Five gigahertz in a phone. I can't wait for all Android manufacturers to resell this number as something unique to them next year, even though the Snapdragon 8 Elite Gen 6 will be present in every flagship phone.<\/p>\n\n\n\n
Or is it? This year marks the first time that Qualcomm has introduced two chips, the "regular" Snapdragon 8 Elite Gen 6 and the Snapdragon 8 Elite Extreme Gen 6. I hate their naming of these chips, but it's worth it for the smile on my face when I see what they can do.<\/p>\n\n\n\n
Both use a 2nm process, both have eight Oryon cores, and both reach 5GHz. So the Extreme isn't interesting because it has a processor that is a class faster. The difference starts elsewhere: in graphics, UI processing, video, and the amount of dedicated hardware for tasks that a few years ago the GPU couldn't even handle.<\/p>\n\n\n\n
Qualcomm claims 13-% higher CPU and 44-% higher graphics performance for the Snapdragon 8 Elite Extreme Gen 6 compared to the previous generation, while the regular Elite Gen 6 has 10 and 35 %, respectively.<\/p>\n\n\n\n
The processor part of both Snapdragons consists of two main Oryon cores at a maximum of 5 GHz and six performance cores operating at up to 4 GHz. So this year Qualcomm has also been left without the classic set of small efficient cores, as we know from many other mobile processors. All cores are relatively powerful, the difference is mainly in their position in the hierarchy and frequencies.<\/p>\n\n\n\n
On paper, 5 GHz is a fantastic number. In reality, a phone is not a desktop computer with a large heatsink and fan, unless we're talking about RedMagic phones, and even then there are limits to what a fan and water cooling can do.<\/p>\n\n\n\n
When I launch a game, export a video, or use a local language model for an extended period of time, I don't care what the processor does in the first ten seconds. I care what it does ten or twenty minutes later.<\/p>\n\n\n\n
Maybe a 2-nanometer manufacturing process will improve efficiency? A smaller manufacturing process does not in itself mean that each transistor is literally two nanometers wide. The names of the manufacturing nodes have long been not a direct physical measurement of the individual part of the transistor. But they are a good indicator of the generation of manufacturing technology that allows for higher transistor density and, as a rule, a better ratio between performance and power.<\/p>\n\n\n\n
Qualcomm claims that the new CPU is up to 37% more power efficient than the previous generation on both chips. If that turns out to be true, I'll be even more excited about that than the 5GHz frequencies.<\/p>\n\n\n\n
And of course, a lot will depend on the phone. The RedMagic with an active fan, the slim Samsung, and the foldable Honor may all use the same chip, but will achieve very different results after twenty minutes of stress.<\/p>\n\n\n\n
What also caught my attention with the new Oryon is the change in cache. Both processors have 16MB of Oryon FlexCache that can be accessed by the processor cores. Qualcomm uses it as a shared pool of second-level cache, rather than individual cores or groups of cores being more strictly tied to their blocks.<\/p>\n\n\n\n
A processor can be fast, but if it has to constantly wait for data from system memory, part of its capacity is simply sitting idle. Cache is a small, very fast storage right next to the processor cores, where the processor stores data that it expects to need again soon. The more data that can stay close to the CPU, the less often it has to access the much slower and more energy-intensive system memory.<\/p>\n\n\n\n
Both get the next-generation Adreno graphics. Qualcomm claims up to 35 % more graphics performance and up to 40 % better efficiency for the standard Elite Gen 6, while the Extreme offers up to 44 % more performance and up to 40-% better efficiency.<\/p>\n\n\n\n
