Hubbry Logo
Intel XeIntel XeMain
Open search
Intel Xe
Community hub
Intel Xe
logo
8 pages, 0 posts
0 subscribers
Be the first to start a discussion here.
Be the first to start a discussion here.
Intel Xe
Intel Xe
Intel Xe
View on Wikipedia
from Wikipedia

Intel Xe
Release dateSeptember 2, 2020; 5 years ago (2020-09-02)
Manufactured byIntel
TSMC
Designed byIntel
Marketed byIntel
Fabrication processTSMC N6
Cards
Entry-levelIris Xe Graphics
High-endIntel Arc
History
PredecessorGen 11
SuccessorIntel Xe 2

Intel Xe (stylized as Xe and pronounced as two separate letters,[1] abbreviation for "exascale for everyone"[2]), earlier known unofficially as Gen12,[3][4] is a GPU architecture developed by Intel.[5]

Intel Xe includes a new instruction set architecture. The Xe GPU family consists of a series of microarchitectures, ranging from integrated/low power (Xe-LP),[6] to enthusiast/high performance gaming (Xe-HPG), datacenter/high performance (Xe-HP) and high performance computing (Xe-HPC).[7][8]

History

[edit]

Intel's first attempt at a dedicated graphics card was the Intel740,[9] released in February 1998. The Intel740 was considered unsuccessful due to its performance which was lower than market expectations, causing Intel to cease development on future discrete graphics products. However, its technology lived on in the Intel Extreme Graphics lineup.[10] Intel made another attempt with the Larrabee architecture before canceling it in 2009;[11] this time, the technology developed was used in the Xeon Phi, which was discontinued in 2020.[12]

In April 2018, it was reported that Intel was assembling a team to develop discrete graphics processing units, targeting both datacenters, as well as the PC gaming market, and therefore competitive with products from both Nvidia and AMD.[13] Rumors supporting the claim included that the company had vacancies for over 100 graphics-related jobs, and had taken on former Radeon Technologies Group (AMD) leader Raja Koduri in late 2017 – the new product was reported to be codenamed "Arctic Sound".[13] The project was reported to have initially been targeting video streaming chips for data centers, but had its scope expanded to include desktop GPUs.[13]

In June 2018, Intel confirmed it planned to launch a discrete GPU in 2020.[14]

The first functional discrete "Xe" GPU, codenamed "DG1", was reported as having begun testing in October 2019.[15]

According to a report by Hexus in late 2019, a discrete GPU would launch in mid 2020; combined GPU/CPU (GPGPU) products were also expected, for data center and autonomous driving applications. The product was expected to initially be built on a 10 nm node (with 7 nm products in 2021) and use Intel's Foveros die stacking packaging technology (see 3D die stacking).[16] During 2020, the first GPUs were released under the name Intel Iris Xe Max, being integrated in the 11th generation Intel Core processors (codenamed "Tiger Lake" and "Rocket Lake"),[4] followed in 2021 by the Iris Xe DG1 card, exclusive to Intel OEM manufacturers.[17] Finally and after some delays, the retail launch of these first discrete graphics cards from the company in over 20 years, known as the Intel Arc series, would occur during 2022.[18]

Architecture

[edit]

Intel Xe expands upon the microarchitectural overhaul introduced in Gen 11 with a full refactor of the instruction set architecture.[19][4] While Xe is a family of architectures, each variant has significant differences from each other as these are made with their targets in mind. The Xe GPU family consists of Xe-LP, Xe-HP, Xe-HPC, and Xe-HPG sub-architectures.

Unlike previous Intel graphics processing units which used the Execution Unit (EU) as a compute unit, Xe-HPG and Xe-HPC use the Xe-core.[20] This is similar to an Xe-LP subslice.[20] An Xe-core contains vector and matrix arithmetic logic units, which are referred to as vector and matrix engines. Other components include L1 cache and other hardware.[20][21]

Xe-LP (Low Power)

[edit]

Xe-LP is the low power variant of the Xe architecture with removed support for FP64.[22] Xe-LP is present as integrated graphics for 11th-generation Intel Core and the Iris Xe MAX mobile dedicated GPU (codenamed DG1),[6] as well as in the H3C XG310 Intel Server GPU (codenamed SG1).[4] Compared to its predecessor, Xe-LP includes new features such as Sampler Feedback,[23] Dual Queue Support,[24] DirectX12 View Instancing Tier2,[25] and AV1 8-bit and 10-bit fixed-function hardware decoding.[26]

Xe-HP (High Performance)

[edit]

Xe-HP is the datacenter/high performance variant of Xe, optimized for FP64 performance and multi-tile scalability.[5]

Xe-HPC (High Performance Compute)

[edit]

Xe-HPC is the high performance computing variant of the Xe architecture.[7][8] An Xe-HPC Xe-core contains 8 vector and 8 matrix engines, alongside a large 512 KB L1 cache.[27] It powers Ponte Vecchio.

Xe-HPG (High Performance Graphics)

[edit]

Xe-HPG is the enthusiast or high performance graphics variant of the Xe architecture. The microarchitecture is based on Xe-LP with improvements from Xe-HP and Xe-HPC.[28] The microarchitecture is focused on graphics performance and supports hardware-accelerated ray tracing,[7][29] DisplayPort 2.0,[30] XeSS or supersampling based on neural networks (similar to Nvidia DLSS), and DirectX 12 Ultimate. [31] Intel confirmed ASTC support has been removed from hardware starting with Alchemist and future Intel Arc GPU microarchitectures will also not support it.[32] An Xe-HPG Xe-core contains 16 vector engines and 16 matrix engines.[21] An Xe-HPG render slice will consist of four Xe-cores, ray tracing hardware, and other components.[21]

Xe-LPG (Low Power Graphics)

[edit]

The Xe-LPG architecture is a low power variant of Xe-HPG designed for the tile-based iGPUs (tGPUs) of Intel's Meteor Lake and Arrow Lake processors. It is based on the same Arc Alchemist graphics (Gen 12.7) used by Intel's Arc A-series graphics cards but is optimized for operation with lower wattage and higher performance per watt.

Intel Xe 2

[edit]
Intel Xe 2
Release dateDecember 13, 2024[33]
CodenameBattlemage
History
PredecessorIntel Xe
SuccessorIntel Xe 3P

A successor to Xe was revealed during Intel Architecture Day 2021, under the name of Xe 2, codenamed Battlemage. In an exclusive interview with HardwareLuxx Tom Petersen confirmed that Xe2 would be segmented into "Xe2-LPG" (Low Power Graphics) for integrated GPUs and "Xe2-HPG" (High Performance Graphics) for discrete GPUs.[34] Xe 2 was first released with Xe2-LPG in Lunar Lake on September 24, 2024.[35] On December 3 2024, Intel announced the Arc B-Series Graphics Cards for desktop, utilizing the Xe2-HPG architecture, which was later released on December 12 2024.[36]

Intel Xe 3

[edit]

Intel Xe 3 is the architecture for the iGPU in the upcoming Panther Lake products. It, among much of the Arc series, will support XeSS 3 at launch, which will include Multi-Frame Generation (MFG). Some Xe based products may support this after launch.

Intel Xe 3P

[edit]
Intel Xe 3P
CodenameCelestial
History
PredecessorIntel Xe 2
Intel Xe 3
SuccessorIntel Xe 4

Intel Xe 3P, codenamed Celestial,[21] is the upcoming successor to the Intel Xe 2 and Xe 3 microarchitecture.

Intel Xe 4

[edit]
Intel Xe 4
CodenameDruid
History
PredecessorIntel Xe 3P

Intel Xe 4, codenamed Druid,[37] is the upcoming successor to the Intel Xe 3 microarchitecture.

Products using Xe

[edit]

Integrated graphics

[edit]
See also: Intel Graphics Technology

Newer Intel processors use the Xe-LP microarchitecture. These include 11th generation Intel Core processors (codenamed "Tiger Lake" and "Rocket Lake"),[4] 12th generation Intel Core processors (codenamed "Alder Lake"), 13th generation Intel Core processors (codenamed "Raptor Lake"), and 14th generation Intel Core processors (codenamed "Raptor Lake Refresh"). The iGPUs in the Intel Core Ultra 100 series processors (codenamed "Meteor Lake") use the Xe-LPG microarchitecture. The Intel Core Ultra 200S and 200H/HX series processors (codenamed "Arrow Lake") also use the Xe-LPG microarchitecture in their iGPUs. Meanwhile, the Intel Core Ultra 200V series processors (codenamed "Lunar Lake") uses the Xe2-LPG microarchitecture. The upcoming Intel Core Ultra 300 series (codenamed "Panther Lake") uses the Xe3 microarchitecture.

