Seems so and it's seemingly not as important as some make out as even Apple dumped LPDDR3 for DDR4 in their latest laptops so they jump from 16 to 32GB; 16GB being the limit for LPDDR3.
Is this a documented limit for LPDDR3? There are 16GB LPDDR3 modules being sold by Crucial. If I put one in each of my laptop's 2 SODIMM slots, wouldn't that give me 32GB?
The fruits of 14nm++, right? I wonder how they’ll deal with the higher idle draw. Is this where those low power displays (1W) come in?
I can’t help but imagine Cannon Lake (10nm) will be a Broadwell (first 14nm arch): bad clocks, non-existent availability for ages, and an overall “forgotten” CPU line. Then Ice Lake (Skylake, in the analogy 14nm timeline) makes it usable.
At this point it is not even clear that Cannon Lake will actually be released (I mean in something more than a rare single low power 2-core sample with a.. disabled iGPU). In case it was not clear the expected high volume release of 10nm CPUs from Intel in the last couple of months of 2019 refers to Ice Lake, not Cannon Lake.
These, as you might recall, are based on Intel's 10nm+ node, not 10nm. So it strongly looks like Intel will treat their first-gen 10nm node and Cannon Lake as nothing more than a beta node / CPU(s). They had serious issues with their (first-gen) 10nm node and ASML helped them overcome them. It appears that they are going to implement what they learnt in 10nm+ and Ice Lake, and thus skip 10nm and Cannon Lake. I would be greatly surprised if they released a second Cannon Lake CPU before the Ice Lake release.
It sounds like they made some processing improvements with Whiskey Lake. For it to get double digit gains - and higher frequency at lower watt then 8705G they must have done something in power to performance ratio. But keep in my 8705G wattage is not just the CPU - but discrete GPU.
I wonder if they will have 6 core version of Whiskey lake.
Considering the current chips easily pull 40-60W while turboing given sufficient cooling and power delivery, you have to wonder how high these will go, and just how short that turbo window is.
Unchanged base clocks are also pretty telling. Still, faster response times are always good.
These are my thoughts also. A massive turbo isn't that useful if it has to throttle back down after only 40ms. I'd rather see sustained base clocks come up instead of battery life killing turbo frequencies being raised. But hey, just so long as that press release has higher numbers to show...
Yep, a lot of what Intel is doing recently has a certain undertone of "See, we're progressing! Really! Look, it's faster than before!" Of course, whether turboing higher means more power draw depends if they manage to balance boost time and power draw - race-to-sleep is a good tactic for smartphones, but they don't scale to 4-5x their TDP. There's no denying that KBL-R is a great CPU series, but it also leaves the question of where Intel can go from there without 10nm. For now, it looks like the answer is "nowhere, really".
I wonder the same thing. If the all-core sustained boost clocks are increased proportionately, and not just the short-term boost, then Whiskey Lake U will be a very good improvement over Kaby Lake R.
I have an i5-8250u laptop and it can boost to 2.3 Ghz all cores, indefinitely. 2.5 Ghz when undervolted. But it can boost to 3.4 Ghz all cores for only 7-8 seconds.
I would like to see an article comparing all of the different * Lake CPUs set at the same clock speed to see how much if any the CPUs have improved in IPC since Sky Lake was released.
I don't trust any testing methodology to stand by <1% differences beyond a margin of error, anyways.
From Intel's roadmap this week, it will be 2015 to 2020 with only core # / clock frequency powering CPU performance improvements. As 2019 Cannon Lake is only a 10nm die shrink of 2015 Skylake, we'll need to wait until 2020 for even a 1% boost in Intel IPC.
Cannon Lake is not a pure die shrink. Just as with Ivy Bridge and Broadwell, at least some minor (<5%) IPC improvement is likely. I think Anandtech is working on an article regarding this.
Ah, good point: I forgot we were so lucky that even ticks had some minor IPC improvements (where "minor" would be remarkable these days). Haswell (2013) to Broadwell (2015) was a ~3.3% IPC improvement: https://www.anandtech.com/show/9482/intel-broadwel...
