A processor architect can battle between two major opposing principles. The one most of us seem to enjoy is performance, which when taken to the extreme exhibits an all-or-nothing approach. At the other end is low-power operation which has become the main focus of the laptop and notebook market where battery capacity and density is at a premium. The position in the middle of this is efficiency, trying to extract the best of performance and power consumption and provide a product at the end of the day which attempts to satisfy both.
Of course processor architects only have control up to the point where the chips leave the fab, at which point the final product design is in the hands of OEMs, who for various reasons will have their own product design goals. It's this latter point that has resulted in an interesting situation developing around the Core M ecosystem, where due to OEM design goals we've seen the relative performance of Core M devices vary much more than usual. In our tests of some of the Core M notebooks since the beginning of the year, depending on the complexity of the test, the length of time it is running and the device it is in, we have seen cases where devices equipped with the lowest speed grades of the Core M processor are outperforming the highest speed grade processors in similar types of devices, an at-times surprising outcome to say the least.
Never content to leave things alone, we wanted to take a look at the performance in Core M in depth and how device design - particularly cooling design - can significantly impact performance. So for today we will be diving deep into Core M, to see what we can test and what can be said about system design as a whole for the mini-PC, notebook, tablet and 2-in-1 ecosystem.
Core M
First, let us loop back to the design of Core M, which historically sits in the ‘Y’ processor stack and goes by the codename Broadwell-Y. Core M only comes in dual core flavors with Hyperthreading, with an official TDP of 4.5 watts. Each model comes with 4MB of L2 cache and Intel’s 8th generation of graphics architecture under the designation HD 5300.
HD 5300 is technically a 24 execution unit design, consisting of a major slice with three sub-slices of eight execution units each. This is double the GT1 / HD (Haswell) design where Intel enables only 12 units (which has benefits when it comes to enhancing yields), or half the full GT3 implementation which keeps the same front end but doubles the major slices. A full on 48-EU design looks something like this, although Core M only has one major slice.
The three main differentiators between each of the SKUs in the Core M line are the processor base frequency, the processor turbo frequency and the integrated graphics frequency range. A number of the processors also support cTDP Up and cTDP Down modes which adjust the base frequency of the processor only. Core M supports Turbo Boost Technology 2.0 which allows individual cores of the silicon to increase in frequency within specified parameters, which may include workload, estimated current/power consumption, and temperature. These two points are key to the rest of the article.
Core M Specifications
Model Number
5Y71
5Y70
5Y51
5Y31
5Y10c
5Y10a
5Y10
Cores/Threads
2 Cores / 4 Threads
CPU Base/Boost Frequency
1.2 GHz2.9 GHz
1.1 GHz2.6 GHz
1.1 GHz2.6 GHz
0.9 GHz2.4 GHz
0.8 GHz2.0 GHz
0.8 GHz2.0 GHz
0.8 GHz2.0 GHz
GPU Base/Boost Frequency
300 MHz900 MHz
100 MHz850 MHz
300 MHz900 MHz
300 MHz850 MHz
300 MHz800 MHz
100 MHz800 MHz
100 MHz800 MHz
TDP
4.5 W
4.5 W
4.5 W
4.5 W
4.5 W
4.5 W
4.5 W
cTDP Down(CPU Base)
3.5 W600 MHz
No
3.5 W600 MHz
3.5 W600 MHz
3.5 W600 MHz
No
No
cTDP Up(CPU Base)
6.0 W1.4 GHz
No
6.0 W1.3 GHz
6.0 W1.1 GHz
6.0 W1.0 GHz
No
No
Intel vPro
Yes
Yes
No
No
No
No
No
Core vs. Atom
In general, Core M is a small departure from Intel (pun intended), bringing its main Core processor architecture, typically used for big core performance, down to power levels and chip sizes better suited for fanless laptops and tablets. Despite the drop in core counts and frequency to reach 4.5 watts, the Core M line typically has a single threaded performance advantage at this power segment as compared the competition, which is no simple achievement.
For Intel, this 4 to 5 watt TDP window for processors has typically been occupied by the Atom line of integrated SoCs. In 2014 this meant Silvermont cores in a Bay Trail system produced at the 22nm process node, but for 2015 will mean Airmont cores in a Cherry Trail system at 14nm. For most of 2014, Atom competed against high powered ARM SoCs and fit in that mini-PC/tablet to sub 10-inch 2-in-1 area either running Android, Windows RT or the full Windows 8.1 in many of the devices on the market.
Despite Intel’s initial long cadence with Atom, we are seeing them step up to the plate and provide an iterative cycle that allows for the latest node technologies combined with the updated graphics technology from the integrated Core ecosystem. Nevertheless, Intel has split this 4 to 5 watt TDP segment into two clear formats based on performance and price.
Atom sits at the lower price band ($50-$100 per chip), typically in a dual or quad core arrangement without hyperthreading and uses ‘modules’ of two discrete cores sharing an L2 cache. The integrated IO is designed to be enough for this market segment, as seen in the recently announced Surface 3, and shows that devices in the $500 region are ripe for the next Atom SoCs. Note that Braswell, which also uses Airmont cores at 14nm but goes under the Celeron/Pentium nomenclature, also sits in this 4 to 6 watt region but is aimed more at the mini-PC arena.
Intel lists all of the Core M processors at $281, and a user will be hard pressed to find a Core M device priced under $700 on the market now; such is the gap that Intel wants to strike with the two platforms. Core M sits at the heart of the new Macbook (read our hands on), as well as most of the devices in this test such as the ASUS UX305, Lenovo Yoga 3 Pro and the Dell Venue 11 Pro 7000. With the Broadwell architecture and hyperthreading under its belt, the results do speak for themselves as Core M attacks the Haswell-U line from the last 18 months in terms of direct performance.
When Intel put its plans on the table for Core M, it had one primary target that was repeated almost mantra-like to the media through the press: the aim for fanless tablets using the Core architecture. In terms of physical device considerations and the laws of phystics themselves, this meant that for any given chassis temperature and tablet size and thickness, there was an ideal SoC power to aim for:
Core M is clocked and binned such that an 11.6-inch tablet at 8mm thick will only hit 41°C skin temperature with a 4.5 watt SoC in a fanless design. In Intel's conceptual graph we see that moving thinner to a 7mm chassis has a bigger effect than moving down from 10mm to 8mm, and that the screen dimensions have a near linear response. This graph indicates only for a metal chassis at 41°C under 25°C ambient, but this is part of the OEM dilemma.
When an OEM designs a device for Core M, or any SoC for that matter, they have to consider construction and industrial design as well as overriding performance. The design team has to know the limitations of the hardware, but also has to provide something interesting in that market in order to gain share within the budgets set forth by those that control the beans.
This, broadly speaking, gives the OEM control over several components that are out of the hands of the processor designers. Screen size, thickness, industrial design, and skin temperature all have their limits, and adjusting those knobs opens the door to slower or faster Core M units, depending on what the company decides to target. Despite Intel’s aim for fanless designs, some OEMs have also gone with fans anyway to help remove those limits, however it is not always that simple.
The OEMs' dilemma, for lack of a better phrase, is heat soak causing the SoC to throttle in frequency and performance.
How an OEM chooses to design their products around power consumption and temperature lies at the heart of the device's performance, and can be controlled at the deepest level by the SoC manufacturer through implementing different power states. This in turn is taken advantage of in firmware by the OEM on the motherboard that can choose to move between the different states through external analysis of battery levels, external sensors for temperature and what exactly is plugged in. Further to this is the operating system and software, which can also be predefined by the OEM by add-ins at the point of sale over the base – this goes for both Windows and OS X. More often than not, the combination of product design and voltage/frequency response is the ultimate play in performance, and this balance can be difficult to get right when designing an ‘ideal’ system within a specified price range.
To say this is a new issue would be to disregard the years of product design up until this point. Intel used to diffentiate in this space by defining the Scenario Design Power (SDP) of a processor, meaning that the OEM should aim for a thermal dissipation target equal to the SDP. In some circles, this was seen as a diversionary tactic away from the true thermal design power properties of the silicon, and was seemingly scrapped soon after introduction. That being said, the 5Y10c model of the Core M line up officially has a SDP of 3.5W, although it still has the same specifications as the 5Y10. Whether this 3.5W SDP is a precautionary measure or not, we are unsure.
For those of us with an interest in the tablet, notebook, and laptop industry, we’ve seen a large number of oddly designed products that either get very hot due to a combination of things, or are super loud due to fans as well as bad design. The key issue at hand is heat soak from the SoC and surrounding components. Heat soak lies in the ability (or lack of) for the chassis to absorb heat and spread it across a large area. This mostly revolves around the heatsink arrangement and whether the device can move heat away from the important areas quickly enough.
The thermal conductivity (measured in watts per meter Kelvin) of the heatpipes/heatsinks and the specific heat capacity (measured in joules per Kelvin per kilogram) define how much heat the system can hold and how the temperature can increase in an environment devoid of airflow. This is obviously important towards the fanless end of the spectrum for tablets and 2-in-1s which Core M is aimed at, but in order to add headroom to avoid heat soak requires fundamentally adding mass, which is often opposite of what the OEM wants to do. One would imagine that a sufficiently large device with a fan would have a higher SoC/skin temperature tolerance, but this is where heat soak can play a role – without a sufficient heat movement mechanism, the larger device can be in a position where overheating happens quicker than in a smaller device.
Examples of Thermal Design/Skin Temperature in Surface Pro and Surface Pro 2 during 3DMark
Traditionally either a sufficiently large heatsink (which might include the chassis itself) or a fan is used to provide a temperature delta and drive heat away. In the Core M units that we have tested at AnandTech so far this year, we have seen a variety of implementations with and without fans and in a variety of form factors. But the critical point of all of this comes down to how the OEM defines the SoC/skin temperature limitations of the device, and this ends up being why the low-end Core M-5Y10 can beat the high-end Core M-5Y71, and is a poignant part of our tests.
Simply put, if the system with 5Y10 has a higher SoC/skin temperature, it can stay in its turbo mode for longer and can end up outperforming a 5Y71, leading to some of the unusual results we've seen so far.
The skin temperature response by the SoC is also at the mercy of firmware updates, meaning that from BIOS to BIOS, performance may be different. As always, our reviews are a snapshot in time. Typically we test our Windows tablets, 2-in-1s and laptops on the BIOS they are shipped with barring any game-breaking situation which necessarily requires an update. But OEMs can change this at any time, as we experienced in our recent HTC One M9 review, which resulted in a new software update giving a lower skin temperature.
We looped back to Intel to discuss the situation. Ultimately they felt that their guidelines are clear, and it is up to the OEM to produce a design they feel comfortable shipping with the hardware they want to have inside it. Although they did point out that there are two sides to every benchmark, and it will heavily depend on the benchmark length and the solution design for performance:
Intel Core M Response
Low Skin/SoC Temperature Setting
High Skin/SoC Temperature Setting
Short Benchmark
Full Turbo
Full Turbo
Medium Benchmark
Depends on Design
Turbo
Long Benchmark
Low Power State
Depends on Design
Ultimately, short benchmarks should all follow the turbo mode guidelines. How short is short? Well that depends on the thermal conductivity of the design, but we might consider light office work to be of the same sort of nature. When longer benchmarks come into play, the SoC/skin temperature, the design of the system and the software controlling the turbo modes can kick in and reduce the CPU temperature, resulting in a slower system.
What This Means for devices like the Apple MacBook
Apple’s latest MacBook launch has caused a lot of fanfare. There has been a lot of talk based on the very small size of the internal PCB as well as the chassis design being extremely thin. Apple is offering a range of different configurations, including the highest Core M bin, the 5Y71, which in its standard mode which allows a 4.5W part to turbo up to 2.9 GHz. That being said, and Apple having the clout they do, it would be somewhat impossible to determine if these are normal cores or special low-voltage binned processors from Intel, but either way the Apple chassis design has the same issue as other mobile devices, and perhaps even more so. With the PCB being small and the bulk of the design based on batteries, without a sufficient chassis-based dispersion cooling system, there is a potential for heat soak and a reduction in frequencies. It all depends on Apple’s design, and the setting for the skin temperature.
Core M vs. Broadwell-U
The OEMs' dilemma also plays a role higher up in the TDP stack, specifically due to how more energy being lost as heat is being generated. But because Core M is a premium play in the low power space, the typical rules are a little relaxed for Broadwell-U due to its pricing, not to mention the fact that the stringent design restrictions associated with premium products are only present for the super high end. None the less, we are going to see some exceptional Core M devices that can get very close to Broadwell-U in performance at times. To that end, we’ve included an i5-5200U data set with our results here today.
Big thanks to Brett for accumulating and analyzing all this data in this review.
For today's article we have run a sampling of devices through several benchmarks which vary in workload substantially. Some are single-threaded and some are multi-threaded - some emphasize burst performance, and some focus on sustained performance. Some involve the GPU and some do not. During all of the benchmarks, CPU frequencies, GPU frequencies, and processor temperature were logged. The devices are all different as well, and offer different takes on Core M.
The first device is the Lenovo Yoga 3 Pro. This is the same device that we reviewed, and it features a Core M-5Y71 processor which is the very top of the Core M range. Lenovo has chosen to include a fan, so this is the only one of the Core M devices being included that is actively cooled. Being a convertible laptop, Lenovo must be more wary of surface temperatures than a traditional laptop since the Yoga 3 Pro can be used in the hand as a tablet.
The second device is the ASUS Zenbook UX305, which was recently reviewed as well. This features a Core M-5Y10 processor, which is the lowest-end model available. The UX305 is passively cooled and features an entirely aluminum chassis, which helps to dissipate the heat generated. As a laptop, higher surface temperatures can be manageable since the device is normally sitting on the raised feet and not in direct contact with skin.
The third device is the Dell Venue 11 Pro 7000, which also features the top end Core M-5Y71. This is a tablet first and foremost, and is also passively cooled. The Venue features a plastic rear casing, and as a tablet surface temperatures must be taken into consideration.
The final device is the Dell Latitude 14 7000, which is powered by the Core i5-5200U processor. Being a much higher TDP part, but largely the same architecture, will give a reference point on what Broadwell will do when given better cooling. The sample received has only one channel of memory, which will mostly affect the GPU scores. Dell does offer dual-channel memory, so this device can perform higher than the sample that we have.
Overall the Core i5-5200U is much less dynamic than Core M, with a base CPU frequency of 2.2 GHz and boost of 2.7 GHz. If you will notice, the boost is actually less than the Core M-5Y71, so assuming adequate cooling, or short enough workloads, Core M could in theory outperform the i5, which is something we did see on some benchmarks in the Yoga 3 Pro review.
The average computing day for anyone is going to be wildly different depending on what tasks they are performing. A lot of tasks however are very much burst workloads. As an example, browsing the web means loading the page, which is mostly done upfront. These kinds of workloads will play well into what Core M can do. Boost up to the maximum frequency, get the work done, and then fall back down to the base frequency and cool off. This is the epitome of Intel's hurry up and get idle philosophy.
However not every workload is like this. Gaming for example is a lot of consistent work, done over a long period of time, so cooling is the key here to keep performance up.
To sample a wide variety of workloads, I have picked a variety of benchmarks which are both short and long, do burst work or sustained work, and some involve the GPU and others do not.
Cinebench R15 Single-Threaded: This benchmark performs rendering on a single CPU core, so it should showcase higher clock speeds and good single-threaded performance. The benchmark lasts roughly ten minutes.
Cinebench R15 Multi-Threaded: The same benchmark, but the work is performed on all available cores, including hyper-threading. This benchmark is roughly three minutes.
PCMark 8 Home and Creative: Both the Home and Creative suites of PCMark 8 feature a variety of workloads. Home includes workloads for web browsing, writing, gaming, photo editing, and video chat. Creative includes web browsing, photo editing, video editing, group video chat, media transcoding, and gaming workloads. Home is around thirty minutes, and Creative takes about an hour to complete.
TouchXPRT 2014: This benchmark performs beautify photos (add filters, HDR, etc), blend photos, convert videos for sharing, create music podcast, and create slideshow from photos. Each task is timed, and a lower time results in a higher score. This benchmark takes about ten minutes to finish.
3DMark Sky Diver and Cloud Gate: 3DMark is a staple of our reviews. Both run through several graphics and physics tests which work both the CPU and GPU. Sky Diver is the more difficult of the tests. Sky Diver is about five minutes, and Cloud Gate is about three minutes.
3DMark Ice Storm Unlimited: This test is completely off-screen, and allows for comparison of the graphics across devices and even platforms. Being that it is available for smartphones and tablets, it is a much lower demand on the GPU, and completes very quickly on a PC with the entire benchmark being complete in about a minute.
DOTA 2: This incredibly popular online multiplayer battle-arena game is our final benchmark. This is the same workload performed for the DOTA 2 benchmark we have for reviews, only we run it for the full length of the recording. The entire run is around 45 minutes.
The following pages are very graph heavy, with some of the graphs being quite wide to show the sustained performance of the device over the benchmark run. Below is a gallery of all of the images, in order, which can be references as larger images in a separate window.
A note about the graphs. Each benchmark will show an entire run on each device, and then some combined graphs with the individual scores compared against the other devices. Due to the sampling rate, it may appear that some devices finished the benchmark before the others, but this is not always the case. Several of the devices were too loaded to always log to the text file, so they may have less entries, and appear to get the work done quicker if just comparing based on the time scale. The important data on the combined graphs is how each device handles the entire workload versus the others. We have also included the scores from each device to see where they finish the benchmark.
Cinebench will run the CPU up to 100% load for the duration of the test. As this is the single-threaded run, only one core will be active, which should in theory provide more headroom for that one core than when all cores (physical and virtual) are loaded. There is no burst workload here at all, and sustained single-threaded performance is the key for this test.
The Core i5 does exactly what would be expected for this benchmark. With just a single core loaded, the cooling system has no issues keeping the CPU from throttling. It maintains an extremely consistent CPU frequency during the run. This cannot be said of the two Core M-5Y71 devices though. The Dell Venue 11 Pro starts off with quite a high frequency, but as the temperature increases, the CPU drops in frequency to keep below the threshold of 90°C set on the SoC. At any opportunity, it increases its CPU frequency to try to increase performance, but generally that does not last for very long, and it ends up falling back down. The Yoga 3 Pro on the other hand, has a much lower allowed SoC temperature, with Lenovo locking in on 65°C as their maximum target temperature. This keeps the frequency down.
The ASUS Zenbook has an entirely flat CPU line though. The excellent heat dissipation of the chassis allows it to run for the duration of the benchmark with no throttling at all. It has to be noted though that the maximum CPU frequency is a quite a bit lower than the 5Y71 devices, topping out at 2.0 GHz versus 2.9 GHz for 5Y71. It would be very interesting to see how the UX305 would do with the faster CPU inside, and if it would run into throttling issues as well.
Looking at the average CPU frequency over the run shows that the i5 clearly has the most headroom, which is not surprising. Averages are only part of the story though, with both of the 5Y71 devices being able to jump past the 5Y10's frequency several times during the test.
Looking at temperatures, it's interesting to note that the Dell Venue 11 Pro has the top-tier Core M-5Y71, but it puts that processor in what is the smallest chassis and with a plastic exterior. Consequently it quickly loads up to its maximum temperature and stays there for the duration. The rest of the devices stay much cooler with just a single core loaded.
Here we have the actual benchmark results. On single-threaded workloads, the 5Y71 can and does outperform 5Y10. Despite the average CPU frequencies being lower on both 5Y71 devices, they had enough headroom when necessary to jump past the very consistent 5Y10. None of them can match the Core i5 in this test. It is actually very interesting that the highest scoring Core M in this test has the lowest average CPU frequency.
Looking at a multi-threaded run of Cinebench, the devices which will perform the best are going to need to have enough thermal headroom to keep all of the cores working at a good pace. All of these devices have four logical cores mapped to two physical cores via Hyperthreading, all of which are run at maximum load for the duration of this test.
The Core i5 once again has no issues maintaining its high CPU frequency, even though the overall SoC temperature does get higher than the single-threaded run. The Dell Venue 11 Pro tablet though starts off really reaching for the stars, but quickly must throttle back until it finds a consistent range that allows it to stay within its cooling constraints. The Yoga 3 Pro is similar, but quickly falls back due to the 65°C limit placed on the processor by the manufacturer. The ASUS UX305 performs just as well in this test as the last, with a very consistent CPU frequency, despite the temperatures getting a bit higher than the last run.
When it comes to average CPU frequency, both the Lenovo Yoga 3 Pro and the Dell Venue 11 Pro once again end up falling behind the ASUS and its much lower turbo speed in this test, though not by a huge margin. The ASUS averages the highest CPU frequency of the Core M contenders just like in the single-threaded workload, with the Lenovo less than 100MHz behind it, and the Dell Venue a ways back again. Neither of the 5Y71 devices turbo much over the 5Y10 in this test though.
Looking at the temperatures, you can see just how conservative Lenovo has been with the Yoga 3 Pro. The overall SoC temperature is quite a bit lower than all of the other devices when the device is under load. The active cooling and low SoC temperatures help the Yoga 3 Pro to keep a cool exterior to the device.
Now we come to the end result of this workload. The 5Y10 device handily outperforms both of the higher ranked models. Unsurprisingly it comes no where near the Core i5, but looking at the CPU frequency graph really demonstrates why it scores higher. Both of the 5Y71 have a lower average score, but unlike the single-threaded result, neither of them can sustain a CPU frequency past the frequency of the ASUS very much.
PCMark 8 Home is a much different workload than Cinebench. Cinebench thrives on sustained performance over the duration of the workload, with the CPU utilization staying around 100% for the duration. While an important metric, most people do not use their computers like that in their day to day lives, so Futuremark has crafted the PCMark suite to perform tasks which are more akin to what the average person will do in a day. Home includes workloads for web browsing, writing, gaming, photo editing, and video chat, and the nature of these loads mean that there is a lot more burst performance needed, so the race to sleep mentality of the Core M can be more effective in this scenario.
The burst nature of this benchmark is apparent just looking at the Core i5. No longer is the CPU frequency consistent across the board, and the temperatures ramp up and down as the work is performed and finished. Even more pronounced is the Dell tablet, which spikes up and down from its maximum temperature, but at the same time ramping clock speeds up quite high as well. The incredible cooling of the ASUS UX305 passive solution makes a big difference here, with the UX305 being able to maintain almost its maximum frequency for the duration of this benchmark. The Yoga 3 Pro really shines here though, with it maintaining quite high speeds for almost the entire duration of the benchmark.
Average CPU frequencies on the other hand show an unexpected disparity between the results we saw above and what the averages end up being. It's the cool Yoga 3 Pro that holds the highest average clockspeeds, followed by the UX305, and finally bringing up the rear is the Venure 11 Pro 7000.
The GPU averages for the three Core M devices are very similar overall, although none are at their maximum. Only the 15 watt Core i5 can maintain its maximum GPU frequency for the duration of this test. As we will see later, GPUs can draw a lot of power.
Moving on to temperature, with the burst nature of this benchmark, all of the devices have a reasonable time to cool off between workloads. The ASUS shows its amazing cooling capabilities again, with a significantly lower temperature than even both of the active cooled devices, but none of them are too close to their maximum allowed temperature over the duration.
Looking at the end result of this benchmark kind of throws everything we have seen in the above graphs on its head. The Yoga 3 Pro, despite sustaining a CPU frequency higher than all of the other Core M devices in this test, ends up scoring the worst, however the overall result by the Yoga 3 Pro is disadvantaged in this benchmark by the gaming test, due to the high resolution display on the Yoga 3 Pro. This is very similar to the results seen in the Dell XPS 13 review, where the QHD+ model only scored 2691 and the FHD model scored 3042 with the same processor. However the ASUS UX305 beats the other Core M devices, although it does so with a much lower resolution display than the Yoga 3 Pro which would certainly beat it otherwise.
The Creative suite for PCMark changes the workloads out a bit, and is overall a much longer benchmark. Creative includes web browsing, photo editing, video editing, group video chat, media transcoding, and gaming workloads, so like the previous test the higher resolution of the Yoga 3 Pro will bring its scores down compared to the 1080p of all of the other devices. Like PCMark 8 Home, the work features high demand followed by low demand.
The Core i5 performs much the same as during the Home benchmark. Clearly the cooling system which is designed to get rid of 15 watts of heat can pretty easily cope with these types of workloads, and it even allows the CPU to turbo quite often to the CPU’s maximum speed of 2.7 GHz. GPU workloads are also no issue for the cooling system. None of the 4.5 watt TDP devices fare so well though, and as with the Home benchmark we see the Yoga 3 Pro having quite good CPU and GPU frequencies. The Dell is limited quite a bit more on temperature, and the ASUS is limited by its lack of maximum turbo frequency.
Looking at the average CPU frequency tells a big story of thermal limitations on the Dell Venue 11 Pro. With an average that is barely over its base of 1.2 GHz, the device spends a significant amount of time below its base frequency. The UX305 keeps its consistency high, with it almost reaching its maximum turbo clock, while the Yoga 3 Pro ends up quite a bit faster.
The GPU story has the ASUS bumping into its maximum GPU speed of 800 MHz, keeping pace with the Dell Venue 11 Pro which can go as high as 900 MHz. The Yoga 3 Pro is over 800 Mhz, here, showcasing its active cooling solution and meaning it is certainly spending time closer to its 900 MHz GPU turbo.
Looking at average temperatures, it is obvious why the ASUS can stay close to its maximum turbo on many workloads. After this one hour benchmark, the CPU average was just over 56°C despite the passive cooling. The active cooled Broadwell-U laptop is a bit higher, and the Yoga 3 Pro kept its average under its 65°C CPU maximum.
The Lenovo Yoga 3 Pro scores the highest of all three Core M devices in this benchmark despite the higher resolution display. On this type of workload where the actual work is much shorter, it keeps its CPU frequency much higher than all of the other Core M devices. The Venue 11 Pro also outscores the UX305 despite its low average CPU frequency. When needed, it was able to turbo well past the 2 GHz maximum of the 5Y10 device.
TouchXPRT performs several tasks, and the workloads very much fall into the race to sleep category. There are several workloads, from adding filters to photos, to creating podcasts. The benchmark takes about ten minutes to complete, but each workload is slightly different.
Looking at the Core i5 graph makes it very obvious where the heaviest lifting is in this benchmark, but even that one is full of bursts of work. The Dell Venue tablet is able to hit a very high frequency for many of these tasks, since it has adequate time to cool off in between. The ASUS is as consistent as always, and the Yoga 3 Pro can really stretch its legs on this benchmark.
Looking at the average CPU frequency, the Yoga 3 Pro beats out the other Core M devices by a lot, and even turbo higher than the Core i5 on many occasions. If the work is short, the higher burst frequency of the 5Y71 can make a big difference. This would be very similar to web browsing, where short bursts of work get the job done. The Dell has the lowest average CPU frequency again, but as we have seen in previous results the fact that it can hit a much higher frequency than the ASUS can help it regain ground, especially on a short workload such as this one.
The GPU average frequencies show the disadvantage of the 5Y10. It is quite a bit under the 800 MHz turbo frequency of the chip, and it cannot turbo to the 900 MHz of the other two chips when it does have thermal headroom to make up the difference. The other two devices can be seen to jump all the way up to the 900 MHz maximum many times.
Temperatures are low, and the ASUS is the lowest again. The Yoga 3 Pro is sitting right at the 65°C target temperature, which means it was not always able to keep within that target during these quick bursts of energy needed.
TouchXPRT is almost a perfect workload for 5Y71, and the Yoga 3 Pro outperforms even the Core i5-5200U in this test. The Venue 11 Pro also comes in right at the score of the Core i5. The ASUS UX305 is certainly hampered by its lack of turbo compared to the other devices in this test. It had the lowest average temperature, and it could not do anything with it.
Most of the previous benchmarks were day to day tasks. Some involved the GPU, but it was never the focus. We will now move on to benchmarks which focus on GPU performance to see what kind of an effect this can have. Remember, the TDP of Core M is 4.5 watts including the integrated graphics, so any thermal room needed for graphics is going to come at the expense of the CPU. 3DMark Sky Diver is aimed for gaming laptops and mid-range PCs, so it is a bit too much load for integrated graphics. But it does feature DirectX 11 and includes both graphics and physics tests. The benchmark is around five minutes long.
We can see that the Core i5 continues shrugging off these tests. While the SoC did heat up, the GPU frequency was flat throughout the results. This is quite a bit different than all of the Core M processors, which had to throttle both the CPU and GPU as needed. It is very interesting especially in the UX305 results to see that the GPU is throttled on high CPU workloads to give more headroom for the CPU, which you can see on the third heat spike in its graph. This would be the physics test, which relies heavily on the CPU. The Dell Venue 11 and Yoga 3 Pro had very different temperature curves, and the Yoga 3 Pro had to throttle the GPU quite a bit to stay at its target SoC temperature.
Looking at the average CPU frequencies reaffirms what we have seen in previous results. The ASUS, despite having the lowest turbo frequency, has the highest average for the Core M devices. But it is the GPU frequencies which are the most important in this test.
All of the Core M devices had to throttle the GPU to some extent, but the ASUS did the least. The Yoga 3 Pro and Dell Venue 11 Pro were basically tied in average GPU frequency for the duration of this test. GPU workloads can pull a lot of power into the SoC, which can raise temperatures as we will see in the next graph.
Looking at the SoC temperatures explains the results. The Yoga 3 Pro has an average of 65.2°C, which is the target temperature for the Yoga. This means it was not able to leverage the breaks in workloads to ramp up its higher turbo frequencies when needed. The Dell Venue 11 is at almost 90°C for the benchmark, and that is also its limit. The ASUS, with its better cooling, manages to basically mirror the Core i5 for SoC temperature.
The excellent cooling of the ASUS form factor shines in the GPU tests. For the overall score, it comes very close to the Core i5. Both of the 5Y71 devices struggle under sustained GPU workloads, as the scores confirm.
3DMark Cloud Gate is a benchmark aimed at notebooks and home PCs, and is quite a bit less demanding. It has a DirectX 11 engine but is limited to Direct3D feature level 10, and is compatible with DirectX 10 hardware. The overall run is about three minutes.
There is not much more to be said about the Core i5 at this point. It does an admirable job keeping the GPU frequency almost flat during this benchmark. You can clearly see the Dell Venue 11 Pro ramping up frequencies on the CPU, which cause temperature spikes when this happens. When it throttles the CPU on this workload, it does free up enough thermal room to allow the GPU frequency to be fairly strong. We see a lot of throttling on the ASUS as well, but not quite as pronounced. Once again, on the physics test the GPU is pushed down in frequency to give the CPU more room. The Yoga 3 Pro tries its best but is once again limited by a much lower SoC temperature set point.
On the CPU side, we have a very similar situation to the Sky Diver benchmark. The ASUS once again keeps a higher average CPU frequency than all of the other Core M devices in this test. The Venue 11 is close though.
On the GPU side, the Zenbook and Venue 11 Pro are basically tied. The shorter and less demanding workload lets the Dell keep up despite not having as good of a cooling solution. But, averages are just averages. Clearly the ASUS keeps a substantially higher GPU frequency for much of this test, as is seen in the graph.
The SoC temperatures are actually quite high on the Zenbook in this test, with it coming close to the Venue 11 Pro, but the cooling system clearly is more efficient since the change in temperature on the ASUS is much more gradual than the spikes seen in the Venue 11 Pro. The Yoga 3 Pro tries to stay around 65°C but near the end the temperature does go above their target.
The overall benchmark results for this test are very similar to the previous 3DMark test. The ASUS comes in very close to the Dell Latitude with its Core i5, and the other devices fall back quite a ways. Long sustained GPU workloads are very difficult for both of the 5Y71 devices to handle.
Ice Storm Unlimited is quite a bit different than the last two benchmarks. The test is built for smartphones and tablets, so is far less demanding than the other GPU benchmarks. There are two GPU tests, and a physics test, and as you will see in the graphs, when those workloads are occurring is very obvious. The overall benchmark is quite short though, which allows the devices that have more thermal issues, but higher overall turbo frequencies, to keep the frequencies up much more. It is basically the equivalent of a CPU burst workload, except mostly run on the GPU.
The Core i5 does not even flinch at this workload, even leveraging its turbo when needed. The Venue 11 Pro is the most interesting graph because it so clearly defines when the actual work is happening. Because the duration is so short, it is able to turbo quite high, and the GPU frequencies are not throttled too much. The ASUS does have to throttle the CPU to keep the GPU frequency up on this test. The Yoga 3 Pro shows quite a strong result in this very short test.
Looking at the average CPU speeds, the Yoga 3 Pro jumps way out in front. The Venue 11 Pro is quite far behind, but as you can see in the graphs, when the work was required, it did have thermal headroom available to turbo.
On the GPU front, the Yoga 3 Pro is almost at the same average as the Core i5 in this test, as both have the same base and turbo frequencies. The Venue 11 is only a bit behind, and the ASUS falls to third due to the 100 MHz frequency deficit that the 5Y10 has on the GPU compared to the 5Y71 processor.
On the SoC temperature side, none of the devices struggle with temperature on such a short test.
On such a short test, the Core M devices all do very well, and the fastest Core M model in the Yoga 3 Pro tops this GPU test. It is quite a bit in front of the rest of the devices, showing that with active cooling, it can still get a lot of work done in a short amount of time. Remember that the Core i5 Dell Latitude is the only device with single-channel memory, which hurts it most in the GPU tests and explains why it is below the Core M devices despite much higher average frequencies for both the CPU and GPU.
DOTA 2 is a multiplayer battle arena game, and for this test we are using the same setup as our Mainstream benchmark, but this time with a full game. At 1600x900, all of the devices should be around 30 fps, and the overall test is about 45 minutes.
The Core i5 once again does a great job throughout this test. The CPU frequencies are dropped to keep the GPU running at full speed. The GPU basically runs at full speed for the duration of this test. The Venue 11 Pro is not so lucky, with it quickly heating up and being forced to throttle both the CPU and the GPU. The ASUS continues its amazing run, and showcases what can happen with a good passive cooling solution. The Yoga 3 Pro is not so lucky, with that pesky 65°C set point rearing its ugly head, which causes a big drop in overall frequency on such a long sustained workload.
The average CPU frequency for this sustained real world gaming workload has even the Core i5 having to give up some CPU headroom to keep the GPU fed with power. The ASUS has a sizable advantage here, and both the 5Y71 devices drop well under their base 1.2 GHz CPU frequency when the GPU is running at maximum.
The GPU is really the story though, since this is a gaming workload. Amazingly the ASUS is only 100 MHz off of its maximum turbo frequency as an average for this 45 minute workload. Both the Dell Venue 11 Pro and the Yoga 3 Pro do not have enough cooling to keep these kinds of sustained GPU loads going.
The Yoga 3 Pro is by far the coolest SoC in almost all of these tests, with its combination of active cooling and a 65°C maximum SoC temperature. The ASUS is far and away the hottest device in this test, but it also does a lot more work than the other Core M devices, and it is not getting any hotter by the end of the test, so the device cooling is doing its job.
It is clear at this point that the ASUS can keep the GPU frequency much higher than the other Core M devices due to the nature of its cooling, and form factor. The DOTA 2 test is really dominated by it. It is much faster in this test than the other Core M devices, and once again due to the single-channel nature of the Core i5, the ASUS even outperforms the Core i5 in this test.
There is a lot to say about Core M performance. We have tested three very different devices, all with Core M inside - each device tackles the design philosophy of Core M from different directions, and it comes across in the results based on where each device stands. We started this analysis to answer the question "how can 5Y10 beat 5Y71?', and the results answer that quite clearly.
Dell Venue 11 Pro 7000 (5Y71)
Starting with the Dell Venue 11 Pro 7000, this is an 11 inch tablet with passive cooling and a small chassis. The plastic exterior helps with skin temperature, but hinders the ability of the device to radiate the heat that it generates.
This chart shows where Intel believes Core M can be done in a passivly cooled device, and it assumes a metal chassis which the Venue 11 Pro 7000 lacks. Device thickness, material, and chassis size all play a big factor in how much heat can be dissipated. Under sustained use, the Venue 11 Pro can get warm to the touch, but as seen in the previous pages the actual SoC temperature can spike very rapidly. This compromises performance, although everything is relative. Compromised Core M CPU performance is still quite a bit more powerful than a Silvermont core in an Atom.
Lenovo Yoga 3 Pro (5Y71)
The Yoga 3 Pro on the other hand is a convertible tablet, and is the only Core M device in our test with active cooling. Lenovo has gone to great lengths to ensure that it does not get hot in the hand, since it most certainly can and will be used when held. It has the lowest SoC temperature of any of the devices by quite a bit, although it is of course helped by the inclusion of a fan. Lenovo has clearly set a target SoC temperature of 65°C as the maximum they are comfortable with in order to keep skin temperatures where they want them. It does not really hurt the device in all workloads, and as we have seen the Yoga 3 Pro can even outperform a Core i5-5200U in cherry-picked scenarios. On sustained maximum performance though, the lower SoC temperature means that the CPU and GPU must cut back sooner than the other devices, which limits performance.