Discrete graphics

[edit]

Intel Iris Xe Max (DG1)

[edit]
Model Launch Process Execution

units

Shading

units

Clock speeds Memory Processing power (GFLOPS) Notes
Boost clock

(MHz)

Memory

(MT/s)

Size

(GB)

Bandwidth

(GB/s)

Bus

type

Bus width

(bit)

Half Single Double INT8
Iris Xe MAX November 1, 2020 Intel 10SF 96 768 1650 4266 4 68 LPDDR4X 128 5069 2534 — 10138  
Iris Xe January 26, 2021 Intel 10SF 80 640 900 4266 4 68 LPDDR4X 128 4224 2112

In August 2020, Intel was reported to be shipping Xe DG1 GPUs for a possible late 2020 release, while also commenting on a DG2 GPU aimed at the enthusiast market (later found out to be the first generation of Intel Arc nicknamed "Alchemist"). The DG1 is also sold as the Iris Xe MAX and as Iris Xe Graphics (stylized as iRIS Xe) in laptops, while cards for developers are sold as the DG1 SDV.[38][39]

The Xe MAX is an entry-level GPU that was first released on November 1, 2020, in China and is similar in most aspects to the integrated GPU found in Tiger Lake processors, the only differences being a higher clock speed, slightly higher performance and dedicated memory and a dedicated TDP requirement. It competes with Nvidia's laptop-level GeForce MX series GPUs. It is aimed at slim and highly portable productivity laptops and has 4 GB of dedicated LPDDR4X-4266 memory with a 128-bit-wide memory bus, has 96 EUs, 48 texture units, 24 ROPs, a peak clock speed of 1650 MHz and a performance of 2.46 FP32 teraFLOPs with a 25w TDP. By comparison, the integrated GPU in Tiger Lake processors has a performance of 2.1 FP32 teraFLOPs.[40][41] The Xe MAX does not replace the system's integrated GPU; instead it was designed to work alongside it, so tasks are split between the integrated and discrete GPUs.[42] It was initially available on only 3 laptops: The Asus VivoBook Flip 14 TP470, the Acer Swift 3X, and the Dell Inspiron 15 7000. Intel Xe MAX GPUs can only be found on systems with Tiger Lake processors.

Intel officially announced Intel Iris Xe Graphics desktop cards for OEMs and system integrators on January 26, 2021. It is aimed at mainstream desktop and business PCs as an improvement over other graphics options in AV1 video decoding, HDR (high dynamic range) video support and deep learning inference, and is not as powerful as its laptop counterpart, with only 80 enabled EUs. The first cards are made by Asus, have DisplayPort 1.4, HDMI 2.0, Dual Link DL-DVI-D outputs and are passively cooled.[43][44][45][46]

Intel Arc Alchemist

[edit]

Intel Arc is a high-performance discrete graphics line optimized for gaming. This will competes with the Radeon and GeForce lines of graphics processing units. The first generation (codenamed "Alchemist"), was developed under the "DG2" name and is based on the Xe-HPG architecture. The second generation (codenamed "Battlemage") was developed under the "DG3" name and is based on the Xe2 architecture. Future generations include Celestial ("DG4", based on Xe3P), and Druid ("DG5").

Desktop
[edit]
Overview of Intel Arc Alchemist GPUs
Branding and Model[47] Launch MSRP
(USD) 		Code name Process Transistors (billion) Die size
(mm2) 		Core config [a] L2 cache Clock rate
(MHz)[b] 		Fillrate Memory Processing power (TFLOPS) TDP Bus
interface
Pixel
(GP/s) 		Texture
(GT/s) 		Type Size (GB) Bandwidth
(GB/s) 		Bus width Clock
(MT/s) 		Half
precision
(base) 		Single
precision
(base) 		Double
precision
(base)
Arc 3 A310 Sep 28, 2022 $110 ACM-G11
(DG2-128) 		TSMC
N6 		7.2 157 6 Xe cores
768:32:16:6
(192:96:2) 		4 MB 2000
2000 		32 64 GDDR6 4 GB 124 64-bit 15500 6.144 3.072 n/a 75 W PCIe 4.0 x8
A380 Jun 14, 2022 $139 8 Xe cores
1024:64:32:8
(256:128:2) 		2000
2050 		64
65.6 		128
131.2 		6 GB 186 96-bit 8.192
8.3968 		4.096
4.1984 		n/a
n/a
Arc 5 A580 Oct 10, 2023 $179 ACM-G10
(DG2-512) 		21.7 406 24 Xe cores
3072:192:96:24
(768:384:6) 		8 MB 1700
1700 		163.2 326.4 8 GB 512 256-bit 16000 20.890 10.445 n/a 175 W PCIe 4.0 x16
Arc 7 A750 Oct 14, 2022 $289 28 Xe cores
3584:224:112:28
(896:448:7) 		16 MB 2050
2400 		229.6
268.8 		393.6
460.8 		29.3888
34.4064 		14.6944
17.2032 		n/a
n/a 		225 W
A770 8GB $329 32 Xe cores
4096:256:128:32
(1024:512:8) 		2100
2400 		268.8
307.2 		537.6
614.4 		34.4064
39.3216 		17.2032
19.6608 		n/a
n/a
A770 16GB $349 16 GB 560 17500
  1. ^ Shading cores (ALU): texture mapping units (TMU): render output units (ROP): ray tracing units
       (tensor cores (XMX): execution units: render slices)
  2. ^ Boost values (if available) are stated below the base value in italic.
Mobile
[edit]
Overview of Intel Arc Alchemist GPUs for mobile devices
Branding and Model[48] Launch Code name Process Transistors (billion) Die size
(mm2) 		Core config[a][b] L2
cache 		Core clock
(MHz)[c] 		Fillrate[d] Memory Processing power (TFLOPS) TDP Bus
interface
Pixel
(GP/s) 		Texture
(GT/s) 		Type Size Bandwidth
(GB/s) 		Bus width Clock
(MT/s) 		Half
precision 		Single
precision 		Double
precision
Arc 3 A350M Mar 30, 2022 ACM-G11
(DG2-128) 		TSMC
N6 		7.2 157 6 Xe cores
768:48:24:6
(96:96:2) 		4 MB 1150
2200 		27.6
52.8 		55.2
105.6 		GDDR6 4 GB 112 64-bit 14000 3.5328
6.7584 		1.7664
3.3792 		0.4416
0.8448 		25–35 W PCIe 4.0 ×8
A370M 8 Xe cores
1024:64:32:8
(128:128:2) 		1550
2050 		49.6
65.6 		99.2
131.2 		6.3488
8.3968 		3.1744
4.1984 		0.7936
1.0496 		35–50 W
Arc 5 A530M Q3 2023 ACM-G12
(DG2-256) 		12 Xe cores
1536:96:48:12
(192:192:3) 		8 MB 1300 4 GB
8 GB 		224 128-bit 65–95 W
A550M Q2 2022 ACM-G10
(DG2-512) 		21.7 406 16 Xe cores
2048:128:64:16
(256:256:4) 		900
1700 		57.6
108.8 		115.2
217.6 		8 GB 7.3728
13.9264 		3.6864
6.9632 		0.9216
1.7408 		60–80 W
A570M Q3 2023 ACM-G12
(DG2-256) 		1300 75–95 W
Arc 7 A730M Q2 2022 ACM-G10
(DG2-512) 		21.7 406 24 Xe cores
3072:192:96:24
(384:384:6) 		12 MB 1100
2050 		105.6
196.8 		211.2
393.6 		12 GB 336 192-bit 13.5168
25.1904 		6.7584
12.5952 		1.6896
3.1488 		80–120 W PCIe 4.0 ×16
A770M 32 Xe cores
4096:256:128:32
(512:512:8) 		16 MB 1650
2050 		211.2
262.4 		422.4
524.8 		16 GB 512 256-bit 16000 27.0336
33.5872 		13.5168
16.7936 		3.3792
4.1984 		120–150 W
  1. ^ Shading cores (ALU): texture mapping units (TMU): render output units (ROP): ray tracing units
       (tensor cores (XMX): execution units: render slices)
  2. ^ Texture fillrate is calculated as the number of texture mapping units (TMUs) multiplied by the base (or boost) core clock speed.
  3. ^ Boost values (if available) are stated below the base value in italic.
  4. ^ Pixel fillrate is calculated as the lowest of three numbers: number of ROPs multiplied by the base core clock speed, number of rasterizers multiplied by the number of fragments they can generate per rasterizer multiplied by the base core clock speed, and the number of streaming multiprocessors multiplied by the number of fragments per clock that they can output multiplied by the base clock rate.
Workstation
[edit]
Overview of Intel Arc Alchemist GPUs for Workstations
Branding and Model[49] Launch Code name Process Transistors (billion) Die size
(mm2) 		Core config[a] L2
cache 		Core clock
(MHz)[b] 		Fillrate[c][d] Memory Processing power (TFLOPS) TDP Bus
interface
Pixel
(GP/s) 		Texture
(GT/s) 		Type Size Bandwidth
(GB/s) 		Bus width Clock
(MT/s) 		Half
precision 		Single
precision 		Double
precision
Arc Pro A30M Aug 8, 2022 ACM-G11
(DG2-128) 		TSMC
N6 		7.2 157 8 Xe cores
1024:64:32:8
(128:128:2) 		4 MB 1550 GDDR6 4 GB 112 64-bit 14000
4.20[49] 		50 W PCIe 4.0 x8
A40 6 GB 192 96-bit 16000
5.02[49]
A50 2050 75 W
A60M June 6, 2023 ACM-G12
(DG2-256) 		16 Xe cores
2048:128:64:16
(256:256:4) 		1300 8 GB 256 128-bit
9.42[49] 		95 W PCIe 4.0 x16
A60 2000 12 GB 384 192-bit
10.04[49] 		130 W
  1. ^ Shading cores (ALU): texture mapping units (TMU): render output units (ROP): ray tracing units
       (tensor cores (XMX): execution Units: render slices)
  2. ^ Boost values (if available) are stated below the base value in italic.
  3. ^ Pixel fillrate is calculated as the lowest of three numbers: number of ROPs multiplied by the base core clock speed, number of rasterizers multiplied by the number of fragments they can generate per rasterizer multiplied by the base core clock speed, and the number of streaming multiprocessors multiplied by the number of fragments per clock that they can output multiplied by the base clock rate.
  4. ^ Texture fillrate is calculated as the number of texture mapping units (TMUs) multiplied by the base (or boost) core clock speed.

