<br><div><p class=”wp–block–paragraph“><a href=”https://mightygadget.com/tag/mediatek/”>MediaTeka> has announced the Dimensity 9600 Pro, its latest flagship smartphone chipset. It is built on a 2nm process and succeeds the Dimensity 9500, with MediaTek describing it as an all–new native AI architecture that pairs pro–level performance with high efficiency.p><p class=”wp–block–paragraph“>According to MediaTek, the Pro name reflects a new design rather than a minor revision of the previous chip. The biggest change is how AI processing is spread across the platform. Instead of leaving AI work to a single dedicated NPU, MediaTek has linked the NPU with the CPU, GPU, image signal processor and connectivity hardware so that each part of the chip can handle AI workloads directly.p><p class=”wp–block–paragraph“>All performance and power efficiency figures quoted by MediaTek are measured against the Dimensity 9500 and come from the company’s own testing. Independent results will only be available once phones using the chip reach reviewers.p><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”A_new_native_AI_architecture“/><strong>A new native AI architecturestrong><span class=”ez–toc–section–end“/>h2><div class=”wp–block–image“><figure class=”aligncenter size–large“><img data–lazyloaded=”1” decoding=”async” width=”1024” height=”756” src=”https://mightygadget.com/wp–content/uploads/2026/09/Dimensity–9600–Pro–Infographic_0909–1024×756.png” alt=”” class=”wp–image–90805” srcset=”https://mightygadget.com/wp–content/uploads/2026/09/Dimensity–9600–Pro–Infographic_0909–1024×756.png 1024w, https://mightygadget.com/wp–content/uploads/2026/09/Dimensity–9600–Pro–Infographic_0909–300×221.png 300w, https://mightygadget.com/wp–content/uploads/2026/09/Dimensity–9600–Pro–Infographic_0909–767×566.png 767w, https://mightygadget.com/wp–content/uploads/2026/09/Dimensity–9600–Pro–Infographic_0909–1536×1133.png 1536w, https://mightygadget.com/wp–content/uploads/2026/09/Dimensity–9600–Pro–Infographic_0909.png 1694w” data–sizes=”(max–width: 1024px) 100vw, 1024px“/>figure>div><p class=”wp–block–paragraph“>MediaTek splits its approach to AI into three layers. The device layer is the hardware itself, which includes an all big core CPU, a GPU aimed at console–grade gaming and a new NPU. The system layer is the Dimensity Agentic AI Engine, which includes the Dimensity Sensing Foundation SDK and the Dimensity NeuroPilot SDK. These give phone makers and developers a tool chain for building their own AI applications on the platform. The final layer is the ecosystem, where MediaTek is working with developers, phone makers and industry partners such as Google to build support for agentic AI applications.p><p class=”wp–block–paragraph“>Inside the chip, MediaTek has given each hardware pairing its own name:p><ol class=”wp–block–list“><li>CPU and NPU: AI Compute Fusion Architectureli><li>Performance NPU and efficiency NPU: Dual–NPU Architectureli><li>GPU and NPU: Neural Graphics Architectureli><li>ISP and NPU: AI Imaging Fusion Architectureli><li>Connectivity and AI: an AI–optimised connectivity frameworkli>ol><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”CPU“/><strong>CPUstrong><span class=”ez–toc–section–end“/>h2><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”All_big_core_233_layout“/><strong>All big core 2+3+3 layoutstrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The CPU has been fully revamped and uses an all big core design with a 2+3+3 layout. There are two C2–Ultra cores, each with 2MB of L2 cache, three C2–Pro cores with 1MB of L2 cache each, and three further C2–Pro cores with 512KB of L2 cache each. All eight cores share a 16MB L3 cache.p><p class=”wp–block–paragraph“>Cache has received a notable upgrade. The L2 cache is 21% larger than on the Dimensity 9500, and the combined L2, L3 and system cache comes to 34.5MB. Based on the core layout above, the L2 cache totals 8.5MB, which leaves around 10MB of system–level cache alongside the 16MB L3. MediaTek says the larger cache pool is particularly beneficial for low power use in everyday scenarios.p><p class=”wp–block–paragraph“>The company claims the following gains over the Dimensity 9500:p><ol class=”wp–block–list“><li>17% faster single–core performanceli><li>15% faster multi–core performanceli><li>37% more efficient single–core power usageli><li>61% more efficient multi–core power usageli>ol><p class=”wp–block–paragraph“>The efficiency figures are the more notable claims here, especially the 61% improvement in multi–core power usage. MediaTek has not yet detailed the benchmarks or test conditions used for these CPU figures, so they are best treated as a guide until independent testing is possible. The CPU also supports Scalable Matrix Extension (SME), which MediaTek says is currently used mainly for camera AI and object tracking.