By IG Share Share Mobile silicon has hit a hardware ceiling. For years, simply adding more cores to the Mali series resulted in diminishing returns due to the “Bandwidth Wall”—where system memory couldn’t feed the shaders fast enough. Note: If you buy something from our links, we might earn a commission. See our disclosure statement. The shift from the brute-force Mali-G710 to the efficiency-focused Immortalis-G720 marked the first real solution: Deferred Vertex Shading (DVS). Now, the Immortalis-G925 pushes further with Fragment Prepass and desktop-class Ray Tracing logic. This report examines the silicon-level changes defining the Dimensity 9400 and 9300, stripping away marketing terms to measure raw rasterization, thermal sustainability, and the real impact of moving from software ray queries to hardware Lumen support. Arm GPU Deep Dive: Immortalis-G925 vs Mali Series | Faceofit F Faceofit Analysis Internals Thermals Specs Access Data Tech Deep Dive Updated Dec 2025 The Silicon Canvas. Immortalis-G925 vs. The World. Arm’s mobile graphics just hit the “Cortex-X” moment. We analyze the aggressive shift from the Bandwidth Wall of G710 to the ray-tracing powerhouse of G925. Performance Trajectory Relative rasterization gains in flagship implementations (Dimensity 9000 to 9400). +211% Gain since G710 Source: Faceofit Analysis Lab The Mobile Graphics Shift Mobile graphics have moved past simple thermal constraints. We now operate in a paradigm where desktop features like hardware ray tracing and mesh shading define the flagship experience. The driving force is the “Bandwidth Wall.” As resolutions standardized at 1440p and 120Hz, traditional pipelines failed to feed shader cores without choking system memory. Architecture Evolution 2021 • Valhall 3 Mali-G710 Introduced Command Stream Frontend (CSF) to reduce CPU driver overhead. Pure rasterization focus. 2022 • Valhall 4 Immortalis-G715 First hardware Ray Tracing Unit (RTU). Bifurcation of Mali (Non-RT) and Immortalis (RT). 2023 • 5th Gen Immortalis-G720 & G725 Deferred Vertex Shading (DVS) solves the bandwidth crisis. G725 appears as the sub-premium variant. 2024/25 • 5th Gen Advanced Immortalis-G925 Fragment Prepass and Cortex-X style performance targeting. 50% faster Ray Tracing. 1. The Bandwidth Solution: G720’s DVS Before the G720, GPUs relied on “Immediate Mode” or simple tiling. The 5th Gen architecture introduced Deferred Vertex Shading (DVS). In high-geometry games like Genshin Impact, older GPUs (G710/G715) would process geometry, write it to main RAM, and read it back for texturing. This round-trip killed battery life. DVS keeps that intermediate geometry data on the GPU silicon itself. Bandwidth Impact: Reduces system memory traffic by up to 40%. Result: Allowed the Dimensity 9300 to run sustained high FPS without thermal throttling. 2. Silicon Floorplan: Inside the Core The shift from Valhall (G715) to 5th Gen (G720/G925) wasn’t just about Ray Tracing; it changed how the Execution Engines (EE) handle math. The Dual-Issue Architecture In previous generations, the ALU (Arithmetic Logic Unit) pipelines were rigid. In G925, the cores act as “Dual-issue” machines. They can process two distinct math operations per clock cycle if the instructions are compatible. G710 Core G925 Core Parallel FP32 Operations 3. Ray Tracing: G715 vs. G925 The G715 brought Ray Tracing to hardware, but the G925 makes it usable for complex global illumination. Feature Immortalis-G715 Immortalis-G925 RT Implementation Basic Hardware RTU Gen 2 Hardware RTU Primary Usage Soft Shadows, Simple Reflections Lumen (UE5), Global Illumination Intersection Engine Standard Double Throughput for Complex Geometry Performance Delta Base +52% Faster 4. Beyond Pixels: AI & Machine Learning While NPUs (like Google’s TPU or MediaTek’s APU) handle heavy AI lifting, the GPU plays a crucial role in real-time inference during gaming, such as DLSS-style upscaling (FSR or MetalFX equivalents). Matrix Multiply Enhancements The G710 introduced dedicated instructions for Matrix Multiplication (MatMul). The G925 significantly expands this with optimized support for Int8 and Int16 precision. This allows camera pipelines to use the GPU for advanced noise reduction and “Magic Eraser” style effects directly in the viewfinder stream, unloading the main CPU. 5. Sustained Performance & Thermals Peak performance is a marketing number; sustained performance is the user reality. The chart below illustrates how the architecture manages heat dissipation over a 20-minute gaming session. The G710 (Red line) starts strong but throttles aggressively after 4 minutes. The G925 (Blue line) maintains 85% of its peak performance due to the bandwidth savings from DVS and Fragment Prepass. 