But the Extreme also gets Adreno Matrix Cores and Adreno Neural Fusion. Matrix cores are dedicated computational units for the operations that power neural networks. If you're familiar with the concept from Nvidia's graphics cards and their Tensor cores, the idea is similar. Until now, it's been logical to think of the GPU and NPU as two fairly separate blocks on a phone. The GPU renders graphics, while the NPU powers UI models. Neural Fusion aims to remove some of that divide.<\/p>\n\n\n\n
Dedicated matrix cores are directly in the graphics part, so the UI can help create the final image without the system having to send data from the graphics pipeline to another processor block every time. Qualcomm talks about neural super-resolution, frame generation and advanced image reconstruction. In practice, this means that the GPU is no longer only responsible for classic drawing of triangles, textures, lighting and effects. Part of the image can be reconstructed or improved by a neural model.<\/p>\n\n\n\n
This is a very similar direction to what we have been able to observe on computers for some time, for example DLSS, FSR...<\/p>\n\n\n\n
Instead of always rendering the game at full target resolution, I can render it at a lower resolution and then use the model to reconstruct a sharper final image. If I can do the same with intermediate frames, I reduce the amount of classical computation that the GPU has to do. On a phone, this approach is perhaps even more important than on a desktop, because every watt saved counts.<\/p>\n\n\n\n
The Extreme also has 18MB of dedicated Adreno High Performance Memory, a fast memory directly next to the graphics section. Its task is similar to that of the CPU cache: to reduce the need to constantly access system RAM.<\/p>\n\n\n\n
Now all we need is games. Without developer support, Matrix Cores are just a very interesting piece of silicon that sits idle most of the time. That's why Qualcomm is working directly with game engines and developers. I hope this collaboration will bear fruit.<\/p>\n\n\n\n
Qualcomm is talking about around 14-% more NPU performance in the standard Snapdragon 8 Elite Gen 6. The Extreme is a much bigger jump. The NPU is said to be 35-% faster and offer up to 33-% better performance-to-power ratio.<\/p>\n\n\n\n
More and more local UI will need to be running all the time, not just when I open an app and type a question. That's why Qualcomm is using a redesigned Sensing Hub that includes smaller processing units and two micro NPUs. The goal is to perform simpler tasks with significantly less power, without having to wake up the main Hexagon NPU or processor.<\/p>\n\n\n\n
With Extreme, Qualcomm also talks about a large enough memory space to run Mixture-of-Experts models with more than 30 billion parameters. In the MoE model, not all 30 billion parameters are used at each step. Only a part or individual \u201cexperts\u201d are activated, so it is possible to have a fairly large shared model without the processor having to calculate through the entire network for each token created.<\/p>\n\n\n\n
Much of the final image is created in silicon. The Snapdragon 8 Elite Gen 6 uses new algorithms and an image pipeline with Intelligent Pixel Control, Elite Color Engine, and Motion Intelligence. Qualcomm calls Intelligent Pixel Control an image analysis at the pixel level, primarily to reduce flare, halos, and other artifacts. Motion Intelligence can use motion information for stabilization.<\/p>\n\n\n\n