Battlemage

[edit]

Battlemage (Xe2) is the second-generation Xe architecture that debuted with its low power variant in Lunar Lake mobile processors that released in September 2024. On December 3, 2024, Intel announced two Arc B-Series desktop graphics cards based on the Xe2-HPG graphics architecture.

Desktop

[edit]
Overview of Intel Arc Battlemage GPUs
Branding and Model Launch MSRP

(USD)

Code name Process Transistors (billion) Die size

(mm2)

Core Cache Memory Fillrate Processing power (TFLOPS) TDP Bus

interface

Core Config Clock

(MHz)

L1 L2 Type Size Bandwidth

(GB/s)

Bus width Clock

(MT/s)

Pixel

(GP/s)

Texture

(GT/s)

Half precision Single precision Double precision
Arc 5 B570 Jan 16, 2025 $219 BMG-G21 TSMC

N5

19.6 272 18 Xe Cores (144) 2304:144:72:18:144

(128:128:5)

1700

2500

4.5 MB 10 MB GDDR6 10 GB 380 160-bit 19000 122.4 200.0 244.8

360.0

23.04 11.52 1.44 150 W PCIe 4.0

x8

B580 Dec 13, 2024 $249 20 Xe Cores (160) 2560:160:80:20:160

(160:160:5)

1700

2670

5 MB 12 MB 12 GB 456 192-bit 136.0

213.6

272.0

427.2

27.34 13.67 1.709 190 W
  1. ^ Pixel fillrate is calculated as the number of render output units (ROPs) multiplied by the base (or boost) core clock speed.
  2. ^ Texture fillrate is calculated as the number of texture mapping units (TMUs) multiplied by the base (or boost) core clock speed.
  3. ^ Xe2-HPG Cores (Xe Vector Engines)

Unified Shaders : Texture Mapping Units : Render Output Units : Ray Tracing Cores : XMX Cores

  1. ^ Boost values (if available) are stated below the base value in italics.

Workstation

[edit]
Overview of Intel Arc Battlemage GPUs for Workstations
Branding and Model Launch MSRP (USD) Code name Process Transistors (billion) Die size

(mm2)

Core L2

cache

Fillrate Memory Processing power (TFLOPS) TDP Bus

interface

Config Clock

(MHz)

Pixel

(GP/s)

Texture

(GT/s)

Type Size Bandwidth

(GB/s)

Bus width Clock

(MT/s)

Half precision Single precision Double precision XMX Half Precision
Arc Pro B50 Sept 3, 2025 $349 BMG-G21 TSMC

N5

198.6 272 16 Xe2-cores

2048:128:64:16:128 (128:128:4)

1700

2600

4 MB 87

133

218

332.8

GDDR6 16 GB 224 128-bit 14000 21.3 10.65 1.33 197 70 W PCIe 5.0 x8
B60 Q3 2025 20 Xe2-cores

2560:160:80:20:160

(160:160:5)

2400 16 MB 192 384 24 GB 456 192-bit 19000 24.5 12.8 1.54 170 120-200 W
  1. ^ Pixel fillrate is calculated as the lowest of three numbers: number of ROPs multiplied by the base core clock speed, number of rasterizers multiplied by the number of fragments they can generate per rasterizer multiplied by the base core clock speed, and the number of streaming multiprocessors multiplied by the number of fragments per clock that they can output multiplied by the base clock rate.
  2. ^ Texture fillrate is calculated as the number of texture mapping units (TMUs) multiplied by the base (or boost) core clock speed.
  3. ^ Shading cores (ALU): texture mapping units (TMU): render output units (ROP): ray tracing units   (tensor cores (XMX): execution Units: render slices)
  4. ^ Boost values (if available) are stated below the base value in italic.

Future generations

[edit]

Intel has revealed future generations of Intel Arc GPUs under development: Celestial (Xe3P), and Druid (Xe4). Additionally, Panther Lake series iGPUs will be based on the Xe3 architecture.

Datacenter

[edit]

Intel H3C XG310

[edit]

On November 11, 2020 Intel launched the H3C XG310 data center GPU consisting of four DG1 GPUs with 32 GB of LPDDR4X memory on a single-slot PCIe card.[50][51] Each GPU is connected to 8 GB of memory over a 128-bit bus and the card uses a PCIe 3.0 x16 connection to the rest of the system. The GPUs use the Xe-LP (Gen 12.1) architecture.

Ponte Vecchio

[edit]

Intel officially announced their Xe general HPC/AI GPU codenamed Ponte Vecchio on November 17, 2019. It was revealed to use the Xe-HPC variant of the architecture[52] and Intel's 'Embedded Multi-Die Interconnect Bridge' (EMIB) and Foveros die stacking packaging on a Intel 4 node (previously referred to as 7 nm). Intel later confirmed at Architecture Day 2021 that Ponte Vecchio would use Compute Tiles manufactured on TSMC N5, Base Tiles and Rambo Cache Tiles manufactured using Intel 7 (previously referred to as 10 nm Enhanced SuperFin) and Xe Link Tiles manufactured on the TSMC N7 process. The new GPU is expected to be used in Argonne National Laboratory's new exascale supercomputer, Aurora, with compute nodes comprising two next generation Intel Xeon (codenamed "Sapphire Rapids") CPUs, and six Ponte Vecchio GPUs.[53][54]

Model[55][56] Launch Code name(s) Process Transistors (billion) Die size
(mm2) 		Core config[a] Cache Core clock
(MHz)[b] 		Fillrate[c][d] Memory Processing power (TFLOPS) TDP Bus
interface
L1 L2 Pixel
(GP/s) 		Texture
(GT/s) 		Type Size Bandwidth
(GB/s) 		Bus width Clock
(MT/s) 		Bfloat16 Single
precision 		Double
precision
Data Center GPU Max 1100 Jan 10, 2023 Xe-HPC
(Ponte Vecchio) 		Multiple[57] 100 1280 7168:448:0:56:448:448 28 MB 204 MB 1000
1550 		0 448.0
694.4 		HBM2E 48 GB 1228.8 3072-bit 3200
352		 14.336
22.221 		300 W PCIe 5.0 x16
Data Center GPU Max 1350 abandoned 14336:896:0:112:896:896 56 MB 408 MB 750
1550 		672.0
1388.8 		96 GB 2457.6 6144-bit
704		 21.504
44.442 		450 W
Data Center GPU Max 1550 Jan 10, 2023 16384:1024:0:128:1024:1024 64 MB 408 MB 900
1600 		921.6
1638.4 		128 GB 3276.8 8192-bit
832		 29.491
54.423 		600 W
  1. ^ shading cores (ALU):texture mapping units (TMU):render output units (ROP):ray tracing units:tensor cores (XMX):execution Units
  2. ^ Boost values (if available) are stated below the base value in italic.
  3. ^ Pixel fillrate is calculated as the lowest of three numbers: number of ROPs multiplied by the base core clock speed, number of rasterizers multiplied by the number of fragments they can generate per rasterizer multiplied by the base core clock speed, and the number of streaming multiprocessors multiplied by the number of fragments per clock that they can output multiplied by the base clock rate.
  4. ^ Texture fillrate is calculated as the number of texture mapping units (TMUs) multiplied by the base (or boost) core clock speed.