I'll be quite interested in that article; anything is better than zero. :(
According to this page: https://cpugrade.com/articles/cinebench-r15-ipc-co... there have been slight IPC increases. It's pretty difficult to extract more IPC without significantly rethinking the entire architecture from scratch and probably breaking most native backwards compatibility
Actually, that page shows that there haven't been any IPC increases. Skylake (6700K) -> Kaby Lake (7700K) -> Coffee Lake (8700K) are performing identically.
Eh, that shows a 0.7% uplift between Skylake and Coffee Lake; I don't think "www.cpupgrade.com" has done the statistical confidence testing to show that 0.76% increase is real and not within margin of error.
I mean, not that they're aiming to be misleading (it's a handy site, FWIW), but confirming a 0.7% increase would take tens of dozens of rounds in identical thermal and system conditions.
I don't think we can stand behind that kind of reasoning, however nice I do admit it "feels right" for a newer product to perform a teeny bit better than an older product.
//
Agreed. There are many reasons *why* we're stuck at essentially no IPC upgrades, but I actually did mean to just point out that it's not a measurable difference, in at least how we test hardware.
4.6GHz turbo for a 15W chip is nothing to balk at. IPC improvement or not, that's insane. My Dell Latitude with an i5-8250U is insanely fast, almost laughably so compared to my 2 year old HP Elitebook i5-6300U. Multithreaded performance is nearly DOUBLE for the same power envelope.
"As 2019 Cannon Lake is only a 10nm die shrink of 2015 Skylake, we'll need to wait until 2020 for even a 1% boost in Intel IPC"
Cannon Lake is not just a 10nm die shrink - it also has serious updates that include architecture changes include AVX-512 for consumer level chips. There are likely additional improvement - especially on Ice Lake version which is like 2019 version - 2018 is cannon lake and probably only limited low power cpus.
One thing I wish people would ignore the nm - part of process - Intel's 10nm is supposedly similar to other company's 7nm - just calling the process with a smaller number does not mean it more dense - you have to consider how it layer. Intel 10nm is taking a longer because it much more challenging process than others.
Larger GPU are coming with Artic Sound - and my guess they are coming in EMIB configurations - similar to i8605 in my XPS 15 2in1 - minus of course the AMD "White Label" GPU.
I think this is just a marketing move from Intel... Without a new architecture or smaller node, is it really possible to reach max boost clock speed without throttling? Well, I guess we need to wait until Whiskey Lake U officially released.
They don't need to change anything on the silicon, really, to hit 4,8GHz on a single core. As all these 14nm i7-****K will prove, they are even good for 5 GHz, if you can afford the juice and have a well binned chip.
Done the same with my i7-7700K, allowing it to go 4.8 on one core, stepping down to 4.4 on all four, I think. With that it stays well within the 85 Watt TDP at all times (and practically silent with Noctua), never asks for extra voltage and I no longer feel stupid about one of these new notebooks CPUs getting higher marks on Geekbench single.
So I guess they take more juice, take it for a shorter time to compensate the unavoidable heat build-up and battery drain and off they go to new marketing heights.
If all you need it burst performance, that's quite allright, just don't like the prices they charge for these 15 Watt i7. And most of the time I don't really feel badly constrained at 3GHz either.
So, essentially, they are taking advantage of minor platform improvements in efficiency to allow the CPU to run a bit faster for quick bursts. I really can't say that I'm impressed. Oh, I'm sure that, in short benchmarks, they will show modest gains in single threaded applications, but sustained multi-threaded applications would be lucky to see more than a precious few percentage points in improvement. Granted, that's not a typical use case for laptops, but, as we've seen with the recent MacBook Pro fiasco, there will be some that find those limits.
More Intel "Bug Inside" chips uniquely requiring OS kernel relocation, such a JOKE (INTENDED according to the Intel ex-CEO) in the processor industry, the 1st and only!
Intel is allowed to keep selling the faulty chip with the design bug inside, unbelievable!
Stupid mass keep buying the "Bug Inside" products, amazing! What a wonderful world that is so forgiving!
