Upcoming Intel Apollo Lake could disappoint going by Broxton Goldmont benchmarks

intel-broxton-apollo-lake-benchmarkIntel’s upcoming Apollo Lake low-end tablet and PC processor could trail ARM-based chips by a big margin if a benchmarking result from a product using similar technology is any indication.

Goldmont, the core inside Apollo Lake is to be the first major update to Intel’s Silvermont architecture and was supposed to take on ARM chips.

When first unveiled in 2013, the Silvermont architecture was more or less comparable in power to high-end offerings from the ARM ecosystem, though it consumed more battery juice.

Since then, ARM processors have increased their processing muscle by around 150%, going from a score of around 1,000 points per core in Geekbench to 2,500 points for the latest designs such as Cortex A72 and Kryo.

In contrast, and going by the only available benchmark score, Intel may be looking at a performance increase of only around 30% as it moves from Silvermont to Goldmont — a woefully inadequate improvement compared with what is happening on the ARM side.

As of now, the only available benchmark score of Goldmont — the core inside Apollo Lake — belongs to a system bearing the name ‘Intel Corp Broxton P‘. It shows a single-core score of 1,301 and a quad-core score of 4,385.

In comparison, the Snapdragon 820 from the ARM family scores close to 2,500 in single-core mode and around 5,700 in quad-core mode. Both chipsets are made on the 14 nm process and are running at comparable clockspeeds.

While the Broxton unit was running at 2.3 GHz, the Snapdragon 820 chipset has a max clockspeed of 2.2 GHz.

Similarly, another ARM-based competitor, Helio X25 (MT6797 Turbo) scores around 2050 in single-core mode and a massive 6750 in all-core mode on Geekbench. Remember that MediaTek is still on the 20 nm process, which hampers the performance of this chipset.


Three factors must be kept in mind while making this comparison between Apollo Lake and its ARM competitors. First, we are using Goldmont benchmarking numbers from Broxton — the smartphone and tablet avatar of Goldmont — and not directly from Apollo Lake — which is targeted at cheaper PCs and two-in-ones.

Second, we have only one test sample for Goldmont, compared to picking the best of hundreds of tests for rivals. It is likely that a more optimized configuration would see higher scores of up to 1,600 points per core.

Third, Apollo Lake supports the X86 instruction set, which means that all the software that are currently designed for PCs will run on tablets, two-in-ones and PCs built on the platform.  This is not the case with two-in-ones and laptops that may potentially be built around ARM chipsets. They will require software that are specifically compiled for the architecture, though increasingly this is not a problem.


So, does all mean that Intel is being left behind in the mobile space? Has Intel finally realized that it simply cannot take on ARM when it comes to computing on a budget (both battery- and price-wise)? Is that why Intel killed the smartphone- and tablet-oriented Goldmont architectures?

The answer is a No.

What seems to have happened is that the Goldmont project seems to be a failure — both in terms of the power that the chip is able to deliver as well as in being ready on time.

With Geekbench scores of 1,300 points, Intel seems to have very wisely decided that there was no way it was going to compete with the Snapdragons and Helios of the world.

However, instead of killing the project completely, it seems to have decided to let Goldmont live on in an area where its X86 compatibility gives it a distinct advantage — in the two-in-one and cheap PC market.


So does this mean Intel is beat squarely in the smartphone and tablet market?

Not really, and the reason lies in Intel’s Core M processor series, which were spun off two years from Intel’s experiments with low-voltage ‘U’ series versions of its mainstream processors.

While U series chips typically consumed between 15 and 30 watts, the Core M series had an upper limit of 4.5 watts, which is less than the projected power consumption of Apollo Lake (6-10 W).


Core M was designed with the ultralight laptop and high-performance tablets in mind and brought with it the power of the mainstream Core i series while being as efficient in battery use as possible. Though they can turbo as high as 3.1 GHz, their typical operating frequency is around the 1 GHz mark — the primary reason for their low power consumption.

The chips perform well compared to the ARM competitors.

For example, the Core m5-6y54 scores about 3,000 points in single-core mode and around 5,900 when both cores are operating. These numbers are clearly superior to the 2,500 points posted by Snapdragon 820.

However, Intel is not quite there when it comes to a crucial requirement for phones — low power consumption.

The Core m5-6y54 is designed with a 4.5 Watt power consumption in mind. Even the biggest phones these days come batteries with capacities no higher than with 15 watt-hour and a Core m5 smartphone would last for only around 2.5 hours of active use on such a battery.

In comparison, a similar phone built around Snapdragon 820 lasts twice as long.

We could see the gap narrowing to about 3.5 hours vs 5 hours with the latest Kaby Lake Core M chips, but the real competition will come the introduction of 10-nm Cannonlake line up next year. Till then, ARM would continue to dominate the smartphone market.