meow

Hands-on with Intel’s i7-1185G7 Tiger Lake prototype laptop

Tiger Lake is a much better challenger for AMD—but it’s still fighting uphill.

Jim Salter – Sep 17, 2020 4:45 pm | 149

This MSI-built reference system is powered with a Tiger Lake i7-1185G7, the highest-end CPU in Intel's upcoming lineup. Credit: Jim Salter

We’ve been very interested in Intel’s upcoming laptop CPUs, codenamed Tiger Lake, since the company’s Architecture Day event in August. Tiger Lake’s official launch event earlier this month didn’t offer much red meat for anyone already up-to-date on the news—but today, we finally have our own hands-on test results to share.

Much as Intel did during Tiger Lake’s launch event, we’re going to focus heavily on Intel versus AMD in our own tests and analysis. In our opinion, the current generation-on-generation within Intel’s own lineup is fairly boring (yes, it’s faster than its old parts). Instead, the real question is whether Intel finally has an answer to AMD’s Renoir architecture—and the answer isn’t as simple as “yes” or “no.”

Our reference system has the top-of-the-line Core i7-1185G7 CPU, tuned for a 28W default TDP—although that, too, gets complicated. For now, we’ll just note that it’s the fastest Tiger Lake CPU to be announced. However, assuming one i7-1185G7 system is much like the next would be a mistake.

Our Tiger Lake reference system

Before we dive into the performance analysis, let’s get to know the prototype laptop supplied to us by Intel. The system was built by MSI and, in many ways, it likely resembles what will become MSI’s new Prestige 14 Evo retail system. That said, please don’t draw any definite conclusions about a final retail system—particularly the Prestige 14 Evo—from what you see here.

Intel warned us that this was a reference system, not retail-ready, and likely not tuned the way retail versions will be. The platform also almost certainly doesn’t have the retail port layout: this system has two USB-C ports, both on the left hand side, and absolutely nothing else.

The system also exhibited noticeable capacitor whine. If you have a good ear for electronics, you can actually hear the device “thinking” from the change in the faint, high-pitched noise as the CPU shifts in and out of turbo frequencies, which it does rapidly.

We also did not test and will not talk about the system’s battery life in this review. Again, this isn’t really a laptop to be reviewed—it’s just a platform that enables us to put the i7-1185G7 to the test. We did, however, check the system power consumption at the wall using a Kill-A-Watt meter. Desktop idle varies from 7.5W-8.2W, and peak consumption (as measured during the first several seconds of a Cinebench R20 run) is about 66W-68W.

Performance

At first blush, you’d think the quad-core, octa-thread Core i7-1185G7 in this reference system is a mostly even match for the octa-core, octa-thread Ryzen 7 4700U in our Acer Swift SF314-42. As tested the way the system ships, it generally runs a little slower on multithreaded CPU tests, and it’s noticeably faster on single-threaded CPU tests.

	Core i7-1185G7	Ryzen 7 4700U
Cores/Threads	4C/8T	8C/8T
configurable TDP	12W-28W	10W-25W
TDP as tested	28W	15W
Base clock	1.2GHz	2.0GHz
Boost/Turbo clock	up to 4.8GHz	up to 4.1GHz

This already isn’t really a match to brag about—the Swift 3 wrapped around our Ryzen 7 4700U is a budget laptop that doesn’t have the best thermals, and the 4700U itself is roughly in the middle of AMD’s Renoir lineup. The 4700U is only second from the top in AMD’s U series, and there’s an entire H series for high-powered systems above that.

Of course, Intel has been getting sand kicked in its face for a while now, and seeing something like performance parity is refreshing, even if we have to compare the company’s top-of-the-line CPU in a high-powered reference system versus a middling-high AMD CPU in a budget laptop. The comparison still has problems we need to iron out, though.

CPU power consumption / thermal budget

Tiger Lake, unlike previous generations of Intel mobile CPUs, only has a few SKUs. This looks like a blessing for consumers, but it’s likely more of a curse. Laptop OEMs have an incredibly wide configurable range for each SKU’s TDP (Thermal Design Power).

This means that one i7-1185G7 system is likely to perform very little like another—so now your less-technical colleagues not only won’t pay attention to the difference between one i7 and another, they won’t notice whether the exact same i7 SKU is configured for TDP of 12W, 28W, or anywhere in between.

Any i7-1185G7 system can briefly reach maximum turbo frequency and hit a power-consumption max of upwards of 50W, but the system doesn’t stay there for more than a few seconds, before dropping clock enough to fall back to its configured TDP.

