This has been a long time coming, as we're finally getting around to testing Intel’s Ice Lake architecture. This is Intel’s first real attempt at a 10nm CPU, and in this review we’ll be comparing it against their 14nm offerings to see how it stacks up in terms of performance.

Last year we did see the 10nm 'Cannon Lake' but it was a single processor, the Core i3-8121U, a measly dual-core clocked at up to 3.2 GHz with no integrated graphics. It was used in a single Lenovo laptop plus a small handful of NUCs. Because of its low-end specs, it never saw much traction.

Thus in our book Ice Lake is Intel’s first true entry to the CPU market with the 10nm process. There are many more SKUs available and we’re already seeing decent uptake in the market across ultraportable laptops. Instead of one CPU, there are 11 in Intel’s Ice Lake line-up, across 9W, 15W and 28W power categories. These are all designed for low-power ultraportables and other mobile devices, at least for now.

Intel’s naming scheme is confusing as usual. Before Intel used U suffixes to denote 15W products, and Y to denote 9W. Now this is relegated to a single number: the 1065G7 for example is a 15W CPU, while the 1060G7 is 9W. All 15W products get a 5 in the fourth digit, and 9W parts get a 0. I personally preferred the use of U or Y, that was clearer for buyers rather than hiding this crucial performance indicator among a bunch of other letters and numbers.

What has become clearer this generation though, is the integrated graphics capabilities. The G suffix tells us exactly what sort of CPU configuration we’re getting: G7 denotes Iris Plus with the full 64 execution units unlocked, G4 is Iris Plus cut down to 48 EUs, and then G1 gives us UHD graphics with 32 EUs. All use Intel’s new Gen 11 GPU, which is one of the bigger changes to Ice Lake compared to previous 14nm generations.

And it’s a much needed change. Since Skylake was introduced, 15W CPUs have largely been stuck with GPUs that pack only 24 execution units with an architecture from 2015, and that continues to be the case with 2019’s 14nm refresh codenamed Comet Lake. Some 28W parts would push this up to 48 execution units, but that was generally the maximum you could find.

Now, the base G1 tier includes 32 EUs, so we’re already seeing a 33% increase to core count (if you want to call execution units “cores”). And the top end also gets a bump, but crucially you can still find 64 EU G7 graphics within a 15W power envelope, so Intel is bringing a significant jump to graphics capabilities within the existing TDP, allowing it to better compete with the beefy graphics AMD provides in Ryzen Mobile.

On top of this, Gen11 graphics provide a number of architectural changes, including support for variable rate shading, adaptive sync, display stream compression, and a faster media encoder. This really is the key benefit to getting Ice Lake over previous generations.

There are also significant changes on the CPU front, thanks to an all-new CPU core in Sunny Cove. Intel is promising an 18% IPC increase over Skylake, support for AVX-512, new dynamic tuning capabilities, and a much better memory controller that supports DDR4-3200 and LPDDR4X-3733 speeds.

While IPC has seen a large increase (according to Intel), this has been offset by lower clock speeds across the Ice Lake line. Take the flagship 15W part that we’re looking at today: the Core i7-1065G7. It gets the full 64 execution unit experience, giving us all the benefits of Gen11 graphics. But it only packs four Sunny Cove CPU cores with a base clock of 1.3 GHz, a single-core turbo of 3.9 GHz and an all-core speed of 3.5 GHz.

This is well below what you get with Intel’s other current-generation CPU, Comet Lake. The top-end Comet Lake chip built on 14nm offers a worse integrated graphics, but bumps up the CPU to 6 cores, clocked at 1.1 GHz base, 4.7 GHz single-core turbo, and 4.1 GHz all core. Alternatively you can get a Core i7 quad-core with a 1.8 GHz base and 4.3 GHz all-core turbo.

