Recent from talks
Nothing was collected or created yet.
Ice Lake (microprocessor)
View on Wikipedia| General information | |
|---|---|
| Launched | September 2019 |
| Marketed by | Intel |
| Designed by | Intel |
| Common manufacturer |
|
| CPUID code | 703E5 |
| Product code | 80689 |
| Performance | |
| Max. CPU clock rate | 4.1 GHz |
| DMI speeds | 8 GT/s |
| Cache | |
| L1 cache | 80 KB (per core):
|
| L2 cache | 512 KB (per core) |
| L3 cache | Up to 8 MB |
| Architecture and classification | |
| Technology node | Intel 10 nm+ |
| Microarchitecture | Sunny Cove |
| Instruction set | x86-64 |
| Instructions | x86-64 |
| Physical specifications | |
| Cores |
|
| GPU | Gen11 |
| Socket |
|
| Products, models, variants | |
| Product code name |
|
| Brand name |
|
| Variant |
|
| History | |
| Predecessors | Cannon Lake (10 nm process) Whiskey Lake (14 nm optimization) |
| Successor | Tiger Lake (10 nm optimization) |
| Support status | |
| Legacy support for iGPU | |
| General information | |
|---|---|
| Launched | April 2021 |
| Performance | |
| Max. CPU clock rate | 3.7 GHz |
| Cache | |
| L1 cache | 80 KB (per core):
|
| L2 cache | Up to 50 MB |
| L3 cache | Up to 60 MB |
| Architecture and classification | |
| Technology node | Intel 10 nm+ |
| Microarchitecture | Sunny Cove |
| Instruction set | x86-64 |
| Instructions | x86-64 |
| Physical specifications | |
| Cores |
|
| Socket | |
| Products, models, variants | |
| Brand name |
|
| History | |
| Predecessor | Cascade Lake (14 nm) |
| Successors | Same generation: Cooper Lake (14 nm, 4S/8S systems) Next generation: Sapphire Rapids |
Ice Lake is Intel's codename for the 10th generation Intel Core mobile and 3rd generation Xeon Scalable server processors based on the Sunny Cove microarchitecture. Ice Lake represents an Architecture step in Intel's process–architecture–optimization model.[1][2][3][4] Produced on the second generation of Intel's 10 nm process, 10 nm+, Ice Lake is Intel's second microarchitecture to be manufactured on the 10 nm process, following the limited launch of Cannon Lake in 2018.[1][5][6][7][8] However, Intel altered their naming scheme in 2020 for the 10 nm process. In this new naming scheme, Ice Lake's manufacturing process is called simply 10 nm, without any appended pluses.[9]
Ice Lake CPUs are sold together with the 14 nm Comet Lake CPUs as Intel's "10th Generation Core" product family.[10] There are no Ice Lake desktop or high-power mobile processors; Comet Lake fulfills this role. Sunny Cove-based Xeon Scalable CPUs (codenamed "Ice Lake-SP") officially launched on April 6, 2021.[11][12] Intel officially launched Xeon W-3300 series workstation processors on July 29, 2021.[13]
Ice Lake's direct successor in mobile is Tiger Lake, a third-generation 10 nm SuperFin processor family using the new Willow Cove microarchitecture and integrated graphics based on the new Intel Xe microarchitecture.[14] Ice Lake-SP was succeeded by Sapphire Rapids, powered by Golden Cove cores.[15] Several mobile Ice Lake CPUs were discontinued on July 7, 2021.[16]
Design history and features
[edit]Ice Lake was designed by Intel Israel's processor design team in Haifa, Israel.[17][18]
Ice Lake is built on the Sunny Cove microarchitecture.[19][20] Intel released details of Ice Lake during Intel Architecture Day in December 2018, stating that the Sunny Cove core Ice Lake would be focusing on single-thread performance, new instructions, and scalability improvements. Intel stated that the performance improvements would be achieved by making the core "deeper, wider, and smarter".[20]
Ice Lake features Intel's Gen11 graphics, increasing the number of execution units to 64, from 24 or 48 in Gen9.5 graphics, achieving over 1 TFLOPS of compute performance.[citation needed] Each execution unit supports 7 threads, meaning that the design has 512 concurrent pipelines. Feeding these execution units is a 3 megabyte L3 cache, a four-fold increase from Gen9.5, alongside the increased memory bandwidth enabled by LPDDR4X on low-power mobile platforms.Gen11 graphics introduces tile-based rendering. It also features Coarse Pixel Shading (CPS), which is Intel's implementation of variable-rate shading (VRS). The architecture also includes an all-new HEVC encoder design.[20] On August 1, 2019, Intel released the specifications of Ice Lake -U and -Y CPUs.[21] The Y-series CPUs lost their -Y suffix and m3 naming. Instead, Intel uses a trailing number before the GPU type to indicate their package power; "0" corresponds to 9 W, "5" to 15 W, and "8" to 28 W. Furthermore, the first two numbers in the model number correspond to the generation of the chip, while the third number dictates the family the CPU belongs to (i3, i5, etc.); thus, a 1035G7 would be a 10th generation Core i5 with a package power of 15 watts and a G7 GPU.
