To say there’s a lot riding on Intel’s upcoming 10th-gen Ice Lake CPU would be an understatement.
Seemingly stuck in the mud at 14nm for several generations, the chip giant as AMD pushing its left flank on desktops and servers, and Qualcomm coming over a hill on its right flank. So yeah, there’s a lot riding on how good Intel’s new 10nm Ice Lake CPU is.
The good news for Intel is that its 10th-gen Ice Lake CPU indeed looks like a win at last. Intel let a handful of hardware reviewers, PCWorld included, poke and prod Ice Lake CPUs ahead of today’s launch. Read my colleague Mark Hachman's companion piece on Intel Ice Lake specs and features for details on the entire family.
The upshot? It’s a decently fast CPU that offers everything from mild to fantastic performance buffs over today’s fastest thin-and-light laptop CPUs.
How we tested
For this test, Intel gave us early access to Software Developer Systems (SDS) laptops, which the company has custom-built as testbeds for software partners. The laptops were equipped with 256GB Intel 7600P SSDs, 8GB of LPDDR4X/3733 RAM in dual-channel mode, and a 10th-gen Core i7-1065G7 Ice Lake chip that was configurable to either 15-watt or 25-watt modes. The laptops had the latest 1903 build of Windows 10 installed.
The Core i7-1065G7 features four cores with Hyper-Threading, a base clock of 1.3GHz and maximum single-core Turbo Boost of 3.9GHz. An all-core boost tops out at 3.5GHz. The graphics core runs at 1.1GHz and features 64 execution units (EUs). The total cache in the chip is 8MB.
There were no testing constraints placed on reviewers. We were allowed to run any test we wanted within the business day we had access to the laptops.
We know cynical people will say the laptops were ringers designed to make the chip look better. To make sure the laptops weren’t hooked up to hidden water chillers in the basement of the building, we asked Intel to crack open an SDS laptop.
Inside we saw what appeared to be basic cooling, with a single heat-pipe and dual fans. No thermal or TDP rating was given for the cooling, but it wasn’t anything to write home about. A Dell XPS 13 9380, for example, features two massive heat pipes inside of it, and Dell likes to brag about how much cooling it jammed in there.
That basically means the performance you see here is a baseline setting. While it will vary depending on each laptop design, there’s likely a ton of headroom when you put this CPU into a laptop such as Dell’s XPS 13 9380.
Finally, the SDS could run between its base 15 watts of thermal limit, or its configurable TDP of up to 25 watts. We tried to run at both settings where time allowed.
Most of our comparisons were run against a Dell XPS 13 9380 and an HP Spectre x360 13. Both had the same 8th-gen Whiskey Lake U Core i7-8565U inside, and both gave us very differing representative performance. The Dell XPS 13 9380 is a traditional clamshell, tuned to swing for the fences in most CPU tasks. The HP Spectre x360 13, however, is a convertible with a 360-degree hinger. In our experience, most convertible tablet designs give up performance to control heat, because they know you’re likely to be holding that tablet in your hands, or on your chest or lap. The vast majority of Whiskey Lake U laptops we’ve seen perform closer to the HP Spectre x360’s moderate pace, while only a few performance-oriented clamshells push the envelope like the Dell.
One last thing: The SDS units we tested did not have Intel’s Dynamic Tuning 2.0 enabled. The feature uses machine learning to wring even more performance out of the chip. Basically it looks at what you’re doing and can decide on the fly to increase performance if it thinks it’ll be a fairly light load. If, however, it can see you’re about to pound it for an hour with a video encode, it won’t push as hard. This means there’s possibly even more performance to be had once we see production laptops with Ice Lake CPUs.
10th-gen rendering performance
As always we’ll kick this off with Maxon’s popular Cinebench R15 and R20 benchmarks. It’s a 3D modelling application that is almost a pure CPU benchmark. The more CPU threads you feed it, the higher the performance and the score you get from it.
These days, Intel frowns at Cinebench, because the company feels it’s not reflective of what people do with a three-pound, thin-and-light laptop. While that’s a reasonable argument, we still see Cinebench as a good reference point for CPU performance.
Ice Lake does reasonably well here when it's given 25 watts of thermal headroom to roam: It’s essentially 6 percent faster than the Dell XPS 9380. What, you may ask—going from 14nm to 10nm nets you 6 percent?! Remember there are also clock differences and cooling differences involved.
Maxon updated Cinebench recently with its R20 version, which we also recorded. Besides adding AVX, AVX2, and AVX512 workloads, one key change from Cinebench R20 is how long it takes to run. While Cinebench R15 takes a minute to run on the Dell XPS 13 9380, Cinebench R20 takes three. That longer run time eats up the Turbo Boost quickly, so most of the run is at a lower clock speed.