While these 15W parts generally don’t run at the advertised clock speeds in sustained workloads, even comparing just these rated frequencies shows a massive discrepancy. Base clocks comparing quad core to quad core are nearly 40% higher for Comet Lake, although this narrows to 23% comparing all-core turbos. But with Ice Lake only receiving an 18% average gain to IPC, you can already see that it is probably going to struggle to outperform what Intel already has on 14nm in CPU workloads.

And again, this whole situation with Intel having both 10nm Ice Lake and 14nm Comet Lake on the market at the same time as part of their 10th generation only serves to confuse customers. With Comet Lake boasting a beefier CPU and Ice Lake taking the GPU crown, regular everyday Joes will have to do even more research than ever before to ensure what they are getting is the right product for their use case.

And with product names like Core i7-10710U and Core i7-1065G7, which chip is better and in what areas is so murky I can’t see how anyone other than a hardcore enthusiast will truly know what they are getting. I feel like regular consumers know the old “higher number is better” trick, but with these names, that sort of deduction is impossible.

So that’s the baseline knowledge you need about Ice Lake to kick off this performance review...

Today we're focusing on the top-end Core i7-1065G7, a quad-core CPU with G7 graphics. This review will be discussing everything you need to know about productivity and compute performance, we’ll be comparing Ice Lake to previous 14nm parts, but gaming is something we’ll be tackling in a separate article.

The laptop we have for testing with Ice Lake inside is the new Razer Blade Stealth, a laptop design that we really like with its super sleek metal design and high-end components. It also tends to provide a really great test platform – the CPU inside this beast can be configured at both its standard 15W TDP, or at a higher 25W TDP, giving us useful data for both. It also packs dual-channel memory at the maximum LPDDR4X-3733 speeds, so we’re getting the full memory bandwidth improvements of Ice Lake in this test system, too, which is ideal.

The laptop also includes a discrete GPU, Nvidia’s GeForce GTX 1650 Max-Q. For most tests, this GPU has been disabled so we can focus on the compute performance of the new integrated Gen11 graphics, but in some cases it’s also enabled. We have test data from an MSI Prestige 14 which packs the 6-core i7-10710U and GTX 1650 Max-Q, so we can also get a decent look at whether Ice Lake or Comet Lake is better when a more powerful discrete GPU is thrown into the picture.

Benchmarks

We’re going to kick things off here as usual with a look at Cinebench R20, which gives us that all important look at multi and single threaded performance. The results for Ice Lake and 10nm are quite disappointing but not entirely unexpected given the lower clock speeds this CPU is able to run at. The Core i7-1065G7 sits among the Core i5-10210U and Core i7-8565U towards the bottom of the charts. These are all quad-core CPUs operating at 15W, there clearly isn’t much to be gained here moving from 14nm to 10nm in this power envelope.

The new Core i7-10710U with six cores demolishes the Core i7-1065G7 in multi-threaded performance here: the 10710U is a good 32 percent faster. However, the 1065G7 does very well with single-threaded performance, outperforming most other CPUs in this table, which bodes well for Ice Lake in other single-threaded benchmarks.

When bumping the 1065G7 up to 25W, only then is it able to outperform the 10710U in its 15W configuration, but when both chips are operating at the same TDP, the six-core option is still far superior.

Looking at clock speeds, there aren’t too many surprises given our rated clock speed discussion earlier. Like all 15W CPUs, under a sustained Cinebench run, clock speeds are well below the maximum all-core turbo frequency so power consumption doesn’t exceed 15W. The 14nm Core i5-10210U sits between 2.2 and 2.3 GHz all-core in this run, while the 10nm Core i7-1065G7 drops down to 1.8 to 1.9 GHz.

All three of these CPUs score roughly the same in this benchmark, but the 14nm chips need to be clocked around 22 percent higher. This ends up being quite similar to the difference in rated boost clocks we saw earlier, which was 23 percent, and also similar to Intel’s IPC improvement claims for Ice Lake, at an 18 percent average.