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”AI_Compute_Fusion_Architecture“/><strong>AI Compute Fusion Architecturestrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The AI Compute Fusion Architecture changes how the CPU and NPU share work when handling multi–step requests from AI agents. Previously, a request would be processed through a single pipeline. With the Dimensity 9600 Pro, the CPU receives the tasks, the NPU breaks them down, and the CPU then executes those smaller tasks while the NPU analyses and infers in parallel. Background applications can continue to run on the CPU while the NPU handles the AI work, and the results are then brought back together to complete the instruction.p><p class=”wp–block–paragraph“>MediaTek says combining this approach with its updated scheduling engine delivers 40% faster LLM token generation and a 27% power saving at equivalent performance.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Dimensity_Scheduling_Engine_30“/><strong>Dimensity Scheduling Engine 3.0strong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The third–generation Dimensity Scheduling Engine is designed to allocate compute, power and memory resources across the system. It is made up of three parts:p><ol class=”wp–block–list“><li>A system–level power management engine that monitors computing demand and allocates suitable resources depending on the usage scenarioli><li>Super zRAM technology, which optimises memory usageli><li>Compute Scheduling Engine 3.0, an upgraded version of the scheduler first introduced with the Dimensity 9300li>ol><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Memory_and_storage“/><strong>Memory and storagestrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The Dimensity 9600 Pro supports LPDDR6 memory, which MediaTek says offers 33% faster effective memory bandwidth compared with LPDDR5X. LPDDR5X remains supported, giving phone makers some flexibility over memory choice. Storage support moves to UFS 5.0, which MediaTek claims doubles read and write speeds compared with 2–channel UFS 4. The company positioned the extra memory bandwidth as useful for agentic AI and large language model workloads.p><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”Dual_NPU_architecture“/><strong>Dual NPU architecturestrong><span class=”ez–toc–section–end“/>h2><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”NPU_1090“/><strong>NPU 1090strong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The main AI processor is the NPU 1090, MediaTek’s tenth–generation super performance NPU. It is split into a performance core and a flexible core and runs Generative AI Engine 3.0. Key features include:p><ol class=”wp–block–list“><li>Support for the latest AI hybrid attention techniquesli><li>Next–generation lossless LLM compressionli><li>Double the INT4 compute capability of the previous generationli><li>On–device inference for Mixture of Experts (MoE) large language modelsli>ol><p class=”wp–block–paragraph“>MediaTek claims a 51% improvement in LLM prefill performance, which is the stage where the model processes the prompt before it begins generating a response.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Super_efficient_eNPU_20“/><strong>Super efficient eNPU 2.0strong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>Alongside the NPU 1090 is the super efficient eNPU 2.0, which has an integrated SensorHub. This handles always–on AI tasks at low power, allowing phone makers and developers to build agentic features that run continuously in the background. MediaTek says the eNPU 2.0 is 40% more power efficient than the equivalent part in the Dimensity 9500.