Intel is set to roll out its latest generation of processors later this year despite a minor setback affecting ultra low-voltage models — the ones destined for super slim notebooks. By normal standards, the launch should mark a new “tick” in the company’s product roadmap, but Intel is going beyond just shrinking the current 32nm Sandy Bridge processor by introducing some fundamental advancements along with its new 22nm process.
For those unfamiliar, Intel follows a “tick tock” model for its processor upgrade cycle. With every “tick” the company moves to a smaller manufacturing process, from 32nm to 22nm in this case, dramatically increasing transistor density while enhancing performance and energy efficiency of the current microarchitecture. Then, with an alternating “tock” cycle Intel introduces a new processor microarchitecture.
Ivy Bridge includes manufacturing and subsystem improvements. It is a shrink of Sandy Bridge and is also the first to us Intel’s Tri-Gate transistors, which use a nonplanar architecture to cram more transistors into less space, therefore consuming less power or delivering more performance within the same power envelope.
There’s been quite a bit of information on Ivy Bridge going around ever since Intel detailed the architecture late last year. We’ll recap some of the major changes and practical implications, while also bringing you up to speed on the latest developments, including expected launch lineup and specs.
Tri-Gate transistors = Improved efficiency, performance
Unlike conventional planar transistors that lay flat, Ivy Bridge’s Tri-Gate transistors use a three-dimensional fin that stands vertically from the silicon substrate. This presents several benefits. For starters, Intel can cram more transistors into less space, which will be incredibly valuable as fabrication tech shrinks to 22nm and beyond.
In addition, the new design allows for essentially three times the surface area for electrons to travel when the transistor is in the ‘on’ state, which paves the way for increased performance.
Transistors carry an electrical signal while gates control that flow by turning the current on and off. Whereas in a typical transistor only the small layer between the channel and the gate becomes active when the transistor is switched on, Intel’s Tri-Gate transistor creates a three-sided silicon fin that the gate wraps around, increasing the surface area where electrical current actually flows. The video below does a better job explaining this.
This design also maximises transistor switching performance between on and off states and decreases power-wasting leakage.
Intel summarises the practical implications by saying the 22nm 3D Tri-Gate transistors provide up to a 37 per cent performance increase at low operating voltages versus Intel’s 32nm planar transistors — a big deal for Atom and ULV chips — or close to 20% at 1V for higher end desktop and mobile parts.
Alternatively, the new 22nm Tri-Gate transistors can consume less than half the power when at the same performance level as 2D planar transistors on 32nm chips.
Intel has also mentioned the possibility to have multiple fins standing vertically from the silicon substrate and connected together, as shown to the right, to increase total drive strength of the transistor for higher performance. They haven’t discussed this in detail but we assume Intel could use it to more finely tune its 22nm process in higher end products, or use it as a fail-safe method to improve yields of individual dies.
The new 22nm Tri-Gate wafers shouldn’t be much more expensive to produce, either. Compared to a hypothetical Intel 22nm planar process, the 3D Tri-Gate process should only add another 2-3 per cent to the total cost, according to Intel’s own estimates.
Other architecture changes
Besides the new transistor design there are no major changes in the Ivy Bridge architecture compared to Sandy Bridge. It continues the 2-chip platform partition (CPU + PCH) and is backwards compatible with existing LGA-1155 motherboards, although there will be new chipsets to enable new features.
The central portion of the die has four x86-64 cores with 256KB dedicated L2 cache each and a shared 8MB L3 cache. To each side of this central portion is the system agent and the graphics core.
All these components are bound by a ring-bus that transports data between them. The system agent has interfaces for the dual-channel DDR3 integrated memory controller, the PCI-Express controller (supporting 16 PCIe 3.0 lanes), the DMI chipset bus, a display controller and FDI link to the PCH.
But there are also a few tweaks here and there. First and foremost the graphics core has been completely redesigned and now supports OpenCL 1.1, DirectX 11 and OpenGL 3.1. This will finally bring the Intel integrated GPU to feature parity with AMD’s. Intel also added a graphics-specific L3 cache, three display outputs (up from two in Sandy Bridge), better anisotropic filtering, more shaders or execution units (either 8 or 16 EUs in Ivy Bridge depending on the GPU versus 6 or 12 in Sandy Bridge), and a few other enhancements.
