Intel Reveals Clock Speeds, GPU Specs for 10nm Ice Lake Mobile SoCs

Last week, Intel announced that it had begun shipping 10nm Ice Lake CPUs to its OEM customers to support a holidays 2019 launch. Today, the company is sharing more details about Ice Lake and the new chips it will launch on 10nm later this year. The company has been revealing details about Ice Lake and its architecture, Sunny Cove, since Architecture Day last winter. As “Holidays 2019” draws closer, we’re starting to find out more information.

Ice Lake packs a grab bag of improvements intended to appeal to multiple different markets. Boosted AI inferencing performance is potentially appealing to those working in the field, though I’m not sure if any practical applications actually use AI or AVX-512 on the desktop yet. Faster Wi-Fi via 802.11ax, aka Wi-Fi 6, should boost download speeds. Thunderbolt 3 is not integrated on-die with Intel Ice Lake, though actually offering that connectivity to customers will still require a degree of external hardware and is therefore somewhat dependent on OEMs to make available. Intel has stated, however, that the costs and amount of external components will be smaller than usual.

Mobile GPU performance is said to be significantly stronger, courtesy of a wider GPU core and more efficient execution. Ice Lake will support either dual-channel DDR4-3200 or LPDDR4X-3733 in four 32-bit channels. Overall power consumption is said to be similar between both standards, though the LPDDR4X systems will top out at 32GB, while the DDR4-3200 rigs will support up to 64GB. Our slideshow on the Sunny Cove architecture and the various improvements baked into the core is presented below:

Compared with Skylake, Sunny Cove will be capable of issuing two loads and two stores per cycle, rather than the 2L,1S of Skylake. There are four AGUs now, up from three, 10 ports (up from 8), a second shuffle unit, and a second LEA (Load Effective Address) unit.

The L1 data cache on Sunny Cove is 50 percent larger than Skylake and previous Intel chips, at 48KB compared with 32KB. Larger L2 caches, a larger uop cache, and expanded L2 TLB cache are other features planned for the architecture.

L1 store bandwidth has been doubled, with a wider array of capabilities for the execution ports. The design also now features 4 LEA (Load Effective Address) blocks.

Finally, Intel claims to have improved branch prediction efficiency, with additional latency improvements across the chip and a new integer divider unit.

One new bit of data Intel is revealing today is the actual SKUs and products. Here’s the lineup of 10th Generation mobile parts, with data on their wattage envelopes, clocks, and GPU configurations.

Let’s start with the 28W CPU. The best comparison to that is the Core i7-8569U, a 28W 4C/8T CPU with a 2.8GHz base clock, 4.7GHz boost clock, and 128MB of onboard EDRAM to improve the performance of its integrated Iris Plus Graphics 655 solution. We will immediately grant that we expect Gen 11 Intel graphics to be faster than the EDRAM-boosted solutions the company has used before. We also note that the 10nm Ice Lake Core i7-1068G7 supports faster RAM (DDR4-3200 / LPDDR4-3733 as compared to DDR4-2400). Intel CPUs typically haven’t benefited as much from faster RAM clocks as AMD CPUs, but the fact that Intel is increasing its RAM clock suport with 10nm may mean this has changed.

Both CPU and GPU maximum frequencies have declined. The Core i7-8569U has a maximum GPU frequency of 1.2GHz, while the Core i7-1068G7 supports a maximum frequency of 1.1GHz. Base frequency for the 10th Generation core CPU has dropped 18 percent. Since we know that Intel TDPs are based solely on boost clocks, the implication is that the company had to give up base clock to make its TDP figures.

The 15W minimum frequencies have also come down. The Core i7-1065G7 has a base frequency of 1.3GHz, while the comparative Core i7-8665U is a 1.9GHz CPU with a 4.8GHz base clock. That’s a 32 percent reduction in minimum frequency. I don’t want to sound negative on Ice Lake before we see the chips. It’s possible that some of the clock pulldown has been to make room for the GPU. But this is a point we’ll be watching closely — Intel may have had to strike a difficult balance between allocating TDP for CPU versus GPU operations.

Word from Intel is that Ice Lake will be built on a 10nm+ process, which puts that question to rest. The little-used Cannon Lake Core i3-8121 will evidently be the sole representative of Intel’s base 10nm process. 10nm+ will debut with Ice Lake. Whether Intel has a 10nm++ process in the works or will proceed directly to 7nm for future CPUs isn’t something the company has disclosed yet.

For more on this and some initial benchmarks, read PCMag’s Intel ‘Ice Lake’ Benchmarked: How 10nm CPUs Could Bring Major GPU Grunt to New Laptops.

Continue reading

NASA Created a Collection of Spooky Space Sounds for Halloween

NASA's latest data release turns signals from beyond Earth into spooky sounds that are sure to send a chill up your spine.

Astronomers Might Finally Know the Source of Fast Radio Bursts

A trio of new studies report on an FRB within our own galaxy. Because this one was so much closer than past signals, scientists were able to track it to a particular type of neutron star known as a magnetar.

Sony’s PlayStation 5 Debuts to Strong Reviews

Reviews have come in for Sony's PlayStation 5, and while they're a bit preliminary for the same reason as the Xbox Series X, they're broadly positive about Sony's latest gaming effort.

How to Build a Face Mask Detector With a Jetson Nano 2GB and AlwaysAI

Nvidia continues to make AI at the edge more affordable and easier to deploy. So instead of simply running through the benchmarks to review the new Jetson Nano 2GB, I decided to tackle the DIY project of building my own face mask detector.