6. Variable Rate Shading (VRS) Explained VRS allows the GPU to reduce visual fidelity in areas where the user isn’t looking—such as the corners of the screen or fast-moving blurred objects—while maintaining 4K clarity in the center. Standard Rendering (1×1) Every pixel is calculated individually. High cost, high quality. 1111 1111 1111 1111 VRS Tier 2 (2×2) Four pixels share one color calculation. 75% less shading work. 1 1 1 1 Technical Specifications Matrix A comprehensive breakdown across five generations of Arm graphics architecture. Specification Mali-G710 Mali/Imm-G715 Mali-G725 Imm-G720 Imm-G925 Architecture Valhall 3 Valhall 4 5th Gen 5th Gen 5th Gen Adv. Ray Tracing None HW Gen 1 Optional (Rare) HW Gen 1 HW Gen 2 VRS Support No Tier 1 Tier 2 Tier 2 Tier 2 Geometry Traditional Traditional DVS DVS DVS + Prepass AI Support FP16 FP16/MatMul Int8/Int16 Int8/Int16 Adv. Matrix Max Cores 16 16 9 16 24 Technical Glossary AFBC (Arm Frame Buffer Compression) A lossless compression format that reduces the size of textures and colors moving between the GPU and memory, saving battery. Overdraw Wasted effort when the GPU draws pixels for an object that is immediately covered up by another object (like a wall covering a chair). Lumen Unreal Engine 5’s global illumination system. It requires heavy hardware ray tracing support, which G925 enables on mobile. CSF (Command Stream Frontend) A small microcontroller inside the GPU that handles scheduling, so the main Android CPU doesn’t have to micromanage the graphics card. Common Questions Is the Mali-G725 just a rebranded G720? It is architecturally identical (5th Gen) but segmented differently. The G720 is the “Immortalis” tier, meaning it mandates Ray Tracing hardware and supports high core counts (10+). The G725 is the “Mali” tier, usually capped at 9 cores and typically shipped without Ray Tracing units to lower costs for mid-range phones. Why does the Pixel 9 use Mali-G715 instead of G720? Google’s Tensor chips often trail the latest Arm IP by one generation due to their custom design cycle. While the Pixel 8 and 9 series use competent GPUs, they often opt for the “Mali” configuration (lower core count, no RTU) to prioritize die space for their NPU (AI Neural Core). Can G925 play PC games via emulation? Yes. The G925’s support for advanced Vulkan extensions and its high raw rasterization performance makes it the best current Android GPU for Windows emulation (via Winlator or Mobox), outperforming the Adreno 750 in raw shader throughput, though driver compatibility varies. What is the ‘Bandwidth Wall’? This refers to the limitation where the GPU is fast enough to render a scene, but the system memory (RAM) is too slow to send it the necessary texture and geometry data. The G720 and G925 solve this with DVS (Deferred Vertex Shading), which keeps data inside the GPU cache, bypassing the slow trip to main RAM. Affiliate Disclosure: Faceofit.com is a participant in the Amazon Services LLC Associates Program. As an Amazon Associate we earn from qualifying purchases. Share What's your reaction? Excited 0 Happy 0 In Love 0 Not Sure 0 Silly 0
The Mobile Graphics Shift Mobile graphics have moved past simple thermal constraints. We now operate in a paradigm where desktop features like hardware ray tracing and mesh shading define the flagship experience. The driving force is the “Bandwidth Wall.” As resolutions standardized at 1440p and 120Hz, traditional pipelines failed to feed shader cores without choking system memory. Architecture Evolution 2021 • Valhall 3 Mali-G710 Introduced Command Stream Frontend (CSF) to reduce CPU driver overhead. Pure rasterization focus. 2022 • Valhall 4 Immortalis-G715 First hardware Ray Tracing Unit (RTU). Bifurcation of Mali (Non-RT) and Immortalis (RT). 2023 • 5th Gen Immortalis-G720 & G725 Deferred Vertex Shading (DVS) solves the bandwidth crisis. G725 appears as the sub-premium variant. 