Rialto Bridge

[edit]

Intel officially announced the successor to Ponte Vecchio, GPU codenamed Rialto Bridge on May 31, 2022.[58] On March 3, 2023 Intel announced the discontinuation of Rialto Bridge in favor of their tile-based flexible and scalable Falcon Shores XPU (CPU + GPU) set to arrive in 2025.[59]

Arctic Sound

[edit]

Under the codename Arctic Sound Intel developed data center GPUs for visual cloud and AI inference based on the Xe-HP architecture (Gen 12.5).[60] The GPUs were supposed to be fabbed on Intel's 10nm node and have a die size of around 190 mm2 with 8 billion transistors.[61] Up to four GPUs tiles could be combined into a single package together with HBM2e memory. In October 2021 Raja Koduri announced that Xe-HP won't be commercialized into a final product.[60] Instead the Arctic Sound cards will be based on the Xe-HPG architecture (Gen 12.7), the same as the Alchemist consumer graphics cards.[62] They were launched on August 24, 2022 as the Intel Data Center GPU Flex series. On March 3, 2023 Intel announced that it would discontinue the development of Lancaster Sound which was supposed to succeed Arctic Sound in 2023 with incremental improvements. Instead Intel will accelerate the development of Melville Sound which will be a significant architectural leap in terms of performance and features.[59]

Intel Graphics Technology

[edit]
Main article: Intel Graphics Technology

Intel Graphics Technology (GT) is a collective name for a series of integrated graphics processors (IGP) produced by Intel,[63][64][65] which are manufactured on the same package or die as the central processing unit (CPU).[66] It was first introduced in 2010 as Intel HD Graphics[67][68] and renamed in 2017 to Intel UHD Graphics.

Before the advent of Intel HD Graphics, Intel integrated graphics were embedded in the northbridge of the motherboard as part of the Intel Hub architecture.[69] They were known as Intel Extreme Graphics and Intel GMA. Within the Platform Controller Hub (PCH) development, the northbridge was eliminated, and graphics processing was moved onto the same die as the central processing unit (CPU). The previous integrated graphics solution from Intel, Intel GMA, had a reputation for insufficient performance and functionality, and therefore was not considered a good choice for more demanding graphics applications, such as 3D gaming. The performance increase provided by Intel HD Graphics made the products competitive with integrated graphics adapters produced by its competitors, Nvidia and ATI/AMD. Intel HD Graphics, demonstrating minimal power consumption, which is important for laptops, was sufficiently powerful that PC manufacturers often discontinued offering discrete graphics options in both lower and higher-end laptop lines where reduced size and low power consumption are important. Nowadays, if a mini-PC for a home office or a portable laptop for work and light entertainment is needed, Iris Xe Graphics may be the best choice.[70]

Intel Iris Graphics and Intel Iris Pro Graphics are series of IGP introduced in 2013[71] with some Haswell processor models as high-performance versions of HD Graphics. Iris Pro Graphics was the first in the series to include embedded DRAM.[72] Since 2016, Intel has called this technology Intel Iris Plus Graphics with the launch of Kaby Lake. In the fourth quarter of 2013, Intel integrated graphics accounted for 65% of all PC graphics processor shipments.[73] However, this percentage does not reflect actual adoption, as some of those shipped devices ultimately ended up in systems with discrete graphics cards.

Intel HD and Iris Graphics are divided into generations, and within each generation into "performance tiers" designated by the "GTx" label. Each generation corresponds to an implementation of the Gen graphics microarchitecture with the corresponding GEN instruction set architecture starting from Gen4.

References

[edit]
Revisions and contributorsEdit on WikipediaRead on Wikipedia

Intel Xe

View on Grokipedia
from Grokipedia
Intel Xe is a unified GPU microarchitecture family developed by Intel Corporation, first announced in 2018, that integrates graphics processing, compute acceleration, and media capabilities into a scalable design spanning low-power integrated solutions, discrete gaming GPUs, and high-performance computing systems. The Xe architecture encompasses multiple microarchitectures tailored to specific use cases, including Xe-LP for efficient integrated graphics in mobile and client processors, Xe-HPG for enthusiast-level gaming and content creation in discrete GPUs like the Intel Arc series, and Xe-HPC for data center and supercomputing workloads in products such as the Ponte Vecchio accelerator. More recent iterations include Xe2 (also known as Battlemage) and Xe3, which enhance performance for AI, HPC, and graphics tasks, with Xe3 providing over 50% performance improvement compared to Xe2 in integrated configurations. This family evolved from Intel's prior Gen architectures to support a "one architecture" approach, enabling consistent development across power envelopes from teraflops to petaflops scale. Key innovations in Intel Xe include the Xe-core as the fundamental building block, comprising up to 16 vector engines (XVEs) and 16 matrix engines (XMX) per core for parallel processing, alongside dedicated hardware for real-time ray tracing, , and variable rate (VRS). It supports advanced technologies such as Intel Xe Super Sampling (XeSS) for AI-driven upscaling, DirectX 12 Ultimate features, and the oneAPI programming model for across CPU, GPU, and FPGA. Performance include up to 17.2 TFLOPs of FP32 compute in Xe-HPG configurations and 275 for using XMX engines. The architecture leverages advanced manufacturing processes, such as TSMC's N6 for discrete variants, and interconnects like EMIB and Xe Link for multi-tile scalability. Intel Xe powers a broad ecosystem, from consumer laptops with Iris Xe graphics in 11th-generation Tiger Lake processors to professional workstations and cloud services, emphasizing energy efficiency, AI acceleration, and ray-traced gaming. Ongoing developments, including Xe3 integration in upcoming Panther Lake processors, continue to expand its role in AI PCs and handheld gaming devices.

History

Announcement and Early Development

In December 2018, Intel announced the Xe at its Architecture Day event as a unified GPU platform intended to replace the existing Gen graphics generations, with plans for scalable implementations spanning integrated solutions in consumer processors to discrete cards for professional and data center applications. This initiative marked Intel's strategic push into the discrete graphics market, building on over a billion integrated GPU shipments to emphasize broad ecosystem integration. Raja Koduri, recruited from AMD in late 2017 as Senior Vice President and Chief Architect of the Core and Visual Computing Group, spearheaded the project and later oversaw it as head of Intel's Architecture, Graphics, and Software division. Under his leadership, Intel assembled a team of over 4,500 engineers, drawing expertise from former AMD and Nvidia talent, to develop Xe as a foundational shift toward a more parallel and adaptable graphics pipeline. The architecture's name, Xe, originated from the concept of "exascale for everyone," underscoring Intel's ambition to democratize exascale-level computing performance across diverse workloads and devices, from mobile to high-end servers. Initial design goals included native support for DirectX 12 to enable efficient rendering and compute tasks, with early planning for ray tracing hardware acceleration—publicly confirmed in May 2019—and dedicated AI acceleration units to handle machine learning inference and training from launch. Early development featured prototypes like a January 2018 demonstration of a discrete GPU die combining two Gen9 graphics slices, validating chiplet-like scalability using Intel's Embedded Multi-die Interconnect Bridge (EMIB) technology. Intel also partnered with Microsoft to align Xe with evolving DirectX features, ensuring compatibility for advanced rendering techniques and API optimizations during the 2018–2020 planning phase. This foundational work paved the way for Xe microarchitectures such as Xe-LP in low-power integrated graphics.

First-Generation Rollout

The first-generation architecture, known as Xe-LP, debuted in integrated form with the launch of 11th-generation processors codenamed on , 2020. These mobile processors targeted thin-and-light laptops, featuring up to 96 execution units in the Iris Xe configuration, enabling support for up to four simultaneous 4K displays and hardware-accelerated AV1 decoding for improved video playback . Early was in laptops, with systems outperforming prior integrated in gaming and tasks by up to 2x compared to competitors like AMD's Renoir series. Intel followed with its initial discrete graphics product, the Iris Xe Max (codenamed DG1), announced on October 31, 2020, and available in laptops starting November 2020 from partners including Acer, , and . This entry-level discrete GPU, built on the same Xe-LP with 96 execution units and 4GB of , focused on workloads like video encoding and photo rather than high-end gaming, integrating via Intel's Deep Link for hybrid CPU-GPU in systems. reception highlighted its in applications but noted limitations in raw against discrete like NVIDIA's MX series. The Xe architecture expanded into the discrete gaming market with the Intel Arc Alchemist series (Xe-HPG variant) in 2022, starting with mobile GPUs on March 30, followed by desktop models including the Arc A770 and A750 on October 12. Market reception was mixed due to launch-period driver instability, which caused crashes, inconsistent frame rates, and compatibility issues in older DirectX 9 titles, leading to delays in desktop availability from an initial Q1 target to mid-2022. Performance benchmarks positioned the flagship Arc A770 as competitive with NVIDIA's GeForce RTX 3060 and AMD's Radeon RX 6600 at 1440p resolution in modern rasterization workloads, delivering similar average frame rates (around 60-80 FPS) in titles like Cyberpunk 2077 and Forza Horizon 5, though ray tracing efficiency lagged behind competitors. In parallel, Intel extended Xe1 to data center applications with the Data Center GPU Flex Series launch on August 24, 2022, targeting cloud gaming, media transcoding, and AI inference. The series included the Flex 170 (up to 32 Xe cores, 16GB GDDR6) for high-performance tasks like supporting 68 simultaneous 1080p cloud gaming streams, and the lower-power Flex 140 for dense deployments, marking Intel's entry into scalable visual cloud computing with early partnerships from Dell, HPE, and others.

