"The others not mentioned will be split between 7nm and external fabs. More on that info in a separate article"
So that means they are going for external fab really. Damn. What a shame. I wonder if that's the CPU or this GPU. Anything is a shame on Intel and that beancounter Bob.
Shame is when you don't acknowledge your problems and how they might limit you. Going with external fabs allows the company to be competitive where it needs the most. But they sure need a long term strategy and and make critical decisions like which processes they would utilize earlier.
It's only GPU. And it's only the Xe-HP (ergo Discreet GPU) variant.
It seems to imply that is only on the 10SFE "enhanced" variant. Basically using TSMC 7nm process for the higher memory and connector chips, named as "Rambo Cache" and "Xe Link". And that's targeting the server market and high-end desktops (ie think Nvidia Titan).
Seems like Intel realise how embarrassing it is to ask for their competitors wafers for their main product, so instead they're taking a half-measure and using it for the GPU only. That way they get to save face and say "well, AMD and Nvidia do it too". Also it seems they couldn't get any good prices, which is why only some chips are made on the 7nm, and rest are made in-house on 10nm. Overall, this is Intel playing hard to catch up to TSMC's 7nm and AMD's Zen2 and AMD's Vega. That's why the whole thing is difficult to understand, as it is purposely obfuscated. Yet these will only ship in (late) 2021, whereas the competition is posied to actually leap ahead in 2020 to +7nm TSMC, Zen3, RDNA2. And when Intel arrives, they might even be facing off against 5nm and a Zen3+ refresh. Just to put things into perspective.
This reads quite promising. Great writeup. "A few angstroms thick" it's hard to even think about the scale of this stuff, that's 0.1 nanometres or 100 picometres. How far from atomic bonding limits?
It's pretty damn close, but that depends on the bond :-). For a perhaps more human comparison, that of O-O in Oxygen is 1.208 Å; and in C=O bonds (e.g. in CO2) it's around 1.16 Å.
Less promising is that they went for a four cores design again. With AMD pushing the desktop to 8 with the first Zen you would think they would realize amd would pull the same when going mobile and prepare for it... so is this arrogance thinking their cores are so much better or they control the market? Lack of strategic insight? Or inability to work out things for the yield and cost math?
If these cores are as good as they're claiming they are, there will be a legitimate dilemma choosing between 4 Very Fast cores and 8 Really Quite Fast cores.
I'd wager that yields are 80% to blame and power is the rest - they have an 8-core Tiger in the works, but it's a 35/45W part and there's nothing firm about a release date yet.
Depends on the atoms, but for reference how about silicon.
"Silicon has the diamond cubic crystal structure with a lattice parameter of 0.543 nm. The nearest neighbor distance is 0.235 nm."
Diatomic hydrogen is the smallest molecule with a bond length of 0.74 angstrom.
Bond length is typically 1-2 angstrom.
"Intel states that this is an industry first/leading design, enabled through careful deposition of new Hi-K materials in thin layers, smaller than 0.1nm, to form a superlattice between two or more material types."
Unless im missing something, the above statement does not make any sense to me. thin layers of atoms less then the average bond length between atoms is nonsensical.
Good news! Hope those announced products will actually ship this year, and not just in drip-drip-drip fashion. This will help keep Team Red on its toes, and that is good for all of us - as we learned during AMD's valley of tears, lack of competition means slower innovation and high prices. And, there is an actual chance those improvements are real; IMO, exhibit A for that is AMD has moved Cezanne's launch date earlier. Unless there is no real competition, why bother? They're barely able to ship enough Renoirs, so this suggests that AMD thinks these new Intel chips might be real, or real enough.
Which high prices are you talking about? Intel's top desktop processor, 2700K through 10700K all had about the same price, usually in the mid $300 range. The lack of AMD competition during the first half dozen years there didn't have almost any impact on price. When AMD was highly competitive at the 1 GHz race, prices were over $1000. There just isn't much impact on prices going from a monopoly to a duopoly.
The high prices per core compared to how much cheaper those cores could have become with noteworthy competition. Also the high absolute prices of the many-core CPUs. Remember the HEDT platform CPU climbing up to $2000 and then the flagship 10th gen one (10980XE) introduced at half the price of the flagship 9th gen one (9980XE)? Sure AMD then went up to $4000 but the price per core is still much better than it was before Ryzen.
The high prices on that artificially-segmented category of "HEDT" for anyone who wanted more than 4 cores, perhaps? Or the fact that for those same prices you now get significantly more cores *and* more Ghz, which prior to the arrival of Ryzen was by no means guaranteed.
You could also check the steadily-increasing prices of high-end laptops, where the lack of competition has been felt more keenly on both CPU and GPU fronts.
You're right that it doesn't drop things *that much* lower, but Intel have been delivering noticeable improvements in PP$ since AMD regained some semblance of competitiveness.
