Intel Nehalem (microarchitecture)

Initial Nehalem processors use the same 45 nm manufacturing methods as Penryn. A working system with two Nehalem processors was shown at Intel Developer Forum Fall 2007, and a large number of Nehalem systems were shown at Computex in June 2008.

The microarchitecture is named after the Nehalem Native American nation in Oregon. At that stage it was supposed to be the latest evolution of the NetBurst microarchitecture. Since the abandonment of NetBurst, the codename has been recycled and refers to a completely different project, although Nehalem still has some things in common with NetBurst. Nehalem-based microprocessors utilize higher clock speeds and are more energy-efficient than Penryn microprocessors. Hyper-Threading is reintroduced along with an L3 Cache missing from most Core-based microprocessors.

The first computer to use Nehalem-based Xeon processors was the Apple Mac Pro workstation announced on March 3, 2009. Nehalem-based Xeon EX processors for larger servers were expected in Q4 2009 based on initial announcements from Intel, but in November 2009 the launch of these processors was pushed back to the first half of 2010.

Mobile Nehalem-based processors were introduced in September 2009.

Technology

Microarchitecture of the quad-core implementation

Various sources have stated the specifications of processors in the Nehalem family:

Two or four
- 731 million transistors for the quad core variant
45 nm manufacturing process
Integrated memory controller supporting two or three memory channels of DDR3 SDRAM or four FB-DIMM channels
Integrated graphics processor (IGP) located off-die, but in the same CPU package
A new point-to-point processor interconnect, the Intel QuickPath Interconnect, in high-end models, replacing the legacy front side bus
Integration of PCI Express and Direct Media Interface into the processor in mid-range models, replacing the northbridge
Simultaneous multithreading (SMT) by multiple cores which enables two threads per core. Intel calls this hyper-threading. Simultaneous multithreading has not been present on a consumer desktop Intel processor since 2006 with the Pentium 4 and Pentium XE. Intel reintroduced SMT with their Atom Architecture.
Native (monolithic, i.e. all processor cores on a single die) quad- and octal-core processors
The following caches:
- 32 KB L1 instruction and 32 KB L1 data cache per core
- 256 KB L2 cache per core
- 4–8 MB L3 cache shared by all cores
33% more in-flight micro-ops than Conroe
Second-level branch predictor and second-level translation lookaside buffer
Modular blocks of components such as cores that can be added and subtracted for varying market segments

Performance and power improvements

It has been reported that Nehalem will have a focus on performance, which accounts for the increased core size. Compared to Penryn, Nehalem will have:

1.1x to 1.25x the single-threaded performance or 1.2x to 2x the multithreaded performance at the same power level
30% lower power usage for the same performance
According to a preview from AnandTech "expect a 20–30% overall advantage over Penryn with a 10% increase in power usage."
Per Core, clock-for-clock, Nehalem will provide a 15–20% increase in performance compared to Penryn.

PC Watch found that a Nehalem "Gainestown" processor has 1.6x the SPECint_rate2006 integer performance and 2.4x the SPECfp_rate_2006 floating-point performance of a 3.0 GHz Xeon X5365 "Clovertown" quad-core processor.

A 2.93 GHz Nehalem "Bloomfield" system has been used to run a 3DMark Vantage benchmark and gave a CPU score of 17,966. The 2.66 GHz variant scores 16,294. A 2.4 GHz Core 2 Duo E6600 scores 4,300.

AnandTech tested the Intel QuickPath Interconnect ("QPI", 4.8 GT/s version) and found the copy bandwidth using triple-channel 1066 MHz DDR3 was 12.0 GB/s. A 3.0 GHz Core 2 Quad system using dual-channel 1066 MHz DDR3 achieved 6.9 GB/s.

Overclocking will be possible with Bloomfield processors and the X58 chipset. The Lynnfield processor will use a PCH removing the need for a northbridge chipset.

The Nehalem processors are the first to incorporate the SSE 4.2 SIMD instructions, adding 7 new instructions to the SSE 4.1 set available in the Core 2 series.