2024/25 • 5th Gen Advanced Immortalis-G925 Fragment Prepass and Cortex-X style performance targeting. 50% faster Ray Tracing. 1. The Bandwidth Solution: G720’s DVS Before the G720, GPUs relied on “Immediate Mode” or simple tiling. The 5th Gen architecture introduced Deferred Vertex Shading (DVS). In high-geometry games like Genshin Impact, older GPUs (G710/G715) would process geometry, write it to main RAM, and read it back for texturing. This round-trip killed battery life. DVS keeps that intermediate geometry data on the GPU silicon itself. Bandwidth Impact: Reduces system memory traffic by up to 40%. Result: Allowed the Dimensity 9300 to run sustained high FPS without thermal throttling. 2. Silicon Floorplan: Inside the Core The shift from Valhall (G715) to 5th Gen (G720/G925) wasn’t just about Ray Tracing; it changed how the Execution Engines (EE) handle math. The Dual-Issue Architecture In previous generations, the ALU (Arithmetic Logic Unit) pipelines were rigid. In G925, the cores act as “Dual-issue” machines. They can process two distinct math operations per clock cycle if the instructions are compatible. G710 Core G925 Core Parallel FP32 Operations 3. Ray Tracing: G715 vs. G925 The G715 brought Ray Tracing to hardware, but the G925 makes it usable for complex global illumination. Feature Immortalis-G715 Immortalis-G925 RT Implementation Basic Hardware RTU Gen 2 Hardware RTU Primary Usage Soft Shadows, Simple Reflections Lumen (UE5), Global Illumination Intersection Engine Standard Double Throughput for Complex Geometry Performance Delta Base +52% Faster 4. Beyond Pixels: AI & Machine Learning While NPUs (like Google’s TPU or MediaTek’s APU) handle heavy AI lifting, the GPU plays a crucial role in real-time inference during gaming, such as DLSS-style upscaling (FSR or MetalFX equivalents). Matrix Multiply Enhancements The G710 introduced dedicated instructions for Matrix Multiplication (MatMul). The G925 significantly expands this with optimized support for Int8 and Int16 precision. This allows camera pipelines to use the GPU for advanced noise reduction and “Magic Eraser” style effects directly in the viewfinder stream, unloading the main CPU. 5. Sustained Performance & Thermals Peak performance is a marketing number; sustained performance is the user reality. The chart below illustrates how the architecture manages heat dissipation over a 20-minute gaming session. The G710 (Red line) starts strong but throttles aggressively after 4 minutes. The G925 (Blue line) maintains 85% of its peak performance due to the bandwidth savings from DVS and Fragment Prepass. 6. Variable Rate Shading (VRS) Explained VRS allows the GPU to reduce visual fidelity in areas where the user isn’t looking—such as the corners of the screen or fast-moving blurred objects—while maintaining 4K clarity in the center. Standard Rendering (1×1) Every pixel is calculated individually. High cost, high quality. 1111 1111 1111 1111 VRS Tier 2 (2×2) Four pixels share one color calculation. 75% less shading work. 1 1 1 1
Tech Posts Arm Mali GPU Analysis: Valhall vs. Bifrost Architecture, Specs & Benchmarks Arm’s transition from the Bifrost architecture to Valhall represents a fundamental restructuring of mobile graphics ...
Tech Posts Immortalis-G1 Ultra vs Adreno 830: Benchmark, Architecture & Ray Tracing Analysis The mobile GPU landscape in 2025 is defined by two opposing philosophies. Qualcomm’s Adreno 830, ...
Tech Posts Immortalis-G1 Ultra vs G925 vs G720: Architecture, Ray Tracing & Gaming Performance The shift from Arm’s Immortalis-G720 to the new Mali-G1 Ultra represents a specific hardware pivot ...
Tech Posts The 2026 Fire TV Stick Battle: Vega OS vs Fire OS in India The days of blindly picking the cheapest Amazon stick to upgrade your TV are over. ...
Tech Posts Ethernet Cable Length vs Speed: The Physics of Cat6, Cat6a, and Distance Limits You bought the fastest fiber plan available. You installed an enterprise-grade router. Yet, your file ...
Tech Posts Best Portable Car Headrest Monitors: 120Hz Gaming, Android & DVD Best Portable Car TV Headrest Monitor 2025 | Faceofit.com Note: If you buy something from ...
Ubiquiti UniFi 5G Max Review: 2.5GbE Failover vs. Cradlepoint & Peplink IGDecember 5, 2025 Tech Posts
