Intel Ice Lake: 10nm, fast GPU, and a new architecture
After many years of yearly reiterations of what was basically the same design over and over again, Ice Lake finally contains major architectural improvements over its predecessors, particularly in regard to CPU cores and integrated GPU. Developed and designed over the last four years and manufactured in 10nm in Intel’s Israeli Fab28, the first available Ice Lake chip features four CPU cores and up to 64 GPU units and will only be available as mobile U and Y-series for the time being.
The chip and an addition PCH can be found on two different packages. The more powerful U-series requires more space and is typically rated at 15 to 28 W. The more energy-efficient Y-series is made for smaller notebooks and tablets and therefore fitted onto a smaller overall package. However, given that it is based on the same core design, it features four physical cores as well.
Basically, Intel’s Ice Lake architecture is characterized by three features:
- First mass-produced 10nm chip by Intel. Cannon Lake can be considered a first test run.
- New CPU and GPU architecture.
- Thunderbolt and Wi-Fi 6 partly integrated into chip and/or package.
|Razer Blade Pro 17 RTX 2080 Max-Q||9750H||GeForce RTX 2080 Max-Q||17.3"||1920x1080||2.8 kg|
|MSI GS75 Stealth 9SG||9750H||GeForce RTX 2080 Max-Q||17.3"||1920x1080||2.3 kg|
|Razer Blade 15 RTX 2080 Max-Q||9750H||GeForce RTX 2080 Max-Q||15.6"||1920x1080||2.2 kg|
New CPU Architecture: Sunny Cove
In order to improve performance, Intel implemented two separate strategies:
- Increase IPC by up to 18%
- Implement new instructions functional units for popular and often-used scenarios
In order to increase overall chip performance the pipelines were both widened and deepened, the memory subsystem received an overhaul, and most caches were increased in size. According to Intel the entire architecture was revised. In other words: Ice Lake is more than a minor tweak.
The result is an up to 18% increased IPC (instructions per clock cycle), which Intel determined by running various benchmarks.
This IPC improvement compensates for the fact that Ice Lake runs at lower clock speeds than the optimized-for-high-clock-speeds 14nm++ Whiskey Lake CPUs.
In addition, a number of new instructions as well as functional units for AI / machine learning (DL Boost), cryptography, and compression/decompression have been implemented. Furthermore, AVX-512 is now supported on Ice Lake, which was previously only available in server-class processors. However, in order to benefit from these new instructions, applications will need to implement them first. We thus only expect newly released applications to benefit from this particular architectural improvement.
In addition, Ice Lake is the first processor to support Intel Dynamic Tuning 2.0 with Machine Learning, a technique developed by Intel using machine learning algorithms to determine future CPU core utilization in order to optimize Turbo Boost and apply it more intelligently. A best-case scenario estimates performance improvements of up to 5%.
More GPU Power: Integrated Gen 11 GPU
The integrated GPU was heavily modified and upgraded. The unified shaders (aka execution units) have been widened, and the pixels per clock cycle have been doubled in order to increase per-core performance even further. Units with dedicated eDRAM, such as the older Iris SKUs (e.g. Iris Plus 650 or Iris Pro 580) were not yet available at the time of writing. However, the most powerful Iris Plus model has access to 3 MB of L3 cache located in direct proximity to the GPU and 0.5 GB of shared memory.
A maximum performance of up to 1.12 TFLOPS in FP32 and 2.25 TFLOPS in FP16 based on the maximum clock speed of 1.1 GHz and 64 execution units compares nicely to the GeForce MX250’s 1.25 TFLOPS in FP32. Intel also offers GPUs with fewer execution units and therefore a lower peak performance. The slowest GPU is referred to as UHD Graphics and can be found in all CPUs suffixed with G1. So far, we were able to identify three separate GPUs within Intel’s U-series:
In addition to more oomph the new GPUs have also been enhanced to incorporate a new feature called variable rate shading, which enables developers to render less eye-catching areas on-screen to be rendered and shaded in a lower resolution in order to improve performance.
A total of three display pipes are available, each with support for [email protected] Hz. Two of these can be combined in order to drive a single 8K display. The GPU supports the VESA Adaptive Sync standard (aka FreeSync) that synchronizes the display's refresh rate to the game’s frame rate for a smoother gaming experience. This technology can help make lower frame rates seem smoother than they normally would. Another new feature is the FP16 HDR display pipeline, which enables support for HDR10 and Dolby Vision.
The media encoders have been upgraded and are now not only supposed to work faster but also offer better quality. Intel integrated two encoders with support for HEVC and VP9 up to 4K60 10b 4:4:4 or 8K30 10b 4:2:0.
In order to lower costs for integrating Thunderbolt 3 and the upcoming USB 4.0 standard, Intel integrated the necessary controller directly into the SoC. The CPU can provide two or four ports, and all manufacturers have to do is to implement a simple retimer per port as well as a Power Delivery (PD) controller. The Thunderbolt controller is supposed to consume no more than 300 mW of energy under full load and will therefore not expose the power supply or cooling solution to any amount of noticeable additional stress. Y-series chips in particular should benefit greatly from this feature due to their heavily restricted thermal and power budgets.
Top 10 Smartphones
Smartphones, Phablets, ≤5-inch, Camera SmartphonesNotebookcheck's Top 10 Smartphones under 160 Euros
Fast Wi-Fi on Package
While not directly on-chip, the Wi-Fi 6 (802.11ax CNVi 2) modem is integrated into the on-package PCH (platform controller hub, the artist formerly known as Southbridge) and requires an additional AX201 RF chip.
Intel is thus one of the first manufacturers to support 802.11ax and intends to provide one of the fastest wireless experiences with its Wi-Fi 6 Gig+ feature with support for 160 Hz and OFDMA. Wi-Fi 6 is particularly interesting in networks with a large number of clients.
While the first Ice Lake chips are not necessarily significantly more powerful regarding CPU performance, the faster GPU, support for Wi-Fi 6, and the integrated Thunderbolt controller are major advantages over their predecessors. In addition, the Y-series benefits from two extra cores. That being said, the new chips will still require proper cooling and integration in order to perform adequately despite their lower power consumption thanks to the modern and slowly maturing 10nm manufacturing process. However, we don’t expect any big surprises in energy efficiency and battery life as Intel conveniently skipped over this aspect in its presentation.
Please find our latest articles referring to Intel’s Ice Lake architecture below.
Intel Ice Lake