Release dates and landmarks for major desktop, server and mobile CPUs
Created by cmclellan on 03/06/2011
Last updated: 24/04/13 at 10:03
Tags: cpu processor Intel AMD Nvidia Qualcomm Samsung ARM
The successor to Haswell (and its 14nm die shrink Broadwell) will be the Skylake microarchitecture. Based on a 14nm process, the first chips will be produced in 2015 and move to a 10nm process, codenamed Skymont, in 2016. Skylake is expected to introduce support for DDR4 memory, which will run at clock speeds of 2,133–4,266MHz.
The sixth-generation Nvidia Tegra 6, codenamed Parker, will pair new 64-bit 'Denver' ARM CPU cores with Nvidia's forthcoming Maxwell GPU architecture. Parker will be the first of Nvidia's processors to use TSMC FinFET 3D transistor technology that reduces die size and power consumption.
Excavator is a fourth-generation modular core offering a performance boost over the third-generation Steamroller microarchitecture. It will surface first in the A-series Fusion APUs, before moving on, in revised form, to the FX-series and Opteron ranges in 2015.
Airmont architecture on 14nm process. Low-power SoC for ultrabooks, tablets and smartphones.
The Cortex-A50 series is ARM's first 64-bit processor core based on the ARMv8 architecture. The A57 is the successor to the Cortex-A15, while the A53 is the successor to the Cortex-A7. This processor pair is designed to be used in a big.LITTLE configuration, where powerful and low-energy cores are paired in a single SoC to combine processing power and battery life.
Cortex-A57 cores have a complex, out-of-order multi-issue pipeline offering what ARM claims will be three times the performance of current top-of-the-range smartphones. The A57 can work in configurations of 16 cores. Cortex-A53 cores have a simple, in-order, eight-stage pipeline and deliver four times the power efficiency of today's high-end smartphones, according to ARM. Both cores are designed for use inside servers, tablets and convertible PCs, and high-end smartphones.
The fifth-generation Nvidia Tegra 5, codenamed Logan, will combine four ARM Cortex-A15 cores with the Kepler GPU. Logan will be Nvidia's first system-on-a-chip with a GPU that supports Nvidia's CUDA parallel computing architecture. The GPU will also support Direct3D and OpenGL 4.3.
The follow-up to Trinity/Richland, the high-end Kaveri APUs will be based around two to four 28nm Steamroller CPU cores and a Graphics Core Next (GCN) GPU. The Steamroller architecture will give each CPU core a 4-wide instruction decoder, which AMD estimates will boost operations per cycle by 30 percent over its predecessor. It will feature a heterogeneous system architecture and reportedly have an 128-bit memory controller supporting DDR3 and GDDR5, increasing memory bandwidth over its predecessor.
A lower power version of the Kabini APU, AMD's ultra-low voltage SoC is based around two to four 64-bit Jaguar CPU cores and AMD's Graphics Core Next (GCN) architecture. Temash is aimed at tablets and other low-power, small form-factor devices.
Avoton is a 64-bit Intel Atom SoC targeted at the microserver market that succeeds the Centerton S1200 series of Atom processors. Made using Intel's 22nm process, the processors are built around a 3D Tri-Gate design that reduces power consumption and boosts performance over earlier planar gate designs, as well featuring the more efficient Silvermont out-of-order microarchitecture. Avoton reportedly features up to eight cores, supports dual-channel DDR3-1600/DDR3-1600L memory and is clocked up to 2.4GHz, as well as including an integrated Ethernet fabric controller. Avoton, which will be available in the second half of 2013, will be used in HP Moonshot microservers later in the year.
Based on the Silvermont out-of-order microarchitecture, Rangeley processors are targeted at the network and communications infrastructure market. The series will reportedly be clocked up to 2.4GHz and run up to eight cores, with each pair of cores sharing 1MB of L2 cache. The range is expected to support SSE4 and AES instructions, and VT-x virtualization, as well having a dual-channel DDR3 memory controller with support for up to 64GB of DDR3-1600 or DDR3L-1600 ECC memory. The chip's Crypto Engine enables Intel QuickAssist Technology that supports 128-,192- and 256-bit AES, 3DES and DES. Rangeley, which is expected to be released in second half of 2013, will be targeted at mid-range routers, switches and security appliances.