Second-Generation Advancements

The second-generation Intel Xe architecture, known as Xe2, marked significant progress in graphics efficiency and integration during 2023-2025, building on the foundational Xe1 designs to better support mobile computing and AI-accelerated workloads. This evolution emphasized power-optimized microarchitectures for consumer devices, enabling broader adoption in laptops and handheld systems where battery life and thermal constraints are critical. Xe2 debuted in Intel's processors, released in 2023, incorporating the Xe-LPG tailored for low-power in mobile environments. This focused on integrated for ultrabooks, delivering up to double the compared to prior Iris Xe solutions while supporting hardware-accelerated ray tracing and AI tasks. Lessons from the Xe1 rollout, particularly in driver optimization, informed Xe2's improved stability for real-time rendering and compute applications. In 2024, the Lunar Lake series advanced Xe2 further with the Xe2-LP microarchitecture, integrating seamlessly with an enhanced Neural Processing Unit (NPU) for synergistic AI processing. This design prioritized power efficiency, achieving up to 1.5 times the graphics performance of Meteor Lake at the same power envelope, alongside 67 TOPS of AI capability for on-device inference in mobile workloads. The architecture's optimizations, including merged vector engines and improved caching, reduced energy consumption for AI-enhanced content creation and video processing. The discrete graphics segment saw Xe2's expansion with the Arc Battlemage lineup in 2024, utilizing the Xe2-HPG to elevate gaming . These GPUs targeted mid-range gaming, offering up to 50% better over Xe1 equivalents in select workloads through refined execution units and media engines. Central to Xe2's advancements are architectural enhancements like up to 192 KB of shared local memory per Xe-core, which accelerates data sharing in compute-intensive tasks. Ray tracing capabilities were bolstered with additional pipelines—up to three per unit—for faster intersection calculations, improving real-time rendering in games and simulations by approximately 30% in integrated scenarios. Overall, these features drove Xe2's adoption in efficient mobile AI and graphics, positioning it as a competitive option for power-constrained devices through 2025.

Third-Generation Emergence

In October 2025, Intel revealed the third-generation Xe3 graphics architecture at its Technology Tour event in Arizona, highlighting its integration into the upcoming Panther Lake client processors. The Panther Lake implementation features an integrated Xe3 GPU with up to 12 Xe cores, delivering over 50% higher performance compared to the Xe2 architecture in Lunar Lake at the same power envelope. A performance-optimized variant, Xe3P, extends the architecture to discrete and GPUs, with the GPU designed specifically for AI inference workloads. This solution incorporates 160 GB of LPDDR5X memory and targets air-cooled enterprise servers for efficient, cost-effective operation in high-volume inference scenarios. Key architectural enhancements in Xe3 include an increase in shared local memory to 256 KB per Xe core, up from 192 KB in Xe2, enabling better handling of complex workloads. These changes optimize the architecture for token-based AI processing, such as surging inference token volumes, while also supporting advancements in handheld gaming performance. Intel's roadmap positions Xe3 for broader , with integration in Panther Lake as a successor to Lunar Lake platforms and planned expansions to desktop lines following Lake. This builds on Xe2's emphasis on low-power for mobile devices.

Architecture

Core Design Principles

The Intel Xe architecture represents a unified graphics and compute platform designed to span integrated, discrete, and data center GPUs, with its foundational principles announced at Intel Architecture Day in December 2018. This approach emphasizes scalability, efficiency, and versatility to address diverse workloads from mobile devices to high-performance computing. At its core, Xe adopts a tile-based design that allows modular integration of compute, graphics, and I/O components, enabling customization across power envelopes while maintaining a consistent architectural framework. A key principle is the unified instruction set architecture (ISA) that supports both graphics rendering and general-purpose computing, fostering portability across APIs such as Vulkan, OpenCL, oneAPI, and DirectX 12. This ISA incorporates advanced features like mesh shading for efficient geometry processing and variable rate shading (VRS) to optimize rendering by applying different shading rates based on screen content complexity, reducing computational overhead without visual quality loss. Scalable execution units further embody this principle, with Xe-cores integrating vector engines for traditional graphics and compute tasks alongside specialized units such as XMX engines for AI matrix operations and RT units for hardware-accelerated ray tracing, allowing seamless workload acceleration. These units scale in number and configuration to match application demands, supporting peak throughputs like up to 17 TFLOPs in FP32 for graphics-intensive scenarios. The memory subsystem adheres to a unified , where CPU and GPU share a common to minimize movement and latency, compatible with high-bandwidth types including HBM for data centers, GDDR for discrete GPUs, and LPDDR for integrated solutions. This incorporates multi-level caching—such as up to 16 MB L2 cache and per-core L1 caches—along with to enhance bandwidth efficiency, achieving effective rates up to 560 GB/s in GDDR6 configurations. Efficiency across power scales is achieved through dynamic power management techniques, including power gating to deactivate idle tiles and dynamic voltage and frequency scaling (DVFS) to adjust operating points in real-time based on workload intensity. These mechanisms, combined with the tile-based modularity, enable Xe to operate effectively from sub-10W integrated graphics to over 300W discrete GPUs, optimizing in varied environments like laptops and servers.

First-Generation Microarchitectures

The first-generation Intel Xe microarchitectures, collectively known as Xe1, introduced a unified execution model and scalable design principles to support diverse applications from mobile integrated graphics to high-performance computing. These variants—Xe-LP, Xe-HPC, and Xe-HPG—share core architectural elements such as Xe-cores with vector engines for parallel processing but are optimized for specific power envelopes, workloads, and form factors. Unveiled at Intel Architecture Day 2020, they mark the transition from previous Gen11 graphics to a more versatile GPU family capable of addressing integrated, discrete, and data center needs. A planned Xe-HP variant for high-performance discrete GPUs was canceled prior to release. Xe-LP, the low-power variant, targets efficient integrated graphics in mobile platforms. It features up to 96 Execution Units (EUs) arranged in slices, enabling significant performance improvements over prior generations while maintaining low power consumption. Integrated into the 10nm SuperFin-based Tiger Lake system-on-chip for laptops, Xe-LP delivers up to 2.95x faster graphics performance compared to the previous Ice Lake architecture, with memory bandwidth reaching approximately 68 GB/s when paired with LPDDR4x-4267 or similar configurations. This microarchitecture relies on shared system memory for graphics allocation, optimizing for thin-and-light devices without dedicated VRAM. Xe-HPC builds on the high-performance tile with a focus on compute-intensive applications, particularly in data centers. It includes native FP64 (double-precision floating-point) support to handle scientific simulations and high-performance computing (HPC) tasks efficiently, alongside tensor cores for AI acceleration. Deployed in products like the Intel Data Center GPU Flex Series, Xe-HPC offers configurations such as 32 Xe-cores on a PCIe Gen4 card for 150W TDP or dual-GPU setups at 75W, utilizing GDDR6 memory for balanced bandwidth and capacity in AI training and inference. This variant underscores Xe's role in exascale computing environments. Xe-HPG, the gaming-oriented variant, enhances graphics rendering with dedicated hardware for modern features. It incorporates ray tracing units (RTUs), with one RTU per Xe-core and up to four Xe-cores per slice, enabling hardware-accelerated real-time ray tracing, mesh shading, and variable rate shading compliant with 12 Ultimate. Used in the Arc Alchemist discrete GPUs (such as the A-series), Xe-HPG supports up to eight slices per GPU in flagship models, paired with GDDR6 memory for high-frame-rate gaming at resolutions up to 1440p. This microarchitecture prioritizes enthusiast-level performance while integrating AI upscaling via XMX engines.

