"Historically, Intel has never mixed and matched multiple microarchitecture designs in the same generation."
That's not exactly accurate. The Core family branding started as a mixed generation with 1st generation Core processors consisting of both 45nm Nehalem microarchitecture and 32nm Westmere microarchitecture processors.
That's a fair point. Intel made it confusing with Nehalem being anything from i5-750S up to i7-975, while Westmere was anything under the i5-680 or anything over the i7-970 (except the 975).
They are not corrupt. Intel has no idea what to do - it's far too big, it's lacking consistent vision, and perhaps the most important Intel's problem, is that its current top management haven't ever faced any big challenges - so they all think they are doing great. They are removed from reality, because middle management is blocking possibly critical problems from showing up on reports going up the chain, and they think any problem can be solved by combination of hard work and throwing money at the problem, but not exactly thinking. And last, but not least, they spent last 15 years collecting ambitious and brave directors, managers and engineers... and they are now occupied with chest-thumping, instead of doing something productive.
Source: my brother works there.
So, in the end, AMD got it's crap together, and here we have the effect.
They were caught illegally damaging AMDs business with the abuse of their monopoly position. That set AMD back during a period where they were more than competitive with Intel, and meant they didn't have the R&D resources to maintain their competitiveness. When Intel finally did pay up, it was too little too late.
Not to defend Intel too much, but meh, it was not considered illegal at the time, and it resulted in a settlement in civil court, not criminal convictions so not they were not convicted of the things you mention even if it seems to us that perhaps they should have been.
It is basically impossible for a company with as many divisions and moving parts to avoid ever making a mistake. Also, activity that would be seen as legal at one stage, such as Intel doing co-branding bonuses with OEM's, can seem predatory at another, such as when they had a valid competitor and their marketshare had crossed 90%. The rules change, and its not always obvious when it happens.
It didn't help that Lynnfield (45 nm) coexisted with Westmere (32 nm CPU + 45 nm chipset) int the same LGA 1156 socket. There was also Havendale (45 nm CPU + 45 nm chipset) that never shipped, though it was sampled around to OEMs.
To further confusion, Intel markets Kaby Lake as a new generation but it is just an advanced stepping of Sky Lake for all intents and purposes.
Ian, I'm assuming you meant Westmere was anything under or including the i5 680 (since the 680 is just a faster 670 as far as I can tell from the specs), but the product naming refers to those i5s as Clarkdale. Funny, I hadn't noticed before that Clarkdale was 32nm while the i5 760 Lynnfield, etc. were 45nm. No wonder the Clarkdales oc so easily, running an i3 550 at 4.7 is trivial; I picked up a 680, hoping to get it over 5.0 just for giggles.
So when exactly will 10nm parts be on the shelves? (yeah I know samples ship to OEMs before then but I mean real I can go out and buy one availability)
I thought they might have been dribbling out some random mobile parts this year a la Broadwell in 2014 but nowt...
So if now the H118 refresh is all 14++ Coffee Lake parts then does 10nm become a 2019 end-user release??
Knowing Intel's marketing guys we will never actually hit 10nm, and "10+" will really be "14+++" which they will put forward as being "As good as 10nm"
Intel 10nm process has 100 billion transistors, about double of what the other companies are producing at 10nm while intel's 14nm has about 40 billion transistors. They will still be way ahead don't worry.
Please provide proof of this - Intel's 10nm is not currently release and saying it is limited at 3Ghz is very pre-mature.
Very likely Cannon Lake will be a trial run of 10nm and 3Ghz Y processor would be quite nice especially running at more core and such. Ice Lake 10nm+ will be the main stream it will be lot more 3Ghz and with Raju on line - hopefully the extra transistors mean higher end graphics also,
@firerod1: That is typical Intel marketing fluff fodder--AKA hyperbolic theoretical best case scenario numerics. The reality is their 10nm process is still not yielding well enough for profitable full-scale production, and it will most likely never come close to its design goals. For the last five long years, this has been typical Intel: overpromise and underdeliver. Remember when Xpoint was delayed and delayed, and Intel promised this ( https://semiaccurate.com/wp-content/uploads/2016/0... ), but we only got this ( https://semiaccurate.com/wp-content/uploads/2016/0... )?
To be fair, Intel is the one who is being consistent with its naming, Samsung and others are the ones blurring the lines on what 10nm and other nodes mean in terms of actual feature sizes...
My guess we will see some ultra weight laptops and tablets using Intel 10nm in 2018, but desktop would probably wait until 2019.
Keep in mind one manufacture's 10nm is not the same as another manufacture - the real count should be number transistors on the die and size of die should be taken in account.
I would predict a lot of 10 nm mobile parts being shown off at CES which is a step closer to consumer availability. Cannon Lake right now appears to be mobile only with the first desktop 10 nm part being Ice Lake a year from now from my current reading of the tea leaves.
What a mess. It's hard to believe Intel threw away a generation's advantage over the entire semiconductor fabrication industry. 10nm is an absolute catastrophe, and it's clear to see they are now behind Samsung, TSMC and Globalfoundries. Not to mention, threw away several generations' lead over AMD.
AMD is going to be on 7nm Zen 2 by H1 2019. Some are speculating end 2018, but let's be a bit conservative. It might be the first time in history AMD actually has a process advantage over Intel.