Based around two to four socket-FT3 64-bit Jaguar cores and AMD's Graphics Core Next (GCN) architecture, the 28nm Kabini APU is targeting ultrathin notebooks and low power consumption (9-15W). AMD says Kabini is expected to deliver a 50 percent performance improvement over the previous generation of Brazos APUs. A reference design for Kabini paired a A6-5200 APU with AMD Radeon HD 8400 GPU and 4GB of DDR3 1600 memory.
Valleyview is the codename for the second-generation ultra low-voltage 64-bit Atom SoC — the first Atom processor built on Intel's 22nm Tri-Gate process technology. Valleyview's Silvermont microarchitecture reportedly allows out-of-order execution that's 35 percent faster per CPU clock cycle than the 32nm Saltwell architecture used in Centerton Atom processors. An integrated Intel Gen7 GPU will support 1080p playback at 60fps, while DirectX 11 support delivers three times the graphics performance of current Atom chips. There will be four types of Valleyview chips: D- and M-series for desktops and netbooks, I-series for embedded and industrial devices, and T-series for tablets. Valleyview processors are expected to be available in single-, dual- and quad-core configurations with clock speeds up to 2.7GHz and support for up to 8GB of 1,066MHz or 1,333MHz DDR3 RAM. Power consumption (TDP) is 3W for the T-series, 4-6.5W for the M-series and 12W for the D-series. The first ValleyView devices will be tablets, which are expected in Q3 2103. Silvermont will be followed by processors based on the 14nm Airmont microarchitecture in 2014.
Silvermont architecture on 22nm (3D Tri-Gate) process. Low-power SoC for ultrabooks, tablets and smartphones.
Haswell is the successor to the Sandy Bridge architecture, and will initially be built on the same 22nm process as Ivy Bridge (the die shrink of Sandy Bridge). Haswell will use 3D Tri-Gate transistors and include Haswell New Instructions (AVX2) for enhanced vector floating-point performance, Transactional Synchronization Extensions to add hardware transaction memory support, better virtualisation performance and EPT (Extended Page Tables). Haswell will move to a 14nm process, codenamed Broadwell, in 2014.
The Haswell range is expected to include nine '4th Generation' Core i5s (up to four cores and 3.4GHz) and five Core i7s (up to eight cores and 3.5GHz) for desktops, the E3, E5 and E7 for servers, and a Core i5 and i7 for mobile. The first Haswell chips are rumoured to have gone to PC makers, who will ship machines in June; the first E7 servers are expected in the fourth quarter of 2013.
The successor to Haswell/Broadwell will be the Skylake microarchitecture.
Intel will introduce the new Intel Xeon processor E3 1200 v3 family of server processors, based on the 22nm Haswell architecture, in June 2013. The v3 upgrade lowers the TDP of Intel's Xeon E3 family to 13W, about one quarter below the previous v2 generation. Leaked specs suggest clock speeds of between 3.1GHz and 3.6GHz, with all processors supporting Turbo Boost and two models lacking Hyper-Threading Technology. The new Xeon E3 includes integrated Intel HD Graphics to improve its video streaming and transcoding capabilities, with Intel claiming a 25 percent improvement in transcoding performance over the v2 family.
The successor to AMD's Trinity APU (Accelerated Processing Unit) is a family of 32nm socket-FM2 processors that combine two to four AMD Piledriver CPUs with a second-generation AMD DirectX 11 GPU and a more efficient power management system. Richland will be targeted at both notebook and desktop markets and will form the basis of new processors in the A10, A8, A6 and A4 families. Four mobile processors launched at the end of March, pairing up to four CPUs with Radeon 8000M-series mobile GPUs. AMD promises that Richland will deliver a 20–40 percent performance boost over earlier Trinity APUs.
Based on Intel's Many Integrated Core (MIC) architecture for massively parallel processors, the Xeon Phi Coprocessor is fabricated at 22nm, and has been announced in two models — the 3100 and 5110P. The 5110P became available in January 2013, while the 3100 family will ship during the first half of 2013. The Intel Xeon Phi Coprocessor 3100 offers more than 1000 Gigaflops (1 TFlops) of double-precision performance, supports up to 6GB memory at 240GB/s bandwidth, and includes memory error correction codes (ECC). The 3100 family will operate within a 300W thermal envelope. The 1.053GHz 5110P has 60 cores and can handle 240 threads. The processor is also capable of 1 teraflops of double-precision performance, supporting up to 8GB of memory with a bandwidth of 320GB/s. The 5110P has a TDP of 225W.
