If they're struggling to get clients, it's because of price and nothing else. As long as there is someone who does it cheaper, or more efficient, you'll lose. I've seen videos explaining how we are getting to the limit of physics with how small a transistor can be. I'd love to know how far are we really, because in most cases 10 nm from other fabs is equivalent to the Intel 14nm. It upsets me they measure it differently and use it as marketing.

Silva:

If they're struggling to get clients, it's because of price and nothing else. As long as there is someone who does it cheaper, or more efficient, you'll lose. I've seen videos explaining how we are getting to the limit of physics with how small a transistor can be. I'd love to know how far are we really, because in most cases 10 nm from other fabs is equivalent to the Intel 14nm. It upsets me they measure it differently and use it as marketing.

It is no equivalent by any means. Each of those manufacturing processes result in different density, are targeting different voltage and are made to reach different clock at different leakage. What's intel's selling point? Likely high achievable clock. But do you think they would accept reasonable dead from AMD to build Zen1? I do not think intel wanted competition of any sort to have access to top tier production. In other words, there is damn good chance that they are selling production capacity only on older tech.

Intel was caught pants down by AMD/Ryzen, so it wouldn't surprise me if a lot of Intel's problems were caused by draconian measures to cut spending in R&D in order to fill the CEO's and stock owners' pockets instead (no doubt they still reported lots of billions used in R&D to avoid paying taxes). The original 10nm Cannon Lakes would have been the same old 4-core things we had seen too many times already, with exceptionally marginal performance increases, only made using the smaller process technology to increase Intel's profit per unit. Naturally they had to abandon such thoughts, and had to make the familiar 8000 series Coffee Lake instead, using Sky/Kaby lake as a basis. Even now Intel is still satisfied with the sales numbers, it's beating AMD nicely, so they aren't going to invest anything extra in the 10nm process development, thus making their progress so slow. Intel is a financial corporation, not a technology corporation, after all.

Kaarme:

Intel was caught pants down by AMD/Ryzen, so it wouldn't surprise me if a lot of Intel's problems were caused by draconian measures to cut spending in R&D in order to fill the CEO's and stock owners' pockets instead (no doubt they still reported lots of billions used in R&D to avoid paying taxes). The original 10nm Cannon Lakes would have been the same old 4-core things we had seen too many times already, with exceptionally marginal performance increases, only made using the smaller process technology to increase Intel's profit per unit. Naturally they had to abandon such thoughts, and had to make the familiar 8000 series Coffee Lake instead, using Sky/Kaby lake as a basis. Even now Intel is still satisfied with the sales numbers, it's beating AMD nicely, so they aren't going to invest anything extra in the 10nm process development, thus making their progress so slow. Intel is a financial corporation, not a technology corporation, after all.

Nah - Intel's problem was Brian Kreznick convincing the board to diversify away from Intel's core competency. Their R&D budget got split and it went into drones, wearables, "new devices" and other random bullshit - most of which they just closed. The problems with 10nm are something that a lot of these foundries are going to face/facing currently as they ship their similarly sized 7nm parts. Currently 14nm+++++++++*/+* goes to 5Ghz and their 10nm process probably doesn't. When there are relatively minor improvements to general IPC - how will you sell a generation of CPU's slower than the previous? They need to achieve both switching speed and yield parity with 14nm before they can move forward - something that's getting increasingly difficult at lower feature sizes. It's pretty clear they are refocusing their engineering though with the hiring of Keller/Kadori. Keller was placed as head of Silicon Engineering and is basically overseeing all the chip teams - he's almost definitely going to have an impact on design 3-4 years from now. Which is also probably when we will start seeing discrete GPU's out of Intel... It's going to look like a much different company in 2022.

I don't believe they are moving away from the foundry at all. Like the article pointed out here Intel's 14nm is closer to others 10nm, although I will not go as far to agree with the 7nm piece, that's a stretch. TSMC has 7nm going now but I suspect we won't see any really high 4Ghz+ parts from their process, basically just like we aren't seeing from Intel's 10nm. Intel isn't really behind everyone else they just are not leading the path anymore. If Intel nails 10nm in 2019 they likely will be right on the same pace as AMD/Gloflo on 7nm which from what all the experts say Intel's 10nm and Gloflos 7nm should be similar densities. They have staffed up heavy on engineers lately grabbing the superstars of the Industry. It appears they are going to do the opposite that this article implies. My prediction is that we'll see AMD roll out Ryzen Zen 2(I guess we have to say it like that due to the EPYC on TSMC and Ryzen on Gloflo split for 7nm) on Gloflos process before Intel has a high speed desktop part out on 10nm but they will be only 6-9 months behind AMD. I also think Intel's 10nm and AMD's 7nm processes will be the only 4Ghz+ capbabe processes at this feature size.