By Intel's own admission in the famous Ice Lake graph, 10+ isn't coming before end 2019, and performance is actually similar to 14++. On the other hand, GloFo's 7nm by all means is set to be a major step over 12nm. It's only with 10++ that Intel makes a major step over the current Coffee Lake, but that's looking like 2020 at best. By then, AMD will be sitting pretty with Zen 2 at 7+. Unless they mess it up as badly as Intel just did.
This^ Intel has just messed up big time, I would still back them to have a better or equivalent process to glofo 7nm SOC, but it will be close indeed and should not have been this close, Intel's 10nm will arrive something like 2 years late? (Volume).
Amd still has a way to go to catch up with coffeelake IPC and frequency with pinnacle ridge and 12lpp, which I don't think they will manage...but it will close the gap and keep them competitive with near intel performance and lower prices = successful business.
If they could squeeze 5% IPC also bug fixes and clock improvement plus 10% process frequency improvement then things could get interesting. A few more agesa updates improving memory comparability and lowering latency for gaming wouldn't go amiss either.
It's worth noting that Samsung/GloFlo node naming is not in sync with Intel's as far as transistor density or true half-pitch measure, but it's at most one generation behind (ahead?), i.e. Samsung 10LPP is ~same densities as Intel 14+. This obviously carries through to AMD as their material is FAB'd with GloFlo, but the gap is closing with IPC as you said. AMD has always been forced to price aggressively to remain competitive due to their size, but this year has really been a boon to them both in their successful Zen architecture and in Intel's many missteps. I hope we see this gap close further so that we have a more competitive desktop/server chip market in the coming 2-3 years.
Tech press by now should know better than use the useless marketing node names. Just post some useful metric like Mbits of SRAM per sq mm as an indication of process.
You can already get a brutal improvement when going DDR 3466 CL14.
Ryzen "issues" to be an even bigger beast is the latency penalty when using more than 4MB of the L3 cache (35ns+ and then jump to main DDR4 latency so we end up with 90-100ns~ when it should be 60-65ns).
AMD only has some IPC disadvantage if you run 1 thread per core. If you run 2 threads, taking advantage of SMT/HT, Zen is on par or even slightly ahead, and in the middle range (new i5s vs Ryzen 5s) where Intel's marketing geniuses decided to disable HT AMD is way ahead in multi-thread IPC.
Since this is principally driven by physics starting to bite back rather hard, you have to imagine that the suffering will be fairly equally spread out :) Great chance for everyone to catch up of course.
-- What a mess. It's hard to believe Intel threw away a generation's advantage over the entire semiconductor fabrication industry. 10nm is an absolute catastrophe, and it's clear to see they are now behind Samsung, TSMC and Globalfoundries. Not to mention, threw away several generations' lead over AMD.
folks generally assumed that the Intel engineers know physics better than anyone else. may be not. God doesn't change the rules to favor one side over the other in war. despite what some combatants say. and when you get down to a few dozen atoms/cell, the physics is really hairy.
I get the angst. I am disappointed in Intel as well. That said, fabricating an ARM chip (or GPU, or RAM, or SSD) at 10nm is a far easier challenge than an x86 CPU simply because they are simpler and more repeated structures (or at least that is my understanding). And THAT said; Intel has been dominating and milking the market. They can afford all of the fab tech and engineers they could possibly want. This should be a solved problem for a company that claimed to produce their first successful 10nm CPU samples some 2-3 years ago. Not saying it is an easy challenge... but they set up the expectations that we would be there already, and Intel is not known for falling behind their own predictions (with the exception of NetBurst lol)
-- That said, fabricating an ARM chip (or GPU, or RAM, or SSD) at 10nm is a far easier challenge than an x86 CPU simply because they are simpler and more repeated structures (or at least that is my understanding).
the thing is: on the hardware, X86 is just as RISC-y as ARM; what the X86 machines have is more complex front end "decoders" that get from X86 MASM code to actual on-the-silicon instruction set. we may be seeing the "limit" to the benefit to ever more slicing and dicing of code to microcode.
A lot of people don't know that x86 chips operate via an internal decoder to RISC micro-ops. It's not correct to say the RISCness is akin to the way ARM works though, as the RISC uop setup used in x86 chips is not the kind of ISA one would want to expose to developers, for various reasons.
That sounds pretty well in line with the rumor of Intel dropping legacy support in the next gen CPU, going more RISC than the previous lines, using more SIMD and sacrificing backwards compatibility.
Odds are we'll still have backwards compatibility through some software/HAL. I can't see it all going away next gen. That would be giving far too much of an opportunity for ARM to step in to the mainstream.
I'm going with 3-400 MHz on the turbos and 200mhz on the base. 200mhz xfr and precision boost 2. I don't know whether they can adjust the memory controller/fabric to decrease latency.. But that is what's pulling Ryzen down compared to coffeelake.. especially in budget where people won't buy high spec ram.
"It might be the first time in history AMD actually has a process advantage over Intel." Do not jump to conclusions based on marketing xxnm numbers alone; the basic parameter is transistor density (for the FEOL stack anyway). Intel's "mistake" is that their 10nm process is very ambitious, at ~101 million transistors per mm^2, which it is expected to be equivalent to ~7.5nm of what the others will manage (it might even be slightly denser than GloFo's "7nm").
Of course design ambition does not always win the day, and it might even leave Intel behind if the others move beyond 10nm first. Intel's 14nm is already a bit less dense than the others' 10nm. Perhaps Intel at 10nm is facing serious leakage of power or even quantum tunneling issues. Intel can only prove they have a viable process if they release CPUs.
This looks like a case of marketing gone haywire. You don't need to release a "new generation" of chips every year if you don't actually have a new generation of chips Intel. Leave it alone, people are not so pudding-headed to buy CPUs or entire systems just because you tell them that they're a generation behind. Enthusiasts keep up on this stuff and average joes don't buy new computers until they destroy their current one.
I used Intel for so long but I wasn't the type to upgrade so quickly...I was hoping my Z170 would last through Coffee Lake...I decided to jump ship rather than buy a new Intel motherboard...was a bit hesitant about Ryzen.. mostly just because I was concerned about stability, etc. My 1800x is sitting at 4.1ghz with no issues... looking forward to the upcoming Ryzen refreshes...I don't know what the Intel is thinking these days...you can't even buy Coffee Lake and another refresh is being discussed? Utterly pathetic...you'll probably need a new $250 motherboard as well.
And as gfx APIs change & game focus moves towards 1440p and 4K, what differences there are atm vs. CFL which get exposed at 1080p will become less relevant.
Perhaps the best case example of why marketing should never have this much control over an engineering company. When you try to eek out profit this much, you just cause market confusion and no one has any idea what the hell you are doing.
This is the company that brought the Pentium 4 to market as they sold chips based on increasing clock speed. This was something their marketing department liked as it was easy to sell while their engineering wing realized quickly that that was a bad move and continued wide/power efficient development with the Pentium M.
And yet those very same clock speeds are what drove a bulk of the improvements between most CPU generations. While architecture plays a large role, there are plenty of cases in the 90s when the same architecture throughout its life had nearly a 100% difference in performance between the slowest and fastest model. And I'm not counting like say comparing a Celeron to a top end Pentium. The Pentium III Coppermine had a 500MHz on launch and ended with a 1GHz model within a year. This slowed down to about 50% difference throughout the 2000s. It's now getting to less than 15%.
If you want to go about architectural differences, you can find plenty of CPU reviews on this site and if you normalize the clock speed to compare IPC, you'll find that architectural improvements may account for something like 10%-20% at best except between the Pentium 4 to Core and Bulldozer/Piledriver to Zen.
Mistake in the table: under Core Generation, Skylake, Kaby Lake, Kaby Lake Refresh and Coffee Lake should all be 6th generation. They all share the Skylake core microarchitecture, just paired with marginally improved GPUs and marginally improved 14nm processes.
100% true...however, Intel has been able to get decent performance increases with their refreshes...I hope AMD's first Ryzen refresh buries Intel...we need a new market leader for a change.
The only area with remarkable performance increases were in Kaby Lake mobile where they were able to hit and maintain their higher turbo frequencies more often vs. Sky Lake counter parts. Even then, this was mainly for burst activity which heavily favors turbo. Prolonged loads put the differences between the generations down to marginal base clock speed differences favoring Kaby Lake.
Agreed - it's a new physical layout with process tweaks, but the CPU microarchitecture is definitely completely the same (as far as Intel, AT reviews and measurements told us).
Actually 4th gen (Nelahem, Sandy Bridge, Haswell, Skylake)
The upgrades to Skylake are mostly regarded to having higher turbo/base clocks and the ability to sustain turbo for more time, specially on mobile, basically the only gain on these "new gens".
Intel just created a mess of their "xxx gen" marketing, it was clear when Nehalem/Westmere shared 1st gen code, which was right.
Let's create a chipset that doesn't support overclocking and another one that doesn't support graphics and then create a third one that supports both... brilliant!
Asus is already selling a laptop with a full phat Ryzen 7 1700 tweaked for mobile, A 1700 with a tiny bit of undervolt with no turbo already draws less than 45w.
Zen arch has a nasty efficiency when you relax the frecuencies (they already have a nice efficiency at stock setting, better than any current intel chip)
Personally I don't mind if Intel delays actual 10nm line - if initiate is revolutionize the technology of the products. When I started out with computers, mobile was non existent or very rare - it was common place to build you own computer, purchase motherboard CPU case and memory and graphics card.
But technology has change mobile is where pretty much 90% of Industry is now at - that is strength of going 10nm - less power and smaller size. Intel is extremely smart - they also aware that not all parts have to be that small - and can join them together is a package.
I also believe that events of a couple weeks ago with Raju is significant and I expect it also going have huge impact in Intel Mobile GPU market. The AMD add-on board is just a way a satisfied a smaller market of gamers which I saw in internet is primary aim at replacement for Intel's Gaming NUC device
The 9th gen parts are most probably still CFL. I suppose we'll see 8c i7s in that lineup, which would probably mean 6c/12t i5s and 4c/8t i3s, which IMO should have been the CFL lineup from the start.
Should be very interesting to see the clockspeeds of the Core i7-8850H 6-core mobile CPU. I want to see it in an MSI GS73VR thin and light gaming/workstation laptop
Process node confusion aside, I can understand why Intel lumped 3 different generation architectures into one. Maybe they're doing it like car manufacturers now, giving model names based on perceived performance: BMW's 328i uses a 2.0 turbo with power similar to the old 2.8 inline-six, for example.
Then again, putting Atom and Core together is just plain nuts. Atom should be relegated to Celeron only while cut-down Core should be Pentium, don't mix them up because the performance difference is huge.
