News

Arm’s Total Compute: Diving Into Its New Mali GPUs

June 01, 2021 by Jake Hertz

Recently, Arm released its Total Compute solutions which included four new GPU offerings. Here is a deeper look into what each one is offering.

Last week, Arm made headlines with its new Total Compute solutions platform, which marked the company's most significant release yet.

This release had so much to dive into; it was too much for just one article. Here is the article on the overall platform, and this is a more in-depth look at the CPUs

As for the newly announced GPUs, Arm appears to be doubling down on its system-level approach with a new suite of Mali GPUs to complement the Armv9 CPUs in the Total Compute package. 

In this final installment on Total Compute, this article will look deeper at these four new GPU offerings.

 

Arm's Mali-G710

The flagship GPU in the release is the Arm Mali-G710, which Arm is calling its highest performing GPU. 

Targeting the premium smartphone market, the G710  claims to have a 20% increase in energy efficiency, 20% in performance, and 35% in ML applications compared to the previous generation Mali-G78

These impressive improvements seem to come from an array of new technical features, including a new command stream fronted (CSF) and a redesigned Shader Core.

 

The new CSF from Arm replaces the Mali job manager. Image from Arm.

 

The new CSF integrated into the G710 marks a significant change as it replaces the conventional Mali job manager in previous Mali GPUs. 

The CSF aligns the GPU to support modern APIs, supporting a draw call rate of 5 million draw calls per second which serves to offload a tremendous amount of work from the CPU to the GPU. 

Mali-G710's new Shader Core has also been heavily redesigned. As opposed to G78, which was configurable up to 24 cores, G710 opts for 7-16, larger and more performant cores. 

Arm also claims the execution engine has been redesigned for energy efficiency improvements, with a second execution engine added to each shader core, doubling the compute capability of each core, allowing for more efficient use of shared resources.

 

Arm's Mali-G610

Following the G710 is the new Mali-G610, a GPU meant to be a sub-premium version of the G710. 

For a sub-premium smartphone market, energy efficiency and cost tend to be more important than performance for this GPU. The G610 inherits some of the improvements offered by G610, including the new CSF. However, it offers less performance due to less configurable cores––up to 6 on the G610 instead of 16 on the G710. 

Despite the initial lackluster of this GPU, especially following the G710, Arm appears to be trying to cover its bases by creating GPUs based on various design needs of its consumers. This becomes even more apparent when looking at its CPU offerings and its Total Compute platform as a whole. 

 

Arm's Mali-G510 & G310

To round out the release of its premium GPUs, Arm has also released the Mali-G510 and Mali-G310. These GPUs claim to be for entry and mid-level applications, providing performance and efficiency. 

According to Arm, both GPUs bring significant performance improvements compared to the previous generations while offering new features that reduce bandwidth for performance and power consumption. 

The Mali-G510 was designed to deliver a strong balance between performance and efficiency, claiming to offer a 100% performance improvement, 22% energy savings, and 100% ML uplift compared to the previous generation of Mali-G57

On the other hand, the Mali-G310 appears to have a massive increase over the Mali-G31 generation with a 6x improvement in texturing performance, 4.5x improvement in Vulkan performance, and 2x improvement in Android UI content.

 

G310 offers major performance improvements over the previous G31. Image from Arm.

 

The stated performance improvements for both of these GPUs result from the inherited premium features from G710 (such as CSF and new Shader Cores). These GPUs are then optimized for different performance, power, and area numbers. 

The G510 also marks the introduction of Arm's visually lossless fixed-rate compression (AFRC) into the market for the first time. The new AFRC technology is the first time Arm has offered lossless compression that guarantees a bandwidth and memory footprint reduction. 

This reduction results in performance increases and energy savings due to less data being read and written to DRAM––a feat that reduces the amount of DRAM needed in the system.

 

Final Thoughts on Total Compute

With a deeper look at each GPU offering from Arm Total Compute, Arm seems to be pushing its technology forward to cater to the needs of new technologies and applications. Their Total Compute GPU suite not only covers a variety of market segments, from premium to entry-level, but it also declares significant improvements in performance, power, and ML to each segment. 

After a deeper look, Total Compute seems to signal a strong future for Arm and its Armv9 architecture. These new offerings could boost the next generation of tech with these proposed advancements in GPUs and CPUs. 

 

Featured image used courtesy of Arm.

1 Comment
  • jpthing June 01, 2021

    ARM is now owned by NVIDIA. What are the chances they are going to continue the support for Mali, an independently developed GPU solution? Mali was bought up by ARM mostly to provide a complete SoC solution. Now it seems redundant.

    Like. Reply