AMD is putting the emphasis on cores as opposed to raw clocks--thinking about GHz clocks and nothing else is single-core thinking, for sure.. The days when the performance of one cpu over another could be measured by raw GHz are long gone. Talking about "5GHz" without reference to core count/IPC is fairly meaningless today--32-cores @ 4GHz will always trounce 16 cores @ 5GHz, assuming software that can run up to 32 threads/cores for both, etc., even if the 16-core cpu is slightly faster per core in IPC performance. IPC always has been more important than GHz, imo--ever since the single-core era, anyway. Best architectures employ an optimum balance between per-core IPC, number of cores, and raw GHz clocks. I agree that Intel is trying to change itself and that in 2020 it will be a different company--at least in its PC chips divisions--it will look a lot more like AMD does, today, I would guess. But then, again, AMD isn't sitting still and waiting on Intel, tapping its R&D foot, eh? Competition is the mother of innovation...;)

waltc3:

AMD is putting the emphasis on cores as opposed to raw clocks--thinking about GHz clocks and nothing else is single-core thinking, for sure.. The days when the performance of one cpu over another could be measured by raw GHz are long gone. Talking about "5GHz" without reference to core count/IPC is fairly meaningless today--32-cores @ 4GHz will always trounce 16 cores @ 5GHz, assuming software that can run up to 32 threads/cores for both, etc., even if the 16-core cpu is slightly faster per core in IPC performance. IPC always has been more important than GHz, imo--ever since the single-core era, anyway. Best architectures employ an optimum balance between per-core IPC, number of cores, and raw GHz clocks. I agree that Intel is trying to change itself and that in 2020 it will be a different company--at least in its PC chips divisions--it will look a lot more like AMD does, today, I would guess. But then, again, AMD isn't sitting still and waiting on Intel, tapping its R&D foot, eh? Competition is the mother of innovation...;)

That is true to a degree. AMD did go for cores but Zen 2 Ryzen is going for high clocks on those cores. I very much expect high clock 8-cores that have around 5Ghz boost when Zen 2 Ryzen is outed. Clock speed still matters but you are right it is not the end all be all like it used to be when everything was single threaded, especially in server workloads. I have a suspicion Intel's 10nm is hitting sub 4Ghz before it starts going up that massive power curve like Ryzens do when you start going over 4.1 Ghz and that is why you only see only low speed/low power 10nm chips out from Intel. Even with nice IPC improvements if you can't hit decent clocks you will have a brand new efficient CPU that performs worse than the prior gen and nobody will buy it. That is to say clock speed still matters a lot in desktop chips because most of the work we ask of Desktop chips to perform is single to lightly threaded and that is how all the reviews primarily judge such chips. This is also why it is such a genius move by LIsa Su to bump the EPYC manufacturing to TSMC as there 7nm process very likely wont hit high clocks but as long as it hits target clocks within a power envelope for say a 32 core CPU, it doesn't matter if the CPU could hit over 4Ghz because it wouldn't be clocked that high due to power/heat anyhow. It does matter in the 8 core and under arena where you can dissipate the heat of higher clocks and Gloflos IBM designed 7nm process was designed ground up for high clocks.

waltc3:

AMD is putting the emphasis on cores as opposed to raw clocks--thinking about GHz clocks and nothing else is single-core thinking, for sure.. The days when the performance of one cpu over another could be measured by raw GHz are long gone. Talking about "5GHz" without reference to core count/IPC is fairly meaningless today--32-cores @ 4GHz will always trounce 16 cores @ 5GHz, assuming software that can run up to 32 threads/cores for both, etc., even if the 16-core cpu is slightly faster per core in IPC performance. IPC always has been more important than GHz, imo--ever since the single-core era, anyway. Best architectures employ an optimum balance between per-core IPC, number of cores, and raw GHz clocks. I agree that Intel is trying to change itself and that in 2020 it will be a different company--at least in its PC chips divisions--it will look a lot more like AMD does, today, I would guess. But then, again, AMD isn't sitting still and waiting on Intel, tapping its R&D foot, eh? Competition is the mother of innovation...;)