Throttle Lake... How can you possibly raise frequency by 600-700MHz and keep all else equal at the same TDP on the same node? Of course the TDP is a lie when their documents say it's only valid at the baseclock the CPU never runs at; Intel's 8th gen mobile CPUs already throttle to hell when not given adequate cooling (i.e., cooling that greatly exceeds the fake TDP rating). Perhaps this way of measuring TDP worked with the older turbo models, but not with the "every CPU has a baseclock more than an entire GHz below what it will ever actually run at in the real world and then a single-core turbo another 2GHz above that" model. It may increase performance somewhat (I expect the all-core turbo increase to be much more moderate) but introduces numerous questions and variables that OEMs have to consider. Intel should be ashamed of themselves for refusing to rate their CPUs' thermal profile accurately.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
40 Comments
Back to Article
0iron - Thursday, August 9, 2018 - link
When we will see LPDDR4 coming to laptop? Why it takes so long...DanNeely - Thursday, August 9, 2018 - link
Apparently it's stuck in 10nm limbo with cannonlake.smilingcrow - Thursday, August 9, 2018 - link
Seems so and it's seemingly not as important as some make out as even Apple dumped LPDDR3 for DDR4 in their latest laptops so they jump from 16 to 32GB; 16GB being the limit for LPDDR3.abufrejoval - Friday, August 10, 2018 - link
How so? Have had 32GB in Skylake notebooks for years with DDR3... Just costs twice as much as it used to.yuhong - Friday, August 10, 2018 - link
LPDDR3 is different from DDR3.LMonty - Saturday, August 11, 2018 - link
Is this a documented limit for LPDDR3? There are 16GB LPDDR3 modules being sold by Crucial. If I put one in each of my laptop's 2 SODIMM slots, wouldn't that give me 32GB?LMonty - Saturday, August 11, 2018 - link
Oops correction, I just realized that LPDDR3 (1.2v) is not the same as DDR3L (1.35v). My laptop uses DDR3L.ikjadoon - Thursday, August 9, 2018 - link
The fruits of 14nm++, right? I wonder how they’ll deal with the higher idle draw. Is this where those low power displays (1W) come in?I can’t help but imagine Cannon Lake (10nm) will be a Broadwell (first 14nm arch): bad clocks, non-existent availability for ages, and an overall “forgotten” CPU line. Then Ice Lake (Skylake, in the analogy 14nm timeline) makes it usable.
Santoval - Friday, August 10, 2018 - link
At this point it is not even clear that Cannon Lake will actually be released (I mean in something more than a rare single low power 2-core sample with a.. disabled iGPU). In case it was not clear the expected high volume release of 10nm CPUs from Intel in the last couple of months of 2019 refers to Ice Lake, not Cannon Lake.These, as you might recall, are based on Intel's 10nm+ node, not 10nm. So it strongly looks like Intel will treat their first-gen 10nm node and Cannon Lake as nothing more than a beta node / CPU(s). They had serious issues with their (first-gen) 10nm node and ASML helped them overcome them. It appears that they are going to implement what they learnt in 10nm+ and Ice Lake, and thus skip 10nm and Cannon Lake. I would be greatly surprised if they released a second Cannon Lake CPU before the Ice Lake release.
HStewart - Thursday, August 9, 2018 - link
It sounds like they made some processing improvements with Whiskey Lake. For it to get double digit gains - and higher frequency at lower watt then 8705G they must have done something in power to performance ratio. But keep in my 8705G wattage is not just the CPU - but discrete GPU.I wonder if they will have 6 core version of Whiskey lake.
Valantar - Thursday, August 9, 2018 - link
Considering the current chips easily pull 40-60W while turboing given sufficient cooling and power delivery, you have to wonder how high these will go, and just how short that turbo window is.Unchanged base clocks are also pretty telling. Still, faster response times are always good.
SquarePeg - Thursday, August 9, 2018 - link
These are my thoughts also. A massive turbo isn't that useful if it has to throttle back down after only 40ms. I'd rather see sustained base clocks come up instead of battery life killing turbo frequencies being raised. But hey, just so long as that press release has higher numbers to show...Valantar - Friday, August 10, 2018 - link
Yep, a lot of what Intel is doing recently has a certain undertone of "See, we're progressing! Really! Look, it's faster than before!" Of course, whether turboing higher means more power draw depends if they manage to balance boost time and power draw - race-to-sleep is a good tactic for smartphones, but they don't scale to 4-5x their TDP. There's no denying that KBL-R is a great CPU series, but it also leaves the question of where Intel can go from there without 10nm. For now, it looks like the answer is "nowhere, really".LMonty - Saturday, August 11, 2018 - link
I wonder the same thing. If the all-core sustained boost clocks are increased proportionately, and not just the short-term boost, then Whiskey Lake U will be a very good improvement over Kaby Lake R.I have an i5-8250u laptop and it can boost to 2.3 Ghz all cores, indefinitely. 2.5 Ghz when undervolted. But it can boost to 3.4 Ghz all cores for only 7-8 seconds.