Y series like the Dell XPS 13 2in1 I am typing this on would be perfect for 10nm - allowing more cores and still having lower power. For most people needs the Y series is fine - not every one needs super powerful graphics - even Intel graphics can handle Photoshop and business level graphics and with Raju on board the future looks bright. I don't believe Intel will just use Raju for high end graphics used by few people - but likely will influence where it counts the most at Intel and that is on Laptops and Tablets
Please just add more PCIe lanes to the standard processors. Having a 4 x 16x slot motherboard sucks when there are really only 16 PCIe to 20 max PCIe lanes to split between the 4x 16x PCIe motherboard. Reminds me of the time when you got one VESA slot and a bunch of really slow slots. Are we really so pathetic you have to by a googlecore processor just to get an adequate number of lanes?
If you need something like that, just buy Zen or used X79. Can't really say anything positive about X99 (even though I have one), the CPU support is too weak, but shove a cheap XEON E5-2680 v2 on an X79 board and you've got a pretty decent system for threaded work (basically the same speed as a stock 5960X).
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.
Facebook
Twitter
RSS
86 Comments
Back to Article
ltcommanderdata - Friday, November 24, 2017 - link
"Historically, Intel has never mixed and matched multiple microarchitecture designs in the same generation."That's not exactly accurate. The Core family branding started as a mixed generation with 1st generation Core processors consisting of both 45nm Nehalem microarchitecture and 32nm Westmere microarchitecture processors.
Ian Cutress - Friday, November 24, 2017 - link
That's a fair point. Intel made it confusing with Nehalem being anything from i5-750S up to i7-975, while Westmere was anything under the i5-680 or anything over the i7-970 (except the 975).IGTrading - Friday, November 24, 2017 - link
Thanks for the synthesis Ian.This means that, if there will be no Intel 8-core mainstream CPUs in 2018, we can prepare for an all-AMD year at our company.
Hehe ... that should make the technical solution designs easier :)
Reminds me of the 2000 ~ 2007 period ....
Anyway, I feel like a sinner when I recommend an Intel solution .... it's like supporting that freakishly corrupt corporate monster :)
Vatharian - Sunday, November 26, 2017 - link
They are not corrupt. Intel has no idea what to do - it's far too big, it's lacking consistent vision, and perhaps the most important Intel's problem, is that its current top management haven't ever faced any big challenges - so they all think they are doing great. They are removed from reality, because middle management is blocking possibly critical problems from showing up on reports going up the chain, and they think any problem can be solved by combination of hard work and throwing money at the problem, but not exactly thinking. And last, but not least, they spent last 15 years collecting ambitious and brave directors, managers and engineers... and they are now occupied with chest-thumping, instead of doing something productive.Source: my brother works there.
So, in the end, AMD got it's crap together, and here we have the effect.
Gigaplex - Sunday, November 26, 2017 - link
"They are not corrupt" - VatharianThey were caught illegally damaging AMDs business with the abuse of their monopoly position. That set AMD back during a period where they were more than competitive with Intel, and meant they didn't have the R&D resources to maintain their competitiveness. When Intel finally did pay up, it was too little too late.
Reflex - Monday, November 27, 2017 - link
Not to defend Intel too much, but meh, it was not considered illegal at the time, and it resulted in a settlement in civil court, not criminal convictions so not they were not convicted of the things you mention even if it seems to us that perhaps they should have been.It is basically impossible for a company with as many divisions and moving parts to avoid ever making a mistake. Also, activity that would be seen as legal at one stage, such as Intel doing co-branding bonuses with OEM's, can seem predatory at another, such as when they had a valid competitor and their marketshare had crossed 90%. The rules change, and its not always obvious when it happens.
Kevin G - Saturday, November 25, 2017 - link
It didn't help that Lynnfield (45 nm) coexisted with Westmere (32 nm CPU + 45 nm chipset) int the same LGA 1156 socket. There was also Havendale (45 nm CPU + 45 nm chipset) that never shipped, though it was sampled around to OEMs.To further confusion, Intel markets Kaby Lake as a new generation but it is just an advanced stepping of Sky Lake for all intents and purposes.
ajp_anton - Saturday, November 25, 2017 - link
So your chart should be:Microarch. Core gen Advertised gen
Nehalem 1st 1st
Westmere 2nd 1st
Sandy Bridge 3rd 2nd
Ivy Bridge 4th 3rd
Haswell 5th 4th
Haswell Ref 5th 4th
Broadwell 6th 5th
Skylake 7th 6th
Kaby Lake 7th 7th
Kaby Lake Ref 7th 8th
Coffee Lake 7th 8th
Cannon Lake ? 8th
Ice Lake ? 9th
ajp_anton - Saturday, November 25, 2017 - link
So your chart should be:Microarch. Core gen Advertised gen
Nehalem 1st 1st
Westmere 2nd 1st
Sandy Bridge 3rd 2nd
Ivy Bridge 4th 3rd
Haswell 5th 4th
Haswell Ref 5th 4th
Broadwell 6th 5th
Skylake 7th 6th
Kaby Lake 7th 7th
Kaby Lake Ref 7th 8th
Coffee Lake 7th 8th
Cannon Lake ? 8th
Ice Lake ? 9th
mapesdhs - Monday, November 27, 2017 - link
Ian, I'm assuming you meant Westmere was anything under or including the i5 680 (since the 680 is just a faster 670 as far as I can tell from the specs), but the product naming refers to those i5s as Clarkdale. Funny, I hadn't noticed before that Clarkdale was 32nm while the i5 760 Lynnfield, etc. were 45nm. No wonder the Clarkdales oc so easily, running an i3 550 at 4.7 is trivial; I picked up a 680, hoping to get it over 5.0 just for giggles.Jon Tseng - Friday, November 24, 2017 - link
So when exactly will 10nm parts be on the shelves? (yeah I know samples ship to OEMs before then but I mean real I can go out and buy one availability)I thought they might have been dribbling out some random mobile parts this year a la Broadwell in 2014 but nowt...