Ivy Bridge also greatly improves Intel Quick Sync Video, the chip giant’s transcoding technology. All told, the end result is up to a 60 per cent increase in GPU performance over Sandy Bridge’s integrated GPU.
Hyper-Threading and CPU instruction set changes On the CPU side there are some changes in the way resource allocation for HyperThreading queue takes place. Ivy Bridge will dynamically allocate resources to threads so that if there is only a single thread active, all resources will be dedicated to that thread rather than some going unused as with SB’s static allocation.
There’s a new random number generation process that improves security, a power management feature that offers more flexibility in setting a system’s thermal envelope (more on this next), and memory and string performance enhancements. Ivy Bridge also reportedly allows for more dynamic overclocking.
Configurable TDP & Other Power Optimisations
The move to a 22nm process and Tri-Gate transistors alone should already account for some pretty significant power savings. But there are a few other changes in Ivy Bridge meant to optimise power consumption.
An important addition brought to mobile Ivy Bridge processors is the inclusion of a configurable TDP that allows them to switch between three different ratings: nominal, a lower configurable TDP and an upper configurable TDP. Ultra low voltage (ULV) parts will be rated at 17W, similar to existing ULV Sandy Bridge processors, but can go up to 25W — with a corresponding rise in performance — when running under mains power, or when it’s connected to a docking station that increases the amount of cooling to dissipate the additional heat.
This goes beyond Intel’s existing Turbo Boost feature as it exceeds the CPU’s nominal TDP, whereas current Sandy Bridge chips are mostly bound by it. Likewise, a 17W ULV processor could go down to a mere 14W to save battery life when running light tasks on the go. Besides ULV chips, extreme edition mobile Ivy Bridge processors will also support configurable TDP, with 55W parts able to go up to 65W or down to 45W.
|cTDP Down||Nominal||cTDP Up|
|Ivy Bridge ULV||14W||17W||25W|
|Ivy Bridge XE||45W||55W||65W|
Configurable TDP will be exclusive to mobile processors as far as we know. All models can go down in terms of TDP, but not all will be able to go up. Also, notebook manufacturers would presumably have some freedom to “configure” the chip around the system they want to build rather than doing it the other way.
For desktops, TDP will come in 35W, 45W, 55W, 65W and 77W options, according to the latest leaked Ivy Bridge roadmaps, down from the current peak of 95W for non-Extreme Sandy Bridge parts.
Other power draw optimisations
- Lower System Agent Voltages: The System Agent is the uncore area of the CPU die containing the display output, memory controller, DMI and PCI Express interfaces. It operates on a separate voltage plane than the rest of the chip and with Ivy Bridge Intel will be able to power optimise some SKUs — likely ULV processor models — even further with lower System Agent operating voltages.
- Power Aware Interrupt Routing (PAIR): This feature is meant to improve Intel’s core sleeping technology by making the CPU aware of which of its cores are asleep and which are awake. It can then send interrupt requests from peripherals or a software application to cores that are up and running, rather than waking a core that has been powered down to handle the interrupt.
- Support for DDR3L on mobile CPUs: Intel upgraded the dual-channel DDR3 memory controller of mobile Ivy Bridge to accept ultra-low voltage DDR3, or DDR3L, which could shave a few extra watts from a system’s total power draw. Ivy Bridge will also switch off the DDR I/O to save power when idle.
- Optimised voltage choice for all operating frequency points: Like Sandy Bridge, Ivy Bridge can vary its operating frequency depending on the task at hand, and Intel will use the lowest voltage possible for each one of those frequencies. The chipmaker says it has fine tuned this even further with Ivy Bridge.
Jose Vilches is the managing editor at TechSpot. TechSpot is a computer technology publication serving PC enthusiasts, gamers and IT pros since 1998. [imgclear]
Republished with permission.