This article is more than 1 year old
Arm says its Cortex-X3 CPU smokes this Intel laptop silicon
Chip design house reveals brains of what might be your next ultralight notebook
Arm has at least one of Intel's more capable mainstream laptop processors in mind with its Cortex-X3 CPU design.
The British outfit said the X3, revealed Tuesday alongside other CPU and GPU blueprints, is expected to provide an estimated 34 percent higher peak performance than a performance core in Intel's upper mid-range Core i7-1260P processor from this year.
Arm came to that conclusion, mind you, after running the SPECRate2017_int_base single-threaded benchmark in a simulation of its CPU core design clocked at an equivalent to 3.6GHz with 1MB of L2 and 16MB of L3 cache.
In effect, Arm believes its Cortex-X3 cores can outperform the P cores of the Core i7 when running common-or-garden application code. That 28-Watt Intel part, aimed at thin and light laptops, has four performance cores and eight efficiency cores, and a competing real-world Arm-based system-on-chip – whenever or if ever it arrives – would need multiple Cortex cores.
The reveal of the X3, which succeeds last year's X2, comes as Arm hopes to gain a greater foothold in the laptop space through silicon partners who license Arm's technologies to create chips that compete against x86-based processors from Intel and AMD.
Arm jumps on ray tracing bandwagon with beefy GPU designREAD MORE
In addition to laptops, the Armv9 X3 is also aimed at high-end Android smartphones, where Arm expects to provide 25 percent faster performance than flagship X2-based devices, according to estimates based on simulations that averaged results from the SPECRate2017_int_base, SPECInt_base2006, and Geekbench 5 benchmarks.
Arm said the 64-bit-only X3's advances are made possible in part by an 11 percent uplift in the instructions per clock (IPC) over the X2, marking the second consecutive year of double-digit IPC growth for the Cortex-X line. This makes the X3 the "highest-performing CPU" in Arm's portfolio.
"It's about the best user experience, whether you have an application where peak performance is needed or the perceived user response time is very critical," said Chris Abernathy, a lead CPU architect and fellow in Arm's central engineering organization, in a briefing with journalists.
Arm plans to offer the X3 design through its Total Compute Solutions program, which provides a collection of "purpose-built" chip design configurations that pull together various technologies, including its expanding portfolio of GPU designs.
That program will also offer a new 64-bit-only Armv9 Cortex-A715 CPU design, which Arm has deemed the "workhorse" among its set of chip blueprints for high-performance devices. Designed with a focus on delivering high sustained clock speeds, the A715 is estimated to provide 20 percent better power efficiency and five percent faster performance than the A710 design introduced last year.
To balance things out, Arm refreshed last year's Armv9 Cortex-A510 Little core to provide five percent greater efficiency while delivering the same performance. The earlier A510 was 64-bit-only; the refreshed one can optionally handle 32-bit code in case that support is needed in a system-on-chip.
We're told these CPUs feature a second-generation Armv9 architecture that brings in an Asymmetric Memory Tagging Extension that is seemingly useful for catching exploitable memory corruption bugs, such as buffer overflows and heap overwrites; and Enhanced Privileged Access Never, or EPAN, which is a step up from PAN that prevents the OS from accidentally accessing unexpected unprivileged data.
Arm wants to use cash from planned IPO to fuel growthMEANWHILE
When the X3, A715, and A510 cores are combined in a processor, Arm said, they can can provide double-digit increases in multi-threaded performance compared to configurations using last year's X2, A710, and A510 cores. This multi-threaded boost can range from 12 percent to 23 percent depending on the core arrangement, which is based on performance estimates for the Geekbench 5 MT benchmark.
And, according to Arm, if you were to go to eight X3 cores and four A715 cores, multi-threaded workloads would be boosted by 125 percent compared to a processor using one X2 core, three A710 cores, and four A510 cores from last year, based on a simulation of the design.
And yes, that's 12 cores versus eight, so a performance boost would be expected. The eight-core arrangement would be aimed at things like smart TVs, and the 12-core at things like laptops and PCs. The DynamIQ Shared Unit, or DSU, Arm's glue logic for its CPU core clusters, has, in line with this, been updated to support up to 12 of these latest cores and 16MB of L3 cache.
"We are offering many ways for our partners to innovate and, in some cases, reach new, previously unimagined levels of performance," said Saurabh Pradhan, director of product management at Arm.
Pradhan said these simulations of core configurations is helping Arm uncover new layouts that can benefit devices of varying requirements.
"For example, we noticed that especially in very thermally constrained environments, the Cortex-A700 series is actually a better ROI for sustained workloads," he said. "Now, this is a statement for one specific benchmark. But the goal here is to make sure that we provide a solid portfolio for our partners so that they can pick and choose what they want." ®