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”On–device_AI_features“/><strong>On–device AI featuresstrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>MediaTek highlighted several examples of what the NPUs can run locally, including proactive suggestions, scene detection and a live–photo eraser. The chip can also run on–device AI–generated content models at 2K resolution, which MediaTek demonstrated with live photo editing that takes around three seconds. The company says this is a 25% time saving compared with a cloud–based solution.p><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”Security_and_privacy“/><strong>Security and privacystrong><span class=”ez–toc–section–end“/>h2><p class=”wp–block–paragraph“>MediaTek has worked with Google on the security and privacy side of the platform. The Dimensity 9600 Pro supports Google’s AI Core framework with the Gemini Nano model running on the phone, which allows for low latency responses while keeping data on the device.p><p class=”wp–block–paragraph“>For protection, MediaTek uses what it calls the AISeal architecture, which is split into three areas:p><ol class=”wp–block–list“><li>Data isolation, using a hardware root of trust (HWRoT) with tamper–resistant protection, AISeal Vault, pKVM hypervisor isolation and an isolated trusted execution environment (TEE)li><li>Verification and traceability, with checks at every boot, anti–rollback counters and post–quantum cryptography (PQC) support designed to defend against future quantum attacksli><li>Data control, with access enforced by AISeal isolation under a single policyli>ol><p class=”wp–block–paragraph“>MediaTek says this is intended to keep data isolated, verify and trace each software version, and give control over how data is accessed by agentic AI applications.p><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”GPU_and_gaming“/><strong>GPU and gamingstrong><span class=”ez–toc–section–end“/>h2><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Mali_G2–Ultra_NX_with_Neural_Graphics_Architecture“/><strong>Mali G2–Ultra NX with Neural Graphics Architecturestrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The Dimensity 9600 Pro uses the Mali G2–Ultra NX GPU, which MediaTek describes as the most powerful Dimensity GPU so far. This is the first generation to integrate neural capabilities directly into the GPU. During the briefing, MediaTek confirmed the GPU contains six neural accelerators.p><p class=”wp–block–paragraph“>MediaTek uses two routes for AI graphics work. The neural accelerators inside the GPU can handle some tasks, while the more powerful NPU 1090 can be used for super–resolution upscaling in other scenarios. The company says it is working with game developers to choose the most suitable route for each title.p><p class=”wp–block–paragraph“>As an example, MediaTek showed a game rendered at a native 540p resolution being upscaled to 1080p using the same power and GPU compute. MediaTek does note that these figures are for reference only and that performance varies by game.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Performance_and_efficiency“/><strong>Performance and efficiencystrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>In 3DMark Steel Nomad Light, MediaTek reports a 27% increase in peak performance over the Dimensity 9500, along with 24% lower power consumption at peak performance. The chip also supports native gameplay at more than 185fps.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Ray_tracing“/><strong>Ray tracingstrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The GPU uses MediaTek’s third–generation ray tracing engine. MediaTek claims up to 18% faster ray tracing performance, and in 3DMark Solar Bay Extreme it quoted a score of 3,029 for the Dimensity 9600 Pro compared with 2,570 for the Dimensity 9500, which works out at just under 18%.p><p class=”wp–block–paragraph“>MediaTek also claims an industry–first console–level ray tracing pipeline for mobile, which supports multi–material reflection and global reflection. Hardware–accelerated Opacity Micro–Maps (OMM) ray tracing has been enhanced too. OMM allows ray tracing to handle detailed, partially transparent surfaces such as foliage more efficiently, and MediaTek’s demonstration showed more accurate light and shadow through a tree with the feature enabled. As with any mobile ray tracing, the benefit depends on game support, and MediaTek says it is working with developers to bring more ray traced titles to the platform.p><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”Camera_and_imaging“/><strong>Camera and imagingstrong><span class=”ez–toc–section–end“/>h2><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Imagiq_1290_ISP“/><strong>Imagiq 1290 ISPstrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The Imagiq 1290 is described by MediaTek as its most powerful ISP to date. It includes a Semantic Analysis Engine, a True Color Engine and a Fast–Focus Engine, and connects to the NPU 1090 through the AI Imaging Fusion Architecture. One use of this link is 4K live photo capture.