This took a long time since we are finally finished testing the Intel Ice Lake architecture. This is Intel's first real attempt to develop a 10nm CPU. In this test, we'll compare it to the 14nm offerings to see how performance improves.
We saw the 10nm Cannon Lake last year, but it was a single processor, the Core i3-8121U, a poor dual-core up to 3.2GHz with integrated graphics. It was used in a single Lenovo laptop as well as a small handful of NUCs. Because of its low-end specifications, it has never seen much traction.
In our book, Ice Lake is therefore Intel's first real entry into the CPU market with the 10 nm process. There are many more SKUs available and we are already seeing good acceptance in the ultra portable laptop market. Instead of a CPU, there are 11 in the Intel Ice Lake product range in the power categories 9 W, 15 W and 28 W. These are designed at least for the moment for ultra-portable devices with low power consumption and other mobile devices.
As usual, Intel's naming scheme is confusing. Before Intel used U suffixes for 15 W products and Y for 9 W products. Now this is referred to a single number: for example, the 1065G7 is a 15 W CPU, while the 1060G7 is 9 W. All 15W products get a 5 in the fourth digit and 9W parts a 0. I personally preferred the use of U or Y, which was clearer to buyers than this key performance indicator among a number of other letters and numbers to hide.
However, what has become clearer in this generation are the integrated graphics functions. The G suffix indicates exactly what type of CPU configuration we get: G7 denotes Iris Plus with the full 64 unlocked execution units, G4 is Iris Plus, which is reduced to 48 EUs, and G1 gives us UHD graphics with 32 EUs . All use the new Gen 11 GPU from Intel, one of the major changes to Ice Lake compared to previous 14nm generations.
And it is an urgently needed change. Since the launch of Skylake, 15-W CPUs have mostly been dealing with GPUs, which from 2015 only contain 24 execution units with one architecture. This continues to apply to the 14nm refresh code name Comet Lake from 2019. Some 28W parts would increase this to 48 execution units, but that was generally the maximum you could find.
Now the base G1 level comprises 32 EUs, so we can already see a 33% increase in the number of cores (if you want to call execution units "cores"). And the top end gets a bump too, but the bottom line is that you can still find 64 EU G7 graphics in a 15W power case, so Intel significantly improves the graphics capabilities within the existing TDP to better deal with the beefy graphics AMD can compete offers in Ryzen Mobile.
In addition, Gen11 graphics offer a number of architectural changes, including support for variable rate shading, adaptive synchronization, compression of display streams, and a faster media encoder. This is really the main benefit to preserving Ice Lake over previous generations.
Thanks to a brand new CPU core in Sunny Cove, there are also significant changes to the CPU front. Intel promises an IPC increase of 18% over Skylake, support for AVX-512, new dynamic optimization functions and a much better memory controller that supports the speeds DDR4-3200 and LPDDR4X-3733.
While IPC (according to Intel) saw a sharp surge, this was offset by lower clock speeds across the Ice Lake line. Take the flagship 15W we're looking at today: the Core i7-1065G7. It offers full experience with 64 execution units and offers us all the advantages of Gen11 graphics. However, it only contains 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 far below what you get with the other current generation Intel CPU, Comet Lake. The top-class Comet Lake chip, built on 14 nm, offers poorer integrated graphics, but increases the CPU to 6 cores, those with a 1.1 GHz base, 4.7 GHz single-core turbo and 4.1 GHz all-core are clocked. Alternatively, you can purchase a Core i7 quad-core with a 1.8 GHz base and 4.3 GHz all-core turbo.
While these 15 W parts generally do not run at the specified clock rates when the workload continues, even the comparison of only these nominal frequencies shows a massive discrepancy. Base clocks that compare quad-core to quad-core are almost 40% higher for Comet Lake, although this drops to 23% compared to all-core turbos. However, since Ice Lake only achieves an average IPC gain of 18%, you can already see that it will likely be difficult to outperform what Intel already has on 14nm CPU workloads.
Again, this whole situation, in which Intel has both 10 nm ice lake and 14 nm comet lake on the market as part of the 10th generation, only serves to confuse customers. With Comet Lake with a more powerful CPU and Ice Lake as the GPU crown, Joes has to do more research than ever before to ensure that what he gets is the right product for his application.
And with product names like Core i7-10710U and Core i7-1065G7, which chip is better and in which areas are so cloudy, I can't see how anyone other than a hardcore enthusiast really knows what they're getting. I have the feeling that ordinary consumers know that the old trick "higher number is better", but with these names such a deduction is impossible.
