Haswell (microarchitecture)
Haswell is the codename for a processor microarchitecture developed by Intel as the "fourth-generation core" successor to the Ivy Bridge. Intel officially announced CPUs based on this microarchitecture on June 4, 2013, at Computex Taipei 2013, while a working Haswell chip was demonstrated at the 2011 Intel Developer Forum. With Haswell, which uses a 22 nm process, Intel also introduced low-power processors designed for convertible or "hybrid" ultrabooks, designated by the "Y" suffix.
Haswell CPUs are used in conjunction with the Intel 8 Series chipsets, Intel 9 Series chipsets, and Intel C220 series chipsets.
Design
The Haswell architecture is specifically designed to optimize the power savings and performance benefits from the move to FinFET transistors on the improved 22 nm process node.Haswell has been launched in three major forms:
- Desktop version : Haswell-DT
- Mobile/Laptop version : Haswell-MB
- BGA version:
- * 47 W and 57 W TDP classes: Haswell-H
- * 13.5 W and 15 W TDP classes : Haswell-ULT
- * 10 W TDP class : Haswell-ULX
Performance
- Approximately 8% faster vector processing
- Up to 5% higher single-threaded performance
- 6% higher multi-threaded performance
- Desktop variants of Haswell draw between 8% and 23% more power under load than Ivy Bridge.
- A 6% increase in sequential CPU performance
- Up to 20% performance increase over the integrated HD4000 GPU
- Total performance improvement on average is about 3%
- Around 15 °C hotter than Ivy Bridge, while clock frequencies of over 4.6 GHz are achievable
Technology
Features carried over from Ivy Bridge
- 22 nm manufacturing process
- 3D Tri-Gate FinFET transistors
- Micro-operation cache capable of storing 1.5 K micro-operations
- 14- to 19-stage instruction pipeline, depending on the micro-operation cache hit or miss
- Mainstream variants are up to quad-core.
- Native support for dual-channel DDR3/DDR3L memory, with up to 32 GB of RAM on LGA 1150 variants
- 64 KB L1 cache and 256 KB L2 cache per core
- A total of 16 PCI Express 3.0 lanes on LGA 1150 variants
New features
- Wider core: fourth arithmetic logic unit, third address generation unit, second branch execution unit, deeper buffers, higher cache bandwidth, improved front-end and memory controller, higher load/store bandwidth.
- New instructions.
- The instruction decode queue, which holds instructions after they have been decoded, is no longer statically partitioned between the two threads that each core can service.
- New sockets and chipsets:
- * LGA 1150 for desktops, and rPGA947 and BGA1364 for the mobile market.
- * Z97 and H97 chipsets for the Haswell Refresh and Broadwell, in Q2 2014.
- * LGA 2011-v3 with X99 chipset for the enthusiast-class desktop platform Haswell-E.
- Intel Transactional Synchronization Extensions for the Haswell-EX variant. In August 2014 Intel announced that a bug exists in the TSX implementation on the current steppings of Haswell, Haswell-E, Haswell-EP and early Broadwell CPUs, which resulted in disabling the TSX feature on affected CPUs via a microcode update.
- Hardware graphics support for Direct3D 11.1 and OpenGL 4.3. Intel 10.18.14.5117 driver is the last planned driver release on Windows 7/8.1.
- DDR4 for enterprise/server segments and for the Enthusiast-Class Desktop Platform Haswell-E
- Variable Base clock like LGA 2011.
- Four versions of the integrated GPU: GT1, GT2, GT3 and GT3e, where GT3 version has 40 execution units. Haswell's predecessor, Ivy Bridge, has a maximum of 16 EUs. GT3e version with 40 EUs and on-package 128 MB of embedded DRAM, called Crystalwell, is available only in mobile H-SKUs and desktop R-SKUs. Effectively, this eDRAM is a Level 4 cache; it is shared dynamically between the on-die GPU and CPU, and serving as a victim cache to the CPU's Level 3 cache.
- Optional support for Thunderbolt technology and Thunderbolt 2.0
- Fully integrated voltage regulator, thereby moving some of the components from motherboard onto the CPU.
- New advanced power-saving system; due to Haswell's new low-power C6 and C7 sleep states, not all power supply units are suitable for computers with Haswell CPUs.
- 37, 47, 57 W thermal design power mobile processors.
- 35, 45, 65, 84, 88, 95 and 130–140 W TDP desktop processors.
- 15 W or 11.5W TDP processors for the Ultrabook platform leading to reduced heat, which results in thinner as well as lighter Ultrabooks, but the performance level is slightly lower than the 17 W version.
- Shrink of the Platform Controller Hub, from 65 nm to 32 nm.
Server processors features
- Haswell-EP variant, released in September 2014, with up to 18 cores and marketed as the Xeon E5-1600 v3 and Xeon E5-2600 v3 series.
- Haswell-EX variant, released in May 2015, with 18 cores and functioning TSX.
- A new cache design.
- Up to 35 MB total unified cache for Haswell-EP and up to 40 MB for Haswell-EX.
- LGA 2011-v3 socket replaces LGA 2011 for the Haswell EP; the new socket has the same number of pins, but it is keyed differently due to electrical incompatibility.