So if now the H118 refresh is all 14++ Coffee Lake parts then does 10nm become a 2019 end-user release??
CaedenV - Friday, November 24, 2017 - link
Knowing Intel's marketing guys we will never actually hit 10nm, and "10+" will really be "14+++" which they will put forward as being "As good as 10nm"firerod1 - Friday, November 24, 2017 - link
Intel 10nm process has 100 billion transistors, about double of what the other companies are producing at 10nm while intel's 14nm has about 40 billion transistors. They will still be way ahead don't worry.Pewzor - Friday, November 24, 2017 - link
100 billion transistor that runs at 3.0ghz max.HStewart - Saturday, November 25, 2017 - link
"100 billion transistor that runs at 3.0ghz max."Please provide proof of this - Intel's 10nm is not currently release and saying it is limited at 3Ghz is very pre-mature.
Very likely Cannon Lake will be a trial run of 10nm and 3Ghz Y processor would be quite nice especially running at more core and such. Ice Lake 10nm+ will be the main stream it will be lot more 3Ghz and with Raju on line - hopefully the extra transistors mean higher end graphics also,
[email protected] - Saturday, November 25, 2017 - link
@firerod1: That is typical Intel marketing fluff fodder--AKA hyperbolic theoretical best case scenario numerics. The reality is their 10nm process is still not yielding well enough for profitable full-scale production, and it will most likely never come close to its design goals. For the last five long years, this has been typical Intel: overpromise and underdeliver. Remember when Xpoint was delayed and delayed, and Intel promised this ( https://semiaccurate.com/wp-content/uploads/2016/0... ), but we only got this ( https://semiaccurate.com/wp-content/uploads/2016/0... )?bananaforscale - Monday, November 27, 2017 - link
"Intel 10nm process has 100 billion transistors"That doesn't even make sense. A process doesn't *have* transistors.
extide - Wednesday, November 29, 2017 - link
He means per square millimeter.extide - Wednesday, November 29, 2017 - link
And millions, not billions. (100mil transistors per sq mm for Intel 10nm)Notmyusualid - Saturday, November 25, 2017 - link
@ CaedenVNonsense.
Reflex - Monday, November 27, 2017 - link
To be fair, Intel is the one who is being consistent with its naming, Samsung and others are the ones blurring the lines on what 10nm and other nodes mean in terms of actual feature sizes...HStewart - Friday, November 24, 2017 - link
My guess we will see some ultra weight laptops and tablets using Intel 10nm in 2018, but desktop would probably wait until 2019.Keep in mind one manufacture's 10nm is not the same as another manufacture - the real count should be number transistors on the die and size of die should be taken in account.
Kevin G - Saturday, November 25, 2017 - link
I would predict a lot of 10 nm mobile parts being shown off at CES which is a step closer to consumer availability. Cannon Lake right now appears to be mobile only with the first desktop 10 nm part being Ice Lake a year from now from my current reading of the tea leaves.Cryio - Friday, November 24, 2017 - link
We'll have Skylake for 4 generations straight? Does Intel's bull never end?Thank you AMD for bringing our savior.
trane - Friday, November 24, 2017 - link
What a mess. It's hard to believe Intel threw away a generation's advantage over the entire semiconductor fabrication industry. 10nm is an absolute catastrophe, and it's clear to see they are now behind Samsung, TSMC and Globalfoundries. Not to mention, threw away several generations' lead over AMD.AMD is going to be on 7nm Zen 2 by H1 2019. Some are speculating end 2018, but let's be a bit conservative. It might be the first time in history AMD actually has a process advantage over Intel.
By Intel's own admission in the famous Ice Lake graph, 10+ isn't coming before end 2019, and performance is actually similar to 14++. On the other hand, GloFo's 7nm by all means is set to be a major step over 12nm. It's only with 10++ that Intel makes a major step over the current Coffee Lake, but that's looking like 2020 at best. By then, AMD will be sitting pretty with Zen 2 at 7+. Unless they mess it up as badly as Intel just did.
trane - Friday, November 24, 2017 - link
*Zen 3 at 7+ in 2020.french toast - Friday, November 24, 2017 - link
This^Intel has just messed up big time, I would still back them to have a better or equivalent process to glofo 7nm SOC, but it will be close indeed and should not have been this close, Intel's 10nm will arrive something like 2 years late? (Volume).
Amd still has a way to go to catch up with coffeelake IPC and frequency with pinnacle ridge and 12lpp, which I don't think they will manage...but it will close the gap and keep them competitive with near intel performance and lower prices = successful business.
If they could squeeze 5% IPC also bug fixes and clock improvement plus 10% process frequency improvement then things could get interesting.
A few more agesa updates improving memory comparability and lowering latency for gaming wouldn't go amiss either.
FullmetalTitan - Friday, November 24, 2017 - link
It's worth noting that Samsung/GloFlo node naming is not in sync with Intel's as far as transistor density or true half-pitch measure, but it's at most one generation behind (ahead?), i.e. Samsung 10LPP is ~same densities as Intel 14+. This obviously carries through to AMD as their material is FAB'd with GloFlo, but the gap is closing with IPC as you said.AMD has always been forced to price aggressively to remain competitive due to their size, but this year has really been a boon to them both in their successful Zen architecture and in Intel's many missteps. I hope we see this gap close further so that we have a more competitive desktop/server chip market in the coming 2-3 years.
