Skip to content

Product1 publisher3 min readPublished

Qualcomm's own numbers give the Extreme Gen 6 a 44 percent GPU gain and 12 percent in single-core

Qualcomm has published first-party benchmarks for its new top Snapdragon tier, measured on its own reference hardware. The generational gain it reports depends almost entirely on which part of the chip a workload leans on.

The Product Desk · Product desk

Photograph accompanying Qualcomm's own numbers give the Extreme Gen 6 a 44 percent GPU gain and 12 percent in single-core
Photo: tomsguide.com

What happened

  • Qualcomm has published its expected benchmark figures for the newly announced Snapdragon 8 Elite Extreme Gen 6, set against last year's Snapdragon 8 Elite Gen 5.
  • The Extreme part sits above a regular Snapdragon 8 Elite Gen 6 in a new two-chip hierarchy, built on a 2 nm process with a reworked Adreno GPU and two prime cores rated to 5 GHz.
  • Qualcomm's official claims for the generation are a 13 percent CPU improvement, a 44 percent GPU increase and a 35 percent faster Hexagon NPU.

Compiled by The Product DeskSomething wrong?How this is made

Why it matters

  • decision A team deciding whether to build an Extreme-tier rendering path gets a very different answer from one optimising CPU-bound code, and the same deck supports both cases.
  • contradiction Whether this CPU is 12 percent or 34 percent quicker than last year's depends on whether the Gen 5 baseline is Qualcomm's bench or a shipping Galaxy S25 Ultra, and Notebookcheck endorses only the first.
  • constraint Anyone wanting to check the AI claims against last year's silicon has to wait for someone to run the new suite versions on a Gen 5 part.
  • capability Upscaling at a fixed frame rate target cuts draw to under a fifth of native, so a thermally limited phone can hold that target for longer.

An engineer setting the minimum device target for next autumn's build has to work out which of Qualcomm's figures describes their bottleneck. The 44 percent GPU claim [5] is about 3.7 times the single-core gain in Qualcomm's own Geekbench 6.5 results [17] and about 7.5 times the multi-core gain of 5.9 percent [18][7]. Both sides of that Geekbench comparison ran on Qualcomm reference platforms [2]. Qualcomm's stated 13 percent CPU improvement sits above both of its own Geekbench measures [23].

One CPU-side result moved more. Speedometer 3.1 in Chrome v139 went from 48 to 49 on the Gen 5 to 59.5 to 60.4 on the Extreme Gen 6, a 23.6 percent improvement [8]. Browser responsiveness is closer to what most phone sessions consist of than multi-core throughput is, so for app UI and web views the picture is better than 5.9 percent suggests. Notebookcheck says both generations ran the same 12-model methodology in that test [8].

The biggest generational number in the coverage comes from a comparison Notebookcheck warns against. Geekbench 7 puts the Extreme Gen 6 at 3,599 to 3,624 single-core [11] against roughly 2,702 for a Galaxy S25 Ultra that Primate Labs tested [12], or about 34 percent ahead, and Notebookcheck says this should not be treated as a strict apples-to-apples generational uplift because the Gen 5 figures came from a retail phone and the new ones are expected results from a reference platform [13]. The same silicon therefore carries a 22.1 point spread in its headline CPU gain depending on which baseline is used [19].

Three of the suites in the package changed major version between generations [20]. AnTuTu moved from v11 to v12, so the Extreme Gen 6's 5,606,990 to 5,619,756 cannot be set against the Gen 5's 4.25 to 4.5 million [10]. AIMark went from V2 to V3 and MLPerf from V2 to V6, and Notebookcheck says Qualcomm's AI figures, including 3,659,031 in AITutu-LLM V3 and 835,052 in AIMark VLM V3, do not all allow direct generational comparisons [14]. The AI comparison that does survive is AITutu-CV V3, where a Gen 5 reference platform scored around 2.5 million and the new chip reaches 2,903,533, about 16 percent [9].

The Traverse Evolve result measures a feature developers have to switch on. Qualcomm reports 17.8 FPS natively and 52.2 FPS with upscaling and frame generation enabled, which it calculates as a 193.3 percent increase, in a test built to show its Adreno Neural Fusion technology [15]. Native 17.8 FPS on a reference platform is the figure a game team has to plan around. Qualcomm also claims that holding a fixed 10 FPS target drops power from 5.6 W to 1.0 W, an 81 percent reduction [16]; the rounded figures give 82.1 percent [21].

Two questions sort the package. Did both sides of a comparison come from the same class of hardware, and did they run the same version of the suite? Four comparisons pass both, and they run from 5.9 to 23.6 percent: Geekbench 6.5 single-core, Geekbench 6.5 multi-core, Speedometer 3.1 and AITutu-CV V3 [25]. Then ask which of those four resembles the session your users actually have. Notebookcheck notes the figures come from Qualcomm reference hardware, so real-world performance will depend on cooling, power limits and individual OEM implementations [2].

What to watch

  • Retail Geekbench 6.5 runs from shipping Extreme Gen 6 handsets: how much of the 11.9 percent survives OEM cooling and power limits.
  • Anyone publishing AnTuTu v12 or MLPerf v6 numbers for a Snapdragon 8 Elite Gen 5 part, which would restore the AI comparison.
  • Figures for the regular Snapdragon 8 Elite Gen 6, which set what the step up to the Extreme part is worth.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories