Intel’s Apollo Lake processors are set to hit the streets in the second half of 2016. They’re low-power, low-cost chips aimed at laptops, 2-in-1 tablets, and some desktop computers, and they’re the follow-up to the Intel “Braswell” processors we’ve been seeing over the past year in devices like Chromebooks, cheap Windows laptops, and mini-desktops.
So what’s the difference between Apollo Lake on Braswell? A few months ago Intel told us the new chips would offer better performance and efficiency. Now the company is providing some numbers.
The new chips are based on Intel’s Goldmont CPU architecture, which replaces the Airmont CPU technology used in Braswell (and Intel Atom “Bay Trail”) chips.
AnandTech forum member Paul Jackson posted a slide from an Intel event this week at Computex, showing what we can expect from the move to Goldmont:
- 30 percent better CPU performance
- 30 percent better graphics performance (with Intel Gen9 graphics)
- Support for DDR3L, LPDDR3, and LPDDR4 memory
- 15 percent longer battery life (all things being equal)
- Support for USB Type-C
The slide also confirms that Apollo Lake chips will be branded under the Intel Celeron and Pentium N and J series, much like today’s Braswell chips. There’s still no word on an exact launch date, clock speeds for the new chips, or other details that will really help us figure out whether we can expect to see a 30 percent performance improvement in real-world conditions.
Leave a Reply
So far behind ARM.
Atleast they are usefull, if they can clock them around 3.0ghz then we will have core 2 duo perfomance in less than 10w TDP.
Silvermont/Airmont is already equivalent to Core2Duo at near equivalent clocks. So a 30% improvement means they can provide that level of performance at lower clocks… and we’re talking less than 6W TDP…
No. Silvermont/airmont, clock for clock has horribly less single thread performance than core 2 duo. A 1.8GHz C2D is nearly 3x the single thread performance of a 1.8GHz silvermont core. A 30% boost does not even reach parity.
Nope, first you’re comparing a architecture that’s optimized for quad core versus a architecture optimized for no more than dual core… Silvermont/Airmont don’t support Hyper Threading, for example… But quad core performance makes up for this and puts them in parity… Second, these are no longer the days that single threaded performance is paramount anymore… Most apps and OS fully support multithreaded processing. Third, let’s not exaggerate the performance of the Core2Duo… The benchmarks today are not the same used to bench those old Core2Duo’s and there are other variables that didn’t exist back then either… You’re far more likely to get a SSD, with far better performance than those available when the Core2Duos were around… These ATOMs fully support QuickSync and a variety of other performance enhancing features that the Core2Duos didn’t have… Etc. So real world experience is better now compared to older hardware of apparently equivalent benchmark… Read more »
3x? No way
I check single thread perfomance of c2d l7500 with 1.6ghz and passmark score was 576 so CyberGusa was right c2d and braswell ipc was same. Apollo will be 30% faster.
http://www.cpubenchmark.net/cp…
Looks like 2x to me (z3735g @350 and c2d t7500 @765).
http://www.cpubenchmark.net/cp…
https://www.cpubenchmark.net/c…
The z3735g has a peak of 1.83GHz, and t7500 is 2.2GHz.
Take away 15% from the t7500 for being 15% faster and it still is 2x.
that cpu does not belong to braswell, and that core 2 duo has base clock of 2.2ghz thats not fair because it will always operate at 2.2ghz you need to compare base to base and not base with turbo
the c2d t7500 was released in Q3 2006. 10 years ago. They didn’t have turbo mode back then. The original 3x statement was about silvermont, not braswell. If the silvermont turbo doesn’t turn on during a benchmark, then I guess it will never turn on. You are absolutely right about it being not 3x, but calling them similar is dead wrong.
those baytrail soc are 2 to 4watts they cant sustain turbo at long time, they will just burst and return to clock speed,and those cpu, you did choose slowest baytrail (1.3ghz one) versus 2.2ghz c2d which is not fair at all and you cant measure ipc in that way.
atom z3795 – baytrail (4w) 1.6ghz
http://www.cpubenchmark.net/cp…
intel core 2 duo t5200 (34w) 1.6ghz
http://www.cpubenchmark.net/cp…
Yup, for some reason some people think comparing worst case with best case is somehow a relevant comparison that establishes how the entire platform compares to each other… Sticking to the unequal comparison, here’s another link that more directly compares the two for a little perspective… http://cpuboss.com/cpus/Intel-… There’s like a 21.5x difference in power efficiency per watt between them… and that link also shows the ATOM is still the better deal despite being a bottom of the range offering and despite the clock speed disadvantage not all the benchmarks are lower for it… Never mind, if anyone actually tried running a C2D in a 8″ Windows tablet or HDMI dongle PC stick it’ll probably throttle a lot, assuming it would even run, let alone fit as it was also a much larger multi-chip with no built in GPU… and let’s not get into how lousy Intel GMA’s were ten years… Read more »
Nice try. First, there are no z3795 notebooks ever made. Second, the turbo frequency is 2.4GHz. I will admit that I selected the fastest c2d cpu (it has the most cache). I will accept the t5200 as the c2d sample with passmark 531, but that still needs to be compared to z3735g with passmark 350. Not quite double, but more than 50% faster. A 10 year old c2d laptop is much faster than a baytrail netbook in single threaded performance.
Again why you want to compare 1.3ghz with 1.6ghz one? Just choose 1.6ghz baytrail and compare them. If you have it then disable turbo clock and see if score will be 350 or more
Don’t even need to, the actual average passmark score for the Z3735G is 916… http://www.cpubenchmark.net/cp… I don’t know where they got 350 from but it’s either made up or from a really badly made system that was throttling badly or it’s not even from the model in question… I might believe the 350 score if it was from a Clover Trail but not a Bay Trail or later… Though, a Quark based Intel product like the Curie might be in that range too but that’s a IoT range product… Well, the Z3735G does have a low single core score but this is a SoC optimized for mobile usage. Silvermont is optimized for multi-core performance as that’s more energy efficient. It’s also one of the reasons why they dropped Hyper Threading for Silvermont, as previous ATOMs had it, as it was a trade off for improved energy efficiency as it let… Read more »
Nope, the Z3735G has a average Passmark score of 916! http://www.cpubenchmark.net/cp… Versus 836 for the C2D T5200… http://www.cpubenchmark.net/cp… Mind, you’re comparing a quad core to a dual core… and the Silvermont architecture is optimized for multi-core performance because it has to work with less than 4.5W TDP… Even when all the cores are active and running at full clock they’re only drawing about 2W for four cores combined The Z3735G isn’t a laptop SoC either, it’s a mobile SoC for the tablet and similar device markets! It’s specifically a Bay Trail T, which means tablet range SoC… The Bay Trail M and D series are the ones that went into Laptops under Celeron/Pentium branding… The Z3735G is also bottom range, it’s one of the lowest end Bay Trail’s meant for dirt cheap products… While the Z3795 did make it into 2 in 1’s like the Asus Transformer T200TA… but for… Read more »
YOU GUYS MUST BE SNORTING JENKUM IF YOU THINK THAT A BRASWELL IS ANYWHERE CLOSE TO A CORE 2. I REGULARLY USE AN OLD CORE 2 GENERATION PENTIUM LAPTOP AND IT IS WAY FASTER THAN MY HP STREAM, LIKE WAAYYY FASTER
First, it’s Core2Duo, not Core 2… There’s no such thing as a Core 2…
Second, you’re comparing to a dirt cheap model that’s way slower than a top of the line ATOM… HP Stream has too little RAM and too slow a eMMC that just cripples whatever performance the SoC could provide as well… There’s a reason they call them Cloudbooks!
Something like the Surface 3 is closer to expected performance, with some high end Braswell/Bay Trail offering a bit more… Especially when coupled with at least 4GB RAM and a real SSD instead of eMMC…
Want to really compare you HP Stream to your old Core2Duo then equip your old system with a OS that boots off a SD card (eMMC=embedded Multi-Media Card and shares more in common with a SD than a SSD), and only 1-2GB of RAM and try running W10 off of it… Good Luck!
They keep the perf way low to protect Core M where they have insane margins . Found the new Atom on Geekbench , here vs the new dual core low end AMD that is higher power but at least offers a lot more in single core perf for cheap https://browser.primatelabs.co…So in single core you got 1.5k in integer, 1.1k in FP. A72 does 2.2k in integer and 1.8k in FP. Qualcomm’s Kryo does about 2.1k in both for the higher clocked cores in SD820 but it is less efficient than A72. A73 on 16ff will bring a bit more perf soon and next year on 10nm some 30% more than A72 on 16ff and that means more than 2x faster than this Atom core in Geekbench. We could see midrange SoCs with 2xA73 at high clocks plus 4xA53 on 14/16ff next year, enabling some really fast midrangers. No wonder Intel… Read more »
I lost confidence on Geekbench when their score show A9x in ipad pro is faster than haswell i5 and core m,
Then anandtech did test using spec2006 and core m completely destroy A9x in both power consumption and perfomance.
From your link, it looks like Apollo Lake may be just a bit behind Stoney Ridge in IPC. The Stony Ridge part is listed at 1.75ghz and the Apollo Lake part is listed at 1.1ghz. Those ARM chips you mention are all 2.0ghz+ parts. If they couldn’t outperform a 1.1ghz part, there’d be something seriously wrong with ARM. 😉
Yup, hard to trust leaked benchmarks as well… Benchmarks may not be optimized for the new chips, the product being tested may not be a optimal or equivalent configuration, each architecture may have different advantages and disadvantages, etc.