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Photography“/><strong>Photographystrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>For stills, MediaTek claims an industry–first full range depth–of–field computational optical imaging feature, along with 60fps instant action capture. Focus tracking has been upgraded with an image focus tracking engine and NPU–assisted action capture, providing 60fps continual subject focus tracking. MediaTek quotes a 200% improvement in both focus tracking speed and autofocus speed compared with the Dimensity 9500, which the presenter described as roughly doubling performance.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Video_recording“/><strong>Video recordingstrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>Video is where most of the imaging upgrades are focused:p><ol class=”wp–block–list“><li>4K60 HDR recording with AI true colour tone and content–aware semantic processing, which MediaTek claims as an industry firstli><li>17EV HDR video capture with support for LOFIC and UFCC sensor technologiesli><li>4K 120fps cinematic log recording with native colourli><li>4K 240fps super slow motion with native output, which MediaTek also claims as an industry firstli><li>100dB high dynamic range imagingli>ol><p class=”wp–block–paragraph“>MediaTek says the 4K 240fps mode comes with a 200% improvement in 4K recording performance compared with the previous generation.p><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”Display_and_multimedia“/><strong>Display and multimediastrong><span class=”ez–toc–section–end“/>h2><p class=”wp–block–paragraph“>The Dimensity 9600 Pro adds a new H.266 hardware video decoder, which MediaTek says allows video playback at very low power. H.266, also known as Versatile Video Coding (VVC), is the successor to H.265 and offers better compression for high resolution video. Video encoding power efficiency has also been improved.p><p class=”wp–block–paragraph“>On the display side, the chip features a studio–grade BT.2020 (Rec. 2020) colour pipeline and multi–frame overdrive, which is designed to eliminate motion blur. The block diagram also lists MiraVision 1290 for display processing.p><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”Connectivity“/><strong>Connectivitystrong><span class=”ez–toc–section–end“/>h2><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”5G“/><strong>5Gstrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The 5G–Advanced modem includes in–modem AI (MMAI) and supports five component carrier aggregation (5CC–CA) with downlink speeds of more than 7.4Gbps. Dimensity AI Network Selection 2.0 uses intelligent handover to select and switch between cellular base stations, which is designed to maintain a reliable connection in busy areas with many users and varying signal strength.p><p class=”wp–block–paragraph“>MediaTek claims a 30% improvement in lag in congested networks, rising to 50% when combined with a phone maker’s own AI model. There are also real–time network optimisations aimed at reducing latency in games.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Wi–Fi_and_Bluetooth“/><strong>Wi–Fi and Bluetoothstrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>The Dimensity Connectivity AI Framework has been applied to Wi–Fi and Bluetooth, with dedicated hardware inside these blocks for running AI models. MediaTek says this delivers 10x faster response speed and 50% lower power consumption compared with running the same models on the CPU. The framework also improves roaming stability. Bluetooth 6.2 is supported, although the briefing materials did not confirm which Wi–Fi standard the chip uses.p><h3 class=”wp–block–heading“><span class=”ez–toc–section” id=”Satellite_positioning“/><strong>Satellite positioningstrong><span class=”ez–toc–section–end“/>h3><p class=”wp–block–paragraph“>Dimensity AI Smart Satellite Selection 2.0 uses an AI–enhanced algorithm to narrow down a user’s likely location, which speeds up the search for