Between high-end gaming, 120Hz displays, and the growth of machine learning, graphics performance and fidelity are increasingly important in the smartphone market. Arm has unveiled two new GPUs for us this year, the high-performance Mali-G78 and the new sub-premium tier Mali-G68.

Both are based on Arm’s Valhall micro-architecture found in the Mali-G77 and Mali-G57, but with a few key improvements. The Mali-G78 aims to push the performance envelope with support for more cores, up from 16 to 24 this generation. The Mali-G68 is designed to slot right into a gap in the market, providing the same high-end features but with a lower cost and up to just six GPU cores. Let’s dive into what’s new and what to expect from next-gen SoCs packing Arm’s latest graphics technology.

More from Arm: Cortex-A78 and Cortex-X1 CPUs explained

Arm Mali-G78 – Valhall gets flexible

Arm Mali-G78

Just like previous Arm GPUs, the Mali-G78 scales in performance and power efficiency depending on the number of shader cores. The Mali-G78 scales from seven to 24 cores, allowing for larger deployments than the previous generation. In theory, this could help Mali close the performance gap on its chief competitor, Qualcomm’s Adreno.

Let’s start out with the prospective performance numbers, although these will vary depending on the exact core configuration and clock speeds. The headline number is a 25% performance improvement, owing to the new design and an anticipated move to 5nm manufacturing next generation. That’s pretty good, but bear in mind that other GPUs will also gain from the move to 5nm.

Arm Mali-G78 improvements

Moving to a more direct process comparison with the G77, the newer GPU offers a 15% improvement to performance density. That means 15% more performance for a given silicon area by switching to G78 cores over G77. The substitution also promises a 10% improvement to energy efficiency and a 15% boost to machine learning performance. Not bad.

Diving deeper into Arm’s numbers, gaming performance improves between 6% and 17% compared to the previous generation. What’s also quite revealing is how these cores scale in performance. Moving from an 18- to 24-core setup yields a further 11% improvement, but that’s a 33% increase in core count and a sizable chunk of silicon area. There are clearly still diminishing returns with more cores. That said, we’re not anticipating such large configurations in smartphones, which will likely target the sweet spot around 10 to 12 cores.

What makes the Mali-G78 tick

The Mali-G78 is built on the Valhall architecture that made its debut last generation, complete with the new execution engine and warp thread execution model. One of the keys to boosting performance this generation lies in the new Asynchronous Top Level. In a nutshell, this now allows for separate Top Level and Shader Core frequency clocks and voltage domains. The Top Level houses the Control Fabric, L2 Cache memories, and the Tiler, while the Shader Cores crunch most of the numbers. Arm envisions the Top Level now running at 2x the Shader Core frequency for faster texture and geometry processing. Although this multiplier value is up to Arm’s partners.

Arm Mali-G78 Asynchronous Top Level

The biggest benefit of the Asynchronous Top Level is that the higher bandwidth keeps the Shader Cores better fed with things to do. This means that the Mali-G78 can scale up to larger core numbers for increased peak performance. Alternatively, a larger number of cores can run at a lower frequency for improved energy efficiency. Although depending on the use case, chip designs may or may not want to take the silicon area hit from an increased core count.

The new Asynchronous Top Level runs at a higher frequency than the Shader Cores, keeping them better supplied with numbers to crunch.

The Mali-G78 offers a few other tweaks and performance improvements. The execution engine’s FMA unit has been completely reworked to consume 30% less power while providing a 15% uplift in machine learning workloads. This is partly achieved by separating FP16 and FP32 paths, with each unit aggressively power gated when not in use. The Mali-G78 boasts increased Tiler throughput with an 8% average reduction in vertex shading and a 2% boost to execution time. Improvements to invalidation tracking in the Shader Cores shave 22% off the internal bandwidth too.

Although Mali-G78 shares the bulk of the same internals as the Mali-G77, a thorough rework of key parts elevates performance and energy efficiency this generation.

Arm Mali-G68: A new performance tier

Arm Mali G68 Graphics Processor Roadmap

In addition to the high-end Mali-G78, Arm announced the Mali-G68. This is the company’s first GPU targeted at the “sub-premium” market, slotting between its high performance and mainstream tiers. The Mali-G68 accelerates the arrival of high-tier features to more affordable price points.

The Mali-G68 retains the FMA, dependency tracking, and tiler improvements from the G78. It also boasts the Asynchronous Top Level clock speed control, making it a feature match for its high-end sibling. The biggest difference between the two appears to be that the Mali-G68 only scales between one and six cores.

Mali-G68 scales to just 6 cores, while the G78 can hit 24.

While clearly not targeting the top end of the market, the Mali-G68 still provides the performance and features required to play the latest mobile games. A smaller silicon area for a lower price point could make this GPU a compelling choice for mid-tier SoCs. In the future, Arm anticipates that its sub-premium tier may diversify a little more from its high-performance roadmap.

What to expect in 2021 smartphones

spacex docking simulator game

It’s a little too early to make any exact performance estimations for 2021 smartphones. Especially given the core counts and clock speeds are highly configurable, so we’ll have to wait for chip announcements. Even so, the prospect of a 25% performance improvement is promising and improved scalability of multiple cores should help Mali close the gap on its competitors and remain a viable GPU. However, the bigger picture depends on what sort of performance gains its GPU rivals have in store for next-gen.

Either way, Mali is a regular fixture in Huawei, MediaTek, and Samsung mobile SoCs and we’re almost certainly going to see the Mali-G78 appear in at least one chip from these companies. Although we’ll have to see how far along Samsung and AMD’s plans are for its alternative GPU. The Mali-G68 could have some serious appeal for more affordable price points too, given the appeal of dedicated gaming chipsets like the Snapdragon 730G and 765G.

Next: Is Qualcomm working on a Snapdragon 865 Plus processor?

Comments
Read comments