--- title: Re-thinking framebuffers in PanVK | daily.dev description: Faith Ekstrand details a new framebuffer abstraction for PanVK, the Vulkan driver in Mesa's Panfrost stack targeting Arm Mali GPUs. The post explains tiled... canonical: https://daily.dev/posts/re-thinking-framebuffers-in-panvk-nfycylfnw twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Re-thinking framebuffers in PanVK | daily.dev og:description: Faith Ekstrand details a new framebuffer abstraction for PanVK, the Vulkan driver in Mesa's Panfrost stack targeting Arm Mali GPUs. The post explains tiled... og:url: https://daily.dev/posts/re-thinking-framebuffers-in-panvk-nfycylfnw og:image: https://api.daily.dev/og/posts/nfyCylfnw.png og:image:alt: Re-thinking framebuffers in PanVK og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Re-thinking framebuffers in PanVK **[Collabora Blog](https://daily.dev/sources/collabora)** · 16 min read · 0 upvotes · 0 comments ## Summary Faith Ekstrand details a new framebuffer abstraction for PanVK, the Vulkan driver in Mesa's Panfrost stack targeting Arm Mali GPUs. The post explains tiled rendering on mobile GPUs versus immediate-mode rendering on desktop, then describes how the old OpenGL-derived framebuffer abstraction limited features like efficient MSAA resolves. The new design splits concerns into five structs: pan_fb_layout, pan_fb_load, pan_fb_store, pan_fb_resolves, and pan_fb_desc_info, fully decoupling load, store, and resolve operations. This enables in-tile multisample resolves that avoid writing the full multisampled result to main memory. Benchmarks on a MediaTek Chromebook show up to 25x speedup for 16x MSAA, close to the theoretical 33x maximum, with the jump from 2x to 16x MSAA costing only 5% in FPS once resolves are optimized. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does tiled rendering on mobile GPUs like Mali differ from immediate-mode rendering on desktop GPUs? Tiled rendering divides the framebuffer into small tiles (e.g., 32x32) and rasterizes the entire render pass per tile using fast on-chip tile memory, avoiding the large main-memory bandwidth that immediate-mode renderers require. Desktop GPUs rely on fast GDDR/HBM memory and large caches; mobile GPUs trade some duplicated geometry work for dramatically reduced memory traffic, making them highly sensitive to render pass splits. _Engineers optimizing Vulkan workloads for mobile GPUs track driver architecture changes like this on daily.dev._ ### What is the theoretical bandwidth speedup of doing MSAA resolves entirely in tile memory on a Mali GPU instead of writing out the full multisampled result? For 16x MSAA, the theoretical maximum speedup is 33x: without tile-memory resolves, 16 samples are stored and then 16 are read back for the resolve plus 1 written out, totalling 33 sample accesses per pixel. With an in-tile resolve using VK_ATTACHMENT_STORE_OP_NONE, only 1 sample is written. Real-world benchmarks on a MediaTek Chromebook showed ~25x speedup. For 8x the max is 17x, 4x gives 9x, and 2x gives 5x. _Developers targeting Mali hardware for Vulkan rendering can follow PanVK driver progress on daily.dev._ ### What are the five structs in the new PanVK framebuffer abstraction and what does each one represent? The new abstraction splits framebuffer concerns into: pan_fb_layout (tile memory layout, formats, sample count, tile size, render area), pan_fb_load (load operations and preload shaders at render pass start), pan_fb_store (store operations and destination image views at render pass end), pan_fb_resolves (optional end-of-pass resolve shaders supporting multisample and cross-attachment copies), and pan_fb_desc_info (remaining hardware descriptor fields like tiling heap and static state bits). _Graphics developers working through Mesa driver internals find deep-dives like this surfaced on daily.dev._ ## Similar posts on daily.dev - [Mesa PanVK Driver Seeing Up To 25.7x Speedup For MSAA](https://daily.dev/posts/mesa-panvk-driver-seeing-up-to-25-7x-speedup-for-msaa-xetj6fdto) · Phoronix · 0 upvotes · 0 comments - [PanVK Mali Vulkan Driver Lands In-Memory Cache & On-Disk Shader Cache Support](https://daily.dev/posts/panvk-mali-vulkan-driver-lands-in-memory-cache-on-disk-shader-cache-support-y2z1wbliv) · Phoronix · 0 upvotes · 0 comments - [Notes