So what would be the most efficient, lower core but highest clock rate CPU for desktop these days? I only would need maybe 4+ cores or so. I guess clock rate is more important to me but I also want low 10 nm process or lower for high efficiency. I'm only interested in air cooling also but want a quiet system overall. I feel so behind in my CPU knowledge these days. From what i've read, it seems AMD is doing very well these days. I'm currently running an Intel I7-4790K and have not felt the need to upgrade yet. Would like a huge bump in perceptible speed in general. I don't game much anymore but just want an overall fast and efficient and cool and quiet desktop replacement that could do gaming if I wanted to. What CPU is good now and what future CPU would be worth waiting for?
11th Gen Core for desktop looks really good. Rocket Lake architecture. Will be the last standard desktop CPU before Intel goes BIG.little core design like smartphones. Remains to be seen how Windows will deal with that.
Not sure I really agree with you there. Rocket Lake looks to be a mediocre improvement over 10th gen. Still is going to be on 14nm (although they are backporting the micro arch from Ice/Tiger Lake). That is going to be challenged by AMD's Vermeer, which will be Zen 3 on a more advanced 7nm(EUV?) node with a unified cache, 8 core CCX, higher clocks, and a 15-20% IPC increase. That should hold the crown for the best chip lineup for your money when released in a few months. Gaming will be a close call (lets see the impact of an 8 core ccx), but unless Intel fundamentally changes their pricing scheme you will likely be able to get the 11th gen i7 for the same price as Ryzen 4900x (same for i5 vs R7). Not looking to be a fanboy for AMD (I'm impressed by what I've read about Tiger lake in this article), but AlderLake seems to be much more promising than Rocket Lake for Intel although yes it will be very interesting to see how the BIGlittle strategy is used by Windows. Just thought you might be overhyping Rocketlake and ignoring Vermeer.
Vermeer will be huge, you're right there. I just think Rocket Lake will be a good upgrade point in the near term until new technologies reach adoption, like DDR5, PCIE5, and the BIGlittle idea. Plus Rocket Lake will have the GPU goodies from Tiger Lake and benefit from the clockspeed on 14nm. I'm finally going to ditch my i7-860 when they arrive. However, I will admit, I do prefer Intel. Plus Rocketlake might still use z490 boards which are a great deal currently, and will (likely) support PCIE4. So if Jeff can wait a bit, compare Vermeer to Rocket Lake and decide then.
Based on your stated requirements, something like an AMD 3300X or 3600X would make most sense *right now* - but it would make even more sense to wait until the end of the year to see what Zen 3 brings with it.
Intel aren't due to bring 10nm to the desktop until some time next year at the earliest.
In addition to some other points already raised here, laptop CPUs (read: lower power CPUs) are more lucrative, so it makes sense for Intel to go after that market first. Renoir is probably a bigger threat to Intel's bottom line than the Zen2 desktop chips.
I'd guess you're right about what Intel CCG would prioritize between mobile and desktop. I'd wager the biggest AMD threat to the bottom line is actually EPYC, but that's for DCG.
When you have epic fails at 10mn and 7nm for anything above 45W what does one do? One so-called changes their three-card Monte scheme to ... Redefining the FinFET!
14++++?
Alder Lake should be renamed Lava Lake to be followed by a 10++++ desktop part circa 2030 called Hades Lake and a 7+++ part circa 2040 called Crater Lake.
In fact. all their nodes should be re-branded with Moon crater names. That way they can stop digging ever deeper holes for themselves.
"As part of Architecture Day 2020, Intel also stated that the intranode update beyond 10SF will be called 10ESF," - so there won't be 10+, 10++, 10+++, but 10SF, 10ESF, 10EESF,... For me + are better
I'm really waiting for Tiger Lake H which I think will go upto 8cores and feature LPDDR 5. I'm really hoping the next Microsoft Surface Book 4 will feature it and finally also support Thunderbolt for the first time. Tiger Lake H would make the Surface Book practical and functional when the screen is detached. If you look at the price tag of Surface Book they can pass on the cost of LPDDR 5 on a $3000 device. Laptops using Tiger Lake H can be paired mobile discrete GPU. The PCIe 4.0x4 available for fast NVME SSD storage directly from the CPU will bridge the gap between next gaming consoles. Booting up will be instant. Unfortunately we have to wait until 2021 for Tiger Lake H. On the other hand Chrome OS is going to really benefit from Tiger Lake U with the iGPU improving things. I expect to see a lot of Chromebooks and Chrome 2in1.
How much more expensive can LPDDR5 really be? S20 series, OnePlus 8 series, Xiaomis Mi 10 (Pro) and at least one phone from Realms and Redmi do ready use LPDDR5.
If that is already possible with phones it should also be possible with premium notebook designs.