Second-Generation Microarchitectures

The second-generation Intel Xe microarchitectures, collectively known as Xe2, represent an evolutionary step focused on enhancing AI acceleration, power efficiency, and graphics performance in consumer devices, building briefly on the gaming-oriented Xe1-HPG foundations. These architectures introduce larger ray tracing units, enhanced vector engines, and optimized front-end designs to deliver up to 50% better performance per Xe-core and improved efficiency. Key variants target integrated and discrete applications, prioritizing low-power mobile scenarios while advancing compute capabilities. The Xe2-LPG variant powers integrated graphics in Intel's Lunar Lake processors (Core Ultra 200V series), optimized for thin-and-light laptops with up to 8 Xe-cores per GPU tile. Each Xe-core features 192 KB of shared L1 cache and shared local memory (SLM), paired with an overall 8 MB L2 cache for the GPU, enabling efficient handling of and compute workloads. Deep integration with the on-chip Neural Processing Unit (NPU 4) allows seamless AI task offloading, achieving up to 67 of INT8 performance across the GPU and NPU for tasks like generation and video enhancement. This design yields up to 1.5x performance over prior generations at similar power levels, emphasizing battery life in ultraportable devices. For discrete graphics, the Xe2-HPG microarchitecture drives the Battlemage (Arc B-Series) GPUs, such as the Arc B580 with 20 Xe-cores, targeting mainstream 1440p gaming. It incorporates larger ray tracing units with three traversal pipelines per core, delivering approximately 30% faster ray tracing performance and up to 2x efficiency in select BVH traversal workloads compared to Xe1-HPG. Support for XeSS 2.0, Intel's second-generation AI upscaling technology, leverages dedicated XMX matrix engines to boost frame rates while maintaining visual fidelity, with up to 233 TOPS of AI compute across the card. These enhancements result in up to 70% better per-Xe-core performance overall. Across Xe2 implementations, general upgrades include 8 enhanced vector engines per Xe-core, supporting 512-bit SIMD operations for doubled FP16 throughput (512 operations per clock per lane) and improved compute scalability. Deeper caches and native support for indirect command execution further optimize AI and graphics pipelines, contributing to 1.2x to 12.5x IP-level performance gains in targeted scenarios like mesh shading and blending.

Third-Generation Microarchitectures

The third-generation Intel Xe microarchitectures, collectively referred to as Xe3, represent a significant evolution aimed at enhancing integrated graphics performance and AI acceleration in mobile and edge computing platforms. Announced in October 2025 and debuting in Intel's Panther Lake processors, the Xe3 integrated variant employs a 12-Xe-core design, enabling up to 50% improvement in rasterization performance compared to the prior Xe2 architecture in Lunar Lake. This uplift stems from architectural refinements, including an increase in shared local memory to 256 KB per Xe core from 192 KB in the previous generation, which reduces latency for workloads spilling over to higher-level caches. Additionally, each Xe core now supports up to 10 threads in flight—25% more than Xe2—bolstering multithreaded efficiency for graphics and compute tasks. A core innovation in Xe3 is the enhanced XMX (Xe Matrix eXtensions) engines, Intel's tensor core equivalents, which deliver up to 120 tera operations per second () of AI in the 12-core configuration, nearly doubling the 67 TOPS of Xe2's 8-core setup. These improvements target large models (LLMs) and other AI workloads, with optimized matrix operations for and at the edge, achieving better power-per-watt metrics suitable for battery-constrained devices. The also incorporates upgraded ray-tracing units and vector engines, with eight per Xe core, to handle complex and AI-driven rendering more effectively. The Xe3P extends these capabilities to performance-oriented applications, powering next-generation Arc discrete and GPUs such as the inference-focused accelerator, announced in 2025. Optimized for AI , Xe3P emphasizes tokens-per-watt through support for diverse types like FP8 and INT4, paired with 160 GB of LPDDR5X to manage high-bandwidth demands in enterprise servers. This prioritizes air-cooled, cost-effective scaling for edge AI deployments, building on Xe3's while targeting datacenter inference volumes. Retaining the tile-based of prior Xe generations, Xe3 enables modular across integrated and discrete implementations, facilitating configurations for expanded core counts in environments. These advancements position Xe3 as a bridge to workloads, with gains over Xe2-LP informing its low-power optimizations for mobile AI PCs.

Products

Integrated Graphics

Intel's integrated graphics based on the Xe represent a significant in embedded GPU , powering a range of mobile processors from the 11th-generation Core series onward. These iGPUs leverage the Xe-LP foundation to deliver improved media processing, gaming, and AI capabilities within power-constrained laptop environments. The first implementation appeared in the 11th-generation Core processors codenamed Tiger Lake, launched in 2020. These featured the Xe-LP microarchitecture with configurations up to 96 execution units (EUs), branded as Intel Iris Xe Graphics for higher-end models or UHD Graphics for entry-level variants. This design enabled playable 1080p gaming performance in titles like Shadow of the Tomb Raider at around 30 frames per second on low settings, marking a substantial uplift over prior Intel integrated solutions. Advancing to the Core Ultra Series 1 (Meteor Lake) in 2023, the Xe-LPG variant introduced up to 8 Xe-cores—equivalent to 128 EUs—optimized for AI-accelerated laptops under the branding. This supported ray tracing and vector engines for tasks, delivering efficient in generative AI workloads such as , where optimized drivers achieved up to 2x faster compared to previous generations on similar power envelopes. In 1080p gaming benchmarks, Meteor Lake iGPUs averaged 40-50 fps in modern titles like at low settings, benefiting from dedicated AI hardware for upscaling. The Core Ultra 200V series (Lunar Lake), released in 2024, employed the Xe2-LP microarchitecture with 4 to 8 Xe-cores, emphasizing ultra-low power consumption for thin-and-light devices while maintaining Arc Graphics integration. This low-power focus enabled sustained performance in battery-constrained scenarios, with the top-end Arc 140V configuration achieving over 50 fps average in 1080p medium gaming suites across 40 titles, outperforming AMD's Radeon 890M by up to 16% in select games. The Arc 140V supports 4K HDR output at 60 Hz with 10-bit color depth via DisplayPort 2.1 or HDMI 2.1 connections, utilizing efficient formats like 4:2:0 for bandwidth-limited scenarios such as HDMI 2.0. For AI tasks, it handled Stable Diffusion generations at rates competitive with discrete entry-level GPUs, leveraging improved XMX engines for faster matrix operations. Announced in October 2025 and expected to launch in early 2026, the Panther Lake processors will incorporate the Xe3 architecture with up to 12 Xe-cores, promising over 50% better overall performance than Lunar Lake equivalents. This iteration includes integrated Thunderbolt 5 support via the platform controller tile, facilitating high-bandwidth connectivity for AI peripherals. Early previews indicate enhanced 1080p gaming at 60+ fps in demanding titles with XeSS upscaling, alongside superior Stable Diffusion throughput—up to 70% faster than prior Arc 140V—positioning it as a strong contender for on-device AI and light gaming in mobile platforms.

Discrete Graphics

Intel's entry into discrete graphics with the Xe architecture commenced with the Iris Xe Max, codenamed DG1, announced on October 31, , and launched for laptops in early 2021. Based on the Xe-LP fabricated on a 10 nm process, it incorporates 96 execution units, 4 GB of GDDR6 on a 128-bit bus, and a TDP of 28 W, positioning it as an entry-level solution for thin-and-light laptops aimed at light gaming and content creation. The DG1 supports DirectX 12 Ultimate features including hardware-accelerated ray tracing via its 8 ray-tracing units, though its performance is modest compared to contemporary competitors, delivering playable frame rates in older titles at 1080p low settings. The second major discrete Xe product family, Arc Alchemist (DG2), debuted on March 30, 2022, marking Intel's full push into mainstream gaming GPUs under the Arc branding. Built on the Xe-HPG microarchitecture using TSMC's 6 nm process, it spans mobile models from the A370M (8 Xe-cores) to the flagship A770 (32 Xe-cores), with memory configurations up to 16 GB GDDR6 on a 256-bit interface and TDPs ranging from 35 W to 225 W. Key features include dedicated ray-tracing hardware with 32 ray-tracing units per Xe-core for efficient BVH traversal, mesh shaders, and variable-rate shading, enabling competitive performance in ray-traced workloads. In Cyberpunk 2077 at 1440p high settings without ray tracing, the A770 delivers approximately 50-60 FPS, improving to over 80 FPS with driver optimizations by mid-2023. Succeeding Alchemist, the Arc Battlemage series (Xe2-HPG) launched in 2024, focusing on mid-range discrete GPUs with refined for better and driver stability. Fabricated on TSMC's , initial models like the B580 (20 Xe2-cores, 12 GB GDDR6 on 192-bit bus, 190 W TDP) and B570 emphasize improved rasterization and ray tracing over predecessors, incorporating XeSS 2.0 for AI-driven frame and upscaling. These advancements yield stronger mid-range , with the B580 achieving around 53 FPS in at 1440p ultra settings without ray tracing in launch benchmarks. Looking ahead, Intel's roadmap outlines the Xe3P-based Arc family, expected in 2026 or later, as the next evolution in discrete graphics with a focus on enhanced AI capabilities. This architecture promises over 50% performance uplift over Xe2 in graphics tasks, including advanced AI upscaling via an improved XeSS implementation, targeting broader consumer and professional adoption in standalone GPUs. While sharing architectural DNA with integrated Xe variants for ecosystem consistency, these discrete cards prioritize dedicated memory and power for demanding standalone use.

