Google is powering the Pixel 11 series with the new Tensor G6 chipset and we are seeing something new after a long while. It's fabricated on a 2nm process node, directly jumps to Arm's C1-series cores and packs a modem that is not made by Samsung. So, we have compared the latest Tensor G6 chipset against last year's Tensor G5 to measure the performance difference. On that note, let's begin.
Tensor G6 vs Tensor G5: CPU
The CPU is where things are interesting, so let us begin there. The Tensor G5 chip came with an 8-core CPU setup in a 1+5+2 layout. It packed one Cortex-X4 prime core at 3.78GHz, 5x Cortex-A725 cores at 3.05GHz and 2x Cortex-A520 cores at 2.25GHz.
The prime core was the weakest part of the Tensor G5. Google chose the Cortex-X4 core from 2023 for the Tensor G5 and skipped the newer and more powerful Cortex-X925 core. It resulted in much lower single-core scores, compared to the flagship chips.
| CPU | Tensor G6 | Tensor G5 |
|---|---|---|
| Core count | 7 | 8 |
| Prime core | 1x Arm C1-Ultra (4.11GHz) | 1x Cortex-X4 (3.78GHz) |
| Performance cores | 4x Arm C1-Pro (3.38GHz) | 5x Cortex-A725 (3.05GHz) |
| Efficiency cores | 2x Arm C1-Pro (2.65GHz) | 2x Cortex-A520 (2.25GHz) |
That said, the Tensor G6 reverses that trend. Google has switched to an unusual 7-core layout, but brings Arm's latest C1 series cores. There is a single Arm C1-Ultra core clocked at 4.11GHz, 4x C1-Pro cores at 3.38GHz and 2x C1-Pro cores at 2.65GHz. This whole cluster is from Arm's new Lumex generation and it's based on the Armv9.3-A design.
The C1 series cores are also part of MediaTek's flagship Dimensity 9500 chip, but it runs its prime core at 4.21GHz (vs 4.11GHz on the Tensor G6). Not to mention, Samsung's Exynos 2600 processor has the same CPU cores as well.
It means that the Tensor G6 finally shares a CPU family with flagship Android chipsets, even if Google is slightly conservative on both clock speeds and the number of cores. In effect, Google has skipped a full generation in one move, from the Cortex-X4 straight to the C1-Ultra. Put simply, the Tensor G6 is expected to gain a lot in CPU performance.
If we go by Google's own numbers, the company says the Tensor G6 is 25% faster in web browsing, 15% quicker in app loading and up to 20% better on power efficiency, compared to the Tensor G5. In our own testing, the Tensor G5 managed 2,285 in Geekbench 6 single-core and 6,191 in multi-core on the Pixel 10 Pro.
The Tensor G6 has also surfaced on Geekbench 7 and it scored 2,112 in single-core and 5,196 in multi-core, which is not in line with our expectations. That said, take these scores cautiously for now. Early listings tend to show lower scores, as we have seen with past Tensor chipsets.
Related Articles
Tensor G6 vs Tensor G5: GPU
While the CPU is exciting, the GPU choice is baffling. The Tensor G6 uses an Imagination PowerVR CXTP-48-1536 GPU and it seems to be running six compute units clocked at 1.3GHz.
It's a C-Series GPU, derived from a 2021 design, which is older than the D-Series DXT-48-1536 GPU used in the Tensor G5. On paper, the GPU looks like a downgrade.
Graphics performance was already quite weak on the Tensor G5. In our AnTuTu run, the Tensor G5's GPU could only muster 382,578 points. So moving to an older architecture reads like bad news at first glance.
That said, there are two redeeming points. The C-Series design should finally bring hardware ray tracing to a Pixel, which the Tensor G5 lacked. And in an early leaked GPU compute run, the Pixel 11 Pro XL scored around 80% higher than the Tensor G5. It seems the GPU is helped by more compute units, a higher clock speed and efficiency brought in by 2nm node.
So compared to the Tensor G5, raw graphics performance on the Tensor G6 may be ahead in spite of the older series. However, when compared to rivals, the Tensor G6 GPU is still significantly behind, as our Snapdragon 8 Elite Gen 5 benchmark highlighted.