The problem as far as I can tell is volume, mobile phones push more devices so they can benefit from the economy of scale. Samsung with their Exynos 990 and Qualcomm Sanpdragon 865 SOCs support LPDDR5 with around a dozen smartphones out (half from Samsung). Samsung who are leading in manufacturing LPDDR5 memory only uses it on their flagship mobile devices because they will sell millions of smartphones compared to laptops. Now that Intel supports LPDDR5 we will see an increased demand and roll out on laptops.
This close to launch and no benchmarks? hmm, seems suspect. I do hope they got the kick in pants they needed to make a jump in performance needed. Looking forward to apples to apples reviews!
if a phone can have LPDDR5 why not a laptop. I am hoping to see premium laptops(XPS, Spectre, Thinkpad etc) to have an option of LPDDR5. Otherwise this sounds exciting. Even a 8 core 65w CPU for desktop should be interesting. I hope Intel does release that.
The 16" MacBook Pro uses 45 W chips (yes, even the high-end I9-9980HK is specified as 45W). I have the lower-end I7-9750H but still feel the cooling is inadequate, especially when using both CPU and GPU at the same time. Apple just wanted to make it as thin as possible, and didn't expect that Intel would fail so miserably at 10 nm. Now I'm not surprised at all that they are moving to in-house ARM CPUs. Personally I would have preferred AMD for x86 compatibility with my development applications at work, but I'll probably still consider an ARM Mac in a few years and in the worst case rely on ssh to workstations for development. HPCs for running anyway.
Does the comment that 10 SFE is optimized for DC and so Alderlake will not see any xtor improvements from TGL and its just microarchitecture changes to improve performance.
Ian eating more silicon...ah...like a breath of fresh air...;) I find silicon wafers are best enjoyed with a set of Unobtanium™ dentures topped with the diamond tooth inserts and platinum alloy tips--chews up nice, goes down smooth! I have to agree in this really nice write up making the most of the sparseness Intel supplied, that all of this stuff looks incremental to me. Bits and pieces improved. Reading between the lines it looks like Intel is still struggling with its process nodes--the fact that they cannot ship even this right now is certainly telling...nor can they even supply a ship date, apparently.
I sure hope to see TGL scaled to 8 cores CPUs. I feel like the biggest drawback of ICL is it was scaled up to 4 cores only making it underpowered to even comet lake u (with up to 6 cores).
"If you’ve skipped to the end of this article without reading the pages in between,...." ^ You caught me - I'm a sucker for summaries. If I have time, I go back & read through the full articles. Alas, there's only so much time in a day!
In the previous driver update I was under the impression AV1 decode is only partly ASIC accelerated. But the slides here seems to imply it is fully Hardware Decoded.
What's Intel playing at here with ignoring the 8-core offering of Renoir Mobile and just going for 4-core with Tiger Lake? This will be in effect for an entire generation (11th Gen) of mobile products, that they'll have a 50% core deficit. I get that Tiger Lake will have ~20% better ST performance over Renoir, and it could be argued that 4 cores is all one needs nowadays on Mobile, but still, it seems like a calculated and potentially dangerous move by Intel to ignore AMD's core-count advantage.
A socketed TGL 8c/16t would destroy the incoming Rocket Lake so I very doubt Intel will release the TGL 8c for retail desktop consumer. Maybe for OEM-only, just like desktop Renoir
Spelling and grammar errors: This means that the new cores in Tiger Lake a built that for any given power or voltage, they will run at a higher frequency. That made no sense. I have no suggestions on how to improve it.
After Ice Lake was set to be Tiger Lake, built on a '10+' manufacturing node. It's a bit confusing as worded. Try an extra ",": After, Ice Lake was set to be Tiger Lake, built on a '10+' manufacturing node.
"At a high level this means that if a laptop is playing a video, on the CPU we have the display engine is on and the video decode on," Excess "is": "At a high level this means that if a laptop is playing a video, on the CPU we have the display engine on and the video decode on,"
What I want to know is where are the desktop 10nm big core parts? where are they? seriously? you're telling me you cant make 10nm octa core? seriously?
10nm with ddr4 and ddr5, pcie 4, thunderbolt, get it done, otherwise im finally switching to amd how long must i wait.
Small error on page 1, start of the 6th paragraph: "Intel first unveiled its Core microarchitecture in Q1 2006, as an offshoot of the more power efficient Pentium Pro products" - It was an offshoot of the Pentium M (although that does of course owe a lot to the P6 architecture).
That Core family chart on the first page is good for a chuckle.
Conroe - _Core 2_ - is apparently now the first generation of Core. 🤪 It beggared belief enough when they gave Lynnfield that title after the release of Sandy Bridge, but it seems we're deep into revising the revisionism now?
Nehalem and Westmere are now part of the same "first generation" as Conroe, despite the fact that they marked two major steps in integration for Intel - first with the PCH on-die, and then with the GPU on-package.