Code names

Each combination of a Nehalem/Westmere processor die and package has both a separate codename and a product code. Typically, the same dies are used for uniprocessor (UP) and dual-processor (DP) servers, but using an extra QuickPath link for the inter-processor communication in the DP server variant. Where the Core microarchitecture used four different processor sockets, one for each market segment, Nehalem now uses Socket 1366 for the high-end of both UP and DP machines, and Socket 1156 for the low end UP machines. The name for the UP version of Gulftown is not yet known; its product code is 80613 and can be found in Intel's product database

	Mobile	Desktop UP Server	DP Server	MP Server
Dual-Core 45 nm Dual-Channel, PCIe, Graphics Core	Auburndale canceled	Havendale canceled
Dual-Core 32 nm Dual-Channel, PCIe, Graphics Core	Arrandale 80617	Clarkdale 80616
Quad-Core 45 nm Dual-Channel, PCIe	Clarksfield 80607	Lynnfield 80605	Jasper Forest 80612
Quad-Core 45 nm Triple-Channel		Bloomfield 80601	Gainestown 80602
Six-Core 32 nm Triple-Channel		Gulftown 80613	Gulftown 80614
Eight-Core 45 nm Triple-Channel				Beckton 80604

Variants

These tables list all the processors of Nehalem microarchitecture to have been leaked so far. The table is ordered roughly by performance, which usually correlates with price and power. Released processors are set in bold.

Notes:

"Extreme" processors have an unlocked clock multiplier. Thermal design power (TDP) values for CPUs with integrated GPUs include the GPU.
All variants have 64 KiB L1 cache per core, and 256 KiB L2 cache per core.

45 nm processors

Codename	Market	Cores (Threads)	Socket	Brand	Processor No.	Clock rate			Turbo	TDP	Interfaces			L3 cache	Release	1k Unit Price
Codename	Market	Cores (Threads)	Socket	Brand	Processor No.	Base	Core	Uncore	Turbo	TDP	Chipset	Memory	PCIe	L3 cache	Release	1k Unit Price
Beckton	MP Server	8 (16)	LGA-1567							130/105/90 W	4x QPI	4x [DDR3 with SMB motherboard]	n/a	24 MB	Q1 2010
Gainestown	DP Server	4 (8)	LGA-1366	Xeon	W5580	133 MHz	3.2 GHz		Yes	130 W	2x QPI 6.4 GT/s	3x DDR3-1333¹	n/a	8 MB	March 29, 2009	$1500
					X5570		2.93 GHz			95 W						$1286
					X5560		2.8 GHz									$1072
					X5550		2.66 GHz									$858
					E5540		2.53 GHz			80 W	2x 5.86 GT/s	3x DDR3-1066¹				$744
					E5530		2.4 GHz									$530
					E5520		2.26 GHz									$373
					L5520		2.26 GHz			60 W						$530
		4 (4)			E5506		2.13 GHz		No	80 W	2x 4.8 GT/s	3x DDR3-800¹		4 MB		$266
					E5504		2.0 GHz			80 W						$224
					L5506		2.13 GHz			60 W						$423
		2 (2)			E5502		1.86 GHz			80 W						$188
Bloomfield	UP Server	4 (8)	LGA-1366	Xeon	W3580	133 MHz	3.33 GHz		Yes	130 W	1x QPI 6.4 GT/s	3x DDR3-1333	n/a	8 MB	August 9, 2009	$999
					W3570		3.2 GHz				1x QPI 6.4 GT/s	3x DDR3-1333			March 29, 2009	$999
					W3550		3.06 GHz				1x QPI 4.8 GT/s	3x DDR3-1066			August 9, 2009	$562
					W3540		2.93 GHz								March 29, 2009	$562
					W3520		2.66 GHz								March 29, 2009	$284
Lynnfield			LGA 1156		X3470		2.93 GHz			95 W	DMI	2x DDR3-1333			September 8, 2009	$589
					X3460		2.8 GHz									$316
					X3450		2.66 GHz									$241
					X3440		2.53 GHz									$215
					L3426		1.86 GHz			45 W						$284
		4 (4)			X3430		2.4 GHz			95 W						$189
Bloomfield	Extreme/Performance Desktop	4 (8)	LGA-1366	Core i7 Extreme	975	133 MHz	3.33 GHz	2.66 GHz	Yes	130 W	1x QPI 6.4 GT/s	3x DDR3-1066	n/a	8 MB	May 31, 2009	$999
				Core i7 Extreme	965		3.2 GHz	2.66 GHz			1x QPI 6.4 GT/s				November 17, 2008	$999
				Core i7	960		3.2 GHz	2.13 GHz			1x QPI 4.8 GT/s				October 20, 2009	$562
					950		3.06 GHz								May 31, 2009	$562
					940		2.93 GHz								November 17, 2008	$562
					930		2.8 GHz								February 28, 2010	$284
					920		2.66 GHz								November 17, 2008	$284
Lynnfield	Performance/Mainstream Desktop	4 (8)	LGA 1156		870	133 MHz	2.93 GHz	2.4 GHz	Yes	95 W	DMI	2x DDR3-1333	1x16 / 2x8	8 MB	September 8, 2009	$562
					860		2.8 GHz	2.4 GHz		95 W					September 8, 2009	$284
					860S		2.53 GHz			82 W					January 7, 2010	$337
		4 (4)		Core i5	750		2.66 GHz	2.13 GHz		95 W					September 8, 2009	$196
		4 (4)		Core i5	750S		2.4 GHz			82 W					January 7, 2010	$259
Clarksfield	Extreme/Performance Mobile	4 (8)	mPGA-989	Core i7 Extreme	920XM		2.0 GHz			55 W					September 23, 2009	$1054
				Core i7	820QM		1.73 GHz			45 W						$546
				Core i7	720QM		1.6 GHz			45 W				6 MB		$364

