores and 14 threads, while the 338H has 12 cores and 12 threads. Neither part uses simultaneous multithreading, so thread counts equal core counts in both cases. The 235HX runs at a base clock of 2.90 GHz and a boost clock of 5.10 GHz, compared to 1.90 GHz base and 4.70 GHz boost for the 338H. The clock speed gap, particularly at base frequency, is substantial and contributes to the 235HX's per-core advantage.

Cache hierarchies are similar in L1 size but diverge at lower levels. Both processors provide 192 KB of L1 cache per core. The L2 cache is 3 MB per core on the 235HX and 2.5 MB per core on the 338H. The shared L3 cache is 24 MB on the 235HX and 18 MB on the 338H. These differences in both per-core and shared cache capacity favor the 235HX, which can hold more working data closer to the cores.

Memory support also differs. The 235HX uses DDR5 memory with a dual-channel bus and a recorded memory bandwidth of 102.4 GB/s. The 338H uses LPDDR5X memory, also dual-channel, but with a higher recorded bandwidth of 136.5 GB/s. The 338H therefore has a memory bandwidth advantage despite having a smaller cache pool, which may explain its competitive showing in the physics test.

PCIe connectivity is another point of divergence. The 235HX provides Gen 5 with 20 lanes (CPU only), while the 338H provides Gen 5 with only 4 lanes (CPU only). The 235HX offers far more expansion capacity for discrete GPUs or storage controllers. The integrated graphics differ as well: the 235HX uses Arc Xe-LPG Graphics with 48 execution units, while the 338H uses Arc B370.

The 235HX has an unlocked multiplier, meaning it can be overclocked, while the 338H is locked. The sockets are different: Intel BGA 2114 for the 235HX and Intel BGA 2540 for the 338H. The 235HX has a TDP of 55 watts, while the 338H has a TDP of 25 watts. The 338H is a lower-power part by a wide margin, which is consistent with its lower clocks and smaller core count.

Release dates also differ. The 235HX was released in January 2025, while the 338H was released in January 2026. Both are listed as Active in production status, and both are mobile market segment parts. Neither has ECC memory support.

Where Each One Wins

The benchmark data indicates a clear workload split, but it is heavily weighted toward the 235HX. The 235HX wins every multi-threaded test in the head-to-head comparison, including Cinebench R15, R20, and R23 multi-core, plus all PassMark multi-threaded workloads. For rendering, video encoding, data compression, encryption, and any task that scales with core count, the 235HX is the stronger choice. Its 14 cores, higher clocks, and larger caches give it a decisive advantage in parallel processing.

Single-threaded performance also favors the 235HX in every recorded test, with deltas ranging from 12% in PassMark single-thread to 139.9% in Cinebench R23 single-core. The 235HX's boost clock of 5.10 GHz, compared to 4.70 GHz for the 338H, provides a meaningful edge in lightly threaded applications such as web browsing, office productivity, and legacy software that relies on one or two cores.

The 338H has one confirmed win: the PassMark physics test, where it scores 2697 vs 2550. This test appears to favor the 338H's memory bandwidth, which is recorded at 136.5 GB/s versus 102.4 GB/s for the 235HX. The physics test often stresses memory access patterns, and the 338H's LPDDR5X support may provide an advantage in that specific scenario. However, this is a single data point, and the margin is only 5.5%.

The 338H also has a power profile that is more than twice as efficient on paper. Its TDP is 25 watts versus 55 watts for the 235HX. For thin-and-light laptops where sustained performance under a low thermal budget matters, the 338H may be the more practical choice despite losing most benchmarks. The database records no battery life or thermal tests, so the practical impact of the TDP difference cannot be quantified here, but the spec sheet clearly positions the 338H as the lower-power part.

FAQ

Q: Which processor has the higher multi-core performance?

A: The Intel Core Ultra 5 235HX wins all multi-core benchmarks. In Cinebench R23 multi-core, it scores 34731 against 16331 for the 338H, a 112.7% lead. The PassMark multithread test shows 40849 vs 28717, a 42.2% advantage.