Pre-orders for laptops featuring Ice Lake CPUs started in August 2019, followed by shipments in September.[22]
CPU
[edit]- Intel Sunny Cove CPU cores[23]
- Dynamic Tuning 2.0 which allows the CPU to stay at turbo frequencies for longer[26][27]
- TAGE-like directional branch predictor (with a global history size of 194 taken branches)[28]
- Hardware acceleration for SHA operations (Secure Hash Algorithms)
- Intel Deep Learning Boost, used for machine learning/artificial intelligence inference acceleration[29][27]
- PCI Express 4.0 on Ice Lake-SP
GPU
[edit]- Gen 11 GPU with up to 64 execution units (From 24 and 48 EU)[30][31]
- 4K at 120 Hz, 5K, 8K display output[32]
- Variable Rate Shading[27][33]
- DisplayPort 1.4a with Display Stream Compression; HDMI 2.0b
- Up to 1.15 TFLOPS of computational performance[citation needed]
- Two HEVC 10-bit encode pipelines, either two 4K 60 Hz RGB/Y′CBCR 4:4:4 streams simultaneously or one 8K 30 Hz Y′CBCR 4:2:2[27]
- VP9 8-bit and 10-bit hardware encoding for all supported platforms as part of Intel Quick Sync Video[34][35][36]
- Integer and nearest neighbor image scaling[37][38]
- 4th Gen IPU[39]
Package
[edit]List of Ice Lake CPUs
[edit]Ice Lake (mobile)
[edit]| Processor branding |
Model | Cores (threads) |
CPU clock
(GHz) |
GPU | L3 cache (MB) |
TDP
(W) |
cTDP (W) | Price
(US$) | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Base | Turbo | Series | EUs | Boost clock (GHz) |
up | down | ||||||
| Core i7 | 1068NG7 | 4 (8) | 2.3 | 4.1 | Iris Plus | 64 | 1.1 | 8 MB | 28 | $426 | ||
| 1065G7 | 1.3 | 3.9 | 15 | 25 | 12 | |||||||
| 1060NG7 | 1.2 | 3.8 | 10 | |||||||||
| 1060G7 | 1.0 | 9 | 12 | |||||||||
| Core i5 | 1038NG7 | 2.0 | 1.05 | 6 MB | 28 | $320 | ||||||
| 1035G7 | 1.2 | 3.7 | 15 | 25 | 12 | |||||||
| 1035G4 | 1.1 | 48 | $309 | |||||||||
| 1035G1 | 1.0 | 3.6 | UHD | 32 | 13 | $297 | ||||||
| 1030NG7 | 1.1 | 3.5 | Iris Plus | 64 | 10 | |||||||
| 1030G7 | 0.8 | 9 | 12 | |||||||||
| 1030G4 | 0.7 | 48 | ||||||||||
| Core i3 | 1005G1 | 2 (4) | 1.2 | 3.4 | UHD | 32 | 0.9 | 4 MB | 15 | 25 | 13 | $281 |
| 1000NG4 | 1.1 | 3.2 | Iris Plus | 48 | 9 | |||||||
| 1000G4 | 12 | 8 | ||||||||||
| 1000G1 | UHD | 32 | ||||||||||
| Pentium | 6805 | 3.0 | 0.85 | 15 | $161 | |||||||
Ice Lake-SP (Xeon Scalable)
[edit]Xeon Platinum series
[edit]| Model | sSpec number |
Cores (threads) |
Clock rate | Turbo Boost all-core/2.0 (/max. 3.0) |
L2 cache |
L3 cache |
TDP | Socket | I/O bus | Memory | Release date | Part number(s) |
Release price (USD) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Xeon Platinum 8351N |
|
36 (72) | 2.4 GHz | 3.5 GHz | 36 × 1.25 MB | 54 MB | 225 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-2933 | 6 April 2021 |
|
$3,466 (equivalent to $4,022 in 2024) |
| Xeon Platinum 8352S |
|
32 (64) | 2.2 GHz | 3.4 GHz | 32 × 1.25 MB | 48 MB | 205 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-3200 | 6 April 2021 |
|
$4,632 (equivalent to $5,375 in 2024) |
| Xeon Platinum 8352V |
|
36 (72) | 2.1 GHz | 3.5 GHz | 36 × 1.25 MB | 54 MB | 195 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-2933 | 6 April 2021 |
|
$3,993 (equivalent to $4,633 in 2024) |
| Xeon Platinum 8352Y |
|
32 (64) | 2.2 GHz | 3.4 GHz | 32 × 1.25 MB | 48 MB | 205 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-3200 | 6 April 2021 |
|
$3,995 (equivalent to $4,636 in 2024) |
| Xeon Platinum 8358 |
|
32 (64) | 2.6 GHz | 3.4 GHz | 32 × 1.25 MB | 48 MB | 250 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-3200 | 6 April 2021 |
|
$4,607 (equivalent to $5,346 in 2024) |
| Xeon Platinum 8358P |
|
32 (64) | 2.6 GHz | 3.4 GHz | 32 × 1.25 MB | 48 MB | 240 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-3200 | 6 April 2021 |
|
$4,523 (equivalent to $5,248 in 2024) |
| Xeon Platinum 8360Y |
|
36 (72) | 2.4 GHz | 3.5 GHz | 36 × 1.25 MB | 54 MB | 250 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-3200 | 6 April 2021 |
|
$5,383 (equivalent to $6,246 in 2024) |
| Xeon Platinum 8362 |
|
32 (64) | 2.8 GHz | 3.6 GHz | 32 × 1.25 MB | 48 MB | 265 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-3200 | 6 April 2021 |
|
$6,236 (equivalent to $7,236 in 2024) |
| Xeon Platinum 8368 |
|
38 (76) | 2.4 GHz | 3.4 GHz | 38 × 1.25 MB | 57 MB | 270 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-3200 | 6 April 2021 |
|
$7,214 (equivalent to $8,371 in 2024) |
| Xeon Platinum 8368Q |
|
38 (76) | 2.6 GHz | 3.7 GHz | 38 × 1.25 MB | 57 MB | 270 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-3200 | 6 April 2021 |
|
$7,719 (equivalent to $8,957 in 2024) |
| Xeon Platinum 8380 |
|