- The already launched Xeon E3 v3 Haswells will get a refresh in spring 2014, together with a refreshed Intel C220 series PCH chipset.
- TDP up to 160 W for Haswell-EP.
- Haswell-EP models with ten and more cores support cluster on die operation mode, allowing CPU's multiple columns of cores and last level cache slices to be logically divided into what is presented as two non-uniform memory access CPUs to the operating system. By keeping data and instructions local to the "partition" of CPU which is processing them, therefore decreasing the LLC access latency, COD brings performance improvements to NUMA-aware operating systems and applications.
Haswell Refresh
The CPUs codenamed Devil's Canyon, covering the i5 and i7 K-series SKUs, employ a new and improved thermal interface material called next-generation polymer thermal interface material. This improved TIM reduces the CPU's operating temperatures and improves the overclocking potential, as something that had been problematic since the introduction of Ivy Bridge. Other changes for the Devil's Canyon CPUs include a TDP increase to 88 W, additional decoupling capacitors to help smooth out the outputs from the fully integrated voltage regulator, and support for the VT-d that was previously limited to non-K-series SKUs. TSX was another feature brought over from the non-K-series SKUs, until August 2014 when a microcode update disabled TSX due to a bug that was discovered in its implementation.
List of Haswell processors
Desktop processors
- All models support: MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, F16C, Enhanced Intel SpeedStep Technology, Intel 64, XD bit, Intel VT-x, and Smart Cache.
- * Core i3, i5 and i7 support AVX, AVX2, BMI1, BMI2, FMA3, and AES-NI.
- * Core i3 and i7, as well as the Core i5-4570T and i5-4570TE, support Hyper-Threading .
- * Core i5 and i7 support Turbo Boost 2.0.
- * Although it was initially supported on selected models, since August 2014 desktop variants no longer support TSX due to a bug that was discovered in its implementation; as a workaround, a microcode update disabled the TSX feature.
- * SKUs below 45xx as well as R-series and K-series SKUs do not support Trusted Execution Technology or vPro.
- * Intel VT-d, which is Intel's IOMMU, is supported on all i5 and i7 SKUs except the i5-4670K and i7-4770K. Support for VT-d requires the chipset and motherboard to also support VT-d.
- * Models i5-4690K and i7-4790K, codenamed Devil's Canyon, have a better internal thermal grease to help heat escape and an improved internal voltage regulator, to help deliver cleaner power in situations like overclocking.
- Transistors: 1.4 billion
- Die size: 177 mm2
- Intel HD and Iris Graphics in following variants:
- * R-series desktop processors feature Intel Iris Pro 5200 graphics.
- * All other currently known i3, i5 and i7 desktop processors include Intel HD 4600 graphics.
- * The exceptions are processors 41xxx, which include HD 4400 graphics.
- * Celeron and Pentium processors contain Intel HD Graphics.
- Pentium G3258, also known as the Pentium Anniversary Edition, has an unlocked multiplier. Its release marks 20 years of "Pentium" as a brand.
SKU suffixes to denote:
- K unlocked
- *The Pentium G3258 CPU is unlocked despite not having the K-suffix.
- S performance-optimized lifestyle
- T power-optimized lifestyle
- R BGA packaging / High-performance GPU
- X extreme edition
Server processors
- All models support: MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, AVX, AVX2, FMA3, F16C, BMI +BMI2, Enhanced Intel SpeedStep Technology, Intel 64, XD bit, TXT, Intel vPro, Intel VT-x, Intel VT-d, hyper-threading, Turbo Boost 2.0, AES-NI, and Smart Cache.
- Haswell-EX models support TSX, while for Haswell-E, Haswell-WS and Haswell-EP models it was disabled via a microcode update in August 2014, due to a bug that was discovered in the TSX implementation.
- Transistors: 5.56 billion
- Die size: 661 mm2
Lists of launched server processors are below, split between Haswell E3-12xx v3, E5-16xx/26xx v3 and E7-48xx/88xx v3 models.
SKU suffixes to denote:
- L low power
Mobile processors
- All models support: MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, F16C, Enhanced Intel SpeedStep Technology, Intel VT-x, Intel 64, XD bit , and Smart Cache.
- * Core i3, i5 and i7 support AVX, AVX2, BMI1, BMI2, FMA3, and hyper-threading .
- * Core i3, i5 and i7 except the Core i3-4000M support AES-NI.
- * Core i5 and i7 except the Core i5-4410E, i5-4402EC, i7-4700EC, and i7-4702EC support Turbo Boost 2.0.
- Platform Controller Hub integrated into the CPU package, slightly reducing the amount of space used on motherboards.
- Transistors: 1.3 billion
- Die size: 181 mm2
- When a cooler or quieter mode of operation is desired, this mode specifies a lower TDP and lower guaranteed frequency versus the nominal mode.
- This is the processor's rated frequency and TDP.
- When extra cooling is available, this mode specifies a higher TDP and higher guaranteed frequency versus the nominal mode.
- M mobile processor
- Q quad-core
- U ultra-low power
- X "extreme"
- Y extreme low-power
- E / H BGA1364 packaging