So that's the basic knowledge you need about Ice Lake to start this performance review …
Today we focus on the top-end core i7-1065G7, a quad-core CPU with G7 graphics. This test covers everything you need to know about productivity and computing power. We'll compare Ice Lake to previous 14nm parts, but gaming is covered in a separate article.
The laptop we have inside for testing with Ice Lake is the new Razer Blade Stealth, a laptop design that we really like with its super-slim metal design and high-end components. It also offers a really great test platform – the CPU in this beast can be configured with both its standard TDP of 15 W and a higher TDP of 25 W, which gives us useful data for both. It also packs two-channel memory at the maximum speed of LPDDR4X-3733, so in this test system, too, we get the improvements in Ice Lake's full memory bandwidth, which is ideal.
The laptop also includes a discrete GPU, the GeForce GTX 1650 Max-Q from Nvidia. Most of the tests disabled this GPU so that we can concentrate on the computing power of the new integrated Gen11 graphics. In some cases, however, it is also activated. We have test data from an MSI Prestige 14 that includes the 6-core i7-10710U and the GTX 1650 Max-Q, so we can also have a decent look at whether Ice Lake or Comet Lake is better if a more powerful discrete GPU is is used in the picture.
We'll start here as usual with a look at Cinebench R20, which gives us the most important look at the performance of multiple and single threads. The results for Ice Lake and 10nm are quite disappointing, but not entirely unexpected given the lower clock speeds that this CPU can work with. The Core i7-1065G7 is located between the Core i5-10210U and the Core i7-8565U at the bottom of the diagrams. These are all quad-core CPUs that run at 15W. There is clearly not much to be gained here if you move from 14 nm to 10 nm in this power envelope.
The new Core i7-10710U with six cores destroys the Core i7-1065G7 in multithreaded performance: The 10710U is a good 32 percent faster. However, the 1065G7 performs very well with single-threaded performance and outperforms most of the other CPUs in this table, which is a good sign for Ice Lake in other single-threaded benchmarks.
When raised to 25W, the 1065G7 can only outperform the 10710U in its 15W configuration. However, if both chips work with the same TDP, the six-core option is still far superior.
With regard to clock rates, there aren't too many surprises given our discussion of the rated clock rate. As with all 15 W CPUs, the clock rates for a sustained Cinebench run are significantly below the maximum all-core turbo frequency, so that the power consumption does not exceed 15 W. The 14 nm core i5-10210U is between 2.2 and 2.3 GHz all-core in this run, while the 10 nm core i7-1065G7 drops to 1.8 to 1.9 GHz.
All three CPUs perform roughly equally in this benchmark, but the 14 nm chips have to be clocked 22 percent higher. This roughly corresponds to the difference in the previously rated Boost watches, which was 23 percent, and also the IPC improvement claims from Intel for Ice Lake with an average of 18 percent.
The downside is that while Ice Lake can do more per watch and power, as well as 14nm parts, efficiency isn't significantly improved. This is quite worrying for a new process, although it also speaks volumes about what Intel has achieved with 14nm and its continuous improvements. After all, this is 10 nm compared to 14 nm ++++, and these plus points make a small difference with these mobile parts.
For those thinking of a desktop Ice Lake processor, Intel would be able to run a 4.0 GHz 10 nm CPU like a 4.8 GHz 14 nm CPU, if we achieved perfect scaling with Cinebench R20 results. But there are many question marks here: Can Intel Ice Lake run with 4.0 GHz all-core, especially over 8 cores? Given the boost watches for mobile parts, this seems to be difficult. Would it even be so efficient at these frequencies? Would it improve performance?
We have to wait for desktop 10nm parts to get answers to these questions, but you can see why Intel remains on the desktop at 14nm 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 just talked about, but compared to the 10710U with its six cores, no significant gains can be made: The 1065G7 is still 24 percent slower. The performance of a single thread is very good, as we saw earlier.
With the handbrake, the boys are separated from the men, as this multi-hour benchmark really emphasizes the long-term performance of these chips in steady state. Here too, the CPU performance at 15 W has no advantage over parts of the previous generation. 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 slightly slower than 14 nm equivalents like the Core i5-10210U. In a multithreaded program like this, the 10 nm process from Intel does not offer many advantages.
However, it's not all bad for Ice Lake. Single-threaded performance is very strong, and if we look at a single instance of Premiere's Warp Stabilizer effect, we can see that the 1065G7 is at least 8 percent faster than last-generation 14nm CPUs and 12 percent faster than the low-clocked 10710U from Intel. We saw evidence of this in Cinebench, but here we can see the material difference since Ice Lake was a minute away from this workload.