40 (80) | 2.3 GHz | 3.4 GHz | 40 × 1.25 MB | 60 MB | 270 W
|
LGA 4189 | 11.2 GT/s QPI | 8×DDR4-3200 | 6 April 2021 |
|
$9,359 (equivalent to $10,860 in 2024) |
Xeon Gold series
[edit]| Model | Cores (threads) |
Base clock | Boost clock (1-core) |
Boost clock (All-core) |
L3 cache | L2 cache | TDP | Price (RCP) |
|---|---|---|---|---|---|---|---|---|
| 6354 | 18 (36) | 3.00 GHz | 3.60 GHz | 39 MB | 22.50 MB | 205W | $2445 US | |
| 6348 | 28 (56) | 2.60 GHz | 3.50 GHz | 3.40 GHz | 42 MB | 35.00 MB | 235W | $3072 US |
| 6346 | 16 (32) | 3.10 GHz | 3.60 GHz | 36 MB | 20.00 MB | 205W | $2300 US | |
| 6338N | 32 (64) | 2.20 GHz | 3.50 GHz | 2.70 GHz | 48 MB | 40.00 MB | 185W | $2795 US |
| 6338T | 24 (48) | 2.10 GHz | 3.40 GHz | 2.70 GHz | 36 MB | 30.00 MB | 165W | $2742 US |
| 6338 | 32 (64) | 2.00 GHz | 3.20 GHz | 2.60 GHz | 48 MB | 40.00 MB | 205W | $2612 US |
| 6314U | 32 (64) | 2.30 GHz | 3.40 GHz | 2.90 GHz | 48 MB | 40.00 MB | 205W | $2600 US |
| 6342 | 24 (48) | 2.80 GHz | 3.50 GHz | 3.30 GHz | 36 MB | 30.00 MB | 230W | $2529 US |
| 6334 | 8 (16) | 3.60 GHz | 3.70 GHz | 3.60 GHz | 18 MB | 10.00 MB | 165W | $2214 US |
| 6330N | 28 (56) | 2.20 GHz | 3.40 GHz | 2.60 GHz | 42 MB | 35.00 MB | 165W | $2029 US |
| 6336Y | 24 (48) | 2.40 GHz | 3.60 GHz | 3.00 GHz | 36 MB | 30.00 MB | 185W | $1977 US |
| 6330 | 28 (56) | 2.00 GHz | 3.10 GHz | 2.60 GHz | 42 MB | 35.00 MB | 205W | $1894 US |
| 5318S | 24 (48) | 2.10 GHz | 3.40 GHz | 2.60 GHz | 36 MB | 30.00 MB | 165W | $1667 US |
| 5320T | 20 (40) | 2.30 GHz | 3.50 GHz | 30 MB | 25.00 MB | 150W | $1727 US | |
| 5320 | 26 (52) | 2.20 GHz | 3.40 GHz | 2.80 GHz | 39 MB | 32.50 MB | 185W | $1555 US |
| 6312U | 24 (48) | 2.40 GHz | 3.60 GHz | 3.10 GHz | 36 MB | 30.00 MB | 185W | $1450 US |
| 5318N | 24 (48) | 2.10 GHz | 3.40 GHz | 2.70 GHz | 36 MB | 30.00 MB | 150W | $1375 US |
| 6326 | 16 (32) | 2.90 GHz | 3.50 GHz | 3.30 GHz | 24 MB | 20.00 MB | 185W | $1300 US |
| 5318Y | 24 (48) | 2.00 GHz | 3.40 GHz | 2.60 GHz | 36 MB | 30.00 MB | 165W | $1273 US |
| 5317 | 12 (24) | 3.00 GHz | 3.60 GHz | 3.40 GHz | 18 MB | 15.00 MB | 150W | $950 US |
| 5315Y | 8 (16) | 3.20 GHz | 3.60 GHz | 3.50 GHz | 12 MB | 10.00 MB | 140W | $895 US |
Xeon Silver series
[edit]| Model | Cores (threads) |
Base clock | Boost clock (1-core) |
Boost clock (All-core) |
L3 cache | L2 cache | TDP | Price (RCP) |
|---|---|---|---|---|---|---|---|---|
| 4316 | 20 (40) | 2.30 GHz | 3.40 GHz | 2.80 GHz | 30 MB | 25.00 MB | 150W | $1002 US |
| 4314 | 16 (32) | 2.40 GHz | 3.40 GHz | 2.90 GHz | 24 MB | 20.00 MB | 135W | $694 US |
| 4310T | 10 (20) | 2.30 GHz | 3.40 GHz | 2.90 GHz | 15 MB | 13.75 MB | 105W | $555 US |
| 4310 | 12 (24) | 2.10 GHz | 3.30 GHz | 2.70 GHz | 18 MB | 12.50 MB | 120W | $501 US |
| 4309Y | 8 (16) | 2.80 GHz | 3.60 GHz | 3.40 GHz | 12 MB | 10.00 MB | 105W | $501 US |
Ice Lake-D
[edit]This section needs expansion. You can help by adding to it. (July 2024) |
Intel announced the next generation of Xeon D, codenamed Ice Lake-D in April 2021.[42] Intel official launched the Xeon D D-2700 series and D-1700 series CPUs at MWC 2022.[43] Xeon D D-2800 series and D-1800 series were announced on Dec 14, 2023.[44]
Workstation processors
[edit]"Ice Lake-W3300" (10 nm)
- PCI Express lanes: 64
- Supports up to 16 DIMMs of DDR4 memory, maximum 4 TB.[45]
| Model number |
Spec number |
Cores (threads) |
Frequency | Turbo Boost all-core/2.0 (/max. 3.0) |
L2 cache |
L3 cache |
TDP | Socket | I/O bus | Memory | Release date | Part number(s) |
Release price (USD) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
| |||||||||||||
| Xeon W-3375 |
|
38 (76) | 2.5 GHz | ?/4.0 GHz | 38 × 1.25 MB | 57 MB | 270 W
|
LGA 4189 | DMI 3.0 | 8× DDR4-3200 | 29 July 2021 |
|
$4,499 (equivalent to $5,221 in 2024) |
| Xeon W-3365 |
|
32 (64) | 2.7 GHz | ?/4.0 GHz | 32 × 1.25 MB | 48 MB | 270 W
|
LGA 4189 | DMI 3.0 | 8× DDR4-3200 | 29 July 2021 |
|
$3,499 (equivalent to $4,060 in 2024) |
| Xeon W-3345 |
|
24 (48) | 3 GHz | ?/4.0 GHz | 24 × 1.25 MB | 36 MB | 250 W
|
LGA 4189 | DMI 3.0 | 8× DDR4-3200 | 29 July 2021 |
|
$2,499 (equivalent to $2,900 in 2024) |
| Xeon W-3335 |
|
16 (32) | 3.4 GHz | ?/4.0 GHz | 16 × 1.25 MB | 24 MB | 250 W
|
LGA 4189 | DMI 3.0 | 8× DDR4-3200 | 29 July 2021 |
|
$1,299 (equivalent to $1,507 in 2024) |
| Xeon W-3323 |
|
12 (24) | 3.5 GHz | ?/3.9 GHz | 12 × 1.25 MB | 21 MB | 220 W
|
LGA 4189 | DMI 3.0 | 8× DDR4-3200 | 29 July 2021 |
|
$949 (equivalent to $1,101 in 2024) |
See also
[edit]References
[edit]- ^ a b Bright, Peter (August 15, 2017). "Intel's next generation chip plans: Ice Lake and a slow 10nm transition". Ars Technica. Retrieved August 15, 2017.