29a - Thursday, August 9, 2018 - link
I would like to see an article comparing all of the different * Lake CPUs set at the same clock speed to see how much if any the CPUs have improved in IPC since Sky Lake was released.ikjadoon - Thursday, August 9, 2018 - link
0% IPC improvements since August 2015: https://us.hardware.info/reviews/7602/22/intel-cor...I don't trust any testing methodology to stand by <1% differences beyond a margin of error, anyways.
From Intel's roadmap this week, it will be 2015 to 2020 with only core # / clock frequency powering CPU performance improvements. As 2019 Cannon Lake is only a 10nm die shrink of 2015 Skylake, we'll need to wait until 2020 for even a 1% boost in Intel IPC.
Brunnis - Thursday, August 9, 2018 - link
Cannon Lake is not a pure die shrink. Just as with Ivy Bridge and Broadwell, at least some minor (<5%) IPC improvement is likely. I think Anandtech is working on an article regarding this.ikjadoon - Thursday, August 9, 2018 - link
Ah, good point: I forgot we were so lucky that even ticks had some minor IPC improvements (where "minor" would be remarkable these days). Haswell (2013) to Broadwell (2015) was a ~3.3% IPC improvement: https://www.anandtech.com/show/9482/intel-broadwel...I'll be quite interested in that article; anything is better than zero. :(
ishould - Thursday, August 9, 2018 - link
According to this page: https://cpugrade.com/articles/cinebench-r15-ipc-co... there have been slight IPC increases. It's pretty difficult to extract more IPC without significantly rethinking the entire architecture from scratch and probably breaking most native backwards compatibilityBrunnis - Thursday, August 9, 2018 - link
Actually, that page shows that there haven't been any IPC increases. Skylake (6700K) -> Kaby Lake (7700K) -> Coffee Lake (8700K) are performing identically.HStewart - Friday, August 10, 2018 - link
Maybe on desktop but not on the mobile sideikjadoon - Thursday, August 9, 2018 - link
Eh, that shows a 0.7% uplift between Skylake and Coffee Lake; I don't think "www.cpupgrade.com" has done the statistical confidence testing to show that 0.76% increase is real and not within margin of error.I mean, not that they're aiming to be misleading (it's a handy site, FWIW), but confirming a 0.7% increase would take tens of dozens of rounds in identical thermal and system conditions.
I don't think we can stand behind that kind of reasoning, however nice I do admit it "feels right" for a newer product to perform a teeny bit better than an older product.
//
Agreed. There are many reasons *why* we're stuck at essentially no IPC upgrades, but I actually did mean to just point out that it's not a measurable difference, in at least how we test hardware.
abufrejoval - Friday, August 10, 2018 - link
The other problem is that too much rethinking can lead to Spectre type leaks....HStewart - Friday, August 10, 2018 - link
"The other problem is that too much rethinking can lead to Spectre type leaks...."These type issue is probably a key reason why Intel 10nm is late.
Samus - Thursday, August 9, 2018 - link
4.6GHz turbo for a 15W chip is nothing to balk at. IPC improvement or not, that's insane. My Dell Latitude with an i5-8250U is insanely fast, almost laughably so compared to my 2 year old HP Elitebook i5-6300U. Multithreaded performance is nearly DOUBLE for the same power envelope.HStewart - Friday, August 10, 2018 - link
Yes indeed - similar as my XPS 15 2in1 - it much faster also especially on CPU.HStewart - Friday, August 10, 2018 - link
"As 2019 Cannon Lake is only a 10nm die shrink of 2015 Skylake, we'll need to wait until 2020 for even a 1% boost in Intel IPC"Cannon Lake is not just a 10nm die shrink - it also has serious updates that include architecture changes include AVX-512 for consumer level chips. There are likely additional improvement - especially on Ice Lake version which is like 2019 version - 2018 is cannon lake and probably only limited low power cpus.