¹ Though there is only one memory controller and it has only three channels, Intel states the Gainestown processors have six memory channels. Gainestown processors have dual QPI links and have a separate set of memory registers for each link; in effect, a multiplexed six-channel system.

The Havendale and Auburndale variants (which contained Gilo and Ironlake) have been cancelled.

32 nm processors

Codename	Market	Cores (Threads)	Socket	Brand	Processor No.	Clock rate			Turbo	TDP	Interfaces			L3 cache	Release	1k Unit Price
Codename	Market	Cores (Threads)	Socket	Brand	Processor No.	Base	Core	GPU	Turbo	TDP	Chipset	Memory	PCIe	L3 cache	Release	1k Unit Price
Gulftown	Extreme/Performance Desktop	6 (12)	LGA-1366	Core i7 Extreme	980 XE	133 MHz	3.33 GHz	n/a	Yes	130 W	2x QPI	3x DDR3-1066	n/a	12 MB	March 16, 2010	$999
Clarkdale	Mainstream/Value Desktop	2 (4)	LGA-1156	Core i5	670	133 MHz	3.46 GHz	733 MHz	Yes	73 W	DMI	2x DDR3-1333	1 x16	4 MB	January 7, 2010	$284
					661		3.33 GHz	900 MHz		87 W						$196
					660		3.33 GHz	733 MHz		73 W						$196
					650		3.2 GHz									$176
				Core i3	540		3.06 GHz		No							$133
				Core i3	530		2.93 GHz									$113
		2 (2)		Pentium	G6950		2.8 GHz	533 MHz				2x DDR3-1066		3 MB		$87
Arrandale	Mainstream/Value Mobile	2 (4)	mPGA-989	Core i7	620M	133 MHz	2.66 GHz	766 MHz	Yes	35 W	DMI	2x DDR3-1066	1 x16	4 MB	January 7, 2010	$332
					640LM		2.13 GHz	566 MHz		25 W						$332
					620LM		2.0 GHz	566 MHz		25 W						$300
					640UM		1.2 GHz	500 MHz		18 W		2x DDR3-800				$305
					620UM		1.06 GHz	500 MHz		18 W		2x DDR3-800				$278
				Core i5	540M		2.53 GHz	766 MHz		35 W		2x DDR3-1066		3 MB		$257
					520M		2.4 GHz	766 MHz		35 W		2x DDR3-1066				$225
					520UM		1.06 GHz	500 MHz		18 W		2x DDR3-800				$241
					430M		2.26 GHz	766 MHz		35 W		2x DDR3-1066				OEM
				Core i3	350M		2.26 GHz	667 MHz	No
				Core i3	330M		2.13 GHz	667 MHz
		2 (2)		Celeron	P4500		1.86 Ghz	500 Mhz						2 MB	Q2, 2010

For the desktop, Gulftown is to be an "Extreme Edition" CPU and so will coexist with Bloomfield. It will have Turbo Boost and similar clock speeds to Bloomfield.

Lynnfield and Clarksfield may make the 32 nm transition in the middle of 2010, sometime after Q2, while Beckton will move to 32 nm at the end of 2010. The 32 nm CPUs will not have significantly different clock speeds compared to 45 nm CPUs. Clarkdale and Arrandale contain the 32 nm dual core processor Hillel and the 45 nm integrated graphics device Ironlake, and support switchable graphics. The lowest-power variant of Arrandale may have a 10 W CPU TDP, and a maximum clock speed of 1.6 GHz. A successor to Bloomfield and entry level server chips are also expected in Q2 2010.