Intel’s Tiger Lake Is Spoiling for a Rematch Against AMD’s Zen 3


Intel unveiled its Tiger Lake platform yesterday, and the company is clearly spoiling for a fight against AMD. An actual tiger roaring somewhere offstage would not have seemed unusual. The company’s overall positioning suggests it’s got something impressive on its hands.

Intel has taken plenty of flak for marching in place as far as feature integrations are concerned, but Tiger Lake adds PCIe 4.0 support, fully integrated Thunderbolt 4 support, Intel’s new Xe graphics engine (debuting for the first time), additional display pipes, a 6th-generation new image processing unit (Icelake was v4), version 2.0 of its “Gaussian Network Accelerator” (Ice Lake had v1), and support for various low-level deep learning features like AVX-512. Toss on a whopping frequency spike courtesy of Intel’s new 10nm “Superfin” technology, and you’ve got Tiger Lake: leaner, meaner, and ready for a showdown with AMD’s Ryzen 3000 family.

Intel was eager to show benchmarks and discuss performance across the various aspects of Tiger Lake. Demos against AMD hardware were plentiful and rarely positive, though you’d scarcely expect anything else from a vendor event. Even so, the strength of Intel’s presentation, and the data it chose to emphasize, point to the underlying strength of the company’s product.

For Intel, Tiger Lake is all about performance.

Intel likes to talk about “real-world” benchmarking these days, and it’s switched its messaging over to focus on what it considers to be “real-world” applications. Note, however, that each of these tests focuses on a single application of a single feature (apart from Gigapixel, which really only has one feature). Also, be advised that Topaz products incorporate Intel-favoring libraries and, while this is of secondary importance when workloads run on the GPU, support for AMD GPUs is very early. The good news is, this support is coming. I’ve been alpha testing builds of Topaz, including one with support for the AMD Radeon 5700. It works quite well.

Chart exports being 1.3x faster in Excel may be true, but it doesn’t tell me anything about which CPU is faster at calculating a series of equations. This is one of the problems with “real-world” benchmarking — in an application with so many different capabilities, testing a range of them is the only way to give any kind of fair comparison between any two products.

Intel Xe

Xe arrives with up to 96 EUs, 1.35GHz of clock, and can output either 48 texels or 24 pixels per clock, which is quite good for an integrated solution. There’s a larger L3 cache to keep memory pressure off the main bus and a new, higher-efficiency software scoreboard. Nvidia and AMD both use software scoreboards, up until now Intel had relied on a hardware algorithm.

Intel’s Xe is a claiming some remarkable speed-ups:

The big question with integrated graphics is always bandwidth: How effectively will Xe be able to use its limited pool of resources? Based on the numbers, pretty good:

There are a few places where Xe seems poised to outperform AMD. We’ll see if Intel’s GPU driver team is ready for big things this generation, though AMD has had its own struggles in that department for a long time now.

Intel even argues that its integrated graphics can trade blows with discrete cards, though the card in question has the approximate performance of a GeForce GTX 960M. Then again, being able to compete against a budget discrete GPU from 4-5 years ago is actually pretty impressive — I remember the days when Intel’s integrated graphics could barely run games that were 10 years old.

The 11th CPU family, in all its glory, is below:

Intel has decided to move away from TDP as a singular metric of thermal dissipation and is now giving data on TDP ranges. This is probably a more fair way to capture the likely power draw of the system at any given moment, and it’s practically old hat by now for AMD or Intel to float some new method of attempting to communicate power efficiency to customers. TDP was an imperfect metric and this one likely won’t be perfect either, but it captures the range more effectively. TDP was never intended to be a metric for power consumption in the first place; it refers to Thermal Design Power — the amount of thermal energy the cooling solution needs to dissipate in order to maintain full performance.

Underneath the hoopla, Tiger Lake looks like a very solid part. This may catch some AMD fanboys off guard, but it shouldn’t. Historically, mobile has been Intel’s strongest market and the company got serious about reducing its overall power consumption footprint years before AMD did. More OEMs have optimized more heavily for Intel and are more experienced with how it controls power. OEMs are more willing to invest in creating top-notch Intel solutions.

We’ve already seen this situation begin to change in the past year with the debut of the Surface Laptop 3 as a major AMD win and other launches of the new 7nm mobile processors this year. AMD is ramping its own efforts and we expect Zen 3 imminently. In fact, mobile is shaping up as the battleground to watch between the companies in the imminent future. Expect Intel to try and build on Tiger Lake’s support for AVX-512 and other deep learning capabilities at every turn, while AMD is likely to eschew these comparisons in favor of other metrics.

Now Read:

Leave a Reply

Your email address will not be published. Required fields are marked *