Notmyusualid - Saturday, November 25, 2017 - link
+1peevee - Tuesday, November 28, 2017 - link
+2.Tech press by now should know better than use the useless marketing node names. Just post some useful metric like Mbits of SRAM per sq mm as an indication of process.
bananaforscale - Monday, November 27, 2017 - link
*GloFo (Global Floundries sounds kinda fishy :D)Lolimaster - Friday, November 24, 2017 - link
You can already get a brutal improvement when going DDR 3466 CL14.Ryzen "issues" to be an even bigger beast is the latency penalty when using more than 4MB of the L3 cache (35ns+ and then jump to main DDR4 latency so we end up with 90-100ns~ when it should be 60-65ns).
peevee - Tuesday, November 28, 2017 - link
AMD only has some IPC disadvantage if you run 1 thread per core. If you run 2 threads, taking advantage of SMT/HT, Zen is on par or even slightly ahead, and in the middle range (new i5s vs Ryzen 5s) where Intel's marketing geniuses decided to disable HT AMD is way ahead in multi-thread IPC.Qwertilot - Friday, November 24, 2017 - link
Since this is principally driven by physics starting to bite back rather hard, you have to imagine that the suffering will be fairly equally spread out :) Great chance for everyone to catch up of course.FunBunny2 - Friday, November 24, 2017 - link
-- What a mess. It's hard to believe Intel threw away a generation's advantage over the entire semiconductor fabrication industry. 10nm is an absolute catastrophe, and it's clear to see they are now behind Samsung, TSMC and Globalfoundries. Not to mention, threw away several generations' lead over AMD.folks generally assumed that the Intel engineers know physics better than anyone else. may be not. God doesn't change the rules to favor one side over the other in war. despite what some combatants say. and when you get down to a few dozen atoms/cell, the physics is really hairy.
CaedenV - Friday, November 24, 2017 - link
I get the angst. I am disappointed in Intel as well.That said, fabricating an ARM chip (or GPU, or RAM, or SSD) at 10nm is a far easier challenge than an x86 CPU simply because they are simpler and more repeated structures (or at least that is my understanding).
And THAT said; Intel has been dominating and milking the market. They can afford all of the fab tech and engineers they could possibly want. This should be a solved problem for a company that claimed to produce their first successful 10nm CPU samples some 2-3 years ago. Not saying it is an easy challenge... but they set up the expectations that we would be there already, and Intel is not known for falling behind their own predictions (with the exception of NetBurst lol)
FunBunny2 - Friday, November 24, 2017 - link
-- That said, fabricating an ARM chip (or GPU, or RAM, or SSD) at 10nm is a far easier challenge than an x86 CPU simply because they are simpler and more repeated structures (or at least that is my understanding).the thing is: on the hardware, X86 is just as RISC-y as ARM; what the X86 machines have is more complex front end "decoders" that get from X86 MASM code to actual on-the-silicon instruction set. we may be seeing the "limit" to the benefit to ever more slicing and dicing of code to microcode.
mapesdhs - Monday, November 27, 2017 - link
A lot of people don't know that x86 chips operate via an internal decoder to RISC micro-ops. It's not correct to say the RISCness is akin to the way ARM works though, as the RISC uop setup used in x86 chips is not the kind of ISA one would want to expose to developers, for various reasons.0ldman79 - Tuesday, November 28, 2017 - link
That sounds pretty well in line with the rumor of Intel dropping legacy support in the next gen CPU, going more RISC than the previous lines, using more SIMD and sacrificing backwards compatibility.Odds are we'll still have backwards compatibility through some software/HAL. I can't see it all going away next gen. That would be giving far too much of an opportunity for ARM to step in to the mainstream.
Lolimaster - Friday, November 24, 2017 - link
We will have a quick upgrade in the next months with Ryzen+, 12nm with a process made for high frequencies while keeping the efficiency.My guess, 300-400Mhz+ on all base frequencies with the 2000 denomination.
french toast - Friday, November 24, 2017 - link
I'm going with 3-400 MHz on the turbos and 200mhz on the base.200mhz xfr and precision boost 2.
I don't know whether they can adjust the memory controller/fabric to decrease latency..
But that is what's pulling Ryzen down compared to coffeelake.. especially in budget where people won't buy high spec ram.
Santoval - Friday, November 24, 2017 - link
"It might be the first time in history AMD actually has a process advantage over Intel."Do not jump to conclusions based on marketing xxnm numbers alone; the basic parameter is transistor density (for the FEOL stack anyway). Intel's "mistake" is that their 10nm process is very ambitious, at ~101 million transistors per mm^2, which it is expected to be equivalent to ~7.5nm of what the others will manage (it might even be slightly denser than GloFo's "7nm").
Of course design ambition does not always win the day, and it might even leave Intel behind if the others move beyond 10nm first. Intel's 14nm is already a bit less dense than the others' 10nm. Perhaps Intel at 10nm is facing serious leakage of power or even quantum tunneling issues. Intel can only prove they have a viable process if they release CPUs.