Nvidia's Tegra 4, codenamed Wayne, uses four ARM Cortex A-15 CPU cores clocked at up to 1.9GHz, with a fifth low-power companion core (possibly based on a Cortex A-7) and 72 GeForce GPU cores, based on unified shaders and with Direct3D 11 and OpenGL 4.0 support. Nvidia claims an average of 2.6 times performance boost for CPU-related tasks over 2011's Tegra 3. Native video encoding and decoding at 2,560 by 1,440 resolution is included, along with USB 3.0 and dual display support.
A cut-down smartphone-focused version of the Tegra 4, the Tegra 4i, was introduced in February 2013. This uses Cortex-A9 rather than A15 CPU cores and incorporates a version of Nvidia's i500 LTE modem.
Texas Instruments' 5th-generation OMAP platform is a 28nm system-on-a-chip with a 2GHz dual-core ARM Cortex-A15 CPU, a pair of ARM Cortex-M4 processors to handle less demanding tasks and save power, a dual-core PowerVR SGX544-MPx GPU, a dedicated TI 2D BitBlt graphics accelerator and an IVA 3 multimedia accelerator. There are currently two chips in the OMAP 5 series: the 5430, which is aimed at smartphones and supports dual-channel LPDDR2 package-on-package memory; and the 5432, which is aimed at mobile and consumer devices and supports dual-channel DDR3 memory.
Announced in February 2011, the OMAP 5 platform is expected to sample in H2 2011 and become available in mobile devices in H2 2012.
The 32-bit Cortex-A15 processor supports 40-bit physical addressing of up to 1TB of RAM, multiple power domains, hardware-level virtualisation, plus DSP (digital signal processing) and NEON media extensions to the ARMv7 instruction set. It has an out-of-order superscalar 15-stage pipeline with a tightly-coupled, ECC-protected, low-latency L2 cache up to 4MB in size. The Cortex-A15 can decode up to three instructions per clock cycle, compared to the two decoded by its A9 predecessor, and can issue eight micro-ops per cycle as compared to the A9's four.
The Cortex-A15 runs at up to 2.5GHz and delivers what ARM says is "twice the performance of smartphones based on A9 processors". SoCs using the A15 include the Nvidia Tegra 4, Samsung Exynos, ST-Ericsson Nova A9600 and TI OMAP 5.
Krishna is an APU for low-power notebooks, netbooks and tablets. Based on 28nm Enhanced Bobcat cores, Krishna will come in 2- or 4-core variants and replace the 40nm E-series Zacate APUs.
Trinity is a Fusion-series APU (Accelerated Processing Unit) for high-performance desktops and notebooks, succeeding the Llano A8-series. It comprises 32nm Enhanced Bulldozer CPU cores and a DirectX 11-capable GPU, and will come in 2- or 4-core variants.
Terramar is aimed at the 2- and 4-processor server markets, and is based around 32nm next-generation Bulldozer cores. It will have up to 20 cores per chip and use a new G2012 socket.
Sepang is aimed at the 1- and 2-processor server markets, and is based around 32nm next-generation Bulldozer cores. It will have up to 10 cores per chip and use a new C2012 socket.
The successor to the 45nm A5 system-on-a-chip will be built on a 28nm process. Having fallen out with Samsung over its Android smartphones and tablets, Apple is reportedly using Taiwanese chipmaker TSMC to produce test runs of the A6 chip, which will power the 'iPad 3' and 'iPhone 6'.
Komodo is based on 32nm Enhanced Bulldozer cores and will ship as an 8-core processor. It's aimed at servers and performance desktops, where it will be coupled with DirectX 11 GPUs. Komodo will replace AMD's Zambezi processor.
Wichita is an APU for low-power notebooks, ultraportables, netbooks and tablets. Based on 28nm Enhanced Bobcat cores, Wichita will come in 1- or 2-core variants and replace the 40nm C-series Ontario APUs.
Ivy Bridge is a 22nm die-shrink of Sandy Bridge using 3D Tri-Gate transistors and incorporating PCI Express 3.0 and DirectX 11 support. Ivy Bridge's companion chipset is codenamed Panther Point.