The length of time which the system is allowed to stay at the highest PL2 consumption is called “tau.” Tau, along with the TDP itself, is configurable by the OEM. We seriously doubt tau will be disclosed on the box or in the advertising copy for many systems—and it can potentially have an even bigger impact than the TDP. A laptop allowed to run at PL2 (and draw 50+W) for minutes at a time will perform considerably higher (and exhibit a hellaciously higher power draw) than a competing system with the same TDP but a tau of only a few seconds.

At its out-of-box defaults, Intel’s reference system—set on the middle of three notches in Windows 10’s “performance” slider, found when clicking on the battery icon in the toolbar—is configured for a 28W TDP. Sliding the performance slider to the left sets TDP to 15W. The tau on the reference system isn’t specified, but it appears to be roughly 25-30 seconds regardless of TDP, judging from our observations and Anandtech’s.

If you slide the performance widget all the way to the right, the TDP remains 28W. But tau, on the other hand, becomes governed by Adaptix, an algorithm that extends turbo time by dynamically regulating clock frequencies down when the CPU itself isn’t the bottleneck. In general, users who enable Adaptix can expect a modest multithreaded performance gain with a somewhat larger corresponding power draw.

We did not directly test Adaptix, due to very limited time available with the reference system.

Multithreaded performance

In its default configuration, with a 28W TDP and roughly 25-second tau, the i7-1185G7 reference system runs pretty much neck-and-neck with a Ryzen 7 4700U-equipped Swift 3. We see the same basic relationship among all systems on both Cinebench R20 and Passmark. Geekbench 5 flattens the differences between all the processors in general, but it puts the 1185G7 at the head of the pack.

However, we suspect many retail 1185G7 systems won’t be configured for a 28W TDP—they’ll likely be configured at 15W TDP, just as both the Ryzen 7 4700U in the Swift 3 and the Ice Lake i7-1065G7 in the Dell XPS 13 are. When we drop the reference-system TDP limit to 15W by moving the slider hard left, we see a significant drop in performance. It actually falls slightly below the Dell’s Ice Lake 1065G7 on Cinebench R20, let alone the Acer’s Ryzen 7 4700U.

Since the i7-1185G7 is the highest-performance SKU announced for Tiger Lake, we felt that comparing it to the Ryzen 9 4900HS in our Asus ROG 14 gaming laptop would only be fair. The 4900HS in the ROG 14 is running at 35W TDP, and it utterly dominates the Tiger Lake i7 in both Cinebench R20 and Passmark testing.

Single-threaded performance

True to form for Intel, single-threaded performance on the i7-1185G7 is beastly—it dominates the pack, including the top-of-the-line Ryzen 9 4900HS, in every single-threaded test. Better yet, single-threaded performance doesn’t drop too much when the TDP is limited to 15W, the TDP we expect to see on ultraportables.

As always, we want to caution people not to make too much of these single-threaded results—the single-threaded results are much closer to flat than the multithreaded results are. Just as importantly, single-threaded truly means single-threaded, and you won’t see results like these if your system is doing significant work on any other threads. Individual tasks which bind on a single thread completing are common, but entire whole-system workloads with no background threads are considerably less so.

Gaming performance

We also tested the i7-1185G7 reference system using 3DMark’s Time Spy and Night Raid benchmarks. Both Time Spy and the less-demanding Night Raid tend to bottleneck much more heavily on GPU than CPU, which makes them a good way to test Tiger Lake’s new Xe integrated graphics.

The short version: Xe delivers on its promises and then some. Whether the i7-1185G7 CPU as a whole is running at 28W TDP or 15W TDP, it beats all the other integrated graphics systems on both Time Spy and Night Raid by a healthy margin.

Xe, in turn, gets its butt handed to it by the ROG 14’s discrete RTX 2060 Mobile GPU. That’s hardly an indictment, though—the RTX2060 in the ROG 14 has its own 65W TDP (downconfigured from the 115W TDP maximum for the RTX 2060 reference design) that’s entirely separate from the CPU’s own 35W TDP. Meanwhile, the Xe is making do with its share of the 28W total CPU + GPU TDP allotted to the reference system’s i7-1185G7.

We will note that the integrated graphics in the Ryzen 7 4700U aren’t AMD’s absolute top of the line—the Ryzen 7 4800U has eight execution units to the 4700U’s seven. With that said, testing around the Web shows the roughly 10 to 15 percent performance boost you’d expect out of the extra EU. That still doesn’t bring top-of-the-line AMD laptop iGPUs up to par with the i7-1185G7’s Xe as tested, even when TDP is downconfigured to 15W.

