Introduction

The MediaTek Dimensity 9000 Plus, Google Tensor G5, and Google Tensor G4 represent three different approaches to flagship-class mobile silicon: MediaTek’s 2022 Cortex-X2 flagship, Google’s newest TSMC-built Tensor chip, and its Samsung-fabricated predecessor. This comparison matters because these three chips still power devices being bought, reviewed, and debated today, and their differences  architecture generation, fabrication partner, GPU vendor  translate into real gaps in speed, gaming, and camera capability. Every figure below comes from NanoReview’s published specifications and benchmark data for each chip.

Processor Overview

Dimensity 9000 Plus, Tensor G5, and Tensor G4 sit three CPU generations apart in Arm core design despite two of them launching within a year of each other. The Dimensity 9000 Plus is a July 2022 flagship built on TSMC’s 4nm node, pairing a single Cortex-X2 prime core with Cortex-A710 performance cores and Cortex-A510 efficiency cores, plus a Mali-G710 MP10 GPU and MediaTek’s own APU 590 NPU. The Tensor G5, announced August 2025, is Google’s first Tensor chip fabricated by TSMC (3nm) rather than Samsung, moving to a Cortex-X4/A725/A520 configuration and a PowerVR DXT-48-1536 GPU sourced from Imagination Technologies instead of Arm’s Mali line. The Tensor G4, its August 2024 Samsung-fabricated predecessor, uses the same Cortex-X4 prime core as the G5 but pairs it with older Cortex-A720 performance cores and a Mali-G715 MP7 GPU. Market position for all three is flagship-class, with the Dimensity 9000 Plus now a value-flagship option and both Tensor chips exclusive to Google’s Pixel line.

CPU Architecture & Processing Power

Tensor G5 has the strongest CPU on paper and in benchmarks among the three. Its Cortex-X4 prime core running at 3.78GHz, backed by five Cortex-A725 cores at 3.05GHz, produces a Geekbench 6 single-core score of 2305 and a multi-core score of 6054 — a lead of roughly 17% in single-core and 12% in multi-core over the Dimensity 9000 Plus (1967 / 5401), and roughly 35% and 37% over the Tensor G4 (1707 / 4431).

What stands out is that the two-year-older Dimensity 9000 Plus actually beats the newer Tensor G4 in both Geekbench metrics, by about 15% single-core and 22% multi-core. The Dimensity 9000 Plus’s Cortex-X2 core config, despite being an older generation, benefits from four efficiency cores versus the Tensor G4’s four A520 cores clocked lower, and from a TSMC 4nm process rather than Samsung’s. NanoReview’s own CPU Performance sub-score reflects this ordering directly: Tensor G5 leads at 69, Dimensity 9000 Plus follows at 62, and Tensor G4 trails at 55.

For everyday use, this means the Tensor G5 opens apps and handles multitasking with the most headroom, the Dimensity 9000 Plus is close behind and still comfortably capable, and the Tensor G4 is the most likely of the three to feel stretched under heavy multitasking. None of this can be inferred from clock speed or core count alone — the Tensor G4’s higher-clocked Cortex-X4 core does not translate into a CPU-benchmark win over the older Dimensity 9000 Plus, which is a useful reminder that architecture generation and node together matter more than a single spec.

GPU & Gaming Performance

Which processor has the better GPU depends on whether you look at theoretical output or measured gaming performance — and here the three chips tell a genuinely mixed story. On paper, GPU compute is close: Tensor G5’s PowerVR DXT-48-1536 rates at 1689.6 GFLOPS, narrowly ahead of Tensor G4’s Mali-G715 MP7 at 1684.4 GFLOPS, with the Dimensity 9000 Plus’s Mali-G710 MP10 trailing at 1216 GFLOPS.

Real-world gaming metrics flip that order. NanoReview’s Gaming Performance sub-score puts Tensor G4 highest at 35, Dimensity 9000 Plus second at 33, and Tensor G5 lowest at 25 — despite the G5 having the newest silicon. The 3DMark Steel Nomad Light graphics test (measured in FPS, as supplied) shows the same pattern: Tensor G4 renders at 8 FPS, Tensor G5 at 7 FPS, and Dimensity 9000 Plus at 6 FPS. The Dimensity 9000 Plus also has a supplied Wild Life Extreme score of 2642 (15 FPS graphics test), but this benchmark wasn’t run on the two Tensor chips, so it isn’t directly comparable across all three.