The downside is that while Ice Lake is able to do more per clock and perform as well as 14nm parts, there is no material improvement to efficiency. This is pretty concerning for a new process, although it also speaks volumes about what Intel has been able to do with 14nm and its continual refinements. After all, this is 10nm up against 14nm++++, and those pluses do make a small difference with these mobile parts.

For those thinking about a desktop Ice Lake processor, if we got perfect scaling with Cinebench R20 results, Intel would be able to make a 4.0 GHz 10nm CPU run like a 4.8 GHz 14nm CPU. But there are many question marks here: can Intel get Ice Lake running at 4.0 GHz all-core, especially over what would need to be 8 cores? Given boost clocks for mobile parts, that seems like it might be tricky. Would it even be as efficient at these frequencies? Would it improve performance?

We’ll have to wait for desktop 10nm parts to get answers to those questions but you can see why Intel is sticking with 14nm on the desktop for now.

Let’s look at some more benchmarks. Cinebench R15 has the 1065G7 slightly ahead of the Core i5 and Core i7 14nm models we’ve just been talking about, but there’s no significant gains to be made compared to the 10710U with its six cores: the 1065G7 is still 24 percent slower. Single thread performance is very good as we saw previously.

Handbrake is where the boys get separated from the men, as this multiple hour long benchmark really stresses the long term steady state performance of these chips. Again, when constrained to 15W, there’s really no advantage to CPU performance here over previous generation parts. We’re about 23 percent slower than the 10710U, while the higher clocked Core i7-8565U actually outperforms the 1065G7 by a few minutes.

Very similar story with x264 encoding, although this time the 1065G7 is marginally slower than 14nm equivalents such as the Core i5-10210U. In anything multi-threaded like this there just aren’t many gains to be had from Intel’s 10nm process.

It’s not all bad for Ice Lake though. Single-threaded performance is very strong, and here looking at a single instance of Premiere’s Warp Stabilizer effect we can see that the 1065G7 is at least 8 percent faster than last-gen 14nm CPUs, and 12 percent faster than Intel’s lower clocked 10710U. We saw hints of that in Cinebench but here we can see the material difference, with Ice Lake shaving a minute off this workload.

What about another good result for Ice Lake? In Adobe Photoshop’s Iris Blur filter we see strong performance out of the Core i7-1065G7. This CPU is 8 percent faster than the 10710U and 17 percent faster than the 10510U, which is a really strong showing. While this isn’t a single-threaded test, it’s quite memory intensive given the size of the photo we’re working with, so I believe we’re seeing good gains from the massive improvement to memory bandwidth we get thanks to Ice Lake’s new memory controller.

But when we move back to these long term multi-threaded tests, it’s not good news for Ice Lake. In this Blender benchmark run which is run on the CPU for all of these processors, the 1065G7 is slower than 14nm equivalents like the Core i7-8565U. It’s not massively slower, but ideally you’d like to see some sort of gain in the 15W power class. That is only provided by the six-core 10710U, which matches the performance of the 1065G7 at 25W.

7-zip continues the story we’ve been showing for a while now. Even though this is a short workload, it’s multi-threaded, and the 1065G7 ends up falling behind the Core i5-10210U. The margins aren’t huge, all within single digits for most of these 10nm vs 14nm quad core comparisons, but still not that impressive.

Few more workloads to go, this time our brand new MATLAB benchmark which is a real world usage of differential equations and fast Fourier transforms, which are common tasks people perform in this engineering tool. This workload hits just a few threads and also can be quite memory and cache intensive, so with Ice Lake seeing gains in these areas, MATLAB performance is higher than other 10th-generation parts. 13 percent gains here over 10th-gen is very decent.

What about Adobe PDF exporting, which is another single-threaded task? Ice Lake performs well here, like we’ve seen in other 1T workloads. 10 percent better performance than the Core i7-10710U is a good result and is in line with what we’ve shown so far.