“Evo Platform” startup times

The Tiger Lake prototype laptop Intel provided us for testing, with i7-1185G7 CPU, cold boots to the Windows 10 desktop in a startling 7.2 seconds. In our experience, few laptops even get past POST in that time frame.

Intel made a lot of promises about the newest iteration of its original Project Athena certification program, which is now known as Evo. In particular, Intel promised extremely fast boot and wake-up times for Evo-certified laptops. Although this laptop can’t be Evo certified—it’s a reference model prototype, not a retail product—we might be seeing some of what Intel’s been talking about.

The laptop shipped to us in sleep mode, not properly shut down—and it startled us when we pushed the power button and it immediately snapped to a ready desktop. Intel promises less than one-second wakeup from sleep on Evo laptops; on this reference system, the wakeup time is better measured in milliseconds—and not many of them.

When we powered the system down and restarted it, we were almost as startled by that process. The entire cold-boot sequence for the reference system is seven seconds, from completely powered-down to sitting at the desktop. Yes, that includes POST (Power On Self Test) time. The highest-end Windows laptop we had on hand to compare it to is Microsoft’s top-of-the-line Surface Book 3, with an Ice Lake i7-1065G7 and NVMe SSD. It can’t even POST in 7.2 seconds, let alone get to the desktop.

We’re hesitant to proclaim this as “the new normal” until we see it on retail systems—there’s no way for us to rule out shortcuts taken in a reference system not intended for end users. But it certainly got our attention.

Time, RUGs, and relevance

We regretfully did not spend much time on Intel’s provided RUGs (Reference Usage Guidelines), which rely on real-world applications such as Adobe Photoshop and Premiere, Topaz Gigapixel AI, and Nero AI Photo Tagger. While we do agree with Intel that real-life workloads are critically important, it’s difficult to separate Intel’s own search for favorable comparisons from the specific features the company chooses to test inside those real world applications.

In particular, any applications running AI inference workloads using OpenVINO will benefit enormously from Intel’s AVX-512 instructions (the company brands this as Deep Learning Boost). This is a real advantage for Intel, in those applications, and we expect this advantage will increase over time as more applications begin integrating inference workflows. You can already see such workflows in applications like Photoshop for some filters and “magic select” tools.

You’ll also find AI inference workloads in tasks such as image recognition or categorization/tagging, natural language processing, voice recognition, and more. For those workloads, you can likely expect a 200 to 500 percent boost in performance on Tiger Lake CPUs as compared to otherwise faster competition without AVX-512 support. This is much like the boost we saw when comparing last year’s i9-10980XE to its otherwise more-powerful Threadripper 3970x competition.

With that said, we don’t think—at least for now—that most people’s workflows depend so heavily on inference workloads that they should be the deciding factor in their CPU selection. And we had very limited time with the reference system, so we chose to limit our testing to more traditional benchmarks, for which there are considerably more reference points and much less question of bias from the CPU vendor.

Conclusions

In our opinion, there’s no one “right” way to feel about Tiger Lake. You can look at Tiger Lake as an exciting improvement from Ice Lake, Comet Lake, and prior architectures, which finally makes it a somewhat reasonable competitor for AMD’s Renoir. Or—just as accurately—you can scoff at the need to nearly double Renoir’s power consumption before the highest-spec Tiger Lake CPU can compete roughly evenly with a middling Renoir on most tests.

If your workload is unusually AI-dependent, Tiger Lake is likely a slam dunk. Its AVX-512 instruction set gives it an enormous advantage over otherwise more powerful and efficient AMD systems. But we need to stress the word unusually: inference workloads are more common and important every year, but they shouldn’t be most people’s deciding factor yet.

Where Tiger Lake shines brightest is in the narrow realm of laptop gaming without a discrete GPU. Although the CPU portion of Tiger Lake has difficulty holding its own against AMD’s Renoir, its new Xe graphics look to be the new gold standard for iGPUs. Xe isn’t ready to compete with high-end discrete GPUs in the Nvidia RTX series, but we suspect it’s rung the death knell for Nvidia’s lower-performance, more power-efficient MX line.

The final question for any new Intel CPU line in 2020, of course, is what availability will be like. But since multiple vendors such as MSI and Acer have already announced entire product lines based on Tiger Lake, we don’t expect to see a repeat of last year’s Ice Lake unobtainium fiasco.

Listing image: Jim Salter

Jim Salter Former Technology Reporter

Jim is an author, podcaster, mercenary sysadmin, coder, and father of three—not necessarily in that order.

149 Comments