The classic ISP used to be mostly concerned with noise reduction, color processing, and HDR. Today, it has to work with the NPU, understand the content of the frame, separate the subject from the background, and do all this dozens or hundreds of times per second in video. The Extreme is significantly more ambitious in this area. It supports 8K recording at 60 frames per second and 4K at 240 frames per second, and Qualcomm also mentions simultaneous support for three 64-MP cameras.<\/p>\n\n\n\n
8K60 is technically impressive, although I still shoot in 4K much more often. I find 4K240 more interesting. At such a high speed, I can get very high-quality slow motion without dropping to Full HD.<\/p>\n\n\n\n
Both processors use the Snapdragon X105 modem, which supports 5G Advanced and theoretical transfer speeds of up to 14.8 Gbps towards the user and up to 4.2 Gbps in the other direction. No, I won't be measuring 14.8 Gbps on my phone tomorrow. For years, the maximum theoretical speeds of modems have been more of a representation of technological maximums than real everyday results.<\/p>\n\n\n\n
Qualcomm says the new RF part uses up to 30 % less power, and the new antenna configuration is said to improve reception and throughput in poor conditions. It's a similar story with the FastConnect 8800. The system supports Wi-Fi 8, 320-MHz channels, 4K QAM, and a maximum theoretical speed of about 11.6 Gbps. Qualcomm also highlights a 4x4 Wi-Fi configuration. Wi-Fi 8 is far from a reason to buy a new phone right now. But it's nice to know that the chip that will power phones in 2027 and probably beyond will also be ready for the next generation of networks.<\/p>\n\n\n\n MediaTek has also switched to 2nm manufacturing with the 9600 Pro chip, but it uses Arm cores and a different CPU configuration. The design is 2 + 3 + 3. Two of the most powerful Arm C2-Ultra cores and two groups of three C2-Pro cores, differing in the amount of L2 cache. So MediaTek also does not have the classic small cores and continues its "All Big Core" philosophy.<\/p>\n\n\n\n MediaTek claims a 17-% increase in single-core and 15-% increase in multi-core performance over its predecessor. It is said to deliver a 37-% improvement in single-core and a 61-% improvement in multi-core.<\/p>\n\n\n\n The Dimensity 9600 Pro has two very interesting aces: LPDDR6 and UFS 5.0. MediaTek claims up to 33 % more effective throughput at the same frequency for LPDDR6, while UFS 5.0 is about twice as fast in reading and writing compared to the previous generation. MediaTek also has an NPU 1090 and a separate Super Efficient NPU 2.0, which is conceptually very similar to Qualcomm's division into the main NPU and smaller units for constantly running tasks. For graphics, it uses Mali-G2 Ultra NX, the latter's performance is said to have grown by 27 %, efficiency by 24 %, and ray tracing by up to 18 %.<\/p>\n\n\n\n Apple introduced the A20 Pro this year, which is also made on a 2-nanometer process. The CPU has six cores, the GPU has seven, and for the UI, Apple uses a new Dual 16-core Neural Engine, which is a total of 32 neural cores. Apple claims twice the computational power of the Neural Engine compared to the A19 Pro, and the GPU is said to be up to 40 % faster. Memory bandwidth has increased by 50 %.<\/p>\n\n\n\n What about cooling? Apple has changed the packaging of the chip so that the memory is no longer directly in the thermal path between the silicon and the cooling system. The chip can be more efficiently connected to the vapor chamber, which in the iPhone 18 Pro has three times the surface area of the previous generation. As a result, Apple promises up to 40% higher sustained performance.<\/p>\n\n\n\n <\/p>\n\n\n\n The Snapdragon 8 Elite Extreme Gen 6 has the most aggressive design for neural graphics and advanced video. The Dimensity 9600 Pro is extremely interesting due to the combination of All Big Core CPU, dual NPU, LPDDR6, UFS 5.0 and very aggressive promises of power efficiency. However, Apple is still playing its game of tight integration of the chip, memory, software and cooling system.