Yeah but the difference is that AMD had nowhere to go but up in performance when they started shipping more cores with Ryzen. The number one seller of these chips isn't core counts or frequencies it's the bars on a graph in Page 5 of Hilbert's review that shows the newly released shiny processors faster than last years model. That's the stuff the kid's crave. Take for example if Intel shipped a 12 Core, 10nm part with similar IPC per clock to current gen but it was only capable of hitting 4.2Ghz, potentially 4.4/4.5 OC. Nearly every game benchmark and a fair number of application benchmarks would show an outright regression in performance. It would look terrible for Intel in general consumer reviews. Then you have to ask, how many more cores or how low of a price would Intel need to add to such a part to make it redeemable to the general consumer? People like "Extreme FPS Connoisseur @las " would literally shit the bed and no price point would ever make that product redeemable to him.. I think most people would just buy the 14nm older chips because the vast majority of people don't need more than 8 cores for the foreseeable future. AMD never experienced this when it scaled core count because Ryzen came with a 65%+ IPC uplift. So it just looked significantly better across the entire spectrum of benchmarks compared to prior products. Intel wasn't shipping 8 core consumer parts than either.. so multicore numbers were higher than Intel's products at the same price point (the one reviewers were comparing too). Those general "all boat" lifting IPC increases are gone now though - there really isn't anything coming down the pipeline in academia either. Most of the newer IPC improvements are done for specific workloads, like AVX for example. So Intel needs to at least achieve clockspeed parity with its current generation before it can rely on core count as a useful differentiator. Most of what I read though shows that with 10nm yields are the primary issue - Intel started their transition to that density early so it's working with older technology. Here is a quote from Semiwiki for those interested:

Multi-Patterning

In the front end of the line all four companies are using Self Aligned Quadruple Patterning (SAQP) with multiple cut masks for Fin formation and Self Aligned Double Patterning (SADP) for gate formation. At contact some versions of Litho-Etch are used, either Litho-Etch-Litho-Etch (LE2), or Litho-Etch-Litho-Etch-Litho-Etch (LE3) or even LE4 (EUV for Samsung). These are all similar between the companies except for Samsung's use of EUV. In the Back End Of Line (BEOL) is where we see a significant differences. GF and TSMC both use SADP for critical metal layers, Intel uses SAQP for 2 metal layers and Samsung is expected to use EUV for critical metal layers. We believe GF and TSMC are both ramping yield on schedule. It is possible that the yield issues Intel is seeing are related to SAQP in the BEOL. BEOL metal layers require multiple block layers and this is complex to implement. The first block layer would remove the layers needed for subsequent block layers, the way this is addressed is block layers are applied as reverse images and then once all the block layers are done, the whole pattern is reversed. Implementing multiple block layers in concert with SAQP versus fewer block layers at SADP is more difficult. This could explain why multi-patterning may be more difficult at Intel. Intel has a 36nm pitch in the BEOL versus GF and TSMC at 40nm and the smaller pitch is more difficult to achieve. We don't know much about Samsung's process yield ramp or exact specifications but certainly their early use of EUV may presents some challenges for them and we wouldn't be surprised if Samsung encounters yield issues as well.

It all depends what are actually issues with Intels10nm process, if its yield that can be in most cases improved relatively easy, if its low clock speed then its harder, but given that Intel will be 4 years late to market if he manage use it in 2019 those problems have to be quite lot of problems in both areas. Another question is what and how quickly Intel can do with their CPU architecture because for their top end 28core they seems to have bad yield even on state of art tuned 14nm+++ and moving to 10nm its likely getting worse. I don't see Intel even keeping pace with AMD for next 2 or so years, unless they already hve their "infinity fabric" alternative and were already working on MCM CPU design past few years.

xrodney:

It all depends what are actually issues with Intels10nm process, if its yield that can be in most cases improved relatively easy, if its low clock speed then its harder, but given that Intel will be 4 years late to market if he manage use it in 2019 those problems have to be quite lot of problems in both areas. Another question is what and how quickly Intel can do with their CPU architecture because for their top end 28core they seems to have bad yield even on state of art tuned 14nm+++ and moving to 10nm its likely getting worse. I don't see Intel even keeping pace with AMD for next 2 or so years, unless they already hve their "infinity fabric" alternative and were already working on MCM CPU design past few years.

Intel already has there "mesh" called EMIB which is there version of Infinity fabric. I would not count intel out, maybee delayed 6 months to year but you can be assured they will pass AMD up again by end of 2020. Hopefully AMD will leapfrog back again after that as that would be healthy competition.