ASUS Zenbook UX305 (5Y10)
The ASUS Zenbook UX305 is a completely different device. It is a laptop, so skin temperatures are not as big of a concern, and while it does get hot at around 48°C on long workloads, that heat is well away from where you would normally be touching the device. Also, being made out of aluminum is clearly a huge benefit for a passively cooled device such as this, as it allowed the ASUS to consistently outperform the other Core M devices despite it being the lowest boost frequencies in the test. The performance of the Zenbook was very consistent, even on extended workloads, and on the DOTA 2 test, the Zenbook even outperformed the Latitude, although that was less to do with thermals and more to do with the lack of memory bandwidth on the Dell laptop.
Core M
So with all of this data, what more do we know about Core M? Clearly, Intel’s goal with Core M is to provide excellent performance on short workloads. It has higher boost frequencies than the Core i5-5200U that was included in this test, and it has 4 MB of L3 cache as well compared to 3 MB of cache on the i5. On certain workloads, performance can even surpass the i5-5200U. Race to sleep is not a new idea, but that is what Core M is designed to do, and it does it well. Run on very little power, and then when tasked with work, get it done as quickly as possible and get back to the low power mode. This is not unique to Core M of course, as the Core i5 does the exact same thing, however the much more restrictive thermal envelope of just a 4.5 watt TDP means that sustained workloads just have to suffer compared to a device with a 15 watt TDP.
We did see that happen too. The Core M scored very well in the PCMark 8 benchmarks, which attempt to emulate real world use rather than just performing a single task until it is done. The burst nature of this allowed Core M to have enough time in between work to keep the temperatures in check. On sustained work, this was not always the case.
On the 5Y71 vs. 5Y10 front, we have some clear lines drawn:
Intel Core M Performance
5Y10 Result
ASUS UX305
5Y71 Result
Lenovo Yoga 3 Pro
5Y71 Result
Dell Venue 11
Pro 7000
Cinebench R15 ST
82.14 pts
90.85 pts
86.00 pts
Cinebench R15 MT
210.66 pts
196.00 pts
175.00 pts
PCMark 8 - Home
2655 pts
2443 pts
2606 pts
PCMark 8 - Creative
3056 pts
3110 pts
3064 pts
TouchXPRT 2014
654 pts
820 pts
764 pts
3DMark Sky Diver
2773 pts
1624 pts
1820 pts
3DMark Cloud Gate
4251 pts
3685 pts
3753 pts
3DMark Ice Storm
47527 pts
49619 pts
44911 pts
DOTA 2
34.0 fps
26.5 fps
25.2 fps
It also shows just how much the individual device plays in how much performance is available. The ASUS has by far the best cooling solution of the three Core M devices, which is helped of course by the form factor. Lenovo could not get away with a 50°C surface temperature on a device that can be used as a tablet. It just would not work. The form factor of the Yoga 3 Pro plays against it on sustained workloads. The Dell Venue 11 Pro allowed a much higher SoC temperature, but it would also spike there very quickly. What we would really love to see is the Core M-5Y71 processor inside of the Zenbook to see just how much the increased boost of the 5Y71 compromises the performance on longer duration workloads, if it does.
Die Shot of a Core M Processor
Is Core M a good processor? Or is it slow? It is clearly slower than a Core i5, but it would be hard to expect it not to be. On many daily workloads, it performs very well. On things like web browsing we are already to the point where the Yoga 3 Pro outscores a Core i7-860 4C/8T 95 watt desktop CPU from 2009 in web benchmarks. Since Core M is mainly aimed at thin and light devices, it can be expected that these are the kinds of workloads that one would perform on them. Many companies have jumped onboard with the smaller processor, as it allows a thinner and lighter device, and the possiblity of having no moving parts. Core M enables this with a much lower TDP than the next tier of mobile processors from Intel, as well as a smaller overall SoC which is thinner as well.
Broadwell-Y (left) vs Broadwell-U (center) vs Haswell-U (right)
With the rise of tablets, the migration from hard desktops to smaller form factors has been unprecedented and Core M gives much more CPU performance than any tablet SoC available right now, at the expense of the additional cost the premium product brings. It is well suited to the types of workloads that many of us do during a typical day. There has been a lot of design wins already for this processor, covering a large range of device types and manufacturers. Even Apple has decided that there is a niche where a thinner and lighter version of their laptop may be of good use to their customers.
That said, there are a few other differences that set Core M apart from the mainline Core processors that need to be noted. For example, even though it shares the Gen8 Graphics with the Core i3 / i5 / i7 models in Broadwell-U, it is not quite as capable. It only has software HEVC decode for instance, which means that it might not be suitable for a Home Theatre PC without additional graphics helping out. Gaming is also a scenario where Core M can come up short, but only if you are comparing it to typical PC games. Tablet style games will have no issues, and Core M can perform similarily to other tablets in the GPU department.
In the end, we will quote one of Brett's favorite Formula 1 announcers and author, the great Steve Matchett. Everything is compromise. In Formula 1 racing, each corner of the track would need a different setup on the car for maximum results for that corner, and on the straights you would want most of the car’s downforce removed. Therefore every race and even every corner of a race is compromised for the maximum overall lap speed. The same can be said of Core M. In order to get something as powerful as the Core architecture inside of a fanless tablet, there is going to be compromise. In a Core M device, that is going to be sustained performance. What you give up in sustained performance though allows a thinner and lighter device, in form factors that would never have been possible with Core even one year ago. But it also means that the Core M SKU designation is only a sign of general performance, rather than absolute positioning. For that, we have to compare and contrast each unit in a review. Luckily, we hope to cover a large number of the important models over the next few months.