satellites. MediaTek claims this improves positioning accuracy by 30%. A new GNSS RF design also provides up to a 65% improvement in power efficiency.p><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”Key_specifications“/><strong>Key specificationsstrong><span class=”ez–toc–section–end“/>h2><figure class=”wp–block–table“><table class=”has–fixed–layout“><thead><tr><th><strong>Featurestrong>th><th><strong>Dimensity 9600 Prostrong>th>tr>thead><tbody><tr><td><strong>Processstrong>td><td>2nmtd>tr><tr><td><strong>CPUstrong>td><td>2x C2–Ultra, 6x C2–Pro (2+3+3 all big core)td>tr><tr><td><strong>Cachestrong>td><td>34.5MB total (L2, 16MB L3 and system cache)td>tr><tr><td><strong>GPUstrong>td><td>Mali G2–Ultra NX with six neural acceleratorstd>tr><tr><td><strong>NPUstrong>td><td>NPU 1090 and super efficient eNPU 2.0td>tr><tr><td><strong>ISPstrong>td><td>Imagiq 1290td>tr><tr><td><strong>Memorystrong>td><td>LPDDR6 and LPDDR5Xtd>tr><tr><td><strong>Storagestrong>td><td>UFS 5.0td>tr><tr><td><strong>Videostrong>td><td>Up to 4K 240fps recording, H.266 hardware decodingtd>tr><tr><td><strong>Cellularstrong>td><td>5G–Advanced, 5CC–CA, more than 7.4Gbps downlinktd>tr><tr><td><strong>Wirelessstrong>td><td>Bluetooth 6.2td>tr><tr><td><strong>Securitystrong>td><td>HWRoT, PQC, AISeal with pKVMtd>tr>tbody>table>figure><h2 class=”wp–block–heading“><span class=”ez–toc–section” id=”Availability_and_pricing“/><strong>Availability and pricingstrong><span class=”ez–toc–section–end“/>h2><p class=”wp–block–paragraph“>MediaTek did not confirm pricing or name the first smartphones to use the Dimensity 9600 Pro in the briefing materials. As MediaTek supplies chipsets to phone makers rather than selling directly to consumers, pricing will be set by individual manufacturers when devices using the chip are announced.p><div class=”sabox–authors“><div class=”saboxplugin–wrap” itemtype=”http://schema.org/Person” itemscope=”” itemprop=”author“><div class=”saboxplugin–tab saboxplugin–tab–about“><div class=”saboxplugin–gravatar“><a href=”https://mightygadget.com/author/jpsmythe_mightygadget_com/”><img data–lazyloaded=”1” src=”http://mightygadget.com/wp–content/uploads/2026/03/98e821b604a5ab9c5d89928902785bc3f6eb3bd6550d340f8d776eb2308a054f.jpg.webp” width=”100” height=”100” alt=”” itemprop=”image“/>a>div><div class=”saboxplugin–desc“><div itemprop=”description“><p>I am James, a UK–based tech enthusiast and the Editor and Owner of <a href=”https://mightygadget.com/”>Mighty Gadgeta>, which I’ve proudly run since 2007. Passionate about all things technology, my expertise spans from computers and <a href=”https://mightygadget.com/category/hardware/networking–hardware/”>networkinga> to <a href=”https://mightygadget.com/category/phones/”>mobilea>, <a href=”https://mightygadget.com/category/fitness/”>wearablesa>, and <a href=”https://mightygadget.com/category/home–automation/”>smart home devicesa>.p><p>As a fitness fanatic who loves running and cycling, I also have a keen interest in fitness–related technology, and I take every opportunity to cover this niche on my blog. My diverse interests allow me to bring a unique perspective to tech blogging, merging lifestyle, fitness, and the latest tech trends.p><p>In my academic pursuits, I earned a BSc in Information Systems Design from UCLAN, before advancing my learning with a Master’s Degree in Computing.p><p>I’m proud to share that Vuelio has consistently ranked Mighty Gadget as one of the <a href=”https://www.vuelio.com/uk/social–media–index/top–10–uk–technology–blogs/”>top technology blogs in the UKa>. With my dedication to technology and drive to share my insights, I aim to continue providing my readers with engaging and informative content.p><p>I work directly with the companies he reviews products for or their chosen PR agency.p><p>For Social Media, you can find me on:p><p><a href=”https://twitter.com/mighty_gadget“>Twitter/Xa> – <a href=”https://www.instagram.com/mighty.gadget/”>Instagrama> – <a href=”https://www.facebook.com/MightyGadget“>Facebooka> – <a href=”https://www.linkedin.com/in/jamespsmythe/”>LinkedIna> – <a href=”https://www.strava.com/athletes/4656172“>Stravaa>p>div>div>div>div>div>div><script async src=”https://platform.twitter.com/widgets.js” charset=”utf–8“>script><script async src=”//www.instagram.com/embed.js“>script>
<br><a href=”https://mightygadget.com/mediatek–dimensity–9600–pro–announced/”>Source link a>