The gap between Tensor G5’s high theoretical GPU output and its weaker measured gaming score points to driver and API maturity rather than raw hardware weakness — PowerVR is a newer GPU vendor in Google’s lineup, and Vulkan/OpenCL optimization on unfamiliar silicon typically lags behind an established Mali or Adreno implementation. For heavy 3D gaming and sustained sessions, the Tensor G4 and Dimensity 9000 Plus currently deliver more consistent real-world output than the specs of the Tensor G5 alone would suggest. None of the supplied data indicates a difference in ray-tracing support across the three chips.

Manufacturing Process & Efficiency

Fabrication is the one category where all three chips take a different path. The Dimensity 9000 Plus uses TSMC’s 4nm process, the Tensor G5 uses TSMC’s 3nm process, and the Tensor G4 uses Samsung’s 4nm process. A smaller node doesn’t automatically mean faster or cooler silicon — architecture, clock strategy, and packaging matter just as much — but moving from Samsung 4nm to TSMC 3nm between the G4 and G5 is a foundry change as well as a node shrink, and TSMC’s flagship nodes have generally shown stronger yields and efficiency in recent generations.

The supplied Battery life sub-scores are nearly identical: Dimensity 9000 Plus and Tensor G5 both score 91, and Tensor G4 scores 90. That near-tie suggests the Tensor G5’s newer node hasn’t yet produced a measurable battery-life advantage over either the older Samsung-built G4 or the TSMC 4nm Dimensity 9000 Plus, at least on the metric NanoReview tracks. TDP is only disclosed for the Dimensity 9000 Plus, at 4W sustained — not disclosed for either Tensor chip, so a direct sustained-power comparison isn’t possible from this data.

RAM, Memory & Storage Support

The Dimensity 9000 Plus supports the highest maximum RAM of the three at 24GB of LPDDR5X, versus 16GB for both Tensor G5 and Tensor G4. However, the two Tensor chips run their LPDDR5X faster, at 4266MHz versus the Dimensity 9000 Plus’s 3750MHz, giving them higher rated bandwidth (68.2 Gb/s versus 60 Gb/s). All three use the same 4x 16-bit memory bus configuration.

For most users, the Dimensity 9000 Plus’s higher RAM ceiling matters more for phones that ship with generous memory configurations aimed at heavy multitasking, while the Tensor chips’ faster memory frequency modestly benefits memory-bandwidth-sensitive tasks like camera processing pipelines. On storage, the Tensor G5 and Tensor G4 both support UFS 3.1 and UFS 4.0, while the Dimensity 9000 Plus is limited to UFS 3.1 — meaning phones built on the Tensor chips can offer faster storage read/write speeds if the manufacturer opts for UFS 4.0 modules.

AI & NPU Capabilities

Only the Dimensity 9000 Plus names its AI accelerator: the MediaTek APU 590. Both Tensor G5 and Tensor G4 are listed simply as having an NPU present, with no model name or TOPS figure disclosed for either. This is a meaningful data gap for a category that’s central to Google’s own marketing around Tensor, but per the supplied specifications, no verified TOPS number exists for any of the three chips.

Without disclosed AI performance figures, no claim can be made about which chip is faster for generative AI, voice processing, or on-device image enhancement based on hardware specs alone. Real-world AI feature availability (like Pixel’s on-device photo and video AI tools) is largely a software layer decision made by Google, not something derivable from the NPU model name alone. Not enough disclosed data to compare AI throughput directly across these three chips.

ISP, Camera & Video Capabilities

The Dimensity 9000 Plus supports the highest maximum camera resolution among the three, listing 1x320MP or 3x32MP simultaneous sensor support. Tensor G5 supports up to 1x200MP. The Tensor G4’s maximum camera resolution is not disclosed in the supplied data. These figures describe chip-level ISP ceilings only — they say nothing about the actual camera sensors, lenses, or image processing tuning that a phone manufacturer pairs with the chip, so a higher supported megapixel count on the SoC doesn’t guarantee a better photo from any specific phone.

Video capability shows a clearer, more consistent gap. Both Tensor G5 and Tensor G4 support 8K recording at 30FPS and 4K at 120FPS for capture and playback, while the Dimensity 9000 Plus tops out at 4K at 60FPS. All three support the same core codec set: H.264, H.265, AV1, and VP9. For anyone recording high-frame-rate slow motion or aiming for 8K footage, either Tensor chip has a clear capability edge over the Dimensity 9000 Plus — though as with camera resolution, actual recording quality still depends on the phone’s sensor and stabilization hardware, not the chip alone.