Meanwhile Skylake and its *extremely minor derivatives* span *four whole generations*, while poor little Palm Cove / Cannon Lake just wasn't any sort of generation at all. It all seems to painfully arbitrary.
In case there was any doubt left after all these decades: Intel is primarily a marketing company.
Now that I've reached the end of the article... My not-at-all-sarcastic executive summary:
10SF is the new 10+, which was the new 10++. This nomenclature is definitely not a way of covering up how badly they messed this up until now.
They've managed to gain a whole node's worth of extra performance from intra-node changes! Yay! Only, this node previously performed worse than the previous node in all important metrics... So they've fixed the performance side of the process (woo), but presumably not yields (boo), as we're still stuck at quad-core and they're not talking dates on the 8-core version.
They can maybe reach something nearly like 5.0Ghz sort-of sometimes perhaps?! But only at a higher power level than Ice Lake, which was already a 25W chip doing a serviceable 15W impersonation. Very promising from a company who have spent the past 5 years selling you mobile chips that struggle to maintain their peak speeds.
The graphics will be powerful! But the memory bandwidth won't be there to support that on launch... so basically you're going to get 3/4 the final finished product, and you'll like it. Please don't ask about the drivers.
So, in summary: Ice Lake Beta 2, or perhaps Cannon Lake Beta 3?
All snark aside, it sounds like they've finally got all their ducks in a row, and I'm honestly glad for that. 15W CPUs aren't a product sector I have any personal interest in for the time being, but when the 8-core variant comes out I'll be casting a curious eye over it.
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.
71 Comments
Back to Article
Quantumz0d - Thursday, August 13, 2020 - link
"The others not mentioned will be split between 7nm and external fabs. More on that info in a separate article"So that means they are going for external fab really. Damn. What a shame. I wonder if that's the CPU or this GPU. Anything is a shame on Intel and that beancounter Bob.
IanCutress - Thursday, August 13, 2020 - link
That's referring to the other tiles of Ponte Vecchio. There are 4. We covered 2.Eliadbu - Thursday, August 13, 2020 - link
Shame is when you don't acknowledge your problems and how they might limit you. Going with external fabs allows the company to be competitive where it needs the most. But they sure need a long term strategy and and make critical decisions like which processes they would utilize earlier.Kangal - Thursday, August 13, 2020 - link
It's only GPU.And it's only the Xe-HP (ergo Discreet GPU) variant.
It seems to imply that is only on the 10SFE "enhanced" variant. Basically using TSMC 7nm process for the higher memory and connector chips, named as "Rambo Cache" and "Xe Link". And that's targeting the server market and high-end desktops (ie think Nvidia Titan).
Seems like Intel realise how embarrassing it is to ask for their competitors wafers for their main product, so instead they're taking a half-measure and using it for the GPU only. That way they get to save face and say "well, AMD and Nvidia do it too". Also it seems they couldn't get any good prices, which is why only some chips are made on the 7nm, and rest are made in-house on 10nm. Overall, this is Intel playing hard to catch up to TSMC's 7nm and AMD's Zen2 and AMD's Vega. That's why the whole thing is difficult to understand, as it is purposely obfuscated. Yet these will only ship in (late) 2021, whereas the competition is posied to actually leap ahead in 2020 to +7nm TSMC, Zen3, RDNA2. And when Intel arrives, they might even be facing off against 5nm and a Zen3+ refresh. Just to put things into perspective.
tipoo - Thursday, August 13, 2020 - link
This reads quite promising. Great writeup."A few angstroms thick" it's hard to even think about the scale of this stuff, that's 0.1 nanometres or 100 picometres. How far from atomic bonding limits?
DrJackMiller - Thursday, August 13, 2020 - link
It's pretty damn close, but that depends on the bond :-). For a perhaps more human comparison, that of O-O in Oxygen is 1.208 Å; and in C=O bonds (e.g. in CO2) it's around 1.16 Å.vortmax2 - Friday, August 14, 2020 - link
Would love to see an article here of the next 10-20 years of processors as we enter deeper into the atomic scale. What the next big breakthrough?jospoortvliet - Saturday, August 15, 2020 - link
Less promising is that they went for a four cores design again. With AMD pushing the desktop to 8 with the first Zen you would think they would realize amd would pull the same when going mobile and prepare for it... so is this arrogance thinking their cores are so much better or they control the market? Lack of strategic insight? Or inability to work out things for the yield and cost math?Spunjji - Monday, August 17, 2020 - link
If these cores are as good as they're claiming they are, there will be a legitimate dilemma choosing between 4 Very Fast cores and 8 Really Quite Fast cores.I'd wager that yields are 80% to blame and power is the rest - they have an 8-core Tiger in the works, but it's a 35/45W part and there's nothing firm about a release date yet.
none12345 - Tuesday, August 18, 2020 - link
Depends on the atoms, but for reference how about silicon."Silicon has the diamond cubic crystal structure with a lattice parameter of 0.543 nm. The nearest neighbor distance is 0.235 nm."
