In August, Intel ran certainly one of its uncommon Architecture Days the place the corporate went into some element about its upcoming Tiger Lake processor. This included goal markets, core counts, graphics counts, a glance into a few of the new acceleration options, and a promise of a product launch later within the 12 months. That product launch is now right here, and Intel is offering Tiger Lake with speeds and feeds, offering element and anticipated benchmark efficiency for Intel’s subsequent era of notebook-class gadgets.
What We Already Know
Recapping what we discovered at Intel’s Architecture Day 2020, the usual type of Tiger Lake is a quad core processor with an up to date graphics structure constructed on Intel’s newest manufacturing course of.
These new 11th Generation Core processors use 4 of Intel’s newest cores constructed with the Willow Cove microarchitecture – a barely modified model of the Sunny Cove microarchitecture present in Intel’s 10th Generation Ice Lake processors. These new cores supply extra efficiency than earlier than, peaking at 4.eight GHz in comparison with the 4.Zero GHz seen within the earlier era, a 20% enchancment. However the underlying clock-for-clock efficiency enhancements are minimal, with Intel as an alternative specializing in that frequency acquire.
The motive for the frequency acquire comes from the manufacturing course of – previously Intel’s 10++ course of, renamed as soon as to 10+ after which once more throughout Architecture Day to its new identify: Intel’s 10nm TremendousFin know-how. The purpose with this manufacturing course of was two-fold: firstly to extend effectivity and scalability to allow increased frequencies, but additionally to enhance yields. As a end result, we’ll see Intel claiming increased frequencies, that result in increased efficiency, at comparable energy ranges to the earlier era. These cores help AVX-512, in addition to Intel’s DL-Boost acceleration libraries.
On the graphics facet, Tiger Lake makes use of Intel’s new Xe-LP graphics structure, which the corporate additionally detailed at Architecture Day – we’ve a separate article explaining the variations. Simply put, Xe-LP will increase the uncooked thread depend and compute depend per execution unit (EU), in addition to the cache hierarchy and a few acceleration options. For Ice Lake we noticed 64 EUs working at 1100 MHz, whereas for Tiger Lake we’ll see 96 EUs (+50%) working at 1350 MHz (+22%). Add in a few of the different advantages and we must always supposedly see a 2x enchancment in graphics efficiency in comparison with the earlier era on paper. For AI compute, the graphics additionally helps DP4A directions for INT8 inference workloads.
The graphics show pipeline has been improved, with help for AV1 decode, in addition to show pipes for as much as 4 4K60 shows or a single 8K60 show. There is a direct show port to reminiscence to enhance latency – reminiscence help on Tiger Lake contains LPDDR4X-4267 (as much as 32 GB), DDR4-3200 (as much as 64 GB), and when out there, we’ll see LPDDR5-5400 gadgets in 2021. For accelerators, Intel has additionally boosted its Gaussian Neural Accelerator to model 2.0, to assist with offloading less complicated AI inference workloads similar to noise cancellation.
Other enhancements to Tiger Lake embrace native Thunderbolt Four help, with the controller embedded into the CPU permitting for as much as 4 TB4 ports per machine. Wi-Fi 6 help can be enabled by means of a CNVi interface. Intel’s energy/frequency algorithms are additionally up to date, permitting for separate scaled energy and clocks on the CPU, GPU, and reminiscence cloth. Intel said that it has doubled the interior cloth bandwidth permitting all these elements to speak with one another with extra information.
Today’s Announcements
Speeds, Performance, New Style
The distinction between the earlier Architecture Day and right this moment’s Tiger Lake launch is that right this moment is extra…