- ^ Cutress, Ian (August 15, 2017). "Intel Officially Reveals Post-8th Generation Core Architecture Code Name: Ice Lake, Built on 10nm+". AnandTech. Archived from the original on August 16, 2017. Retrieved April 7, 2023.
- ^ Shilov, Anton; Cutress, Ian. "Intel Server Roadmap: 14nm Cooper Lake in 2019, 10nm Ice Lake in 2020". AnandTech. Archived from the original on August 8, 2018. Retrieved September 3, 2018.
- ^ Cutress, Ian (March 22, 2016). "Intel's 'Tick-Tock' Seemingly Dead, Becomes 'Process-Architecture-Optimization'". AnandTech. Archived from the original on March 23, 2016. Retrieved October 22, 2018.
- ^ Dave James (May 30, 2017). "Intel Coffee Lake - 8th Gen Core >30% faster than Kaby Lake and here by the holidays". PCGamesN.
- ^ Garreffa, Anthony (January 20, 2016). "Intel teases its Ice Lake & Tiger Lake family, 10nm for 2018 and 2019". TweakTown. Retrieved June 3, 2016.
- ^ "What's the Name of Intel's Third 10-Nanometer Chip?". The Motley Fool. January 18, 2016. Retrieved June 3, 2016.
- ^ "Cannon Lake stumbles into the market: The IdeaPad 330-15ICN is the first laptop with a 10-nm-CPU". Notebookcheck. Retrieved October 22, 2018.
- ^ a b Cutress, Ian (September 25, 2020). "What Products Use Intel 10nm? SuperFin and 10++ Demystified". AnandTech. Archived from the original on September 25, 2020. Retrieved September 29, 2020.
- ^ "10th Gen Core: Intel verwirrt mit 1000er- und 10000er-Prozessoren - Golem.de". www.golem.de (in German). Retrieved January 7, 2020.
- ^ "Intel Launches Its Most Advanced Performance Data Center Platform".
- ^ "New Intel Processors Accelerate 5G Network Transformation". Intel Newsroom. Retrieved April 7, 2021.
- ^ "Intel Announces New Xeon W-3300 Processors".
- ^ Cutress, Ian. "Intel's 11th Gen Core Tiger Lake SoC Detailed: SuperFin, Willow Cove and Xe-LP". AnandTech. Archived from the original on August 13, 2020. Retrieved September 29, 2020.
- ^ Pirzada, Usman (October 7, 2020). "Intel Sapphire Rapids: MCM Design, 56 Golden Cove Cores, 64GB HBM2 On-Board Memory, Massive IPC Improvement and 400 Watt TDP". Wccftech. Retrieved April 6, 2021.
- ^ Evenden, Ian (July 7, 2021). "Intel Retires Lakefield and 10th Gen Low Power CPUs". Tom's Hardware. Retrieved April 7, 2023.
- ^ "Intel launches 10th gen core processor developed in Israel". Globes (in Hebrew). May 28, 2019. Retrieved October 6, 2019.
- ^ Solomon, Shoshanna (May 28, 2019). "Intel launches new processors that bring AI to the PC, sired by Haifa team". The Times of Israel. Retrieved October 6, 2019.
- ^ "5-Level Paging and 5-Level EPT". Intel Corporation. May 2017.
- ^ a b c Cutress, Ian (December 12, 2018). "Intel's Architecture Day 2018: The Future of Core, Intel GPUs, 10nm, and Hybrid x86". AnandTech. Archived from the original on December 12, 2018. Retrieved January 14, 2019.
- ^ "Intel Launches First 10th Gen Intel Core Processors: Redefining the Next Era of Laptop Experiences". Intel Newsroom. Retrieved August 1, 2019.
- ^ "Dell taking orders for XPS 13 2-in-1 featuring Intel's 10nm Ice Lake". TechSpot. August 11, 2019. Retrieved August 12, 2019.
- ^ "Sunny Cove - Microarchitectures - Intel - WikiChip". en.wikichip.org. Retrieved February 8, 2021.
- ^ Schor, David (May 28, 2019). "Intel Sunny Cove Core To Deliver A Major Improvement In Single-Thread Performance, Bigger Improvements To Follow". WikiChip Fuse. Retrieved May 28, 2019.
- ^ Schor, David (May 28, 2019). "Intel Announces 10th Gen Core Processors Based On 10nm Ice Lake, Now Shipping". WikiChip Fuse. Retrieved May 28, 2019.
- ^ "Dynamic Tuning - Intel - WikiChip". en.wikichip.org. Retrieved May 28, 2019.
- ^ a b c d Cutress, Ian. "Examining Intel's Ice Lake Processors: Taking a Bite of the Sunny Cove Microarchitecture". AnandTech. Archived from the original on July 30, 2019. Retrieved August 1, 2019.
- ^ Yavarzadeh, Hosein; Taram, Mohammadkazem; Narayan, Shravan; Stefan, Deian; Tullsen, Dean (May 2023). Half&Half: Demystifying Intel's Directional Branch Predictors for Fast, Secure Partitioned Execution. 2023 IEEE Symposium on Security and Privacy (SP). IEEE. pp. 1220–1237. doi:10.1109/SP46215.2023.10179415. ISBN 978-1-6654-9336-9. S2CID 259255212.
- ^ "Intel® Deep Learning Boost". Intel AI. Retrieved August 1, 2019.
- ^ "Intel Processor Graphics Gen11 Architecture" (PDF). Intel.
- ^ "Developer and Optimization Guide for Intel® Processor Graphics Gen11".
- ^ "Intel Ice Lake 10nm CPU Benchmark Leak Shows More Cache, Higher Performance". HotHardware. HotHardware. October 23, 2018. Retrieved November 9, 2018.
{{cite news}}: CS1 maint: others (link) - ^ Cutress, Dr Ian. "The Ice Lake Benchmark Preview: Inside Intel's 10nm". www.anandtech.com. Archived from the original on August 1, 2019. Retrieved August 1, 2019.
- ^ "Release Media SDK 20.2.1 · Intel-Media-SDK/MediaSDK". GitHub.
- ^ "VP9 encode support from Kaby Lake+ · Issue #630 · intel/media-driver". GitHub.
- ^ "Feature request: Expose VP9 encode support on Kabylake+ with the iHD driver · Issue #771 · intel/media-driver". GitHub.
- ^ @IntelGraphics (August 31, 2019). "Our community suggested it and we are..." (Tweet) – via Twitter.
- ^ "Integer Scaling Support on Intel® Graphics".
- ^ "Ice Lake (client) - Microarchitectures - Intel - WikiChip". en.wikichip.org. Retrieved February 8, 2021.
- ^ Sag, Anshel (June 19, 2019). "Intel Charts A New Course With 10th Gen Core And Project Athena".
- ^ "Intel Takes Steps to Enable Thunderbolt 3 Everywhere, Releases Protocol".
- ^ "New Intel Processors Accelerate 5G Network Transformation". Archived from the original on April 6, 2021. Retrieved July 20, 2024.
- ^ "Intel Launches Xeon D Processor Built for the Network and Edge". December 31, 2022.
- ^ "Intel Announces Xeon E-2400 & Xeon D-1800/D-2800 CPUs". www.phoronix.com. Retrieved December 16, 2023.
- ^ Cutress, Ian (29 July 2021). "Intel Launches Xeon W-3300: Ice Lake for Workstations, up to 38 Cores". AnandTech. Archived from the original on 10 May 2023. Retrieved 17 May 2023.
All the processors will support 64 lanes of PCIe 4.0, 8-channel DDR4-3200 memory (up from 6-channel), and with 256 GB LRDIMMs up to 4 TB per socket (16 modules).
Ice Lake (microprocessor)
View on GrokipediaDevelopment History
Announcement and Design Goals
Intel first publicly detailed its Ice Lake microprocessor family during a keynote at Computex 2019 in Taipei, Taiwan, positioning it as the company's inaugural 10nm-based client processor platform for mobile computing.[9] The codename "Ice Lake" follows Intel's longstanding convention of drawing from North American geographic features, particularly lakes, to evoke themes of power and fluidity in processor design.[10] As a successor to the 14nm Whiskey Lake and Coffee Lake architectures, Ice Lake addressed longstanding delays in Intel's 10nm roadmap, which had originally projected high-volume manufacturing by the end of 2017 but faced repeated setbacks due to yield challenges and process refinements.[11] The shift to 10nm aimed to deliver transistor density improvements of approximately 2.7x over the 14nm node, enabling greater integration and power efficiency for thin-and-light laptops.[7] Key engineering targets included an average 18% uplift in instructions per clock (IPC) through the new Sunny Cove core microarchitecture, compared to the Skylake baseline, to enhance single-threaded performance in productivity and creative workloads. Ice Lake's design emphasized mobile efficiency, integrating Gen11 graphics capable of up to 2x the performance of prior-generation integrated solutions in gaming and content creation scenarios, such as Total War: Three Kingdoms.[7] Additional goals focused on AI acceleration via DL Boost, delivering up to 2.5x faster inference throughput compared to the previous generation in benchmarks like AIXPRT, and native support for Thunderbolt 3 to streamline high-speed connectivity in ultrabook form factors.[12] These features targeted the premium thin-laptop segment, aiming to sustain Intel's leadership amid rising competition in power-efficient mobile processors.[13]Release Timeline
Intel's Ice Lake mobile processors, part of the 10th Generation Core family, were officially launched on August 1, 2019, marking the company's first 10nm client products, with the Core i7-1065G7 serving as the inaugural SKU.[14] These processors were showcased at IFA Berlin in early September 2019, and the first laptops featuring them, including the Dell XPS 13 2-in-1, became available starting that month. By the 2019 holiday season, Ice Lake had been integrated into over 30 laptop designs from various OEMs, enabling widespread adoption in ultrathin and 2-in-1 devices.[13] For server variants, Intel began sampling Ice Lake-SP processors, the third-generation Xeon Scalable family, to select partners in June 2019.[15] The general availability of these processors was delayed from initial projections and occurred on April 6, 2021, with the 40-core Xeon Platinum 8380 as the flagship model offering 60 MB of cache and support for eight-channel DDR4 memory. The postponement was influenced by production challenges, including disruptions from the COVID-19 pandemic that affected supply chains and qualification timelines.[16] In July 2021, Intel completed the workstation rollout with the launch of the Xeon W-3300 series, providing up to 38 cores for professional workloads. The Ice Lake-D variant, targeted at edge computing and networking applications as the Xeon D-1700 series, was announced on February 24, 2022, featuring integrated 100GbE capabilities and up to 20 cores. Client mobile Ice Lake processors reached end-of-life status with final shipments concluding in April 2022, while server variants maintained active support with select models reaching end of life in 2025 and end-of-servicing updates concluding by mid-2025 for many.[17][18][19]Architecture and Features
CPU Microarchitecture