Diatomic hydrogen is the smallest molecule with a bond length of 0.74 angstrom.
Bond length is typically 1-2 angstrom.
"Intel states that this is an industry first/leading design, enabled through careful deposition of new Hi-K materials in thin layers, smaller than 0.1nm, to form a superlattice between two or more material types."
Unless im missing something, the above statement does not make any sense to me. thin layers of atoms less then the average bond length between atoms is nonsensical.
eastcoast_pete - Thursday, August 13, 2020 - link
Good news! Hope those announced products will actually ship this year, and not just in drip-drip-drip fashion. This will help keep Team Red on its toes, and that is good for all of us - as we learned during AMD's valley of tears, lack of competition means slower innovation and high prices.And, there is an actual chance those improvements are real; IMO, exhibit A for that is AMD has moved Cezanne's launch date earlier. Unless there is no real competition, why bother? They're barely able to ship enough Renoirs, so this suggests that AMD thinks these new Intel chips might be real, or real enough.
dullard - Thursday, August 13, 2020 - link
Which high prices are you talking about? Intel's top desktop processor, 2700K through 10700K all had about the same price, usually in the mid $300 range. The lack of AMD competition during the first half dozen years there didn't have almost any impact on price. When AMD was highly competitive at the 1 GHz race, prices were over $1000. There just isn't much impact on prices going from a monopoly to a duopoly.ZoZo - Thursday, August 13, 2020 - link
The high prices per core compared to how much cheaper those cores could have become with noteworthy competition. Also the high absolute prices of the many-core CPUs. Remember the HEDT platform CPU climbing up to $2000 and then the flagship 10th gen one (10980XE) introduced at half the price of the flagship 9th gen one (9980XE)? Sure AMD then went up to $4000 but the price per core is still much better than it was before Ryzen.Spunjji - Monday, August 17, 2020 - link
The high prices on that artificially-segmented category of "HEDT" for anyone who wanted more than 4 cores, perhaps? Or the fact that for those same prices you now get significantly more cores *and* more Ghz, which prior to the arrival of Ryzen was by no means guaranteed.You could also check the steadily-increasing prices of high-end laptops, where the lack of competition has been felt more keenly on both CPU and GPU fronts.
You're right that it doesn't drop things *that much* lower, but Intel have been delivering noticeable improvements in PP$ since AMD regained some semblance of competitiveness.
DannyH246 - Thursday, August 13, 2020 - link
WOW!!! More slides and paper talk from Intel!!! How exciting!shabby - Thursday, August 13, 2020 - link
Look at the bright side... it looks like they got a new person making those slides!Samus - Friday, August 14, 2020 - link
Hey, at least they are actually mass producing something on a new process!Jeff72 - Thursday, August 13, 2020 - link
So what would be the most efficient, lower core but highest clock rate CPU for desktop these days? I only would need maybe 4+ cores or so. I guess clock rate is more important to me but I also want low 10 nm process or lower for high efficiency. I'm only interested in air cooling also but want a quiet system overall. I feel so behind in my CPU knowledge these days. From what i've read, it seems AMD is doing very well these days. I'm currently running an Intel I7-4790K and have not felt the need to upgrade yet. Would like a huge bump in perceptible speed in general. I don't game much anymore but just want an overall fast and efficient and cool and quiet desktop replacement that could do gaming if I wanted to. What CPU is good now and what future CPU would be worth waiting for?fogifds - Thursday, August 13, 2020 - link
11th Gen Core for desktop looks really good. Rocket Lake architecture. Will be the last standard desktop CPU before Intel goes BIG.little core design like smartphones. Remains to be seen how Windows will deal with that.TouchdownTom9 - Thursday, August 13, 2020 - link
Not sure I really agree with you there. Rocket Lake looks to be a mediocre improvement over 10th gen. Still is going to be on 14nm (although they are backporting the micro arch from Ice/Tiger Lake). That is going to be challenged by AMD's Vermeer, which will be Zen 3 on a more advanced 7nm(EUV?) node with a unified cache, 8 core CCX, higher clocks, and a 15-20% IPC increase. That should hold the crown for the best chip lineup for your money when released in a few months. Gaming will be a close call (lets see the impact of an 8 core ccx), but unless Intel fundamentally changes their pricing scheme you will likely be able to get the 11th gen i7 for the same price as Ryzen 4900x (same for i5 vs R7). Not looking to be a fanboy for AMD (I'm impressed by what I've read about Tiger lake in this article), but AlderLake seems to be much more promising than Rocket Lake for Intel although yes it will be very interesting to see how the BIGlittle strategy is used by Windows. Just thought you might be overhyping Rocketlake and ignoring Vermeer.fogifds - Thursday, August 13, 2020 - link
Vermeer will be huge, you're right there. I just think Rocket Lake will be a good upgrade point in the near term until new technologies reach adoption, like DDR5, PCIE5, and the BIGlittle idea. Plus Rocket Lake will have the GPU goodies from Tiger Lake and benefit from the clockspeed on 14nm. I'm finally going to ditch my i7-860 when they arrive. However, I will admit, I do prefer Intel. Plus Rocketlake might still use z490 boards which are a great deal currently, and will (likely) support PCIE4. So if Jeff can wait a bit, compare Vermeer to Rocket Lake and decide then.Jeff72 - Friday, August 14, 2020 - link
Thanks for the reply!Jeff72 - Friday, August 14, 2020 - link
Thanks for the reply!Spunjji - Monday, August 17, 2020 - link
He said 10nm or lower for high efficiency. You're trying to sell him on 14nm. 😬Spunjji - Monday, August 17, 2020 - link
Based on your stated requirements, something like an AMD 3300X or 3600X would make most sense *right now* - but it would make even more sense to wait until the end of the year to see what Zen 3 brings with it.Intel aren't due to bring 10nm to the desktop until some time next year at the earliest.