<\/p>","protected":false},"excerpt":{"rendered":" Pet gigahercev v telefonu. Komaj \u010dakam, da bodo vsi Android proizvajalci v naslednjem letu preprodajali to \u0161tevilko kot nekaj edinstvenega samo za njih, \u010deprav bo Snapdragon 8 Elite Gen 6 prisoten v \u010disto vsakem \u00bbflagship\u00ab telefonu. Ali pa\u010d? Letos je prvi\u010d, da je Qualcomm predstavil dva \u010dipa, \u010disto \u00bbobi\u010dajni\u00ab Snapdragon 8 Elite Gen 6 in […]<\/p>","protected":false},"author":2,"featured_media":0,"comment_status":"","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[36,32,4,16,12],"tags":[1187,1189],"class_list":["post-8217","post","type-post","status-publish","format-standard","hentry","category-procesorji","category-komponente","category-racunalnistvo-telefonija","category-oprema","category-triki-nasveti","tag-qualcomm","tag-snapdragon-cipi"],"acf":{"subtitle":"","heading":"","summary":"Novo leto, nov Snapdragon \u010dip oziroma kar dva nova \u010dipa. Eden \"obi\u010dajen\", drugi pa bolj \"ekstremen\". Zmogljivosti bodo zagotovo nore, ampak to pomeni, da bodo proizvajalci telefonov samo \u0161e pod ve\u010djim pritiskom, da kon\u010dno poskrbijo za bolj\u0161e hlajenje.","thumbnail_small":"https:\/\/racunalniske-novice.com\/wp-content\/uploads\/2026\/09\/Snapdragon-8-Elite-Extreme-Gen-6-560x315.webp","thumbnail_large":"https:\/\/racunalniske-novice.com\/wp-content\/uploads\/2026\/09\/Snapdragon-8-Elite-Extreme-Gen-6-1024x512.webp","thumbnail_caption":"","gallery":"","video_gallery":null,"author":"Avtor: Peter \u010cebron","links":null,"sources":null,"skip_language":[]},"yoast_head":"\n<\/th> Snapdragon
8 Elite Gen 6<\/strong><\/th>Snapdragon
8 Elite Extreme Gen 6<\/strong><\/th>MediaTek
Dimensity 9600 Pro<\/strong><\/th>Apple
A20 Pro<\/strong><\/th><\/tr><\/thead>Production<\/strong><\/td> 2 nm<\/td> 2 nm<\/td> 2 nm<\/td> 2 nm<\/td><\/tr> CPU<\/strong><\/td> Orion<\/td> Orion<\/td> Arm C2<\/td> Apple CPU<\/td><\/tr> Cores<\/strong><\/td> 2 + 6<\/td> 2 + 6<\/td> 2 + 3 + 3<\/td> 6 cores<\/td><\/tr> Max. frequency<\/strong><\/td> 5GHz<\/strong><\/td> 5GHz<\/strong><\/td> 4.55GHz<\/td> 4.93GHz<\/td><\/tr> Cache<\/strong><\/td> 16 MB FlexCache<\/td> 16 MB FlexCache<\/td> 16MB L3 + 10MB SLC<\/td> 16MB<\/td><\/tr> GPU<\/strong><\/td> Adreno<\/td> Adreno<\/strong><\/td> Mali-G2 Ultra NX<\/td> 7-core Apple GPU<\/td><\/tr> Neural graphics<\/strong><\/td> More basic version<\/td> Matrix Cores + Neural Fusion<\/strong><\/td> GPU-NPU Neural Graphics<\/td> Neural Accelerators in GPUs<\/td><\/tr> NPU<\/strong><\/td> Hexagon<\/td> More powerful Hexagon<\/strong><\/td> NPU 1090<\/td> Dual 16-core Neural Engine<\/td><\/tr> Low-power UI<\/strong><\/td> Sensing Hub + Micro NPU<\/td> Sensing Hub + Micro NPU<\/td> Super Efficient NPU 2.0<\/td> Integrated architecture<\/td><\/tr> 8K \/ 60fps<\/strong><\/td> No<\/td> Yes<\/strong><\/td> Yes<\/strong><\/td> Not exposed<\/td><\/tr> 4K \/ 240fps<\/strong><\/td> No<\/td> Yes<\/strong><\/td> Yes<\/strong><\/td> Not exposed<\/td><\/tr> APV<\/strong><\/td> Not exposed<\/td> Yes<\/strong><\/td> \u2014<\/td> \u2014<\/td><\/tr> 5G<\/strong><\/td> Snapdragon X105<\/td> Snapdragon X105<\/td> Integrated 5G Advanced<\/td> Apple C2<\/td><\/tr> Wi-Fi<\/strong><\/td> Wi-Fi 8<\/strong><\/td> Wi-Fi 8<\/strong><\/td> Wi-Fi 7<\/td> Wi-Fi 7 via N1<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n MediaTek Dimensity 9600 Pro: Qualcomm has no easy opponent this year<\/strong><\/h3>\n\n\n\n
Apple has fewer cores, but that hasn't meant less performance for a long time<\/strong><\/h3>\n\n\n\n