Flunk - Friday, November 24, 2017 - link
This looks like a case of marketing gone haywire. You don't need to release a "new generation" of chips every year if you don't actually have a new generation of chips Intel. Leave it alone, people are not so pudding-headed to buy CPUs or entire systems just because you tell them that they're a generation behind. Enthusiasts keep up on this stuff and average joes don't buy new computers until they destroy their current one.ACE76 - Friday, November 24, 2017 - link
I used Intel for so long but I wasn't the type to upgrade so quickly...I was hoping my Z170 would last through Coffee Lake...I decided to jump ship rather than buy a new Intel motherboard...was a bit hesitant about Ryzen.. mostly just because I was concerned about stability, etc. My 1800x is sitting at 4.1ghz with no issues... looking forward to the upcoming Ryzen refreshes...I don't know what the Intel is thinking these days...you can't even buy Coffee Lake and another refresh is being discussed? Utterly pathetic...you'll probably need a new $250 motherboard as well.Lolimaster - Friday, November 24, 2017 - link
With AM4 you just got upgrades all the way to 2020, maybe AMD will release new chipsets with pci-e 4.0, usb 3.2 but not something YOU MUST HAVE.mapesdhs - Monday, November 27, 2017 - link
And as gfx APIs change & game focus moves towards 1440p and 4K, what differences there are atm vs. CFL which get exposed at 1080p will become less relevant.Lolimaster - Friday, November 24, 2017 - link
If you like to toy aroud, newer batches of Ryzen achieve 4Ghz+ with lower voltage (some on 1.3v-1.35v while getting 3.9Ghz at 1.25v)trparky - Saturday, November 25, 2017 - link
How would you be able to tell the difference between an older batch and a newer batch based Ryzen chip?jeffkibuule - Friday, November 24, 2017 - link
Perhaps the best case example of why marketing should never have this much control over an engineering company. When you try to eek out profit this much, you just cause market confusion and no one has any idea what the hell you are doing.Kevin G - Saturday, November 25, 2017 - link
This is the company that brought the Pentium 4 to market as they sold chips based on increasing clock speed. This was something their marketing department liked as it was easy to sell while their engineering wing realized quickly that that was a bad move and continued wide/power efficient development with the Pentium M.xenol - Thursday, February 8, 2018 - link
And yet those very same clock speeds are what drove a bulk of the improvements between most CPU generations. While architecture plays a large role, there are plenty of cases in the 90s when the same architecture throughout its life had nearly a 100% difference in performance between the slowest and fastest model. And I'm not counting like say comparing a Celeron to a top end Pentium. The Pentium III Coppermine had a 500MHz on launch and ended with a 1GHz model within a year. This slowed down to about 50% difference throughout the 2000s. It's now getting to less than 15%.If you want to go about architectural differences, you can find plenty of CPU reviews on this site and if you normalize the clock speed to compare IPC, you'll find that architectural improvements may account for something like 10%-20% at best except between the Pentium 4 to Core and Bulldozer/Piledriver to Zen.
Lolimaster - Friday, November 24, 2017 - link
In fact we are only at the 4th gen post Core/Core2 arch (Nehalem-Sandy Bridge-Haswell-Skylake) and even with that Haswell was not much of a jump.r3loaded - Friday, November 24, 2017 - link
Mistake in the table: under Core Generation, Skylake, Kaby Lake, Kaby Lake Refresh and Coffee Lake should all be 6th generation. They all share the Skylake core microarchitecture, just paired with marginally improved GPUs and marginally improved 14nm processes.ACE76 - Friday, November 24, 2017 - link
100% true...however, Intel has been able to get decent performance increases with their refreshes...I hope AMD's first Ryzen refresh buries Intel...we need a new market leader for a change.IGTrading - Friday, November 24, 2017 - link
I second that!french toast - Friday, November 24, 2017 - link
ThirdKevin G - Saturday, November 25, 2017 - link
The only area with remarkable performance increases were in Kaby Lake mobile where they were able to hit and maintain their higher turbo frequencies more often vs. Sky Lake counter parts. Even then, this was mainly for burst activity which heavily favors turbo. Prolonged loads put the differences between the generations down to marginal base clock speed differences favoring Kaby Lake.MrSpadge - Friday, November 24, 2017 - link
Agreed - it's a new physical layout with process tweaks, but the CPU microarchitecture is definitely completely the same (as far as Intel, AT reviews and measurements told us).Lolimaster - Friday, November 24, 2017 - link
Actually 4th gen (Nelahem, Sandy Bridge, Haswell, Skylake)The upgrades to Skylake are mostly regarded to having higher turbo/base clocks and the ability to sustain turbo for more time, specially on mobile, basically the only gain on these "new gens".
Intel just created a mess of their "xxx gen" marketing, it was clear when Nehalem/Westmere shared 1st gen code, which was right.