Silma - Thursday, August 13, 2020 - link
If Tiger Lake is scalable, why begin with 15 W instead of 45 W ?xenol - Thursday, August 13, 2020 - link
It's easier to scale up than it is to scale down and still keep target perf/watt, is my presumption.shabby - Thursday, August 13, 2020 - link
What they mean is when their process is mature enough, ie. in a few years, then they will scale it up.eastcoast_pete - Thursday, August 13, 2020 - link
In addition to some other points already raised here, laptop CPUs (read: lower power CPUs) are more lucrative, so it makes sense for Intel to go after that market first. Renoir is probably a bigger threat to Intel's bottom line than the Zen2 desktop chips.proflogic - Thursday, August 13, 2020 - link
I'd guess you're right about what Intel CCG would prioritize between mobile and desktop. I'd wager the biggest AMD threat to the bottom line is actually EPYC, but that's for DCG.Everett F Sargent - Thursday, August 13, 2020 - link
When you have epic fails at 10mn and 7nm for anything above 45W what does one do? One so-called changes their three-card Monte scheme to ... Redefining the FinFET!14++++?
Alder Lake should be renamed Lava Lake to be followed by a 10++++ desktop part circa 2030 called Hades Lake and a 7+++ part circa 2040 called Crater Lake.
In fact. all their nodes should be re-branded with Moon crater names. That way they can stop digging ever deeper holes for themselves.
TristanSDX - Thursday, August 13, 2020 - link
"As part of Architecture Day 2020, Intel also stated that the intranode update beyond 10SF will be called 10ESF," - so there won't be 10+, 10++, 10+++, but 10SF, 10ESF, 10EESF,...For me + are better
shabby - Thursday, August 13, 2020 - link
I think intel is just tired of the 14nm++++++++++ jokes so they're making a new one.jospoortvliet - Saturday, August 15, 2020 - link
So it is Enhanced Super Fins, Extra Enhanced, Enormous Extra Enhanced, Epically Enormous Extra Enhanced...Linustechtips12#6900xt - Wednesday, April 7, 2021 - link
I think you forgot about EXTREME Xtra enormous enhanced PRO MAX SUPER finsArbie - Thursday, August 13, 2020 - link
An excellent article as usual, and clearly a great deal of work. Thanks.Meteor2 - Tuesday, August 18, 2020 - link
SecondedKimGitz - Thursday, August 13, 2020 - link
I'm really waiting for Tiger Lake H which I think will go upto 8cores and feature LPDDR 5. I'm really hoping the next Microsoft Surface Book 4 will feature it and finally also support Thunderbolt for the first time. Tiger Lake H would make the Surface Book practical and functional when the screen is detached. If you look at the price tag of Surface Book they can pass on the cost of LPDDR 5 on a $3000 device. Laptops using Tiger Lake H can be paired mobile discrete GPU. The PCIe 4.0x4 available for fast NVME SSD storage directly from the CPU will bridge the gap between next gaming consoles. Booting up will be instant. Unfortunately we have to wait until 2021 for Tiger Lake H. On the other hand Chrome OS is going to really benefit from Tiger Lake U with the iGPU improving things. I expect to see a lot of Chromebooks and Chrome 2in1.Stahlkocher - Thursday, August 13, 2020 - link
How much more expensive can LPDDR5 really be? S20 series, OnePlus 8 series, Xiaomis Mi 10 (Pro) and at least one phone from Realms and Redmi do ready use LPDDR5.If that is already possible with phones it should also be possible with premium notebook designs.