What about another good result for Ice Lake? In the Iris blur filter from Adobe Photoshop we see a strong performance 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 performance. While this isn't a single-threaded test, it is quite memory-intensive considering the size of the photo we're working with. I think we are seeing good gains from the massive memory bandwidth improvement we get thanks to the new Ice Lake memory controller.
However, when we return to these long-term multithreaded tests, this is not good news for Ice Lake. In this Blender benchmark run, which runs all of these processors on the CPU, the 1065G7 is slower than 14nm equivalents like the Core i7-8565U. It's not very slow, but ideally you want to see a win in the 15W power class. This is only provided by the 10710U with six cores, which corresponds to the power of the 1065G7 at 25 W.
7-zip continues the story that we have been showing for some time. Although this is a short workload, it is multi-threaded and the 1065G7 falls behind the Core i5-10210U. Margins aren't huge, all in the single digits for most of these 10nm and 14nm quad-core comparisons, but still not that impressive.
Only a few workloads left, this time our brand new MATLAB benchmark, which uses differential equations and fast Fourier transformations in practice. These are common tasks that are performed in this engineering tool. This workload only affects a few threads and can also be very memory and cache intensive. Because Ice Lake is growing in these areas, MATLAB performance is higher than other 10th generation parts. 13 percent growth here compared to the 10th generation is very decent.
What about exporting Adobe PDF, another single thread task? Ice Lake does well here, as we've seen on other 1T workloads. A 10 percent better performance than the Core i7-10710U is a good result and corresponds to what we have shown so far.
Ice Lake is not very fast with VeraCrypt decryption, corresponds to the 8565U and surprisingly falls significantly behind the Core i5-10210U. Comet Lake may have improvements to improve accelerated AES performance that other architectures do not offer. By testing further 10th generation CPUs, you will gain further insights into this benchmark.
Let's take a look at some computer workloads as all CPU-limited tests are out of the way, because here Ice Lake will really shine thanks to its much faster Gen11 GPU. Our new Premiere benchmark is a prime example of this. We can use Premiere's hardware accelerated encoding in addition to the GPU accelerated effects to achieve significant gains.
In its standard configuration without a dedicated GPU, the Core i7-1065G7 outperforms most other configurations, including the Core i5-10210U with an MX250 GPU. It also destroys 14nm parts: 75% faster than the i7-10710U and more than twice as fast as the i5-10210U, all thanks to enormous GPU gains.
It's also impressive to see that with a constant GPU, in this case the GTX 1050 Max-Q, Ice Lake is still slightly faster in this workload than the 10710U with six cores. I think a lot of it depends on better accelerated coding. If Premiere is your main workload, Ice Lake is the way to do better Warp Stabilizer performance and better coding performance.
Using our older Premiere benchmark, which is GPU-intensive, the gains aren't that significant, especially when compared to some of the discrete GPU options. For example, pairing a 10210U with an MX250 is much faster, but uses a lot more power. And Ice Lake falls behind Comet Lake with 6 cores compared to 4 cores if the GTX 1650 Max-Q is kept constant.
We are also seeing tremendous increases in performance at CompuBench Optical Flow, with a more than two-fold advantage over Gen11 graphics with older 14nm options. If the GPU is completely constrained in this way, it's no surprise that the much larger number of execution units on Ice Lake takes control and delivers a huge performance improvement.
The Smart Sharpen filter from Photoshop, which also runs on the GPU, offers similar advantages. Ice Lake's 130% faster performance compared to Comet Lake and other Skylake derivative CPUs is a huge win, and Nvidia's MX250 is easily surpassed by an integrated option. I suspect higher memory bandwidth for the GPU plays a role there, as we just saw that Ryzen performed well in Optical Flow.
What we have learned
All the data in, that's many hours of benchmarking that you just saw right there. We haven't seen too many surprises given all the information we had prior to this release. We didn't expect much in terms of CPU performance, but we expected decent improvements in GPU performance, and we've mostly seen that.
The Core i7-1065G7 offers roughly the same performance as the Core i5-10210U for multithreaded CPUs. Basically nothing was gained when comparing quad cores on the 10 nm and 14 nm nodes from Intel. The performance of a single thread is slightly higher and is around 10 percent. However, this is offset by the occasionally slower result with multiple threads. In general, it's fair to say that performance is roughly even.