Denial:

Yeah but the difference is that AMD had nowhere to go but up in performance when they started shipping more cores with Ryzen. The number one seller of these chips isn't core counts or frequencies it's the bars on a graph in Page 5 of Hilbert's review that shows the newly released shiny processors faster than last years model. That's the stuff the kid's crave. Take for example if Intel shipped a 12 Core, 10nm part with similar IPC per clock to current gen but it was only capable of hitting 4.2Ghz, potentially 4.4/4.5 OC. Nearly every game benchmark and a fair number of application benchmarks would show an outright regression in performance. It would look terrible for Intel in general consumer reviews. Then you have to ask, how many more cores or how low of a price would Intel need to add to such a part to make it redeemable to the general consumer? People like "Extreme FPS Connoisseur @las " would literally crap the bed and no price point would ever make that product redeemable to him.. I think most people would just buy the 14nm older chips because the vast majority of people don't need more than 8 cores for the foreseeable future. AMD never experienced this when it scaled core count because Ryzen came with a 65%+ IPC uplift. So it just looked significantly better across the entire spectrum of benchmarks compared to prior products. Intel wasn't shipping 8 core consumer parts than either.. so multicore numbers were higher than Intel's products at the same price point (the one reviewers were comparing too). Those general "all boat" lifting IPC increases are gone now though - there really isn't anything coming down the pipeline in academia either. Most of the newer IPC improvements are done for specific workloads, like AVX for example. So Intel needs to at least achieve clockspeed parity with its current generation before it can rely on core count as a useful differentiator. Most of what I read though shows that with 10nm yields are the primary issue - Intel started their transition to that density early so it's working with older technology. Here is a quote from Semiwiki for those interested:

There is more then just one problem on Intel side. They could easily make 4 or 6 core CPUs on 10nm if making more cores would cause them to lose clock speed or use it for lower power chips that anyway don't clock over 4GHz. As we realy don't see into what issues Intel have in 10nm its hard to comment. but generaly if you make small chips on process its easier to improve it then when you just try to tune it up on test wafers without real production. Likely 10nm CPUs would cost little bit more to make and I suspect this is one of reasons why we did not see anything commingof it for so long. As for multipatterning for different layers, each company use their own tested methods, but also they adjust in case they need to. 36 vs 40nm pitch is not that big difference and as well 40nm is just gen1 for both TSMC and GloFo, their gen2 is planned to use 36nm as well. Btw if this is what is troubling intell they can still chose to use EUV (they already got some) which will almost make it piece of cake. I dont believe this is only problem with Intels 10nm process as this could have been fixed years ago by adjusting their process.

JamesSneed:

Intel already has there "mesh" called EMIB which is there version of Infinity fabric. I would not count intel out, maybee delayed 6 months to year but you can be assured they will pass AMD up again by end of 2020. Hopefully AMD will leapfrog back again after that as that would be healthy competition.

EMIB is just plain stupid bridge/interconnect, it have same function as is interposer for AMD, Infinity fabric on other hand is more about full data and control interface and can run over many different interconnects, inside chip, through interposer (can even use intels EMIB if they want to), motherboard PCB etc.

JamesSneed:

That is true to a degree. AMD did go for cores but Zen 2 Ryzen is going for high clocks on those cores. I very much expect high clock 8-cores that have around 5Ghz boost when Zen 2 Ryzen is outed. Clock speed still matters but you are right it is not the end all be all like it used to be when everything was single threaded, especially in server workloads. I have a suspicion Intel's 10nm is hitting sub 4Ghz before it starts going up that massive power curve like Ryzens do when you start going over 4.1 Ghz and that is why you only see only low speed/low power 10nm chips out from Intel. Even with nice IPC improvements if you can't hit decent clocks you will have a brand new efficient CPU that performs worse than the prior gen and nobody will buy it. That is to say clock speed still matters a lot in desktop chips because most of the work we ask of Desktop chips to perform is single to lightly threaded and that is how all the reviews primarily judge such chips. This is also why it is such a genius move by LIsa Su to bump the EPYC manufacturing to TSMC as there 7nm process very likely wont hit high clocks but as long as it hits target clocks within a power envelope for say a 32 core CPU, it doesn't matter if the CPU could hit over 4Ghz because it wouldn't be clocked that high due to power/heat anyhow. It does matter in the 8 core and under arena where you can dissipate the heat of higher clocks and Gloflos IBM designed 7nm process was designed ground up for high clocks.