KimGitz - Thursday, August 13, 2020 - link
The problem as far as I can tell is volume, mobile phones push more devices so they can benefit from the economy of scale. Samsung with their Exynos 990 and Qualcomm Sanpdragon 865 SOCs support LPDDR5 with around a dozen smartphones out (half from Samsung). Samsung who are leading in manufacturing LPDDR5 memory only uses it on their flagship mobile devices because they will sell millions of smartphones compared to laptops. Now that Intel supports LPDDR5 we will see an increased demand and roll out on laptops.anonomouse - Friday, August 14, 2020 - link
A Lot™psyclist80 - Thursday, August 13, 2020 - link
This close to launch and no benchmarks? hmm, seems suspect. I do hope they got the kick in pants they needed to make a jump in performance needed. Looking forward to apples to apples reviews!eek2121 - Thursday, August 13, 2020 - link
Benchmarks will likely land on launch day.Martimus - Thursday, August 13, 2020 - link
You accidentally posted the wrong slide for the third slide on the "What is in a Willow Cove Core?" page.LiveWell - Thursday, August 13, 2020 - link
Are they incorporating Mears Silicon Technology into their chips?drexnx - Thursday, August 13, 2020 - link
I guess after "enhanced super fin" they can always go to "enhanced super duper fin"? 10ESDF?Linustechtips12#6900xt - Wednesday, April 7, 2021 - link
the marketing can be called ESD-F safetrivik12 - Thursday, August 13, 2020 - link
if a phone can have LPDDR5 why not a laptop. I am hoping to see premium laptops(XPS, Spectre, Thinkpad etc) to have an option of LPDDR5. Otherwise this sounds exciting. Even a 8 core 65w CPU for desktop should be interesting. I hope Intel does release that.vFunct - Thursday, August 13, 2020 - link
With 65W parts, does that mean we're going to see it in a 16" MacBook Pro?AdditionalPylons - Friday, August 14, 2020 - link
The 16" MacBook Pro uses 45 W chips (yes, even the high-end I9-9980HK is specified as 45W). I have the lower-end I7-9750H but still feel the cooling is inadequate, especially when using both CPU and GPU at the same time. Apple just wanted to make it as thin as possible, and didn't expect that Intel would fail so miserably at 10 nm. Now I'm not surprised at all that they are moving to in-house ARM CPUs.Personally I would have preferred AMD for x86 compatibility with my development applications at work, but I'll probably still consider an ARM Mac in a few years and in the worst case rely on ssh to workstations for development. HPCs for running anyway.
Mark242 - Thursday, August 13, 2020 - link
Is the SuperFin tech really a generational impovement of the 10nm process or is it a backport from 7nm?Sahrin - Thursday, August 13, 2020 - link
So basically Intel had to re-engineer the entire technology stack to get 10nm to work.Are they still using EUV on all layers?
IanCutress - Thursday, August 13, 2020 - link
EUV for Intel is on 7nm. There's no EUV on 10 or 10SF.trivik12 - Thursday, August 13, 2020 - link
Does the comment that 10 SFE is optimized for DC and so Alderlake will not see any xtor improvements from TGL and its just microarchitecture changes to improve performance.Thunder 57 - Thursday, August 13, 2020 - link
"As for the L3 cache on a quad-core Willow Cove system, Intel has moved from an 8 MiB non-inclusive shared L3 cache to a 12 MiB shared L3 cache."Pretty sure you meant inclusive L3, which the "Cache Comparison" chart got right.
WaltC - Thursday, August 13, 2020 - link
Ian eating more silicon...ah...like a breath of fresh air...;) I find silicon wafers are best enjoyed with a set of Unobtanium™ dentures topped with the diamond tooth inserts and platinum alloy tips--chews up nice, goes down smooth! I have to agree in this really nice write up making the most of the sparseness Intel supplied, that all of this stuff looks incremental to me. Bits and pieces improved. Reading between the lines it looks like Intel is still struggling with its process nodes--the fact that they cannot ship even this right now is certainly telling...nor can they even supply a ship date, apparently.Eliadbu - Thursday, August 13, 2020 - link
I sure hope to see TGL scaled to 8 cores CPUs. I feel like the biggest drawback of ICL is it was scaled up to 4 cores only making it underpowered to even comet lake u (with up to 6 cores).harobikes333 - Saturday, August 15, 2020 - link
"If you’ve skipped to the end of this article without reading the pages in between,...."^ You caught me - I'm a sucker for summaries. If I have time, I go back & read through the full articles. Alas, there's only so much time in a day!