Related Articles
Tensor G6 vs Tensor G5: AI and the TPU
In the AI department, Google plays to its strengths. The whole point of a Tensor chip is to power Google's on-device AI and the Tensor G6 carries it forward. It packs a fresh in-house TPU with 50% more compute and 2x the memory bandwidth of the Tensor G5's Edge TPU.
Google built the chip in line with DeepMind's newest models. The advantage is that routine on-device tasks such as notification summaries and voice transcripts are up to 3.5x faster and consume up to 3.5x lower power. The on-device Gemini Nano also responds with less lag on the phone itself.
When I tested on-device AI performance on Android, Google's tight software integration with the TPU felt more polished than rival chipsets, which featured more powerful NPUs on paper. And with more horsepower, the Tensor G6 TPU will offer even better AI experience on the Pixel 11 lineup.
Related Articles
Tensor G6 vs Tensor G5: Process Node, Modem and Efficiency
The move to TSMC's 2nm (N2) node is one of the biggest changes. The Tensor G5 was fabbed on the 3nm N3E process and the Tensor G6 brings Gate-All-Around transistors, built on 2nm node. It should improve efficiency and battery life and fix the heating issue.
By the way, the Tensor G6 is the first mass-produced 2nm smartphone chip, arriving even before Apple's A20 chipset, which is due in the iPhone 18 series this September. It's the first time Google is not behind Apple in fabrication process.
Next, the modem has seen another notable change. For the first time ever, Google has walked away from a Samsung Exynos modem, replacing the Exynos 5400 with MediaTek's M90 5G modem. The M90 modem can deliver peak download speeds up to 12 Gbps and reduces average power use by up to 18%.
For Pixel owners, this might be the single most important change of the lot. Exynos modems have long been blamed for heat and weak reception on Pixels, so if the M90 holds up, it will be a big relief. There is also a new Titan M3 security chip for stronger protection. You can find the full breakdown in our Tensor G6 spec sheet.
Related Articles
Tensor G6 vs Tensor G5: Specs Comparison
| Specification | Tensor G6 | Tensor G5 |
|---|---|---|
| Process Node | TSMC 2nm (N2) | TSMC 3nm (N3E) |
| CPU | 7-core Arm C1-series, Armv9.3-A | 8-core Arm Cortex |
| CPU Cores | 1x C1-Ultra at 4.11GHz, 4x C1-Pro at 3.38GHz, 2x C1-Pro at 2.65GHz | 1x Cortex-X4 at 3.78GHz, 5x Cortex-A725 at 3.05GHz, 2x Cortex-A520 at 2.25GHz |
| GPU | Imagination PowerVR CXTP-48-1536 (C-Series), 1.3GHz | Imagination PowerVR DXT-48-1536 (D-Series), no ray tracing |
| TPU | New Google TPU, 50% more compute, 2x memory bandwidth | Google Edge TPU |
| ISP | Upgraded ISP with a dedicated video portrait accelerator | New ISP, 4K video at 60fps |
| Security | Titan M3 security chip | Titan M2 security chip |
| Storage and Memory | LPDDR5X, UFS 4.0 | LPDDR5X, UFS 4.0 |
| Modem | MediaTek M90 5G, Up to 12Gbps peak download | Samsung Exynos 5400 5G |
| Connectivity | Wi-Fi 7, Bluetooth 6.0 | Wi-Fi 7, Bluetooth 6.0 |
Related Articles
Verdict
All in all, the Tensor G6 looks like a meaningful upgrade over the Tensor G5 and the Pixels have gained the most where it has historically struggled. The new C1-series cores are expected to uplift the CPU performance, the 2nm node improves efficiency and the MediaTek modem finally retires the Exynos modem.
The GPU is the only exception. While the Tensor G6 may be better than the Tensor G5 in graphics performance, it's far behind and can't match rival chipsets. For gaming and raw graphics, you should avoid this chipset.
On the whole, Google has put together a chip that should feel quick, remains cool and power its Gemini AI features better than any Pixel before it. We will run our full benchmark suite as soon as we get our hands on the Pixel 11 series so stay tuned. Meanwhile, you can check out how the new chip lines up against flagship rivals in our Tensor G6 vs Snapdragon 8 Elite Gen 5 comparison.
-356w.webp "Realme 16x 5G")