Q: Does the 338H win any benchmark at all?

A: Yes, the 338H wins the PassMark physics test with a score of 2697, which is 5.5% higher than the 235HX's 2550. This is the only win among the 17 recorded head-to-head comparisons.

Q: What explains the 338H's physics test win?

A: The 338H has a higher recorded memory bandwidth of 136.5 GB/s, compared to 102.4 GB/s for the 235HX. The physics test is likely sensitive to memory throughput, which gives the 338H a specific advantage in that workload.

Q: Are both processors unlocked for overclocking?

A: No. The 235HX has an unlocked multiplier, while the 338H is locked. This means the 235HX can be overclocked, but the 338H cannot.

Q: How do the two processors compare in single-threaded performance?

A: The 235HX wins every single-threaded test. The largest margin is in Cinebench R23 single-core, where it scores 4903 vs 2044, a 139.9% lead. The smallest margin is in PassMark single-thread, where it scores 4683 vs 4180, a 12% lead.

Q: What are the socket requirements for each processor?

A: The 235HX uses Intel BGA 2114, while the 338H uses Intel BGA 2540. They are not socket-compatible.

Specification Differences

| Specification | Intel Core Ultra 5 235HX | Intel Core Ultra 5 338H |

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

| Series | Core Ultra Series 2 | Core Ultra Series 3 |

| Architecture | Arrow Lake | Panther Lake |

| Codename | Arrow Lake-HX | Panther Lake |

| Process node | 3 nm | 3 nm |

| Foundry | TSMC | Intel |

| Transistors | 17,800 million | Not recorded |

| Die size | 243 mm² | Not recorded |

| Cores | 14 | 12 |

| Threads | 14 | 12 |

| Base clock | 2.90 GHz | 1.90 GHz |

| Boost clock | 5.10 GHz | 4.70 GHz |

| TDP | 55 W | 25 W |

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

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

| Memory support | DDR5 | LPDDR5X |

| Memory bandwidth | 102.4 GB/s | 136.5 GB/s |

| PCIe | Gen 5, 20 Lanes (CPU only) | Gen 5, 4 Lanes (CPU only) |

| Integrated graphics | Arc Xe-LPG Graphics 48EU | Arc B370 |

| Socket | Intel BGA 2114 | Intel BGA 2540 |

| Multiplier unlocked | Yes | No |

| Release date | January 2025 | January 2026 |

The Verdict

The data indicates that the Intel Core Ultra 5 235HX is the stronger processor in nearly every measurable performance category. It wins 16 of 17 head-to-head benchmarks, with an average benchmark score of 52073 versus 33989 for the 338H. Its advantages are largest in multi-core workloads, where it leads by as much as 112.7% in Cinebench R23, and in single-core workloads, where it leads by up to 139.9% in the same test. The 235HX also offers more PCIe lanes (20 vs 4), a larger L3 cache (24 MB vs 18 MB), a higher boost clock (5.10 GHz vs 4.70 GHz), and an unlocked multiplier for overclocking.

The 338H has a narrower set of advantages, but they are real. It records a 5.5% win in the PassMark physics test, which aligns with its higher memory bandwidth of 136.5 GB/s. It operates at a TDP of 25 watts, less than half the 55 watts of the 235HX, making it the more appropriate choice for power-constrained mobile designs. It also uses a newer socket (BGA 2540) and a newer architecture (Panther Lake), though the recorded benchmark results do not show a performance benefit from that newer generation.

For users who prioritize raw compute throughput across rendering, compression, encryption, and math-heavy tasks, the 235HX is the clear pick based on the data. For users who need a lower-power processor with a memory-bandwidth edge in specific physics simulations, the 338H offers a targeted advantage, but it sacrifices substantial multi-core and single-core performance to achieve it. The database's percentile rankings reinforce this: the 235HX sits at the 91st percentile of all CPUs, while the 338H sits at the 84th percentile.