Pork@III - Friday, November 24, 2017 - link
Ice Lake Enthusiast(HEDT) + PCI-E 4.0 controller&motherboards not before Q4 2018 a few months before RiZen 2MrSpadge - Friday, November 24, 2017 - link
Those "B" CPUs might be the ones with AMD GPU. They didn't want to make it too obvious, so they chose B over A.Ian Cutress - Sunday, November 26, 2017 - link
Other sub-leaks suggest they'll be G. News post next week on some of thisSonic01 - Friday, November 24, 2017 - link
still no new M series / 5W's? :(limitedaccess - Friday, November 24, 2017 - link
Z68 was not released due to the SATA bug. The SATA bug was addressed via a new B3 stepping and fixed chipsets had that designation.The reason for Z68 was to enable both OCing and graphics support. P67 did not support graphics while H67 did not support OC.
shabby - Saturday, November 25, 2017 - link
Let's create a chipset that doesn't support overclocking and another one that doesn't support graphics and then create a third one that supports both... brilliant!mapesdhs - Monday, November 27, 2017 - link
IIRC it was all a giant mess over Quicksync support.Ian Cutress - Sunday, November 26, 2017 - link
Ah yes, you're right. My bad, will update when I'm home.esntxt - Friday, November 24, 2017 - link
very good.Ej24 - Friday, November 24, 2017 - link
The B sku processors may be one's without igpu's? Though typically they only disable the igpu on Xeon e3 partsIan Cutress - Sunday, November 26, 2017 - link
That's another potentialrepoman27 - Sunday, November 26, 2017 - link
My money would be on the B SKUs supporting the CNL PCH vs the KBL Refresh PCH.sharath.naik - Friday, November 24, 2017 - link
Finally a 6 core mobile CPU. Almost bought a laptop this thanksgiving sale, now it is easier to wait.Lolimaster - Friday, November 24, 2017 - link
Asus is already selling a laptop with a full phat Ryzen 7 1700 tweaked for mobile, A 1700 with a tiny bit of undervolt with no turbo already draws less than 45w.Zen arch has a nasty efficiency when you relax the frecuencies (they already have a nice efficiency at stock setting, better than any current intel chip)
jabbadap - Friday, November 24, 2017 - link
Any news/info about Coffee lake Xeons?HStewart - Friday, November 24, 2017 - link
Personally I don't mind if Intel delays actual 10nm line - if initiate is revolutionize the technology of the products. When I started out with computers, mobile was non existent or very rare - it was common place to build you own computer, purchase motherboard CPU case and memory and graphics card.But technology has change mobile is where pretty much 90% of Industry is now at - that is strength of going 10nm - less power and smaller size. Intel is extremely smart - they also aware that not all parts have to be that small - and can join them together is a package.
I also believe that events of a couple weeks ago with Raju is significant and I expect it also going have huge impact in Intel Mobile GPU market. The AMD add-on board is just a way a satisfied a smaller market of gamers which I saw in internet is primary aim at replacement for Intel's Gaming NUC device
eddman - Friday, November 24, 2017 - link
The 9th gen parts are most probably still CFL. I suppose we'll see 8c i7s in that lineup, which would probably mean 6c/12t i5s and 4c/8t i3s, which IMO should have been the CFL lineup from the start.danwat1234 - Friday, November 24, 2017 - link
Should be very interesting to see the clockspeeds of the Core i7-8850H 6-core mobile CPU. I want to see it in an MSI GS73VR thin and light gaming/workstation laptopvFunct - Friday, November 24, 2017 - link
Are any of these parts the ones with AMD GPUs?HStewart - Friday, November 24, 2017 - link
I don't believe so - only thing I heard is that isolated Intel NUChttp://techreport.com/news/32822/rumor-intel-nuc-w...
Possibly a notebook could have - but my guess it just for Intel NUC division only.
My guess there will be Intel dGPU coming with addition of Raju.
serendip - Friday, November 24, 2017 - link
Process node confusion aside, I can understand why Intel lumped 3 different generation architectures into one. Maybe they're doing it like car manufacturers now, giving model names based on perceived performance: BMW's 328i uses a 2.0 turbo with power similar to the old 2.8 inline-six, for example.Then again, putting Atom and Core together is just plain nuts. Atom should be relegated to Celeron only while cut-down Core should be Pentium, don't mix them up because the performance difference is huge.
Notmyusualid - Saturday, November 25, 2017 - link
Now THIS is what I've been waiting for - a 6C/12T mobile chip.But only 45W? Hmm... I'm on my 4th Extreme mobile CPU, and they were all 55/57W each, as far as I remember.
Oh, and if they charge $2k for it, they can shove it - my 4940XM is chugging along nicely.
wr3zzz - Saturday, November 25, 2017 - link
Is this list complete? I need to upgrade my work laptop but cannot wait until 2H'18 for the 10nm Y-series CannoLake (fanless).HStewart - Saturday, November 25, 2017 - link
Y series like the Dell XPS 13 2in1 I am typing this on would be perfect for 10nm - allowing more cores and still having lower power. For most people needs the Y series is fine - not every one needs super powerful graphics - even Intel graphics can handle Photoshop and business level graphics and with Raju on board the future looks bright. I don't believe Intel will just use Raju for high end graphics used by few people - but likely will influence where it counts the most at Intel and that is on Laptops and Tabletsboe - Saturday, November 25, 2017 - link
Please just add more PCIe lanes to the standard processors. Having a 4 x 16x slot motherboard sucks when there are really only 16 PCIe to 20 max PCIe lanes to split between the 4x 16x PCIe motherboard. Reminds me of the time when you got one VESA slot and a bunch of really slow slots. Are we really so pathetic you have to by a googlecore processor just to get an adequate number of lanes?mapesdhs - Monday, November 27, 2017 - link
If you need something like that, just buy Zen or used X79. Can't really say anything positive about X99 (even though I have one), the CPU support is too weak, but shove a cheap XEON E5-2680 v2 on an X79 board and you've got a pretty decent system for threaded work (basically the same speed as a stock 5960X).peevee - Tuesday, November 28, 2017 - link
Intel marketing screwed up their lineup SO MUCH...Marketoids must be shot.