One thing I wish people would ignore the nm - part of process - Intel's 10nm is supposedly similar to other company's 7nm - just calling the process with a smaller number does not mean it more dense - you have to consider how it layer. Intel 10nm is taking a longer because it much more challenging process than others.
gsalkin - Thursday, August 9, 2018 - link
Soooo is Intel abandoning the 15W 2+3 configurations?Valantar - Friday, August 10, 2018 - link
No reason to suspect that; larger iGPUs always launch later than their "standard" siblings.HStewart - Friday, August 10, 2018 - link
Larger GPU are coming with Artic Sound - and my guess they are coming in EMIB configurations - similar to i8605 in my XPS 15 2in1 - minus of course the AMD "White Label" GPU.boozed - Thursday, August 9, 2018 - link
I wouldn't mind a lake of whiskyGreenReaper - Thursday, August 9, 2018 - link
Would it be enough to quench a meltdown, I wonder?boozed - Friday, August 10, 2018 - link
If nothing else at least it'll stop you worrying about itFritzkier - Thursday, August 9, 2018 - link
I think this is just a marketing move from Intel...Without a new architecture or smaller node, is it really possible to reach max boost clock speed without throttling?
Well, I guess we need to wait until Whiskey Lake U officially released.
eddman - Friday, August 10, 2018 - link
Kaby lake refresh is 14nm+ and whiskey lake is expected to be 14nm++.abufrejoval - Friday, August 10, 2018 - link
They don't need to change anything on the silicon, really, to hit 4,8GHz on a single core. As all these 14nm i7-****K will prove, they are even good for 5 GHz, if you can afford the juice and have a well binned chip.Done the same with my i7-7700K, allowing it to go 4.8 on one core, stepping down to 4.4 on all four, I think. With that it stays well within the 85 Watt TDP at all times (and practically silent with Noctua), never asks for extra voltage and I no longer feel stupid about one of these new notebooks CPUs getting higher marks on Geekbench single.
So I guess they take more juice, take it for a shorter time to compensate the unavoidable heat build-up and battery drain and off they go to new marketing heights.
If all you need it burst performance, that's quite allright, just don't like the prices they charge for these 15 Watt i7. And most of the time I don't really feel badly constrained at 3GHz either.
lightningz71 - Friday, August 10, 2018 - link
So, essentially, they are taking advantage of minor platform improvements in efficiency to allow the CPU to run a bit faster for quick bursts. I really can't say that I'm impressed. Oh, I'm sure that, in short benchmarks, they will show modest gains in single threaded applications, but sustained multi-threaded applications would be lucky to see more than a precious few percentage points in improvement. Granted, that's not a typical use case for laptops, but, as we've seen with the recent MacBook Pro fiasco, there will be some that find those limits.wow&wow - Friday, August 10, 2018 - link
More Intel "Bug Inside" chips uniquely requiring OS kernel relocation, such a JOKE (INTENDED according to the Intel ex-CEO) in the processor industry, the 1st and only!Intel is allowed to keep selling the faulty chip with the design bug inside, unbelievable!
Stupid mass keep buying the "Bug Inside" products, amazing! What a wonderful world that is so forgiving!
HStewart - Friday, August 10, 2018 - link
It has been proven that both AMD and ARM both have Spectra issues..Most people don't care about this stuff - just go into your local BestBuy and you can see what people are buying Laptop mostly with Intel.
ibnmadhi - Saturday, August 11, 2018 - link
Throttle Lake... How can you possibly raise frequency by 600-700MHz and keep all else equal at the same TDP on the same node? Of course the TDP is a lie when their documents say it's only valid at the baseclock the CPU never runs at; Intel's 8th gen mobile CPUs already throttle to hell when not given adequate cooling (i.e., cooling that greatly exceeds the fake TDP rating). Perhaps this way of measuring TDP worked with the older turbo models, but not with the "every CPU has a baseclock more than an entire GHz below what it will ever actually run at in the real world and then a single-core turbo another 2GHz above that" model. It may increase performance somewhat (I expect the all-core turbo increase to be much more moderate) but introduces numerous questions and variables that OEMs have to consider. Intel should be ashamed of themselves for refusing to rate their CPUs' thermal profile accurately.