Display Support

Tensor G5 and Tensor G4 both support display output up to 3840 x 2400, while the Dimensity 9000 Plus is limited to 3200 x 1440. This is a chip-level ceiling for internal panel driving and external display output, not a guarantee that any specific phone using these chips ships with a panel at that resolution. Refresh rate figures are not disclosed for any of the three chips, so no comparison can be made on that specific point.

Modem, 5G & Connectivity

The Dimensity 9000 Plus is the only one of the three with explicit theoretical modem speeds disclosed: up to 7000Mbps download and 2500Mbps upload on its integrated LTE Cat.24 / 5G modem. Neither Tensor G5 nor Tensor G4 has download/upload speed figures listed — both use a Samsung-made Exynos 5400c external modem, and NanoReview’s data doesn’t include a rated throughput for it. Real-world 5G speeds rarely reach any manufacturer’s theoretical maximum regardless of chip, so the Dimensity 9000 Plus’s disclosed numbers should be read as a ceiling, not an expectation.

On short-range connectivity, both Tensor chips have a generational edge: Wi-Fi 7 and Bluetooth 6.0, versus the Dimensity 9000 Plus’s Wi-Fi 6 and Bluetooth 5.3. For anyone on a Wi-Fi 7 router, this means a real gap in peak local wireless throughput and latency in the Tensor chips’ favor. Satellite/GNSS support is broader on the Dimensity 9000 Plus, which adds QZSS, SBAS, and NAVIC on top of GPS, GLONASS, Beidou, and Galileo — both Tensor chips list GPS, GLONASS, Beidou, Galileo, and QZSS, without SBAS or NAVIC support in the supplied data.

Benchmark Comparison

All three chips were tested with the same benchmark versions — AnTuTu 11 and Geekbench 6 — so the following comparisons are valid.

AnTuTu 11 total score: Tensor G5 leads at 1,549,798, followed by Dimensity 9000 Plus at 1,540,240 (a gap of about 0.6%), and Tensor G4 at 1,519,937 (Tensor G5 leads by about 2%). This is a tight three-way race at the total-score level, but the CPU sub-scores diverge sharply: Tensor G5’s AnTuTu CPU sub-score of 734,567 dwarfs the Dimensity 9000 Plus’s 600,873 and Tensor G4’s 536,742. The GPU sub-score tells the opposite story — Tensor G4 actually posts the highest AnTuTu GPU sub-score at 414,930, ahead of the Dimensity 9000 Plus’s 274,706 and Tensor G5’s 183,228, which is consistent with the real-world gaming pattern already seen above.

Geekbench 6: Tensor G5 leads single-core (2305) and multi-core (6054) by a clear margin over both rivals, as detailed in the CPU section. Tensor G5’s GPU Compute score is 4116, well behind Tensor G4’s 8744 — again reflecting the PowerVR driver gap rather than a raw hardware deficit, since Tensor G5’s theoretical GFLOPS figure is actually higher than Tensor G4’s.

3DMark Steel Nomad Light: Tensor G4 scores highest at 1049, Tensor G5 follows at 1012 (about 3.7% behind), and Dimensity 9000 Plus trails at 861 (about 18% behind Tensor G4). The Dimensity 9000 Plus also has a Wild Life Extreme score of 2642, but since neither Tensor chip has a Wild Life Extreme result in the supplied data, that figure cannot be compared across all three.

PCMark 3.0: Only Tensor G5 (15,332) and Tensor G4 (13,174) have this benchmark supplied — Tensor G5 leads by about 16%. No PCMark data is available for the Dimensity 9000 Plus, so a three-way PCMark comparison isn’t possible with this data set.

Real-World Performance

Everyday use: Tensor G5’s CPU lead should make it the snappiest of the three for app launches and multitasking, with the Dimensity 9000 Plus close behind and Tensor G4 the least responsive under heavy load, based on the CPU benchmark gaps above.

Gaming: Tensor G4 and Dimensity 9000 Plus both outperform Tensor G5 in measured gaming metrics (Gaming Performance sub-score and 3DMark FPS), despite Tensor G5 having higher theoretical GPU output — a sign that GPU driver maturity matters as much as raw compute for actual frame rates on this trio.

AI: No TOPS or benchmark data is disclosed for any of the three chips’ AI accelerators, so no real-world AI performance comparison can be made from the supplied specifications.

