Yesterday, earlier than its Government Summit, MediaTek introduced its flagship SoC, the Dimensity 9200, the successor of the MediaTek Dimensity 9000. Alongside higher efficiency and enhanced energy effectivity, the brand new chipset brings many “firsts” to the cell semiconductor business.
The Dimensity 9200 is the primary chip made with the TSMC 2nd Gen 4nm Course of and to function an Arm Cortex-X3 main core clocked at 3.05 GHz primarily based on the Gen 2 ARMv9 structure.
8 Cores CPU for efficiency and energy effectivity
The guts of this system-on-chip (SoC) is a heterogeneous CPU cluster combining 4 efficiency cores and 4 effectivity cores. The Cortex-X3 (3.05 GHz) and three Cortex-A715 (2.85 GHz) are used for the workload that requires efficiency, whereas the 4 Cortex-A510 (1.8GHz) are chosen by the cluster scheduler and OS thread scheduler to course of much less compute-intensive duties.
As measured by MediaTek utilizing GeekBench 5.0, the single-core efficiency has elevated by 12%, and the multi-core efficiency is 10% higher than the Dimensity 9000. The improved energy effectivity is essentially the most spectacular achieve over the earlier technology. The brand new chip consumes 25% much less energy regardless of the elevated efficiency.
The CPU cores entry speedy LPDDR5X RAM with as much as 8,533Mbps of bandwidth, an business first provided by MediaTek within the Dimensity 9200.
For storing knowledge completely, the UFS 4.0 with Multi Round Queue (MCQ) storage communication protocol provides the quickest knowledge transmission at the moment obtainable for direct entry to storage.
Immortalis-G715, 11 Cores GPU with {hardware} ray tracing
Equally to the Dimensity 9000’s graphics processor Mali G710, the Immortalis-G715 GPU options {hardware} ray tracing (RT).
We have now not seen particular ray tracing efficiency indicators, however MediaTek’s messaging means that RT may very well be utilized in 3D video games for fundamental use circumstances corresponding to comfortable shadows, reflections, and ambient occlusion.
MediaTek’s GFXBench Manhattan benchmark numbers present that the GPU is 32% sooner than the earlier technology whereas utilizing 41% much less energy.
“THE GPU IS 32% FASTER THAN THE PREVIOUS GENERATION WHILE USING 41% LESS POWER.”
Gaming and Viewing Expertise Enhancements
On the gaming facet, the brand new SoC options MediaTek’s HyperEngine 6.0 Gaming Expertise that mechanically optimizes system settings to make sure most gaming efficiency.
In accordance with MediaTek, the chipset’s MiraVision 890 show know-how intelligently adjusts many show and video stream parameters and supplies {hardware} and software program optimizations that enhance viewing high quality.
Moreover, for a greater gaming expertise, the brand new {hardware} helps Full HD+ as much as 240Hz, WHQD as much as 144Hz, and 5K (2.5Kx2) as much as 60Hz, with an adaptive refresh price.
APU 690, sixth Gen AI Processing Unit
The AI compute unit named APU 690 is the sixth technology AI Processing Unit developed by MediaTek.
On paper, the APU 690 ought to carry a 35% efficiency enchancment over the earlier technology within the ETHZ 5.0 benchmark and 45% decrease energy consumption for AI-super decision duties (AI-SR).
For instance, while you play a 1080p video saved in your smartphone on a 4K monitor or TV, the APU will assist the graphics system upsample the video with larger high quality. This activity is compute-intensive, so it’s nice that MediaTek made it considerably extra energy environment friendly.
The AI structure has been improved from the earlier APU with combined precision. The fifth Gen used the standard integer 16 computing methodology, which signifies that every complete quantity (integer) was all the time encoded with 16 bits. With mixed-precision computing, the APU 690 can now course of each floating level (fractional quantity) and integer (complete quantity) with varied quantity binary codecs, 4 bits, 8 bits, or 16 bits.
For example, 4 bits encode numbers from 0 to 7 while16 bits encoding reaches 65,535. For numbers from 0 to 7 or from 0 to 255, there isn’t a want to make use of the 16 bits format. Permitting the arithmetic models within the {hardware} to function on extra knowledge blocks concurrently (parallelism). MediaTek claims that the introduction of combined precision for the APU 690 boosts the computing velocity by 98%.
“MIXED PRECISION BOOSTS THE APU 690’S COMPUTE SPEED BY 98%”
Digital camera options and Imagiq 890 picture sign processor (ISP)
The APU often performs a significant function for AI-driven picture segmentation. The GPU and the ISP may take part in that activity.
The brand new {hardware} permits real-time detection of picture and video content material to establish the totally different areas of the picture, corresponding to folks, sky, ocean, buildings, greenery, foreground, and background. As soon as picture segmentation is executed, the system can apply totally different filters or algorithms to the varied elements of the scene in actual time. It’s used for saturation, colour, and distinction tuning.
AI-driven picture segmentation enhances picture high quality for each the digicam and the show.
MediaTek makes use of AI within the Dimensity 9200 to allow the Twin Stream AI Shutter function, eliminating the blur in motion and sports activities pictures. A function that MediaTek nicknamed “Movement Unblur.”
The Imagiq 890 is the primary RGBW ISP to help the brand new RGBW sensors. Along with the standard RGB pixels, the brand new sensors have white pixels to seize brightness no matter colour. This helps to shoot higher evening footage as MediaTek claims a 30% enhance in brightness for evening pictures.
Connectivity
The Dimensity 9200 is the primary SoC to help WiFi 7 with as much as 6.5 Gbps knowledge velocity.
The chipset helps mmWave 5G and sub-6GHz connectivity and options mmWave SMART Beamforming which is 25% sooner than common beamforming, in response to MediaTek. The brand new mmWave know-how delivers a formidable 7.9 Gbps peak velocity for knowledge switch!
Filed in . Learn extra about MediaTek, Processors, SoC and TSMC.
