Along with today's announcement of the new Cortex-A77 CPU microarchitecture, perhaps the bigger announcement is the unveiling of the new Valhall GPU architecture and the new Mali G77 GPU by arm. It's been three years since the bifrost architecture was introduced, and as the industry and workload evolve, the company's GPUs must evolve.
Valhall and the new Mali-G77 are adding significant improvements in performance, density and efficiency to Mali's last three GPU generations. While the G76 made some major changes to the execution engine's computational architecture last year, the G77 goes much further and deviates from Arm's relatively unusual computational core design.
Looking Back on Bifrost ̵
1; The Third Time Is The Charm
It's It's not too big a secret that Arm's GPU IP offerings have not been particularly well received in recent years. When the first Bifrost GPU – the Mali G71 – was announced in 2016 and later manufactured in the same year in the Kirin 960 and Exynos 8895 models, we had expected good performance and efficiency gains.
Bifrost was Arm's first scalar GPU architecture, the previous generation vector command design (Midgard: T-600, 700 & 800 series). The change was fundamental and was similar to the introduction of desktop GPU vendors such as AMD and Nvidia with their new GCN and Tesla architectures in the last decade.
Unfortunately, the first two generations of Bifrost, the Mali G71 and the subsequent G72, were not very good GPUs. The two leading licensees of Arm, HiSilicon and Samsung have achieved quite disappointing SoCs on the GPUs of these two generations. In particular, the Kirin 960 and 970 were extremely poor in this regard, and I would argue that this had a significant impact on Huawei's and Honor's product planning and marketing.
| GFXBench Manhattan 3.1 Off-screen Energy Efficiency
|Mfc. Process||FPS|| Avg. Performance
| Perf / W
|iPhone XS (A12) Warm||7FF||76.51||3.79||20.18 fps / W|
|iPhone XS (A12) Cold / Tip||7FF||103.83||5.98||17.36 fps / W|
|Galaxy 10+ (Snapdragon 855)||7FF||70.67 [196590154881446fps/W|
|Galaxy 10+ (Exynos 9820)||8LPP||68.87||5 , 10||13.48 fps / W|
|Galaxy S9 + (Snapdragon 845)||10LPP||61.16||5.01||11.99 fps / W|
|Huawei Mate 20 Pro (Kirin 9805) 19659||54.54||4.57||11.93 fps / W|
|Galaxy S9 (Exynos 9810)||10 LPP  46.04||4.08||11.28 fps / Sb 835)||10LPE||38.90||3.79||10.26 fps / W|
|LeEco Le Pro3 (Snapdragon 821)||14LPP||33.04||4.18||7.90 fps / W|
|Galaxy S7 (Snapdragon 820)||14 LPP||30.98||3.98||7 , 78 fps / W|
|Huawei Mate 10 (Kirinps / W 19659015) 10FF||37.66||6.33||5.94 fps / W|
|Galaxy S8 (Exynos 8895)  10LPE||42.49||7.35 [19659015/5S7(Exynos8890)||14LPP||29.41||5.95||4.94 fps / W|
|Meizu PRO 5 (Exynos 7420)||14659016PE||14.45 [19659fps/W|
|Nexus 6P (Snapdragon 810 v2.1)||20Soc||21.94||5.44||4.03 fps / W|
|Huawei Mate 8 (Kirin 950) [1965901516FF] 10.37||2.75||3.77 fps / W|
|Huawei Mate 9 (Kirin 960)||16FFC||32 , 49||8.63||3.77 fps / W|
|Huawei P 959)||16FF +||10.59||2.98||3.55 fps / W|
The last iteration of bifrost architecture, the Mali -G76, a was a major leap for Arm and the IP was largely able to solve some of the critical problems of its predecessors, resulting in relatively good results for the Exynos 9820 and Kirin 980 chipsets.
Regrettably, Arm Bifrost's problems were resolved and corrected. The competition not only kept quiet and squeezed the envelope. Qualcomm's Adreno GPU architecture has been a leader in the mobile landscape for several years. While the Adreno 640 has not made quite as impressive a year's improvements, it still leads the pack in terms of performance, efficiency and density. Even more worrying is the fact that Apple's GPU was an absolute leap in performance in the A12, massively outperforming Qualcomm's best GPUs, not to mention Arm's own Mali GPUs.
Introducing Valhall – A major revamp  Today we'll look at Arm's brand new GPU architecture: Valhall (Anglicised version of the old Norse Valhoel, also known as Valhalla). The new architecture includes a brand new ISA and compute core design that seeks to address the major flaws in the bifrost architecture, and seems more similar to the design approaches used by other GPU vendors.
The first iteration of the Valhall GPU is the new Mali-G77, which implements all the architectural and microarchitectural enhancements that we will discuss today.
What is promised is a 30% increase in energy efficiency and areal density (in ISO performance and process) and an increase in machine learning performance to derive workloads on the GPU.
Interestingly, SoCs are expected to see a 40% increase in performance by 2019 and 2020 compared to 2019. For next-generation SoCs, only minor improvements to the process nodes are predicted. Most of the improvements mentioned here are due to the architectural and microarchitectural leaps achieved with the new Mali G77 GPU.