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

Head-to-Head Benchmarks

The recorded head-to-head data shows a decisive overall victory for the Intel Core Ultra 5 235H, which claims 14 of the 17 compared benchmark tests. The Intel Core Ultra 5 338H takes the remaining 3 tests, but those wins are concentrated in specific computational niches rather than general throughput.

The single largest margin comes in the Cinebench R23 single-core test, where the 235H scores 3613 against the 338H's 2044, a 76.8% advantage. This is an enormous gap for processors from the same product family. The Cinebench R23 multi-core test tells a similar story: 25598 versus 16331, a 56.7% difference. These two results alone suggest the 235H has a fundamentally stronger execution core design or much more aggressive clock behavior under sustained load.

The 235H also wins the older Cinebench R15 single-core test by 19.3% (364 versus 305) and the R20 single-core test by 5.3% (1517 versus 1441). The R15 multi-core margin is smaller at 3% (2580 versus 2504), while R20 multi-core shows a 5.3% edge (10751 versus 10213). The pattern shifts slightly across Cinebench versions, with the newer R23 test exposing a much larger chasm than the older ones.

In PassMark workloads, the 235H leads in seven of ten categories. The largest PassMark win is in integer math, where it scores 74247 against 64934, a 14.3% margin. Floating-point math follows with 93509 versus 84067, an 11.2% advantage. Data compression shows 301979 versus 276539, a 9.2% edge, and data encryption delivers 23121 versus 21367, an 8.2% win. Random string sorting favors the 235H by 6.2% (36208 versus 34082). The PassMark multi-thread score shows a 4.8% lead (30091 versus 28717), and single-thread performance is 4.3% higher (4359 versus 4180).

The 338H's three wins are notable for their specificity. The physics test is its biggest victory: 2697 versus 1985, a 26.4% margin. The find prime numbers test shows 304 versus 239, a 21.4% advantage. Extended instructions deliver 23906 versus 23354, a smaller 2.3% edge. These results indicate the 338H has particular strengths in physics simulation, prime-number computation, and certain extended instruction workloads, even though it loses most general-purpose tests.

Architecture Differences

The two processors come from different generations and use different internal designs. The 235H belongs to the Core Ultra Series 2, uses the Arrow Lake architecture with the Arrow Lake-H codename, and is fabricated on a 3 nm process by TSMC. The 338H belongs to the Core Ultra Series 3, uses the Panther Lake architecture with the Panther Lake codename, and is fabricated on the same 3 nm node but by Intel's own foundry.

Core counts differ: the 235H has 14 cores and 14 threads, while the 338H has 12 cores and 12 threads. Neither processor uses simultaneous multithreading, so thread counts equal core counts. The 235H's two additional cores likely explain much of its multi-core benchmark advantage, though the Cinebench R23 single-core gap suggests architectural clock efficiency also plays a role.

Clock speeds favor the 235H. Its base clock is 2.40 GHz with a boost clock of 5.00 GHz. The 338H starts at 1.90 GHz and boosts to 4.70 GHz. Thermal design power is 28 watts for the 235H and 25 watts for the 338H. The higher boost clock of the 235H aligns with its stronger single-core results.

Cache hierarchies differ in L2 sizing. Both have 192 KB of L1 per core and 18 MB of shared L3. The 235H carries 3 MB of L2 per core, while the 338H carries 2.5 MB per core. With 14 cores, the 235H's total L2 capacity is larger in aggregate, which may contribute to its data-heavy workload wins.

Memory support diverges. The 235H supports both DDR5 and LPDDR5X, while the 338H supports only LPDDR5X. Both use dual-channel memory buses, but the 338H achieves higher memory bandwidth: 136.5 GB/s versus 102.4 GB/s. This bandwidth advantage does not translate into benchmark wins for the 338H in the recorded tests, suggesting the 235H's core count and cache configuration matter more.

PCI Express lanes also differ. The 235H provides Gen 5 with 8 lanes (CPU only), while the 338H provides Gen 5 with 4 lanes (CPU only). Integrated graphics use different parts: Arc Graphics 140T on the 235H versus Arc B370 on the 338H. Sockets differ as well, with the 235H using Intel BGA 2049 and the 338H using Intel BGA 2540. The release dates show the 235H arriving in January 2025 and the 338H in January 2026.

FAQ

Q: Which processor wins the majority of benchmark comparisons?

A: The Intel Core Ultra 5 235H wins 14 of 17 head-to-head tests, while the Intel Core Ultra 5 338H wins 3.

Q: What is the largest performance difference between the two?

A: The Cinebench R23 single-core test shows the 235H ahead by 76.8% (3613 versus 2044). The R23 multi-core test shows a 56.7% margin (25598 versus 16331).

Q: Where does the 338H outperform the 235H?

A: The 338H wins the PassMark physics test by 26.4% (2697 versus 1985), the find prime numbers test by 21.4% (304 versus 239), and the extended instructions test by 2.3% (23906 versus 23354).

Q: How do the core counts and clocks compare?

A: The 235H has 14 cores and 14 threads with a 2.40 GHz base and 5.00 GHz boost. The 338H has 12 cores and 12 threads with a 1.90 GHz base and 4.70 GHz boost.

Q: Which processor has higher memory bandwidth?

A: The 338H offers 136.5 GB/s versus the 235H's 102.4 GB/s, even though the 235H supports a wider range of memory types including DDR5 and LPDDR5X.

Q: Are both processors currently in production?

A: Yes, both are listed as Active production status. They use different sockets (Intel BGA 2049 for the 235H, Intel BGA 2540 for the 338H).

Specification Differences

The two processors differ in several key specification fields. Core count: 14 versus 12. Thread count: 14 versus 12. Base clock: 2.40 GHz versus 1.90 GHz. Boost clock: 5.00 GHz versus 4.70 GHz. TDP: 28 watts versus 25 watts. Socket: Intel BGA 2049 versus Intel BGA 2540. Architecture: Arrow Lake versus Panther Lake. Codename: Arrow Lake-H versus Panther Lake. Generation: Ultra 5 (Arrow Lake-H) versus Ultra 5 (Panther Lake-H). Foundry: TSMC versus Intel. L2 cache per core: 3 MB versus 2.5 MB. Memory support: DDR5 and LPDDR5X versus LPDDR5X only. Memory bandwidth: 102.4 GB/s versus 136.5 GB/s. PCIe lanes: 8 versus 4 (both Gen 5, CPU only). Integrated graphics: Arc Graphics 140T versus Arc B370. Release date: January 2025 versus January 2026. Part number: SRQAP versus SA4REQ9EW.

The L1 cache is identical at 192 KB per core, and the L3 cache matches at 18 MB shared. Both processors use dual-channel memory buses, do not support ECC memory, have locked multipliers, and target the mobile market segment. Both are fabricated on a 3 nm process.

The Verdict

The benchmark data presents a clear hierarchy. The Intel Core Ultra 5 235H is the stronger processor in nearly every measured workload, with particularly large advantages in Cinebench R23 tests. Its 14 cores, higher base and boost clocks, larger L2 cache, and TSMC fabrication combine to deliver superior single-thread and multi-thread performance.

The Intel Core Ultra 5 338H cannot match the 235H in general-purpose computation. Its wins are confined to physics simulation, prime-number finding, and extended instruction workloads. The 338H does offer higher memory bandwidth and a more recent release date, but these do not overcome the core count and clock disadvantages in the recorded benchmarks.

The average benchmark scores reinforce this conclusion. The 235H has an average score of 37522 across all tests, while the 338H averages 33989. The 235H sits in the 85th percentile of all CPUs, the 338H in the 84th. The nearest rivals for the 235H include the Intel Core i9-13900HK with a 0.3% higher average score and the Intel Core i5-13600K with a 0.4% lower score. The 338H's nearest rivals include the Intel Core Ultra 7 165H at 0.3% lower, the Intel Core i7-12800HX at 0.3% higher, and the AMD EPYC 4244P at 0.7% lower.

Where Each One Wins

The 235H dominates in rendering and content-creation workloads. Cinebench R15, R20, and R23 multi-core tests all go to the 235H, with margins ranging from 3% to 56.7%. The R23 single-core win of 76.8% makes it the clear choice for applications that depend on single-thread responsiveness. PassMark integer math, floating-point math, data compression, data encryption, random string sorting, multi-thread, and single-thread tests all favor the 235H. This processor suits general productivity, software compilation, data processing, and any workload that scales with core count.

The 338H takes a different role. Its physics score of 2697 against 1985 indicates strength in simulation and physics-based computation. The find prime numbers result (304 versus 239) points to mathematics-heavy algorithms. The extended instructions win (23906 versus 23354) suggests it handles certain specialized instruction sets more efficiently. The 338H also carries the Arc B370 integrated graphics and supports only LPDDR5X memory, positioning it for systems where the platform specification matters more than raw CPU throughput.

The data does not support a recommendation for the 338H as a general-purpose upgrade. Its release date is later, but its benchmark results trail the 235H in most categories. The 235H's higher TDP of 28 watts versus 25 watts reflects its greater performance delivery. For workloads that match the 338H's three winning categories, that processor has a place. For everything else, the 235H is the stronger option according to every recorded measurement.