This Atom processor will implement the Saltwell architecture on a 32nm process. Forming the basis of the Medfield platform, Saltwell is a low-power system-on-a-chip (SoC) aimed at 'ultrabooks', tablets and smartphones.
Based on a new 'Krait' CPU microarchitecture and a 28nm fabrication process, Qualcomm's quad-core APQ8064 is the flagship of the next generation of Snapdragon SoC processors. With speeds of up to 2.5GHz per core and a quad-core Adreno 320 GPU, the APQ8064 will deliver improved performance and lower power consumption for devices that don't require Snapdragon MSM89x0-series' integrated mobile broadband connectivity. Announced in February 2011, the APQ8064 is expected to be available early in 2012.
Based on a new 'Krait' CPU microarchitecture and a 28nm fabrication process, Qualcomm's 4th-generation Snapdragon SoC processors for smartphones will come in single-core (MSM8930) and dual-core (MSM8960) variants and feature upgraded Adreno GPUs (225 in the dual-core 8960, 305 in the single-core 8930). Both models will support next-generation LTE mobile broadband. Announced in February 2011, the MSM8960 is expected to be available in Q4 2011 and the MSM8930 in Q3 2012.
The Xeon E5 series takes the 32nm Sandy Bridge architecture into higher-end and more specialised sever designs, and also covers the workstation market. There are four Xeon E5 families: E5-1600 for high-performance single-socket (LGA 2011) servers and workstations; E5-2400 for dual-socket (LGA 1356) servers; E5-2600 for dual-socket (LGA 2011) servers and workstations; and E5-4600 for 4-socket (LGA 2011) servers. The E5-1600 and E5-2600 families will launch in Q4 2011, with the E5-2400 and E5-4600 following in Q1 2012.
Teamed with the Patsburg (C600) series chipset, these 2-, 4-, 6- and 8-core CPUs form a range of Romley server/workstation platforms.
Nvidia's next-generation Tegra system-on-a-chip, codenamed Kal-El, is the first quad-core processor designed for tablets and other mobile devices. Built on a 40nm process, it includes an ARM Cortex-A9 MPCore running at up to 1.5GHz, a 12-core Nvidia GPU with an ultra-low-power mode and HD (1080p) H.264 High Profile video decoding. Announced in February, the Tegra 3 will ship in Q4 2011, and will be about five times faster than the dual-core Tegra 2. In June, Microsoft showed an early build of Windows 8 running on the Kal-El/Tegra 3 chip.
Nvidia recently revealed the existence of a fifth low-power 'companion' core that operates when a device is performing simple tasks; the main cores are turned on progressively as workloads increase. The first shipping device to run the Tegra 3 is expected to be the Transformer Prime from Asus.
Three high-performance desktop Sandy Bridge processors will launch towards the end of 2011. The entry-level member of the Core i7-3000 family is the multiplier-locked quad-core 3820, running at 3.6GHz (3.9GHz with Turbo Boost). Next up is the 6-core 3930K running at 3.2/3.8GHz. The top-end Core i7 is the 6-core 3960X, running at 3.3/3.9GHz. All three CPUs use Socket LGA 2011 and are supported by the X79 chipset.
Designed as a successor to ARM's Cortex-A8 CPU core, the Cortex-A7 can run at more than 1GHz when manufactured using a 28nm process. ARM claims that the Cortex-A7 delivers five times the energy efficiency and 50 percent better performance than the Cortex-A8.
Cortex-A7-based processors can support up to 64KB of L1 cache, while Large Physical Address Extensions allow up to 1TB of memory to be addressed. ARM Thumb-2 technology reduces the memory required to store instructions, and the architecture includes support for Neon SIMD media and digital signature processing extensions.
The Cortex-A7 is designed to be used with Cortex-A15 CPU cores in a big.LITTLE configuration, where powerful and low-energy cores are paired in a single SoC to optimise the combination of processing power and battery life.
Codenamed Desna, AMD's Z-Series APUs are designed for use in tablets. This low-power chip will be based on one or two Bobcat cores and a DirectX 11-capable GPU. TDP is expected to be around 6.4W, with typical tablet battery life quoted at 10.5 hours. The first Desna-based device is likely to be the MSI Windpad 110W, but no official release date has been announced.