For example, this link shows the A9-9410 in a less advantage light vs Braswell’s that are more competitively clocked…
http://laptopmedia.com/highlig…
This will probably go back and forth a lot before we get the final real numbers we can actually really judge/compare…
I’m with you. I can speculate all I want about Apollo Lake’s performance, but until I can have a sample in my hands to test, speculation is all it is.
Geekbench is perhaps the worst benchmark to compare processors of different architectures due to their having client applications for all the OSs out there, comparing a windows result to an android or ios result is relatively meaningless, due to significant OS operational difference. Perhaps best to compare tests running them all from the same full-fat linux distro.
Also, notice the registered clock speeds for A9-9410 in geekbench. Wikipedia states the processor starts at 2.9ghz and turbos up to 3.5 at 10-25W, which is a fairly broad TDP range.
ARM means no Windows / x86…
and that’s a bad thing how??
Not necessarily a bad thing but for some it means a lack of flexibility, more limited range of software you can get to run. Lack of legacy support as most ARM devices have rapid end of life and lack long term support… Not all ARM devices have as reliable history of support for Linux as Intel does, etc…
Any word on Apollo Lake NUCs? Last year there were roadmap and spec leaks for a few months prior to release IIRC
Nothing official yet… Mind, last minute changes may be requiring a change to their planned lineup…
Apollo Lake wasn’t originally suppose to overlap Broxton for Tablets but with Broxton Cancelled that means Apollo Lake now has to cover Tablets too…
I’m curious to see the first HEVC benchmarks. I hope there is at least one model with HDMI 2.0
ALL I KNOW IS BRASWELL IS A STINKER, LIKE DIARHEA THAT DRIPS FROM UR BUT.
will run linux at least, if you look at the picture on stage.
Yes, by the time this is out it should work with Linux just fine… Intel generally has good Linux support, at least for their latest products.
More a question of what else will be in the system… Like the WiFi chip may be from another company with not good Linux support… So still wait for some reviews in that regard…
When they say 30% faster that’s a complete joke they just lowered the speed of Braswell by 30% relative to the 2850 so if you increase the speed of Taswell by 30% you get the 2850 again
Before you get all excited consider that with Boswell they lowered the clock speed by 30% relative to the 2840 so if the new chip is 30% faster than Braswell you get the 2840 again
I think you meant Braswell, but they didn’t really lower clocks… let alone by 30%… The Airmont architecture is essentially the same as the Silvermont architecture, except for the GPU. The primary purpose of the Airmont release was to help close the massive gap in GPU performance with the competition and thus that’s what they focused on… This left the CPU side essentially unchanged, but they had to deal with a much more powerful GPU that would generate more heat and use more power in a SoC that was still intended to be used in mobile devices. So they didn’t lower clocks, they just didn’t raise them and left them at base level for the first releases… But much of the base clocks were still higher than Bay Trail, the burst clocks were just not as high with a initial max of around 2.4GHz instead of the originally expected 2.7GHz….… Read more »
Base clock of 2840 is 2.1thx. Base clock of 3050 is 1.6ghx
The predecessor to the N3050 is the N2815, not the N2840! The N2815 has a base clock of 1.86 GHz, with a burst max of 2.13 GHz vs the N3050’s 1.6-2.16 GHz range… along with about the same TDP… But the N2840 operates at a higher TDP of 7.5W and a SDP of 4.5W vs the N2815/N3050’s 6W TDP and 4W SDP… So they aren’t equivalent… You’d have to go quad core to see any of the Braswell’s go up to even 6.5W TDP… but a Braswell can still have pretty good burst clocks and unlike the Core’s Turbo Clocks, Burst clocks can be maintained almost indefinitely as long as the SoC doesn’t overheat and throttle. So the lower base clocks are more for power efficiency… The J3710 is a quad core with a base clock of 1.6GHz but it can go up to 2.64 GHz and is one of… Read more »
There’s no way I’d buy a cheap Windows laptop with only 32gb of storage.
Minimum storage is probably going to go up along with RAM… We’ll see but there’s a lot less reason for them to stick to 32GB or less unless it’s for a Cloud based product like Chromebooks that they don’t intend for you to put too many local files on…
Educational market and people close to the poverty line. Just like braswell, just like baytrail.
It’s not just for education and people close to the poverty line, but otherwise yes… This is just a update that brings in some overdue improvements to this product range.
So better but doesn’t cross into other product ranges…
Would also make for a nice CPU for a software based network appliance, a sub-segment of the IoT market that hasn’t been getting much attention until recently.
They need to come out with some higher-end models with regular core I chips he’s right right now these Adam based chrome books r for ghetto monkeys