Data Center GPUs

Intel's Data Center GPU offerings based on the Xe architecture target (HPC), (AI), and cloud workloads in server environments. These GPUs leverage the Xe-HPC microarchitecture, which emphasizes scalable compute tiles interconnected via advanced packaging technologies like EMIB and Foveros to deliver high memory bandwidth and efficiency for enterprise-scale applications. The Intel Data Center GPU Flex Series, formerly codenamed Arctic Sound-M and launched in 2022, utilizes the first-generation Xe-HPC microarchitecture to support cloud-based visual computing tasks such as virtual desktop infrastructure (VDI) and media transcoding. Configurations include the Flex 170 model with 32 Xe-cores, 16 GB of GDDR6 memory, and up to 576 GB/s bandwidth, enabling up to five times the media transcode throughput compared to previous generations in cloud scenarios. Designed for Ethernet-centric data center deployments, it supports scalability through PCIe connectivity and focuses on low-power, half-height form factors for dense server integration. The Intel Data Center GPU Max Series, previously codenamed Ponte Vecchio and introduced in 2022, represents a flagship Xe-HPC implementation for exascale HPC and AI training. It features multi-tile designs with up to 128 Xe-cores and 128 GB of HBM2e memory, providing bandwidth exceeding 3 TB/s to handle memory-intensive simulations and large-scale models. The architecture includes 47 active tiles in its full configuration, connected via high-speed Xe Links for multi-GPU scaling, and supports ray tracing units for advanced rendering in scientific visualizations. This series powers the Aurora supercomputer at Argonne National Laboratory, where each compute blade integrates six Max GPUs alongside Xeon Max processors, achieving exascale performance with over one exaFLOP of AI capability upon full deployment in 2025. Looking ahead, the Intel GPU codenamed , announced in , employs the third-generation Xe3P optimized for AI workloads, particularly large models (LLMs). It includes 160 GB of LPDDR5X to support high token-per-watt and handling, cost-effective scaling in inference-heavy centers without the overhead of training-focused hardware. This inference-centric prioritizes performance-per-watt, positioning it as a bridge to Xe generations for sustainable AI deployment.

Software and Ecosystem

Drivers and Compatibility

Intel Graphics Drivers for Xe-based hardware began with version 30.0.100.xx series for first-generation Xe (Xe-LP) implementations in processors, providing support for integrated features. Subsequent updates in the 30.xx , such as 30.0.101.1404, stability and for and 11, with support handled through the open-source Mesa drivers via the i915 kernel module and Intel's media stack. macOS integration relies on Apple's built-in drivers, updated through releases, ensuring compatibility for Xe-LP on supported Intel-based Macs up to the transition to . Second-generation Xe (Xe-HPG) in Alchemist/Arc discrete GPUs shifted to the 31.xx driver series, starting with versions like 31.0.101.3430 for launch support on Windows, including optimizations for discrete graphics workloads. These drivers also extended to Linux via the Mesa RADV Vulkan driver and Intel's compute runtime, while macOS support remained limited to integrated variants through OS updates. Third-generation Xe (Xe2-HPG) in Battlemage/Arc B-series utilizes the 32.xx series, such as 32.0.101.8247, focusing on enhanced efficiency and compatibility across Windows and Linux environments. Xe hardware achieves full compatibility with key APIs, including 12 for advanced features like mesh shaders and variable rate across all generations, 1.3 for cross-platform rendering, 4.6 for legacy applications, and 3.0 for compute tasks. This unified support enables seamless integration in gaming, , and scientific workloads without requiring hardware-specific tweaks. From their respective launches, Xe architectures incorporate Resizable BAR () support to improve CPU-GPU transfer , particularly beneficial for discrete Arc GPUs in gaming scenarios, and hardware-accelerated AV1 encoding and decoding for efficient video processing—decode available in Xe-LP from , with full encode/decode in Xe-HPG from Alchemist onward. These features enhance bandwidth utilization and media workflows without additional software overhead. Early driver releases for Arc discrete GPUs in 2022 encountered stuttering issues attributed to driver immaturity and resource management. Intel addressed these through iterative updates in late 2022 and 2023, significantly improving stability and user experience in various games, including reductions in frame-time variability. Updating to the latest Intel graphics drivers is recommended to ensure optimal support for advanced display modes, such as 4K resolution with 10-bit color on products like the Intel Arc Graphics 140V, which leverages HDR enabling for automatic switching to 10-bit color depth. Cross-generation driver support ensures backward compatibility, with unified packages covering Xe-LP through Xe2-HPG; however, as of September 2025, first- and second-generation integrated Xe variants (11th-14th Gen Core) transitioned to legacy status, receiving only quarterly critical fixes and security updates rather than new features or day-zero game optimizations, extending reliable operation through at least the Windows 10 end-of-life period.

Development Tools and APIs

The oneAPI suite provides a unified for across CPUs and GPUs, including the Xe architecture, enabling developers to write portable using Parallel C++ (DPC++) and for cross-architecture execution. This approach supports and API-based programming for GPUs, facilitating optimization and deployment of applications on Xe-based hardware without vendor-specific lock-in. The Intel Graphics Compute Runtime serves as an open-source implementation for compute workloads on Linux systems, supporting oneAPI Level Zero and OpenCL APIs to enable efficient execution on Xe GPUs. It integrates with the Linux kernel mode driver to handle low-level graphics and compute operations, allowing developers to target Xe hardware for parallel processing tasks in environments like high-performance computing clusters. Key development tools include , which offers advanced sampling and profiling for GPU , helping identify bottlenecks in Xe-accelerated applications through metrics like compute and utilization. Complementing this, provides roofline and offload modeling to optimize GPU kernels, visualizing limitations and suggesting improvements for and on Xe architectures. Xe-specific extensions enhance AI and workloads, with Intel Xe Matrix Extensions (XMX) introducing specialized instructions for on 2D systolic arrays, accelerating operations like GEMM kernels in SYCL applications. Additionally, support for DP4a instructions in shaders enables efficient INT8 accumulation, optimizing AI and features like super-resolution in pipelines on Xe-HPG and subsequent microarchitectures. The has expanded through integrations with major AI frameworks, such as the Extension for , which optimizes and on Xe data center GPUs using features like mixed precision and XMX . Similarly, the Extension for enables Xe GPU offloading for model execution, supporting scalable AI workloads in datacenter environments with oneAPI compatibility.