Intel CPUs also typically lower clock speeds under AVX workloads even further. Interestingly, while the Ice Lake chip had the lead over the fastest Whiskey Lake U in Cinebench R15, in R20 it’s a little slower. Is AVX512 slowing it down just just enough to lose? We don’t know, and we’ll have to wait for production units to dive in even deeper.
The good news for Intel is the 15-watt setting is plenty to outpace the HP Spectre x360’s Whiskey Lake U chip.
We’ll close out the Cinebench runs with a single-threaded run using Cinebench R15. For the most part, we’d call it a tie. Some will see that as a negative, but again remember the 10th-gen Ice Lake has a maximum single-thread clock of 4.1GHz, vs the 4.6GHz single core Turbo of an 8th-gen Whiskey Lake U.
The chip also does relatively well at the 15-watt TDP, with the Ice Lake U scoring about 8 percent higher in Cinebench.
Our last rendering benchmark is the venerable POV-Ray 3.7. It’s an oldie (since the Amiga) but still a popular way to measure CPU performance. Intel, again, would like argue that few people using thin-and-lights are doing CPU-based ray tracing, but more data can’t hurt.
Again, we see the Dell’s high-performance cooling and aggressive power settings tying it with the Ice Lake set to 25 watts. Not bad when you consider the differences in platforms. We also again see the Ice Lake chip set to 15 watts score about 9 percent higher than the HP Spectre x360 13.
10-gen Audacity Export Performance
Acknowledging that few people use ultralight laptops to perform 3D animation, we decided to try something far more mainstream on the content creation side: edit a podcast. We took at 700MB WAV file ripped from a video podcast, imported it into the free audio editor Audacity, and exported it to an MP3 using the Standard preset with a variable speed set to “fast,” and in joint stereo mode. We then timed how it took to write the file, and averaged the result of three tries.
The clear winner was Ice Lake, with a 9-percent advantage over the Dell and a 25-percent bump over the HP. Where does the advantage come from? Audacity recently integrated the free LAME MP3 encoder, which is optimized for AltiVec (on PowerPC Macs) and all the versions of SSE.
10th-gen Ice Lake Encoding Performance
Next we move onto the encoding performance of the 10th-gen Ice Lake chip. As our time was limited, we skipped most of the conventional encoding using what is arguably old-school H.264. Our standard test, for example, takes about an hour to run on a laptop and uses an older version of HandBrake with a 30GB 1080p file. For this test we installed the latest July 29 Nightly build of HandBrake and measured how long it took to convert a 6GB 4K file using the H.265 preset. We also selected the Quick Sync profile to enable the hardware encode features of the chip. As Intel touts improved Quick Sync performance, it should show up here.
Fortunately for Intel’s sake, it does in a big way. The 25-watt Ice lake vs. the XPS 13’s Whiskey Lake U yields a 42 percent advantage for Ice Lake. Set to a more conservative 15 watts, the gap is about 35 percent against the Spectre x360 13.
Our next encoding test is Cinegy’s Cinescore benchmark. It’s a stand-alone test designed to measure encoding performance on CPU and CUDA-based GPUs in various popular broadcast industry formats from SD to 8K.
The standout score is the 10th-gen Ice Lake set to a 25W TDP. Its Cinescore performance is about 35 percent better than the Dell's. The Dell, oddly, was slightly slower than the HP. This won’t be the only time either: We found the HP often outperformed the Dell XPS 13 in graphics tests, while underperforming it on CPU tests.
The 15-watt setting for the Ice Lake chip was confusing as well, as we would have expected it to outperform the Whiskey Lake units. Unfortunately, we ran out of time to rerun the test.
10th-gen PCMark 10 Performance
Our next test is UL’s latest version of PCMark. Among the changes PCMark 10 makes is a move to open-source applications such as Libre Office for office apps and GIMP, Chrome, and Firefox for application load time measurements.
The overall prize goes to the 10th-gen Ice Lake chip, but it’s not exactly a game-changer.
We’ll break out the PCMark result that shows where most of Ice Lake’s advantages come from: Content Creation. That’s where PCMark delves into “visualization,” photo editing, and video encoding. All these, if you haven’t guessed, give the advantage to the Ice Lake chip.
10th-gen Ice Lake Geekbench Performance
We’ve had our off and on moment’s with Primate Lab’s Geekbench but it’s mostly over how people like to take random results generated on a different platform or OS and then bash each other over the heads with them like a Fred Flintstone cartoon.
For this section, we omit the 25-watt scores from the Ice Lake chip which we had incorrectly recorded or run but after comparing notes with another reviewer, we discovered there’s just not difference. Why? We surmise the very short nature of the 25 different tests Geekbench runs just isn’t enough of a load to push the laptops out of their Turbo Boost windows.
The results you see are the 15-watt Ice Lake U against the Dell XPS 13 which is among the fastest of the Whiskey Lake U laptops we’ve seen.