Ice Lake isn’t very fast in VeraCrypt decryption, matching the 8565U and surprisingly falling well behind the Core i5-10210U. Comet Lake possibly has enhancements to improve accelerated AES performance that other architectures don’t, testing further 10th-gen CPUs will give more insights to this benchmark.

With all the CPU limited tests out of the way, let’s take a look at some compute workloads, because this is where Ice Lake will really shine thanks to its much faster Gen11 GPU. Our new Premiere benchmark is a prime example of this. We’re able to take advantage of Premiere’s hardware accelerated encoding, on top of GPU accelerated effects, to deliver significant gains.

In its stock configuration with no dedicated GPU, the Core i7-1065G7 outperforms most other configurations, including the Core i5-10210U with an MX250 GPU. It also demolishes 14nm parts: 75% faster than the i7-10710U, and more than twice as fast as the i5-10210U, all thanks to huge GPU gains.

What’s also impressive to see is that when the GPU is kept constant, in this case the GTX 1050 Max-Q, Ice Lake is still marginally faster in this workload compared to the six-core 10710U, I think a lot of that comes down to better accelerated encoding. But overall, if Premiere is your main workload, Ice Lake is the way to go with better Warp Stabilizer performance and better encoding performance.

When using our older Premiere benchmark, which is more GPU intensive, gains aren’t as significant especially compared to some of the discrete GPU options. Pairing a 10210U with an MX250 for example, is much faster, although consumes a lot more power. And Ice Lake falls behind Comet Lake with 6 cores versus 4 cores when the GTX 1650 Max-Q is held constant.

We also see huge performance gains in CompuBench Optical Flow, with a more than 2x advantage comparing Gen11 graphics to older 14nm options. When fully GPU limited like this, it’s no surprise to see the much higher execution unit count on Ice Lake take control and deliver a big performance improvement.

Similar gains are available with Photoshop’s Smart Sharpen filter, which is also run on the GPU. 130% faster performance from Ice Lake compared to Comet Lake and other Skylake derivative CPUs is a huge gain, and it slightly nudges out the MX250 from Nvidia which is awesome from an integrated option. It also ends up handily beating AMD’s Ryzen Mobile offerings in this test, I suspect higher memory bandwidth to the GPU is playing a part there, given we just saw Ryzen performing well in Optical Flow.

What We Learned

All the data in, that's many hours of benchmarking you just witnessed right there. We haven’t seen too many surprises given all the information we had in the lead up to this release. We weren’t expecting much in the realm of CPU performance, but we did expect decent gains to GPU performance, and that’s largely what we observed.

The Core i7-1065G7 delivers roughly equivalent multi-threaded CPU performance to the Core i5-10210U. So when comparing quad-cores on Intel’s 10nm and 14nm nodes, basically nothing has been gained here. Single thread performance is somewhat higher, around the 10 percent mark, but this is offset by the occasional slower multi-threaded result. In general, it’s fair to say performance is about even.

And it’s similar comparing the Core i7-1065G7 to the Whiskey Lake Core i7-8565U. There are some larger than expected gains for workloads like MATLAB and Photoshop Iris Blur, but in longer tests like Handbrake, Ice Lake actually clocks in slower. On average, the i7-1065G7 is a few percent faster but it’s not a mindblowing difference and certainly in these CPU limited workloads, there’s not much of a reason to upgrade from any of the 8th-generation parts or newer.

When you compare Ice Lake to the best Intel has to offer within 15W, the Core i7-10710U smokes the Core i7-1065G7 in multi-threaded workloads, so if you want to use your ultraportable for anything intensive like video encoding, a Comet Lake six-core CPU is a better option. At the same time, Ice Lake is generally faster in single-threaded tests, so it will depend what sort of things you do with your laptop as to what processor makes more sense.

The margins don’t change too much when comparing 25W configurations, more power does equal more performance and in this case Ice Lake is about 20 to 25 percent faster in its 25W mode compared to 15W, but so is Comet Lake. If you were expecting to see notably better efficiency at these higher power targets, so far that doesn’t seem to be the case on the mobile side.