That's expectation is bit too much. Reasonable is to expect around 5% clock uplift from current 4.35GHz max One-core boost => 4.57~4.6GHz, and reasonable boost in power efficiency allowing all core boost improvement by 10% from around 3.9~4GHz to 4.3~4.4GHz. And on top of it 10% improved IPC as AMD claims 15% improvement. That brings us to pretty neat boost of performance over Zen+ (+15% ST; +20%MT). Expecting 5GHz from 1st 7nm just because foundry target such clock is bit unrealistic (but would be damn brutal jump). I do expect closer to 5GHz or more with Zen 2+. And while there were talks about denser nodes than 7nm, hints say that 7nm is here to stay for a while. Maybe even as long as intel's 14nm. You may wonder why is that. So here is why: If we kindly ignore release dated of all current Zen lines, we'll have to check all 3 candidates for release: Ryzen, Threadripper, EPYC. - Requirements for Ryzen: Improved Single threaded performance via maximum clock and IPC - Requirements for Threadripper: Some improvement on ST, but main requirement being MT performance via clock and IPC + some efficiency to stay within certain TDP - Requirements for EPYC: Just MT performance and power efficiency If 7nm clocked to 5GHz, AMD would release Ryzen 1st no matter what as that would be just flawless marketing for TR & EPYC. But it seems that EPYC is coming as 1st 7nm product line.

Fox2232:

That's expectation is bit too much. Reasonable is to expect around 5% clock uplift from current 4.35GHz max One-core boost => 4.57~4.6GHz, and reasonable boost in power efficiency allowing all core boost improvement by 10% from around 3.9~4GHz to 4.3~4.4GHz. And on top of it 10% improved IPC as AMD claims 15% improvement. That brings us to pretty neat boost of performance over Zen+ (+15% ST; +20%MT). Expecting 5GHz from 1st 7nm just because foundry target such clock is bit unrealistic (but would be damn brutal jump). I do expect closer to 5GHz or more with Zen 2+. And while there were talks about denser nodes than 7nm, hints say that 7nm is here to stay for a while. Maybe even as long as intel's 14nm. You may wonder why is that. So here is why: If we kindly ignore release dated of all current Zen lines, we'll have to check all 3 candidates for release: Ryzen, Threadripper, EPYC. - Requirements for Ryzen: Improved Single threaded performance via maximum clock and IPC - Requirements for Threadripper: Some improvement on ST, but main requirement being MT performance via clock and IPC + some efficiency to stay within certain TDP - Requirements for EPYC: Just MT performance and power efficiency If 7nm clocked to 5GHz, AMD would release Ryzen 1st no matter what as that would be just flawless marketing for TR & EPYC. But it seems that EPYC is coming as 1st 7nm product line.

It actually would not make sense to make 7nm Ryzen 1st. - As you know production just started and until GloFo and TSMC ramp up to volume production there would not be enough chips and they cost a lot more to make. - Even If AMD went with 7nm it would not be available until Q1 2019 so it make more sense release 2nd gen Ryzen this year and 3rd gen next year, probably quater or so after Epyc. 2nd gen Ryzen is already close to Intel IPC and next year on 7nm it shoulsget a lot better. Even if they did not do architectual changes and we know there did, just move to 7nm allows AMD to increase cache and reduce latency between CCX which should on its own have decent positive effect on IPC as Latency between CCX (and software optimalizations) is whats currently holding it down. Its funny that you bring up TDP as Intel is far worse in this area.

@xrodney All that is true. Honestly I expect 6 months at a max a year of AMD slaughtering Intel starting sometime in 2019, until Intel outs Ice Lake on 10nm in 2020. AMD moving EPYC to TSMC must have Intel freaking out as EPYC will be in mass production much sooner since TSMC is a good bit ahead of Gloflo on 7nm. I'm pretty sure Intel are going to lose a ton of data center sales to EPYC on 7nm in 2019. I don't think Ryzen Zen 2 will be terribly ahead ahead of Intel as getting 7nm to scale up frequencies(they need that in the desktop space) likely will take a few iterations and time to get that down with acceptable yields.

1) Intel is playing catch-up with TSMC for fab customers. 2) Intel is playing catch-up with AMD for chip design. 3) the o.e.m fab business is relatively new to Intel, but is TSMC's bread and butter. 4) complacency across the board is the source of all of Intel's woes. 5) fighting complacency is the cure to all of its ills.