The situation is similar with the Core i7-1065G7 and the Whiskey Lake Core i7-8565U. There are some bigger gains than expected for workloads like MATLAB and Photoshop Iris Blur, but in longer tests like Handbrake, Ice Lake actually runs slower. On average, the i7-1065G7 is a few percent faster, but it's not an overwhelming difference, and with these CPU-limited workloads, there's certainly no reason to upgrade parts of the 8th generation or newer versions.
When you compare Ice Lake with the best that Intel has to offer within 15W, the Core i7-10710U smokes the Core i7-1065G7 in multithreaded workloads. So if you want to use your ultraportable for something intense like video encoding, a Comet Lake six-core CPU is a better option. At the same time, Ice Lake is generally faster in single-thread tests. So it depends on what kind of things you do with your laptop to determine which processor makes more sense.
The margins do not change too much when comparing 25W configurations. More performance means more performance. In this case, Ice Lake in 25 W mode is about 20 to 25 percent faster than 15 W, but also Comet Lake. If you expected significantly better efficiency with these higher-performing goals, so far this doesn't seem to be the case on the mobile side.
On the other hand, Ice Lake is significantly faster if you need GPU acceleration. In pure GPU-limited situations, the unlocked Ice Lake Gen 11 GPU is more than twice as fast as the crappy integrated GPU we got from Skylake derivatives. And all within the same 15 W power envelope. With a mixed workload like Premiere, this can lead to huge performance improvements.
Overall, our first impressions of the Intel Ice Lake processor are mixed. There are a few positive aspects to testing the fastest available 15W configuration: faster single-thread performance, vastly improved GPU performance – but this is affected by problems in other areas.
For the new 10nm node from Intel, it is not important to improve multithreaded performance. Given that the CPUs we tested are all set to 15 W, this means that we do not achieve any performance increases of 10 nm compared to 14 nm, which also means no efficiency gains. Ice Lake appears to have improved the IPC considerably, but this has been completely offset by the CPU running at lower clock speeds. Lower clock rates, higher IPC, same performance, same performance.
In this way, Comet Lake can get started easily and offers a six-core CPU in the same performance shell to significantly improve multithreaded performance. Again, this doesn't apply to single-thread applications where Ice Lake has a head start, but Comet Lake is the way to go for everything that loads the CPU.
Granted, we haven't done any battery life tests with this new platform as it's almost impossible to get a good apple-to-apple comparison, but the same performance at the same level of performance shouldn't lead to major improvements. If anything, there are benefits from other platform benefits such as changes in the way power gating is handled, various boost technologies, more efficient storage, and so on.
Getting a much faster GPU with Ice Lake is nice, but it's also not a revolutionary upgrade as AMD has been offering this type of performance with Ryzen Mobile since late 2017. It's more about getting Intel to do its part on a competitive graphics standpoint. In some cases, the 64-unit Ice Lake GPU will still lag behind first-generation Ryzen Mobile, depending on performance limitations. In other cases, it is faster depending on how much memory bandwidth is required, because Ice Lake has a big advantage in this department.
Intel also has to compete with discrete GPU offerings, including Nvidia's popular MX150 and MX250, and newer, more powerful options like the GeForce GTX 1650 Max-Q that was included in our Razer Blade Stealth test system. This depends heavily on the exact workload, but Ice Lake's built-in GPU is not much faster with an MX250 than Comet Lake. The exception to this rule is Premiere, where a combination of factors makes Ice Lake a much better option.
The currently fastest hardware combination I've seen in ultra-portable form factors seems to be Intel's Core i7-10710U in combination with a GTX 1650 Max-Q, which you can get in MSI's Prestige 14, for example. If Razer I suspect the i7-1065G7 has been replaced with an i7-10710U in this latest Blade Stealth. It would be superior to most workloads, except for a few cases with one thread.
Although it's neat that Ice Lake has a much improved integrated GPU, as this doesn't improve CPU performance, we don't see any improvements over what was already possible with 14 nm. OEMs are increasingly opting for discrete GPUs in their ultraportables, and Ice Lake isn't doing much to improve these configurations. If you bought an 8th generation laptop with an MX150, Ice Lake is not a major upgrade. You'll see gains on laptops that don't use a discrete GPU. If an OEM decides that it only wants to include one chip, something like the Core i7-1065G7 will achieve a large gain in computing power compared to an 8th generation Core i7-8565U, for example.
The bottom line is a humble start for Intel's 10nm series because it lacks efficiency gains, which is likely to have to do with Intel's 10nm struggles. It is possible that with a 10 nm revision or just a full step forward to 7 nm this will improve significantly. Meanwhile, on the other side of the fence, AMD is working intensively on a next-generation Ryzen Mobile part on TSMCs 7nm.