On the other hand, Ice Lake is clearly much faster when you need GPU acceleration. In pure GPU limited situations, Ice Lake’s unlocked Gen 11 GPU is more than twice as fast as the crappy integrated GPU we’ve had from Skylake derivatives. And that’s all within the same 15W power envelope. In a mixed workload like Premiere, that can lead to huge performance improvements.

Overall, our early impressions on Intel’s Ice Lake processor are mixed. From testing the fastest 15W configuration available, there are some positive takeaways: faster single-thread performance, hugely improved GPU performance – but this is spoiled by issues in other areas.

Getting no improvement to multi-threaded performance is concerning for Intel’s new 10nm node. Considering the CPUs we’ve tested have all been locked to 15W, getting no performance gains from 10nm versus 14nm means we’re also seeing no gains to efficiency. Ice Lake does appear to have much improved IPC, but this has been entirely offset by the CPU running at lower clock speeds. Lower clock speeds, higher IPC, same power, same performance.

This handily allows Comet Lake to swoop in, offering a six-core CPU in the same power envelope, for significant gains to multi-threaded performance. Again, doesn’t apply to single-thread applications where Ice Lake holds a lead, but for anything that loads up the CPU, Comet Lake is the way to go.

Granted, we haven’t done any battery life testing with this new platform as it’s nearly impossible to get a good apples-to-apples comparison, but same performance at same power level shouldn’t translate to major improvements. If anything, gains will come from other platform advantages like changes to the way power gating is handled, different boost technologies, more efficient memory and so on.

Getting a much faster GPU with Ice Lake is nice, but it’s also not a revolutionary upgrade given AMD has been offering this sort of performance since late 2017 with Ryzen Mobile. This is more about Intel bringing their part up to a competitive graphics standpoint. In some instances Ice Lake’s 64 execution unit GPU still falls behind first-gen Ryzen Mobile, depending on the power limitations. Other times, it’s faster, depending on how much memory bandwidth is required given Ice Lake has a big advantage in that department.

Intel also has to compete with discrete GPU offerings, including Nvidia’s popular MX150 and MX250, as well as newer, more powerful options like the GeForce GTX 1650 Max-Q that was included with our Razer Blade Stealth test system. It depends heavily on the exact workload in question, but Ice Lake’s integrated GPU isn’t much faster than Comet Lake with an MX250, the exception to this rule being Premiere, where a combination of factors make Ice Lake a much better option.

As it stands right now, the outright fastest hardware combination I’ve seen in ultraportable form factors appears to be Intel’s Core i7-10710U paired with a GTX 1650 Max-Q, which you can get in MSI’s Prestige 14. If Razer, for example, had swapped out the i7-1065G7 for a i7-10710U in this latest Blade Stealth, I suspect in most workloads it would be superior, aside from some single-threaded edge cases.

While it’s neat that Ice Lake has a much improved integrated GPU, because it doesn’t bring forward CPU performance with it, we’re not seeing improvements over what was already possible with 14nm. OEMs have been increasingly opting for discrete GPUs in their ultraportables, and Ice Lake doesn’t do much to improve over these configurations. If you bought an 8th-gen laptop with an MX150, Ice Lake won't be much of an upgrade. Where you will see gains is in laptops that don’t use a discrete GPU. If an OEM decides that they just want to include one chip, then something like the Core i7-1065G7 is going to give a large compute performance gain over say, an 8th-gen Core i7-8565U by itself.

Bottom line, it's a modest start for Intel’s 10nm series due to a lack of efficiency gains which probably has to do with Intel’s struggles at 10nm. It's possible that with a 10nm revision or just a full step forward to 7nm, this will improve significantly. Meanwhile on the other side of the fence AMD is working hard on a next-gen Ryzen Mobile part on TSMC’s 7nm.

Shopping Shortcuts
Further Reading