References

  1. https://www.hc32.hotchips.org/assets/program/conference/day1/HotChips2020_GPU_Intel_Xe_David_Blythe.pdf
  2. https://www.intel.com/content/www/us/en/docs/oneapi/optimization-guide-gpu/2025-0/intel-xe-gpu-architecture.html
  3. https://www.tomshardware.com/pc-components/gpus/intels-xe3-graphics-architecture-breaks-cover-panther-lakes-12-xe-core-igpu-promises-50-percent-better-performance-than-lunar-lake
  4. https://www.intel.com/content/www/us/en/developer/articles/technical/introduction-to-the-xe-hpg-architecture.html
  5. https://www.digitaltrends.com/computing/intel-xe-gen-11-announcement/
  6. https://www.nextplatform.com/2018/12/16/intel-unfolds-roadmaps-for-future-cpus-and-gpus/
  7. https://www.intc.com/news-events/press-releases/detail/193/raja-koduri-joins-intel-as-chief-architect-to-drive-unified
  8. https://www.techspot.com/article/1898-intel-xe-graphics-preview/
  9. https://www.nextplatform.com/2020/09/02/intel-puts-its-xe-gpu-stakes-in-the-ground/
  10. https://www.forbes.com/sites/marcochiappetta/2020/08/28/microsoft-divulges-next-generation-amd-intel-and-qualcomm-gpu-features-in-new-directx-12_2-blog-post/
  11. https://www.intel.com/content/www/us/en/docs/oneapi/optimization-guide-gpu/2023-2/intel-xe-gpu-architecture.html
  12. https://www.intc.com/news-events/press-releases/detail/1411/intel-launches-worlds-best-processor-for-thin-and-light
  13. https://www.phoronix.com/review/intel-tigerlake-launch
  14. https://www.pcmag.com/news/intel-launches-its-first-discrete-gpu-for-laptops-the-iris-xe-max
  15. https://www.notebookcheck.net/Intel-Iris-Xe-MAX-DG1-Mobile-GPU.439516.0.html
  16. https://download.intel.com/newsroom/2022/client-computing/Intel-Arc-Fact-Sheet-033022.pdf
  17. https://game.intel.com/us/stories/intel-arc-graphics-release/
  18. https://www.tomshardware.com/news/intel-blames-poor-software-for-arc-delays-shipments-miss
  19. https://www.pcworld.com/article/1361825/nvidia-geforce-rtx-3060-vs-intel-arc-a770.html
  20. https://www.tomshardware.com/features/intel-arc-graphics-performance-revisited-dx9-steps-forward-dx12-steps-back
  21. https://www.tomshardware.com/news/intel-launches-data-center-gpu-flex-series
  22. https://chipsandcheese.com/p/lunar-lakes-igpu-debut-of-intels
  23. https://www.intel.com/content/www/us/en/content-details/824434/2024-intel-tech-tour-xe2-and-lunar-lake-s-gpu.html
  24. https://www.techpowerup.com/review/intel-meteor-lake-technical-deep-dive/5.html
  25. https://www.tomshardware.com/news/intel-meteor-lake-integrated-graphics-doubles-performance-per-watt
  26. https://www.gamersnexus.net/news/intel-fights-back-arc-battlemage-xe2-gpus-changing-hyper-threading
  27. https://www.tomshardware.com/pc-components/cpus/intel-unwraps-lunar-lake-architecture-up-to-68-ipc-gain-for-e-cores-16-ipc-gain-for-p-cores/4
  28. https://cdrdv2-public.intel.com/824434/2024_Intel_Tech%20Tour%20TW_Xe2%20and%20Lunar%20Lakes%20GPU.pdf
  29. https://www.tomshardware.com/pc-components/gpus/intel-arc-xe2-battlemage-gpus-rumored-to-arrive-next-month-ahead-of-amd-rdna-4-and-nvidia-blackwell
  30. https://wccftech.com/intel-xe2-gpus-50-percent-uplift-new-ray-tracing-cores-lunar-lake-arc-battlemage-discrete/
  31. https://videocardz.com/newz/intel-claims-arc-xe2-lunar-lake-graphics-are-worlds-best-built-in-gpu
  32. https://www.phoronix.com/review/intel-panther-lake
  33. https://videocardz.com/newz/intel-details-xe3-gpu-architecture-for-panther-lake-up-to-12-xe-cores-and-50-performance-vs-lunar-lake
  34. https://newsroom.intel.com/artificial-intelligence/intel-to-expand-ai-accelerator-portfolio-with-new-gpu
  35. https://www.tomshardware.com/pc-components/gpus/intel-unveils-crescent-island-an-inference-only-gpu-with-xe3p-architecture-and-160gb-of-memory
  36. https://wccftech.com/intel-crescent-island-gpu-next-gen-xe3p-graphics-160-gb-lpddr5x-ai-inference/
  37. https://www.gaminglaptopreport.com/news/intel-panther-lake-an-xe3-deep-dive/
  38. https://gamersnexus.net/gpus/intels-new-gpu-xe3-architecture-changes-handheld-gaming-cpus-xess3
  39. https://www.tomshardware.com/pc-components/cpus/intel-takes-the-wraps-off-panther-lake-first-18a-client-processor-brings-the-best-of-lunar-lake-and-arrow-lake-together-in-one-package
  40. https://www.tomshardware.com/news/intel-axes-xe-hp-gpus-for-datacenters
  41. https://download.intel.com/newsroom/2020/client-computing/intel-2020-architecture-day-fact-sheet.pdf
  42. https://www.intel.com/content/www/us/en/products/details/discrete-gpus/data-center-gpu/flex-series.html
  43. https://game.intel.com/anz/stories/battlemage-enters-the-game-intel-arc-b580-out-now/
  44. https://www.tweaktown.com/news/100335/intel-says-xe2-battlemage-gpu-inside-of-lunar-lake-cpus-is-the-worlds-best-built-in/index.html
  45. https://www.intel.com/content/www/us/en/products/docs/processors/core/11th-gen-processors-brief.html
  46. https://www.tomshardware.com/news/intel-launches-11th-gen-tiger-lake-48-ghz-lpddr4-memory-iris-xe-graphics
  47. https://www.tomshardware.com/features/intel-11th-gen-tiger-lake-superfin-10nm-benchmarks
  48. https://videocardz.com/newz/intel-details-arc-xe-lpg-meteor-lake-gpu-architecture-up-to-8-xe-cores-and-ray-tracing-acceleration
  49. https://www.tomshardware.com/laptops/i-tested-intels-meteor-lake-cpus-on-ai-workloads-and-amds-chips-sometimes-beat-them
  50. https://www.pcgamesn.com/intel/core-ultra-200v-lunar-lake-guide
  51. https://www.tomshardware.com/pc-components/gpus/we-benchmarked-intels-lunar-lake-gpu-with-core-ultra-9-drivers-still-holding-back-arc-graphics-140v-performance
  52. https://cdrdv2-public.intel.com/829568/829568-004.pdf
  53. https://www.intel.com/content/www/us/en/support/articles/000057279/graphics.html
  54. https://www.techpowerup.com/review/intel-panther-lake-technical-deep-dive/9.html
  55. https://newsroom.intel.com/client-computing/intel-unveils-panther-lake-architecture-first-ai-pc-platform-built-on-18a
  56. https://www.intel.com/content/www/us/en/products/sku/211014/intel-iris-xe-dedicated-graphics-card-80-eu/specifications.html
  57. https://www.techpowerup.com/gpu-specs/iris-xe-max-graphics.c3737
  58. https://www.intel.com/content/www/us/en/products/details/discrete-gpus/arc.html
  59. https://www.tomshardware.com/news/intel-demoes-arc-a770-gpu-leaves-old-apis-in-the-dust
  60. https://www.techpowerup.com/gpu-specs/arc-b580.c4244
  61. https://gamersnexus.net/gpus/intel-arc-b580-battlemage-gpu-review-benchmarks-vs-nvidia-rtx-4060-amd-rx-7600-more
  62. https://wccftech.com/intel-xe3-graphics-official-50-percent-faster-than-xe2-xe3p-next-gen-arc-family/
  63. https://www.intel.com/content/www/us/en/products/sku/230019/intel-data-center-gpu-flex-170/specifications.html
  64. https://www.intel.com/content/www/us/en/products/sku/232873/intel-data-center-gpu-max-1550/specifications.html
  65. https://www.intel.com/content/www/us/en/products/docs/processors/max-series/overview.html
  66. https://www.tomshardware.com/tech-industry/supercomputers/aurora-supercomputer-is-now-fully-operational-available-to-researchers
  67. https://www.dell.com/support/home/en-us/drivers/driversdetails?driverid=ddtn5
  68. https://videocardz.com/driver/intel-graphics-driver-30-0-101-1404
  69. https://www.intel.com/content/www/us/en/support/articles/000022440/graphics.html
  70. https://www.techpowerup.com/305334/intel-releases-arc-graphics-drivers-31-0-101-4146-whql
  71. https://www.intel.com/content/www/us/en/download/785597/intel-arc-graphics-windows.html
  72. https://www.intel.com/content/www/us/en/support/articles/000005524/graphics.html
  73. https://www.gamedev.net/forums/topic/710288-intel-nuc-11-pro-kit-intel-i5-1135g7-intel-iris-xe-graphics-and-directx-12-ultimate-mesh-shader/
  74. https://www.intel.com/content/www/us/en/support/articles/000090831/graphics.html
  75. https://www.phoronix.com/news/Intel-Gen12-Xe-AV1-Decode-Media
  76. https://videocardz.com/newz/intel-confirms-meteor-lake-has-av1-video-encoding-and-decoding-support
  77. https://gamersnexus.net/gpus/one-year-later-intel-arc-gpu-drivers-bugs-huge-improvements
  78. https://www.intel.com/content/www/us/en/support/articles/000101986/graphics.html
  79. https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit.html
  80. https://www.intel.com/content/www/us/en/docs/oneapi/optimization-guide-gpu/2025-0/configuring-gpu-device.html
  81. https://www.intel.com/content/www/us/en/docs/oneapi/installation-guide-hpc-cluster/2025-0/step-2-install-gpu-kernel-driver-and-intel.html
  82. https://www.intel.com/content/www/us/en/developer/articles/tool/opencl-drivers.html
  83. https://www.intel.com/content/www/us/en/docs/oneapi/optimization-guide-gpu/2023-0/gpu-analysis-with-vtunetm-profiler.html
  84. https://www.intel.com/content/www/us/en/docs/oneapi/optimization-guide-gpu/2023-0/intel-advisor-gpu-analysis.html
  85. https://www.intel.com/content/www/us/en/docs/oneapi/optimization-guide-gpu/2024-1/xmx.html
  86. https://www.intel.com/content/www/us/en/developer/articles/technical/introducing-intel-extension-for-pytorch-for-gpus.html
  87. https://www.intel.com/content/www/us/en/developer/articles/technical/introduction-to-intel-extension-for-tensorflow.html
Add your contribution
Related Hubs
User Avatar
No comments yet.