Skip to content

Build1 publisher2 min readPublished

Apple's A20 Pro reportedly hits 4,006 single-thread in early Geekbench 7 test

One early Geekbench 7 submission reported by Tom's Hardware puts the six-core chip 32% ahead of Intel's Core i9-14900KS in single-thread, while its multi-thread total works out to 2.86 times that score.

The Engineer · Build desk

Photograph accompanying Apple's A20 Pro reportedly hits 4,006 single-thread in early Geekbench 7 test
Photo: apple.com

What happened

  • An early Geekbench 7 submission for Apple's A20 Pro reports 4,006 points in single-thread and 11,460 in multi-thread, the highest single-thread result recorded for the benchmark.
  • The six-core layout pairs two desktop-class performance cores with four efficiency cores, alongside a memory interface Tom's Hardware reports as delivering 50% more bandwidth than the A19 Pro.
  • On single-thread the chip leads AMD's 16-core Ryzen 9 9950X3D by 26% and Intel's Core i9-14900KS by 32%, while both desktop parts stay dramatically faster in multi-threaded work.
  • Tom's Hardware cautions that the submission may or may not demonstrate the performance of actual A20 Pro-based products.
  • Qualcomm's Snapdragon 8 Elite Gen5 and Xiaomi's XRing O3 trail by 31.5% to 33.7% in single-thread, putting both roughly level with Apple's two-year-old A18 Pro.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

  • constraint Teams moving parallel work onto the phone should budget under three times single-thread throughput, because the extra workers land on four efficiency cores rather than four more big ones.
  • decision Choosing between on-device and server inference turns on memory bandwidth, and the interface credited with the 50% gain is reported only as alleged, so this run cannot settle the choice.
  • precedent The next Snapdragon and XRing parts will be measured against 4,006 single-thread, a bar that one unverified submission has set for them.

Multi-thread lands at 2.86 times single-thread [14], out of two performance cores and four efficiency cores [4]. A thread pool here does not scale the way a desktop core count does. The submission shows that at the top of the mobile field: 12.2% ahead of the eight-core Snapdragon 8 Elite Gen5 in multi-thread, and roughly level with the 10-core XRing O3 [8].

That single-thread record was set on the benchmark's own workload. It transfers to your code if the hot path is genuinely one thread, and if it finishes while the two big cores are still at the 4.93 GHz peak the submission reports [4].

Neither condition describes local model inference. I would size an on-device model against memory bandwidth before single-thread score. The bandwidth here is given as 50% above the A19 Pro, produced by a memory interface Tom's Hardware describes as allegedly 96-bit [4]. That interface is unconfirmed, and it is the number the sizing decision turns on.

The generational figure is the more interesting one. When Apple moved the A17 Pro from N4 to N3B in 2023, single-thread went up 9.8% [12]. This N2 part reports 23.3% [2], roughly 2.4 times that gain [15], and the core count is still six, as it has been for many generations [19]. Working backwards from those deltas, the A19 Pro sits at about 3,249 single-thread and about 9,016 multi-thread [16].

Against Apple's own laptop silicon, the same run puts the phone 7% ahead of the M5, 20% ahead of the M4 and 43% ahead of the M3 in single-thread, while trailing the eight-core M3 by 5% in multi-thread [10]. Set beside Intel's Panther Lake, it is 48.7% faster than the Core Ultra X9 388H on one thread, and the 16-core part is 63% faster on many [11].

For a team planning mobile-first work, latency on a single hot path is now a desktop-class number in a phone. Anything that spawns workers or streams weights is bounded by six cores and by a memory interface reported second-hand. On Android the distance is wider: the same comparison puts Google's Tensor G5 99% behind in single-thread and 96% behind in multi-thread [9].

What to watch

  • Whether retail iPhone units reproduce 4,006 and 11,460, given Tom's Hardware's caveat about the early submission.
  • Confirmation of the 96-bit memory interface and the 50% bandwidth gain from a source other than an early benchmark run.
  • Whether Qualcomm's next flagship closes the 31.5% to 33.7% single-thread gap to the A20 Pro.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories