Intel Celeron J4125 vs AMD Ryzen 7 3750H

Intel Celeron J4125

The Intel Celeron J4125 is a quad-core SoC primarily for inexpensive mini PCs and was announced late 2019. It runs at 2-2.7 GHz (Single Core Burst) and is based on the Gemini Lake platform (2019 refresh). Compared to the predecessor, the refresh offers slightly higher clock speeds. Similar to the Apollo Lake predecessors, the chip is manufactured on a 14 nm process with FinFETs but offers slightly improved processor cores, double the amount of L2 cache, partial Wi-Fi 5 support, all in a smaller package. Besides four CPU cores, the chip also includes a DirectX 12 capable GPU as well as a DDR4/LPDDR4 memory controller (dual-channel, up to 2400 MHz). The processor is not replaceable as it is directly soldered to the mainboard.

Architecture

The processor architecture is still called Goldmont Plus. Compared to the older Goldmont cores in Apollo Lake, they feature an increased level 2 cache (to 4 MB). That means the per-clock-performance should be a bit better, but not near the Core CPUs like Kaby Lake Y.

Performance

The average J4125 in our database outguns the AMD 3050e just slightly, the latter being a dual-core, Zen-based, entry-level processor of nearly the same energy efficiency, as far as multi-thread benchmark scores are concerned. This means the Celeron is a very basic CPU that's good for basic day-to-day activities only.

The Beelink GK Mini is one of the fastest systems built around the J4125 that we know of. It can be roughly 70% faster in CPU-bound workloads than the slowest system featuring the same chip in our database, as of August 2023.

GPU Performance

The UHD Graphics 600 (Gemini Lake) is based on Intel's Gen9 architecture, which supports DirectX 12 and is also used for the Kaby Lake / Skylake / Apollo Lake graphics adapters (like HD Graphics 520). Equipped with 12 EUs and a clock of up to 750 MHz, the performance should be roughly on par with the older HD Graphics 500 (Apollo Lake).

The chip also includes an advanced video engine with hardware support for the playback of VP9 and H.265 material (8-bit color-depth).

Power consumption

Like most J-class Intel processors, the Celeron has a default TDP, also known as the long-term power limit, of 10 W. This is low enough, allowing laptop makers to ditch fans while building a passively cooled laptop, tablet or mini-PC around the chip.

The Intel Celeron J4125 is built with one of the old 14 nm Intel processes for poor, as of early 2023, energy efficiency.

AMD Ryzen 7 3750H

The AMD Ryzen 7 3750H is a mobile SoC that was announced in January 2019. It combines four Zen+ cores (8 threads) clocked at 2.3 - 4 GHz with a Radeon RX Vega 10 graphics card with 10 CUs (640 Shaders) clocked at up to 1400 MHz. Compared to the similar Ryzen 7 3700U, the 3750H offers a 20 Watt higher TDP and therefore a better performance under long periods of load. The integrated dual-channel memory controller supports up to DDR4-2400 memory. As the features of the Picasso APUs are the same compared to the Raven Ridge predecessors, we point to our Raven Ridge launch article.

Picasso SoCs use the Zen+ microarchitecture with slight improvements that should lead to a 3% IPS (performance per clock) improvements.

Performance

The average 3750H in our database is in the same league as the Core i7-10610U, as far as multi-thread benchmark scores are concerned. This is a fairly good result, as of early 2022.

The ROG Zephyrus G15 GA502DU is among the fastest laptops built around the 3750H that we know of. It can be about 20% faster in CPU-bound workloads than the slowest system featuring the same chip in our database, as of August 2023.

Power consumption

This Ryzen 7 has a default TDP of 35 W (also known as the long-term power limit). Laptop manufacturers are free to change that value to anything between 12 W and 35 W with clock speeds and performance changing accordingly. These values are comparatively high, making an active cooling solution pretty much a necessity.

Last but not the least, this AMD APU is built with a rather old, as of late 2022, 12 nm process for lower-than-average energy efficiency.