ksec - Saturday, August 15, 2020 - link
In the previous driver update I was under the impression AV1 decode is only partly ASIC accelerated. But the slides here seems to imply it is fully Hardware Decoded.Farfolomew - Saturday, August 15, 2020 - link
What's Intel playing at here with ignoring the 8-core offering of Renoir Mobile and just going for 4-core with Tiger Lake? This will be in effect for an entire generation (11th Gen) of mobile products, that they'll have a 50% core deficit. I get that Tiger Lake will have ~20% better ST performance over Renoir, and it could be argued that 4 cores is all one needs nowadays on Mobile, but still, it seems like a calculated and potentially dangerous move by Intel to ignore AMD's core-count advantage.Fr@nk - Saturday, August 15, 2020 - link
A socketed TGL 8c/16t would destroy the incoming Rocket Lake so I very doubt Intel will release the TGL 8c for retail desktop consumer. Maybe for OEM-only, just like desktop Renoirballsystemlord - Saturday, August 15, 2020 - link
Spelling and grammar errors:This means that the new cores in Tiger Lake a built that for any given power or voltage, they will run at a higher frequency.
That made no sense. I have no suggestions on how to improve it.
After Ice Lake was set to be Tiger Lake, built on a '10+' manufacturing node.
It's a bit confusing as worded. Try an extra ",":
After, Ice Lake was set to be Tiger Lake, built on a '10+' manufacturing node.
"At a high level this means that if a laptop is playing a video, on the CPU we have the display engine is on and the video decode on,"
Excess "is":
"At a high level this means that if a laptop is playing a video, on the CPU we have the display engine on and the video decode on,"
attaran - Saturday, August 15, 2020 - link
Missing Bloomfield that came before er Lynnfield & Clarksfield and used the same process.xtess3ractx - Sunday, August 16, 2020 - link
What I want to know is where are the desktop 10nm big core parts? where are they? seriously? you're telling me you cant make 10nm octa core? seriously?10nm with ddr4 and ddr5, pcie 4, thunderbolt, get it done, otherwise im finally switching to amd how long must i wait.
Meteor2 - Tuesday, August 18, 2020 - link
Why haven't you switched already? Everyone else has. Look at the sales charts.AbRASiON - Monday, August 17, 2020 - link
Well when will we see a 15w 'lite' version (4 core, no HT) version as a "tiger lake celeron" or "tiger lake pentium" in an Intel NUC?Sounds like a nice mobile chip
Spunjji - Monday, August 17, 2020 - link
Small error on page 1, start of the 6th paragraph:"Intel first unveiled its Core microarchitecture in Q1 2006, as an offshoot of the more power efficient Pentium Pro products" - It was an offshoot of the Pentium M (although that does of course owe a lot to the P6 architecture).
Spunjji - Monday, August 17, 2020 - link
That Core family chart on the first page is good for a chuckle.Conroe - _Core 2_ - is apparently now the first generation of Core. 🤪
It beggared belief enough when they gave Lynnfield that title after the release of Sandy Bridge, but it seems we're deep into revising the revisionism now?
Nehalem and Westmere are now part of the same "first generation" as Conroe, despite the fact that they marked two major steps in integration for Intel - first with the PCH on-die, and then with the GPU on-package.
Meanwhile Skylake and its *extremely minor derivatives* span *four whole generations*, while poor little Palm Cove / Cannon Lake just wasn't any sort of generation at all. It all seems to painfully arbitrary.
In case there was any doubt left after all these decades: Intel is primarily a marketing company.
xpclient - Monday, August 17, 2020 - link
So it will be a while before on-chip AV1 encoding arrivesMeteor2 - Tuesday, August 18, 2020 - link
Yes that's my take-away, which is frustratingSpunjji - Monday, August 17, 2020 - link
Now that I've reached the end of the article... My not-at-all-sarcastic executive summary:10SF is the new 10+, which was the new 10++. This nomenclature is definitely not a way of covering up how badly they messed this up until now.
They've managed to gain a whole node's worth of extra performance from intra-node changes! Yay! Only, this node previously performed worse than the previous node in all important metrics... So they've fixed the performance side of the process (woo), but presumably not yields (boo), as we're still stuck at quad-core and they're not talking dates on the 8-core version.
They can maybe reach something nearly like 5.0Ghz sort-of sometimes perhaps?! But only at a higher power level than Ice Lake, which was already a 25W chip doing a serviceable 15W impersonation. Very promising from a company who have spent the past 5 years selling you mobile chips that struggle to maintain their peak speeds.
The graphics will be powerful! But the memory bandwidth won't be there to support that on launch... so basically you're going to get 3/4 the final finished product, and you'll like it. Please don't ask about the drivers.
So, in summary: Ice Lake Beta 2, or perhaps Cannon Lake Beta 3?
All snark aside, it sounds like they've finally got all their ducks in a row, and I'm honestly glad for that. 15W CPUs aren't a product sector I have any personal interest in for the time being, but when the 8-core variant comes out I'll be casting a curious eye over it.