about VK\_EXT\_custom\_resolve](https://daily.dev/posts/notes-about-vk-ext-custom-resolve-vrb2y0jaf) · Khronos · 0 upvotes · 0 comments - [Intel's Vulkan Linux Driver Merges Shader VMA Allocator For Ray-Tracing Capture/Replay](https://daily.dev/posts/intel-s-vulkan-linux-driver-merges-shader-vma-allocator-for-ray-tracing-capture-replay-grab740ce) · Phoronix · 0 upvotes · 0 comments --- Tags: [#vulkan](https://daily.dev/tags/vulkan), [#mesa](https://daily.dev/tags/mesa) [View this post on daily.dev](https://daily.dev/posts/re-thinking-framebuffers-in-panvk-nfycylfnw) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Re-thinking framebuffers in PanVK","url":"https://daily.dev/posts/re-thinking-framebuffers-in-panvk-nfycylfnw","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/re-thinking-framebuffers-in-panvk-nfycylfnw"},"datePublished":"2026-08-11T08:57:24.225Z","dateModified":"2026-08-11T09:40:08.556Z","description":"Faith Ekstrand details a new framebuffer abstraction for PanVK, the Vulkan driver in Mesa's Panfrost stack targeting Arm Mali GPUs. The post explains tiled...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/281ac013d9a07ce3ad0fa037a773c33c?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/281ac013d9a07ce3ad0fa037a773c33c?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collabora Blog","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collabora Blog","logo":"https://media.daily.dev/image/upload/s--ZGZ0FjqS--/f_auto,q_auto/v1786438306/logos/collabora?_a=BAMAMicg0","url":"https://daily.dev/sources/collabora"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/re-thinking-framebuffers-in-panvk-nfycylfnw","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"vulkan,mesa","timeRequired":"PT16M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collabora Blog","item":"https://daily.dev/sources/collabora"},{"@type":"ListItem","position":3,"name":"Re-thinking framebuffers in PanVK"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/re-thinking-framebuffers-in-panvk-nfycylfnw#faq","mainEntity":[{"@type":"Question","name":"How does tiled rendering on mobile GPUs like Mali differ from immediate-mode rendering on desktop GPUs?","acceptedAnswer":{"@type":"Answer","text":"Tiled rendering divides the framebuffer into small tiles (e.g., 32x32) and rasterizes the entire render pass per tile using fast on-chip tile memory, avoiding the large main-memory bandwidth that immediate-mode renderers require. Desktop GPUs rely on fast GDDR/HBM memory and large caches; mobile GPUs trade some duplicated geometry work for dramatically reduced memory traffic, making them highly sensitive to render pass splits. Engineers optimizing Vulkan workloads for mobile GPUs track driver architecture changes like this on daily.dev."}},{"@type":"Question","name":"What is the theoretical bandwidth speedup of doing MSAA resolves entirely in tile memory on a Mali GPU instead of writing out the full multisampled result?","acceptedAnswer":{"@type":"Answer","text":"For 16x MSAA, the theoretical maximum speedup is 33x: without tile-memory resolves, 16 samples are stored and then 16 are read back for the resolve plus 1 written out, totalling 33 sample accesses per pixel. With an in-tile resolve using VK_ATTACHMENT_STORE_OP_NONE, only 1 sample is written. Real-world benchmarks on a MediaTek Chromebook showed ~25x speedup. For 8x the max is 17x, 4x gives 9x, and 2x gives 5x. Developers targeting Mali hardware for Vulkan rendering can follow PanVK driver progress on daily.dev."}},{"@type":"Question","name":"What are the five structs in the new PanVK framebuffer abstraction and what does each one represent?","acceptedAnswer":{"@type":"Answer","text":"The new abstraction splits framebuffer concerns into: pan_fb_layout (tile memory layout, formats, sample count, tile size, render area), pan_fb_load (load operations and preload shaders at render pass start), pan_fb_store (store operations and destination image views at render pass end), pan_fb_resolves (optional end-of-pass resolve shaders supporting multisample and cross-attachment copies), and pan_fb_desc_info (remaining hardware descriptor fields like tiling heap and static state bits). Graphics developers working through Mesa driver internals find deep-dives like this surfaced on daily.dev."}}]} ```