The Sunny Cove microarchitecture powers the Ice Lake family of processors and marks Intel's first major CPU core redesign since Skylake, enabling a transition to the 10 nm process node for increased transistor density and efficiency. This shrink from the previous 14 nm node allows for more complex structures within the same die area, supporting advanced features while maintaining compatibility with the x86-64 instruction set. Sunny Cove emphasizes deeper out-of-order execution, wider pipelines, and smarter resource allocation to boost single-threaded performance, with configurations scaling from dual-core mobile designs to multi-core server setups.[20] A core highlight is the pipeline enhancements, including a 5-wide decode stage that can process up to five instructions per cycle, paired with a 4-wide execution engine capable of dispatching instructions to a broader array of functional units, such as additional integer and floating-point ALUs and FP multipliers. These changes contribute to an average 18% increase in instructions per cycle (IPC) compared to Skylake cores, as measured across a range of workloads at iso-frequency. Branch prediction is improved via a larger branch target buffer (BTB) expanded to approximately 5,000 entries from Skylake's 4,000, reducing misprediction penalties and enhancing front-end efficiency. Caches are also scaled up, with each core featuring a private 512 KB L2 cache—double the size of Skylake's—and shared L3 cache organized in 12 MB slices per cluster of four cores, minimizing latency for data-intensive tasks.[21][22] Sunny Cove introduces full support for AVX-512 vector instructions across all cores, including the Vector Neural Network Instructions (VNNI) extension that accelerates bfloat16-based AI inference by fusing multiply-accumulate operations, potentially delivering up to 10x faster deep learning performance over prior generations without VNNI. This is complemented by Intel Deep Learning Boost (DL Boost), which optimizes matrix multiplications for neural networks. Power management is advanced with Intel Speed Shift 2.0, allowing finer-grained dynamic frequency and voltage adjustments based on workload demands, while core counts range from 2 to 4 in client mobile variants for balanced efficiency and up to 40 cores in Xeon Scalable processors for high-throughput server applications. Specific innovations include a two-level translation lookaside buffer (TLB) for improved virtual-to-physical address translation efficiency, reducing page walk overheads in memory-bound scenarios, alongside enhanced hardware prefetchers that better anticipate data access patterns to proactively load cache lines. Security is bolstered by Software Guard Extensions 2.0 (SGX 2.0), which adds dynamic enclave loading and key sharing capabilities for secure computation in untrusted environments. These elements collectively position Sunny Cove as a versatile core for diverse computing demands, from edge AI to cloud workloads.[20][21]Graphics and Media Processing
The integrated graphics processing unit (iGPU) in Ice Lake processors is based on Intel's Gen11 architecture, featuring up to 64 execution units (EUs) and clock speeds reaching 1.1 GHz in high-end configurations.[23] This design supports DirectX 12 with key features like variable rate shading (VRS) Tier 1, serving as a precursor to full hardware ray tracing capabilities introduced in subsequent Xe architectures.[24] The Gen11 iGPU delivers approximately twice the graphics performance compared to the prior Gen9.5 architecture, enabling smoother handling of gaming and visual workloads at 1080p resolutions while maintaining efficiency in mobile scenarios.[25][26] Ice Lake's media processing is powered by an enhanced Quick Sync Video engine, supporting 4K video decode and encode for formats including HEVC (H.265) at 10-bit color depth and VP9, facilitating efficient playback and transcoding for streaming applications.[27] A standout capability is the dedicated hardware support for AV1 decode, marking Intel's first implementation of this next-generation codec, which offers superior compression efficiency for 4K and 8K content with up to 30% better bandwidth savings over HEVC.[28][29] The integrated image signal processor (ISP) handles up to 4K60 camera inputs, supporting multi-stream processing for advanced imaging tasks like computational photography in laptops.[30] Specific graphics enhancements include VESA Adaptive Sync for tear-free displays and variable rate shading to optimize rendering by applying lower shading rates to less critical screen areas, improving frame rates in games by up to 30% in supported titles.[14][24] The iGPU integrates with the Sunny Cove CPU cores to enable AI-accelerated features, such as enhanced video upscaling via shared DL Boost capabilities for improved quality in low-resolution content.[31] Configuration variants adapt to platform needs: mobile client processors typically feature 32 to 64 EUs (e.g., 32 EUs in UHD Graphics or 64 EUs in Iris Plus Graphics), optimized for 15W TDP envelopes with dynamic frequency scaling for battery life.[32] Server processors like Xeon Scalable do not include an iGPU, focusing on compute workloads.Memory and I/O Capabilities
Ice Lake processors incorporate an integrated memory controller that supports dual-channel DDR4-3200 or LPDDR4x-3733 memory configurations in client and mobile variants, delivering peak bandwidths of up to approximately 60 GB/s in dual-channel setups. This design enables maximum capacities of up to 128 GB in mobile implementations, while server variants, such as the Xeon Scalable family, feature an eight-channel DDR4-3200 controller with ECC support for enhanced reliability in enterprise environments, supporting up to 6 TB when paired with Intel Optane persistent memory.[2] The memory controller's architecture optimizes latency and throughput for diverse workloads, including those in AI and data analytics. The cache hierarchy in Ice Lake builds on the Sunny Cove core design, with each core featuring a private 512 KB L2 cache in client variants to reduce access times for frequently used data. In server configurations, the L2 cache expands to 1.25 MB per core, paired with a scalable shared L3 cache reaching up to 38.5 MB in high-core-count models like the 28-core Xeon Platinum. These server implementations employ a second-generation mesh interconnect to link cores and cache slices, enabling efficient scalability across multi-socket systems while maintaining low latency for inter-core communication.[33][34][35] Ice Lake's I/O subsystem in mobile processors supports up to 16 lanes of PCIe 3.0. Server variants, such as Xeon Scalable, support up to 64 lanes of PCIe 4.0 per socket, facilitating extensive expansion for data center connectivity. Additional interfaces include USB 3.2 Gen 2 for peripheral integration and up to four Thunderbolt 3 ports in high-end mobile models, enabling versatile docking and display options at 40 Gbps per port. Compatibility with Intel Optane memory further enhances I/O performance by accelerating access to persistent storage in bandwidth-intensive applications.[36][2][37] Key innovations in memory and I/O include the integration of Intel Deep Learning Boost (DL Boost), which leverages the memory controller and cache hierarchy to accelerate memory-bound AI inferences and training tasks through optimized vector processing. In server processors, Intel Speed Select Technology provides granular control over core frequencies and power allocation, allowing dynamic tuning for specific workloads to balance performance and efficiency across I/O-bound scenarios.[2][38]Manufacturing and Packaging
