Arm’s new Immortalis G720 should yield big gains in performance and efficiency

Arm has announced the Immortalis G720, Mali G720, and Mali G620, and they look super promising.

Arm-5th-Gen-GPU-Feature

Alongside the launch of Arm’s new Total Compute Solution for 2023 in the form of the Cortex-X4, A720, and A520, we’re also getting a look at Arm’s GPUs for the next year too. This includes a new Immortalis G720, a Mali G720, and a Mali G620. These are Arm’s 5th Generation GPUs, and that’s also the new naming convention for Arm GPUs. No more “Valhall” or “Bifrost”; it’s just “5th Gen.”

Each of these GPUs are relatively straightforward increments over their predecessors, with the Immortalis G720 being the best GPU from Arm yet. Like in their cores, the GPUs here are particularly being touted for their efficiency improvements, though there are some pretty big performance gains to be had as well. As Arm puts it, a third of the power draw from high-power gaming comes from memory access, and memory bandwidth usage has been decreased by up to 40%.

With how good the Immortalis G715 was last year, it’s clear that Arm is really starting to become a formidable contender when it comes to GPUs.

Distinguishing between Immortalis and Mali

Arm-Immortalis-G720-GPU-Layout

Before diving into Arm’s new GPUs, it’s important to distinguish where the differences are in Immortalis and Mali, particularly as the lines have been somewhat blurred giving the shared part names. When I asked Arm, I was told that when OEMs equip their chipsets with Arm GPUs, GPUs marketed as Immortalis must have a ray-tracing unit, whereas the Mali G720 can but is not required to.

On top of that, the Mali G720 can have between six and nine cores, whereas the Immortalis G720 can have up to 16 cores. As for the Mali G620, it’s limited to up to five cores, though it’s also a lot more of a budget GPU.

Arm’s Immortalis G720 is its best GPU yet

Arm-Immortalis-Efficient-GPUs

The big revolution of Arm’s newest GPUs is Deferred Vertex Shading, or DVS. Arm has already been leading in Variable Rate Shading and ray-tracing, and the next step for improvement was in bandwidth and power consumption. As a result of these improvements, Arm touts up to 15% more performance per watt, along with an average of a 15% increase in peak performance.

Arm-Immortalis-DVS

In real-world performance, Arm has measured up to a 20% performance improvement over the Immortalis G715, where a large part of that can be attributed to the bandwidth savings from DVS. Because of these bandwidth savings, PC-level effects such as real-time dynamic lighting, blooming, depth of field, and screen space ambient occlusion can be added. In fact, Unreal Engine 5’s desktop renderer will be coming to Android later this year as a result of this.

Arm-Immortalis-DVS-Explanation

Deferred vertex shading is the “headline feature” of the G720, and it allows the renderer to defer the shading of vertices until (per-tile) fragment shading. The order is still maintained in the polygon list, and framerates increase (by 15% on average) compared to G715, meaning that the user can benefit from reduced bandwidth usage thanks to the rendering pipeline not needing to store intermediate vertex data in memory. Arm also assured me that this doesn’t affect frametimes either, so you’ll have a consistent gaming experience. This is built into the GPU, so developers won’t need to do anything to incorporate its usage.

In turn, while Arm did not share any specific numbers, this will also benefit performance when using ray-tracing. Given Arm’s lead in this field already with its Immortalis GPUs, this will likely only further the gap unless competitors make massive leaps. Arm also added a 2x MSAA module, as previously when a developer would request 2x MSAA from the GPU, it would automatically jump to 4x MSAA.

Arm-G720-Throughput

Other improvements were made in the GPU too, but these are smaller in scale. The command stream front-end (CSF) saw some improvements with extra working registers for streaming processing, and other units across the board saw improvements in their throughput. The CSF handles scheduling and draw-calls with its own computational capabilities.

At the same power constraints as last year’s GPU, Arm says that you’ll have a more powerful GPU, which scales down to a more efficient GPU for the same amount of power.

Like with its cores, GPU efficiency is a big deal

Arm seems to be big on efficiency this year, and with its GPUs, it’s the exact same. That’s not to say better performance isn’t a goal, as evidently, it is (and the company is still somewhat playing catch up), but efficiency is important to most people. While the potential for performance gains is extremely large this time around, most people will probably hope for better battery life overall.

Having said that, those 40% bandwidth improvements can’t be understated, and a 15% improvement in average framerate is nothing to scoff at. We’ll be excited to see what chipsets may incorporate this GPU going forward, and how it compares to upcoming GPUs in chipsets like the Snapdragon 8 Gen 3 later this year.