Based on the 32nm Bulldozer core and aimed at the 2- and 4-processor server markets, the 8-, 12- or 16-core Interlagos processor will ship under the Opteron 6200 Series brand for Socket G34.
Zambezi is a Socket AM3+ desktop processor aimed at the 'enthusiast' segment. Based on 32nm Bulldozer cores, Zambezi will come in 4-, 6- or 8-core variants.
Based on the 32nm Bulldozer core and aimed at the 1- and 2-processor server markets, the 6- or 8-core Valencia processor will ship under the Opteron 4200 Series brand for Socket C32.
Bulldozer is a ground-up CPU core design that replaces AMD's K10 architecture. Aimed primarily at servers and high-performance desktops, the modular 32nm Bulldozer core is claimed to deliver dramatic improvements in performance per watt.
Llano is a series of AMD Fusion APUs (Accelerated Processing Units) based on 32nm K10 CPU cores and a DirectX 11-capable GPU. Available in 2- or 4-core variants, Llano is aimed at the performance/mainstream notebook (Sabine platform) and mainstream desktop (Lynx platform) segments.
Samsung's 45nm system-on-a-chip for smartphones, tablets and mobile internet devices is based on a dual-core ARM Cortex-A9 CPU running at 1.2GHz and an ARM Mali-400 MP GPU. The Exynos 4210, formerly codenamed Orion, first appeared in one iteration (the GT-i9100) of the Samsung Galaxy S II Android 2.3 smartphone.
Intel's 32nm Sandy Bridge server processors are sold under the Xeon brand and start with the Xeon E3-12xx series. Teamed with the Cougar Point chipset, these dual- and quad-core Socket LGA 1155 CPUs form the Bromolow platform, and are aimed at entry-level single-processor servers.
Apple's A5 system-on-a-chip is the successor to 2010's A4, and made its debut in the iPad 2. The Samsung-made A5 is built on a 45nm process, has a dual-core ARM Cortex-A9 CPU and a dual-core PowerVR SGX543MP2 GPU, and runs at up to 1GHz. The A5 chip is expected to power the forthcoming iPhone 5.
The 45nm, dual-core APQ8060 is targeted at tablets and large-display devices that don't require the Snapdragon MSM8x60-series' integrated mobile broadband connectivity. This processor debuted in HP's now-discontinued TouchPad tablet (pictured).
Texas Instruments' 4th-generation OMAP platform is a 45nm system-on-a-chip using a dual-core ARM Cortex-A9 CPU, a PowerVR SGX54x GPU and an IVA 3 multimedia accelerator. The OMAP4430 runs the CPU at 1GHz and the GPU at 304MHz, and powers RIM's BlackBerry Playbook and LG's Optimus 3D. The OMAP4440, which will be available in H2 2011, runs the CPU at 1.5GHz and the GPU at 384MHz. The OMAP4470, expected in Q4 2011, runs the CPU at 1.8GHz and has a PowerVRSGX544 running at 384GHz.
Intel's 32nm desktop and mobile Sandy Bridge processors use the same Core i3/i5/i7 brand names as the previous-generation Nehalem chips but have four-digit model numbers of the format 2xxx[letter]. The letter suffixes are K for unlocked multiplier, S for low-power, T for ultra-low-power and M for Mobile. Desktop chips use Socket LGA 1155, while mobile variants use Socket BGA 1023. The Sandy Bridge chipset family is codenamed Cougar Point.
Codenamed Zacate, AMD's E-Series APUs (Accelerated Processing Units) are aimed at mainstream notebooks, all-in-ones and small form-factor desktops. Based on one or two 40nm Bobcat cores and a DirectX 11-capable GPU, the E-Series has a TDP (Thermal Design Power) of 18 watts.
Codenamed Ontario, AMD's C-Series APUs (Accelerated Processing Units) are aimed at HD netbooks and 'other emerging form factors'. Based on one or two 40nm Bobcat cores and a DirectX 11-capable GPU, the C-Series has a TDP (Thermal Design Power) of 9 watts.
Qualcomm's ARM-based MSM8260 and MSM8660 mobile SoC (system-on-a-chip) processors use a 45nm dual-core Scorpion CPU and an Adreno 220 GPU. The MSM8260 first appeared in the HTC Sensation (pictured). The 82xx models support GSM/GPRS/EDGE and mobile broadband up to HSPA+, while the 86xx models add support for CDMA and EV-DO as well.