Process Technology
Ice Lake processors are fabricated on Intel's 10 nm+ process, an enhanced iteration of the company's initial 10 nm node that incorporates optimizations for drive current and transistor density. This process achieves a transistor density of approximately 67 million transistors per square millimeter for Ice Lake dies, representing about a 1.8-fold increase over the 14 nm node. While Intel's 10 nm family offers enhancements in FinFET design and interconnect scaling, the 10 nm+ variant used in Ice Lake has lower density than TSMC's 7 nm process (around 96.5 MTr/mm²).[39][40] Originally slated for production in 2016, the 10 nm process encountered substantial delays due to complexities in multi-patterning lithography and yield optimization, with high-volume manufacturing for Ice Lake beginning in 2019 using deep ultraviolet (DUV) lithography rather than extreme ultraviolet (EUV), which Intel reserved for subsequent nodes. These delays stemmed from the need to refine aggressive scaling targets, including narrower fin pitches and improved cobalt interconnects to mitigate resistance.[41][42] The 10 nm+ node delivers up to 25% higher performance or up to 60% lower power consumption compared to the initial 14 nm process at equivalent speeds, with 10 nm+ optimizations enabling higher clocks; it facilitates more efficient operation in power-constrained environments and supporting thermal design powers as low as 9 W for mobile variants.[39][43] Early production faced yield challenges, prompting a strategy of prioritizing smaller mobile dies for Ice Lake client processors before scaling to larger server configurations in Xeon variants. Subsequent Xeon implementations, such as Ice Lake-SP, benefited from process refinements, paving the way for the introduction of 10 nm SuperFin enhancements in later products with improved epitaxial growth, source/drain architecture, and metal-insulator-metal capacitors for better capacitance density.[44][45] Intel's work on the 10 nm node included explorations of gate-all-around transistor precursors through advanced FinFET optimizations and early research into backside power delivery to reduce voltage drop, though full implementation occurred in future nodes. These efforts contributed to scaling benefits for the Sunny Cove cores in Ice Lake.[46]Die Layout and Package Variants
The Ice Lake client processors for mobile devices feature a monolithic die design, integrating up to four Sunny Cove cores alongside the Gen11 integrated graphics processing unit (iGPU) on a single silicon substrate. This layout enables a compact form factor suitable for laptops and ultrabooks, with the die measuring approximately 122 mm² for a typical quad-core configuration. The cores are clustered in a central arrangement, with the iGPU positioned adjacent to facilitate efficient data sharing via the on-die interconnect.[36] In contrast, the Xeon Scalable (Ice Lake-SP) processors employ a multi-tile architecture on a monolithic die, dividing functionality into modular blocks such as compute tiles containing 4 to 8 Sunny Cove cores each, dedicated I/O tiles for memory controllers and PCIe interfaces, and mesh interconnect tiles for inter-tile communication. This design supports up to 40 cores across configurations with up to 8 tiles, resulting in a larger die size of about 628 mm² for the 40-core variant, optimizing yield and scalability on the 10 nm process. The tiled approach serves as a precursor to Intel's Embedded Multi-Die Interconnect Bridge (EMIB) technology, laying groundwork for future multi-die interconnects by modularizing components while maintaining a single-die package.[47][35] Package variants differ by form factor and application. Mobile client processors use a Ball Grid Array (BGA) package, specifically the FCBGA1509 for soldering directly onto the motherboard, supporting thermal design powers (TDP) from 9 W to 28 W. Server-oriented Xeon Scalable processors adopt the LGA4189 socket (FCLGA4189 package) for easy installation and upgrades, accommodating TDPs up to 205 W with integrated heat spreaders. Workstation variants, such as the Xeon W-3300 series, utilize the FCLGA4189 package (LGA4189 socket), supporting thermal design powers up to 270 W with integrated heat spreaders for enhanced heat dissipation in dense computing environments.[48][49] The Ice Lake-D variant for edge servers features a monolithic SoC design integrating up to 20 cores, Ethernet, and other peripherals, packaged in a high-density BGA for rugged, space-constrained deployments with optimized cooling via BGA mounting for industrial and embedded applications. This configuration enables up to 20 cores while integrating Ethernet and other peripherals.[50]Processor Families
Client Mobile Processors
The Ice Lake client mobile processors encompass Intel's 10th-generation Core i3, i5, and i7 families designed for low-power laptops, featuring the Sunny Cove CPU microarchitecture and Gen11 integrated graphics. These processors target ultrabooks and convertible devices, emphasizing balanced performance, efficiency, and integrated features for consumer and business mobility. Key SKUs include the Core i7-10xxxG7 series, such as the i7-1065G7 with 4 cores and 8 threads, a base frequency of 1.3 GHz, turbo up to 3.9 GHz, and configurable TDP from 12.5 W to 28 W for U-series variants optimized at 15 W base power. The Core i5-10xxxG7 series, like the i5-1035G7, offers similar 4-core/8-thread configurations with base frequencies around 1.2 GHz and turbo up to 3.7 GHz, also at 15 W TDP, while i5-10xxxG4 and G1 variants provide Iris Plus or UHD Graphics with 48 execution units (EUs) for lighter workloads. Core i3-10xxxG1 models, such as the i3-1005G1, feature 2 cores and 4 threads, base 1.2 GHz with turbo to 3.4 GHz, and UHD Graphics at 48 EUs, suiting entry-level mobile use. Y-series configurations, including the i7-1060G7 and i5-1030G7, operate at 9 W TDP for fanless designs, with reduced base frequencies (e.g., 1.0 GHz for i7) but retaining up to 64 EUs in Iris Plus G7 graphics for thin-and-light form factors.