Camera: The Dimensity 9000 Plus’s higher supported camera resolution and both Tensor chips’ 8K/120FPS video ceiling are chip-level capabilities only; actual photo and video quality depends on the sensor and software Google or a Dimensity 9000 Plus phone maker pairs with the chip, not the chip alone.

Battery/efficiency: Battery life sub-scores are nearly tied (91, 91, 90), so none of the three shows a clear, data-backed efficiency advantage over the others despite the Tensor G5’s smaller 3nm node.

Strengths & Weaknesses

Dimensity 9000 Plus

• Strength: Highest supported RAM ceiling (24GB LPDDR5X) of the three.
• Strength: Only chip with disclosed 5G modem speed ratings (7000/2500 Mbps) and the broadest GNSS support (adds SBAS, NAVIC).
• Weakness: Lowest theoretical GPU output (1216 GFLOPS) and lowest 3DMark Steel Nomad Light score (861).
• Weakness: Capped at 4K/60FPS video and Wi-Fi 6, both behind the two Tensor chips.

Tensor G5

• Strength: Clear CPU leader — highest Geekbench 6 single-core (2305) and multi-core (6054) scores, and highest CPU Performance sub-score (69).
• Strength: Newest process node (TSMC 3nm) and most modern I/O: Wi-Fi 7, Bluetooth 6.0, UFS 4.0 support, 8K/30FPS video.
• Weakness: Lowest Gaming Performance sub-score (25) of the three, despite the highest theoretical GPU GFLOPS — pointing to PowerVR driver immaturity.
• Weakness: Lowest maximum RAM support (16GB) tied with Tensor G4.

Tensor G4

• Strength: Highest Gaming Performance sub-score (35) and highest 3DMark Steel Nomad Light score (1049) of the three.
• Strength: Highest AnTuTu GPU sub-score (414,930) and highest Geekbench GPU Compute score (8744).
• Weakness: Weakest CPU of the three — lowest CPU Performance sub-score (55) and lowest Geekbench 6 single/multi-core scores (1707 / 4431).
• Weakness: Built on Samsung’s 4nm process rather than TSMC, and camera resolution specs are not disclosed in the supplied data.

Best Processor For Different Users

• Best for Gaming: Tensor G4 — highest measured Gaming Performance score and 3DMark result of the three.
• Best for AI: Not enough disclosed data (no TOPS or AI benchmark figures for any chip) to name a winner.
• Best for Efficiency: Tensor G5 — newest 3nm node, tied for the highest battery-life sub-score.
• Best for Camera (chip-level ISP): Dimensity 9000 Plus — highest supported max camera resolution.
• Best for Connectivity: Tensor G5 and Tensor G4 (tied) — Wi-Fi 7 and Bluetooth 6.0 versus the Dimensity 9000 Plus’s Wi-Fi 6/Bluetooth 5.3.
• Best Balanced Choice: Dimensity 9000 Plus — competitive CPU and gaming scores together, without Tensor G5’s GPU driver weak point.

Final Verdict

1. Tensor G5 — Best Overall Its Geekbench 6 leads (2305/6054) and CPU Performance sub-score of 69 are the clearest, largest gap in this comparison, and it pairs that with the newest process node and the most modern display, video, and wireless I/O of the three.

2. Dimensity 9000 Plus — Best Balanced It trails Tensor G5 in raw CPU output but beats it in real-world gaming performance (33 vs 25) and offers the highest RAM ceiling and the only disclosed 5G speed ratings, making it the more evenly matched all-rounder despite being the oldest chip here.

3. Tensor G4 — Best For Gaming Its CPU is the weakest of the three by a clear margin, but it posts the highest measured Gaming Performance score and 3DMark result, making it a legitimate pick specifically for gaming-focused use even though it trails both rivals elsewhere.

Best for Gaming: Tensor G4 Best for AI: Not enough disclosed data to name a winner Best for Efficiency: Tensor G5 Best Balanced Choice: Dimensity 9000 Plus

Bottom Line: Choose the Tensor G5 if CPU speed and a modern feature set (Wi-Fi 7, UFS 4.0, 8K video) matter most to you, accept that its GPU currently underperforms its own spec sheet. Choose the Dimensity 9000 Plus if you want the most balanced profile with higher RAM headroom and a modem with disclosed 5G speed ratings. Choose a Tensor G4 device only if gaming performance is the priority and you’re comfortable with the weakest CPU of the three.