First up is Geekbench 4.4.1 multi-core performance where we see the Ice Lake U is 20 percent faster. Single core Geekbench closes the gap with the Ice Lake chip “only” about 10 percent faster than the Whiskey Lake U.
What’s more interesting about the Geekbench results are the subscores of the individual test algorithms Primate Labs uses. In many of the 25 different tests, it’s mostly meh. The higher-clock speeds of the Whiskey Lake U make it mostly a give and take in many of the tests with Whiskey Lake U also occasionally winning by decent margins in SGEMM (general matrix multiplication).
But some of the tests do touch on the increased performance of the 10th Ice Lake chip enough to greatly boost its overall score and worth calling out.
The SFFT test, for example, performs Fast Fourier Transforms and is hand-tuned for AVX512, AVX2, AVX, alone with SSE3 and SSE2 on x86. The Ice Lake U chip is 49 percent faster than Whiskey Lake U. In the AES encryption module, Geekbench’s algorithms score the Ice Lake U at 58 percent faster Whiskey Lake U.
10th-gen Ice Lake LPDDR4X memory bandwidth
One of the key changes in Ice Lake is the adoption of LPDDR4X memory. One of key changes that brings is increased memory capacity. LPDDR3 used in most laptop tops out at 16GB while LPDDR4X should easily shatter that density limit eventually.
The other key advantage is an increase in memory bandwidth. LPDDR3 typically tops out at 2133MHz. The LPDDR4X in Ice Lake laptops hit 3,733MHz. As memory bandwidth is one of the constraints on integrated graphics performance, this is also one of the contributors to Ice Lake’s better gaming performance.
10th-gen Ice Lake Gaming performance
Intel has touted Gen11 graphics as a game-changer. Despite being skeptical, we’d have to agree for the most part now. First up is UL’s 3DMark Sky Diver. It’s a lighter gaming test in the 3DMark Suite and doesn’t exactly push the GPU, but you can see a huge performance spike for Intel’s Gen 11 graphics over today’s UHD620 graphics.
The 25-watt setting puts Ice Lake about 75 percent faster than the Dell’s score. Perhaps more notable are the two laptops at the bottom of the chart. Both are Huawei Matebooks with discrete graphics. One features Nvidia’s GeForce MX150 set to 10 watts (the 1D12 version), and the other features the MX150 set to 25 watts (the 1D10 version).
The Iris Plus graphics with 64 EUs is dead even with the 10-watt MX150 variant. While it still loses to the 25-watt MX150 variant, it’s awfully close for comfort.
We decided to push the GPUs with a tougher test, so we also ran 3DMark Time Spy and looked only at the graphics performance. For the most part, the results are the same and a big win for Intel Iris Plus graphics.
While Intel is likely hoping to eat into Nvidia’s MX150-range customer base, the truth is that’s just icing on the cake. What Intel is likely more concerned with is this company called Qualcomm and its upcoming SnapDragon 8cx chip running Windows 10. At Computex, Qualcomm gave PCWorld a preview of its graphics performance, showing it eating even top Whiskey Lake U CPUs.
Today, Intel would like to see Qualcomm’s 8cx and raise it, with Ice Lake U’s graphics performance about 45 percent higher. Yup, it’s press-preview laptop vs. press-preview laptop time.
The last 3DMark result we want to show off is the new Variable Rate Shader performance in the graphics core. Microsoft’s VRS basically enables performing high-quality shader effects on areas you care about. For areas you don’t care about or you won’t be looking at? They get lower settings.
Using a preview version of 3DMark, we were able to test the feature on Ice Lake U. The results suggest that games supporting VRS could see significant increase in graphics performance on 10th-gen Ice Lake U chips.
With what time we had, we ran a few games as well. The first one was World of Tanks Encore. It’s a benchmark based on the popular free-to-play game, but it adds further enhancements. For the most part, it validates what we saw in 3DMark.
We also ran Counter Strike: Global Operations on the laptops, set to a more playable 720p resolution. We unfortunately didn’t have time to run it in the 15-watt mode on Ice Lake, but at 25 watts it again torches the older UHD 620 graphics in today’s laptops. The MX150 variants did pull ahead further, however, so Iris Plus doesn’t match discrete on all fronts.
First, we’ll remind everyone that this is a performance preview. We’ll say that again: a PREVIEW. Unlike desktop CPUs, which you can mostly control, laptop CPUs are about the total package. Intel’s 10th-gen Ice Lake won’t be final until we see the first batch of laptops and what each OEM does with it.
With that said, our general takeaway is that the CPU is on a par with, or somewhat faster than, today’s laptops in most conventional applications that don’t touch the special sauce of improved encryption, AI or encoding features. In applications that touch those features, though, it’s a major upgrade over existing 14nm chips.
The real surprise is the performance of the Gen11 graphics. Intel’s integrated graphics have been the butt of jokes for years, but Iris Plus is a turning point. It is a generational performance uplift over UHD graphics and might just be the surprise killer feature of the CPU.