[1][51][52][30] These processors integrate Iris Plus Graphics, with G7 variants delivering up to 64 EUs for enhanced visual performance, achieving approximately 1.1 TFLOPS in FP32 operations at peak clocks around 1.1 GHz, enabling light gaming and content creation without discrete GPUs. Business-oriented models support Intel vPro technology for remote management and security, while native integration with Wi-Fi 6 (802.11ax) provides faster, more reliable wireless connectivity in dense environments. The lower TDP envelopes and architectural efficiencies contribute to improved battery life, with real-world tests in ultrabooks showing up to 16-20 hours of mixed-use runtime compared to prior generations.[53][54] Ice Lake mobile SKUs powered premium devices like the Dell XPS 13 and HP Spectre x360 convertibles, targeting professionals and consumers seeking portability with 4K display support and Thunderbolt 3 connectivity. Although Intel discontinued production of these processors in 2021 with final shipments in 2022, software and graphics driver support extended through 2024, allowing legacy use in 2025 IoT applications for stable, low-power embedded systems.[55][56][57]| Series | Example SKU | Cores/Threads | Base/Turbo Freq. (GHz) | TDP (W) | Graphics (EUs) |
|---|---|---|---|---|---|
| i7 G7 (U/Y) | i7-1065G7 | 4/8 | 1.3/3.9 | 15 (12.5-28) | Iris Plus (64) |
| i5 G7/G4/G1 (U) | i5-1035G7 | 4/8 | 1.2/3.7 | 15 | Iris Plus (64/48) |
| i3 G1 (U) | i3-1005G1 | 2/4 | 1.2/3.4 | 15 | UHD (48) |
Xeon Scalable Processors
The Ice Lake-SP, known as the 3rd Generation Intel Xeon Scalable processors, represents Intel's first 10 nm server CPU family, offering up to 40 cores and 80 threads per socket with thermal design power (TDP) ratings ranging from 205 W to 270 W.[18] These processors utilize the LGA 4189 socket and incorporate the Sunny Cove microarchitecture, enabling enhanced performance for data center workloads such as cloud computing, high-performance computing (HPC), and artificial intelligence (AI).[58] The family is divided into series including Platinum (83xx models, e.g., the 8380 with 40 cores at 2.3 GHz base and up to 3.4 GHz turbo), Gold (63xx and 62xx models, e.g., the 6330 with 28 cores at 2.0 GHz base and up to 3.1 GHz turbo), and Silver (43xx models, e.g., the 4314 with 16 cores at 2.4 GHz base and up to 3.4 GHz turbo), targeting varying levels of performance and cost efficiency.[59] Configurations span 8 to 40 cores per processor, supporting 8-channel DDR4-3200 ECC memory with up to 6 TB capacity per socket when combining DRAM and Optane persistent memory.[58] The on-die mesh interconnect facilitates multi-socket scalability, with up to 3 Ultra Path Interconnect (UPI) links at 11.2 GT/s for configurations supporting up to 8 sockets, though higher socket counts limit maximum cores per processor to prioritize density in large-scale systems.[60] Key optimizations include built-in AI accelerators via Intel Deep Learning Boost (DL Boost), delivering up to 10x performance gains in TensorFlow deep learning workloads using Intel-optimized software compared to the default distribution on these processors.[61] Intel Speed Select Technology enables dynamic frequency scaling for cloud bursting, allowing processors to temporarily boost performance during peak demands while maintaining efficiency in variable workloads.[62] Additionally, support for bfloat16 precision in vector neural network instructions (VNNI) enhances machine learning training efficiency by reducing memory bandwidth needs without sacrificing accuracy.[63] Precursors to Compute Express Link (CXL) 1.1 are enabled through 64 lanes of PCIe 4.0, facilitating coherent memory expansion in future data center designs.[64] Released to general availability in April 2021, Ice Lake-SP processors have been adopted in high-impact HPC environments, such as the Aurora exascale supercomputer at Argonne National Laboratory, where Xeon Gold 5320 models are utilized in storage subsystems.[65] As of 2025, these processors continue to receive support in legacy data centers, providing reliable scalability for single- and dual-socket servers focused on enterprise and AI inference tasks.[66]Xeon D and Workstation Processors
The Intel Ice Lake-D processors power the Xeon D-1700 and D-2700 series, targeted at embedded and edge computing applications such as telecommunications and 5G infrastructure. Released in February 2022, these system-on-chip (SoC) designs offer 4 to 20 cores based on the Sunny Cove microarchitecture, with thermal design power (TDP) ratings ranging from 65 W to 129 W depending on the model. They utilize a ball grid array (BGA) package optimized for dense, space-constrained deployments, supporting up to four channels of DDR4 memory at speeds up to 3200 MT/s and up to 64 lanes of PCIe 4.0 for enhanced I/O connectivity. For example, the Xeon D-2796NT configuration provides 20 cores and 40 threads, with a base frequency of 2.00 GHz and a maximum turbo frequency of 3.10 GHz, making it suitable for network edge processing in telco environments. As the successor to the Skylake-based Xeon D-2100 series from 2018, Ice Lake-D introduces improved per-core performance and integrated Ethernet options up to 100 GbE, while Intel has announced Granite Rapids-D as its follow-on for edge SoCs in 2025. The Xeon W-3300 series, also based on Ice Lake, caters to single-socket workstation needs in professional workflows like computer-aided design (CAD), 3D rendering, and media production. Launched in July 2021, this lineup scales up to 38 cores and 76 threads in models such as the W-3375, with a TDP of up to 270 W and the LGA4189 socket for compatibility with high-end workstation motherboards. These processors support error-correcting code (ECC) memory configurations up to 4 TB of DDR4 across eight channels, along with 64 lanes of PCIe 4.0 to enable multi-GPU setups for accelerated rendering and AI tasks. ISV certifications ensure optimized performance in professional applications from vendors like Adobe and Autodesk, providing reliability for engineering and content creation workloads. Unlike broader scalable variants, the W-3300 emphasizes single-socket efficiency for professional desktops and towers.References
- https://en.wikichip.org/wiki/intel/microarchitectures/ice_lake_(client)
- https://en.wikichip.org/wiki/intel/speed_select_technology
- https://en.wikichip.org/wiki/10_nm_lithography_process
