yep. all these clickbait articles and vids going around awhile back about amd outselling intel 2 to 1. but thats at 1 retailer that only sells boxed cpus right after the launch of amd's lineup and months after the launch of intels (when everybody that wanted one already has one).
for the price of a boxed i5 9400 a dell factory in asia can probalby build a complete i5 9400 system. and none of the factories makes these machines with amd parts because all amd has are 14nm\12nm apus with only 4 cores. while intel has a complete lineup with bundled nic, chipset, cpu etc that actually means intel machines are cheaper to make.
i would say intel outsells amd 20 to 1 at least. probably 50 to 1. its just you dont see those machines. yet they outnumber people almost everywhere we work
all intel has to do, is release cometlake with a 6\12 i5 for 250 ish and a 8\16 i7 (350 ish). get the k- model i5, get a 100 dollar z490 board and get some ram (my z370 pro4 was 110 at launch). overclock (actually near 5ghz will probably be stock) and you have gaming performance that beats amd's whole lineup.
i know not everybody just games but in desktops gaming is the driving force, the only expanding segment in a shrinking market. if you need a 12 core on mainstream then get amd. but what i just said has 90% of the population covered.
and all those people that are only gaming who said that the difference in performance between an intel or amd system in gaming are all of a sudden buying 500 dollar 12 core cpus....that game worse than 330 dollar 8700ks, and soon to be 250 dollar i5's.
Even if all they gain from another take on 14nm is better yeld, hey should feel free to do so. Maybe next AMD product will force them to lower the price.
intel should lower consumer prices (boxed processors etc) but their real customers, oems etc, actually get really good prices, buying in bundles for the billions of office pc's and laptops. a dell factory, making the cheapest entry business i5 9400 system can probalby make that whole system for the price of an i5 9400.
and yields on 14nm are awesome compared to anything else.
and remember those clickbait vids\articles about zen 2 yielding 70% and intel 35%. i mean sure the full 28 core die might only yield 35%, but take a core off and its still sellable so that actually yielded something. its not 65% of 28 core dies are thrown in the trash. plus that is including all the mesh etc
geeze what a ridiculous statement. comparing a 70mm2 8 core chiplet (not even factoring in io die etc) against a 700mm2 intel 28 core die
honestly....really....without a doubt...amd HAS to have a PR\marketing firm or division that just goes around and posts this same comment on EVERY SINGLE ARTICLE AND VIDEO KNOWN TO MAN, even when IT ISNT EVEN ABOUT INTEL
If true, had to start without pellicles (not perfected yet), in order to not be totally shamed by having to ship international Galaxy S10's with 7nm Snapdragon 8150 from competitor TSMC. Already confirmed to ship S10's to all US and other major countries using 7nm Snapdragon 8150 TSMC chips (as usual).
Are you sure "pretty well" ? Your statement is vague. What is "pretty well" ? A bunch of wafers at 50% yields (no pellicles at all because not available not even in upcoming years) done only because Samsung "need" to manufacture the next phone SOC instead of Apple will destroy the high end market? If true your definition of "pretty well" do not fit to Intel or GloFo business structure.
Ps. no mention to yields on this article, only marketing fluff
cpus running from from 65 watts to a few hundred are a whole different story. if all intel had to do was make a 1 watt soc they would have had it out a couple years ago
Apple always hurt SS one way or another. Directly SS need to compete with A12 7nm, also Apple put a lot of money into TSMC allow them to develop 7nm so quickly and benefit another SS competitor, name Qualcomm Snapdragon.
It means same as written approximately 40% area reduction for same complexity for all included logical blocks including complex one's as the transistor. It's probably even 2x for transistor but much less then 40% for SRAM.
AFAIK, and if so (as the article implies from the likely apps for the new tech), it seems worth noting, that wonderful as it is, Samsung are in a slightly different line of work to TSMC.
For CISC like intel & amd cpuS, tsmc are the go to guys.
Samsung has by far the largest portfolio of IP's and actual products that can be paird & quicker. This makes it ahead of anyone else including TSMC and larger than even Intel. LPP ain't something that you would prefer if you are trying to push oscillator frequency that CISC designs are doing for more than a decade. On the other hand it's very suitable for everything else including FPGA's, GPU's, mobile, servers.
Not much ahead, in fact "Sammy" lost Qualcomm on 7nm and in pratics right now Samsung is delivering the new process for internal use only. A big failure for Foundry division. Your "very suitable" is not supported by facts.
Looks like the entire Foundry world is in panic, they see the bulk of customers on 14nm or 12nm for yields reasons and they try to move something with absurd announcements about 5nm or 3nm. Pretty ridicule i have to say, knowing the very bad shape of EUV development. They are at the point that the operative profit forecast is smaller that R&D costs and equipment costs. IMO GloFo was right, waiting for a really suitable litographic solution for smaller geometries.
I don't know where you found your magic crystal ball that gives you insight into Foundry client acquisitions for Samsung. The foundry world seems to be doing just fine judging by the expansion plans and picking up all of Global Foundary's volume for leading edge devices.
The cost of development & testing increased significantly and FinFET structures aren't suitable for analog or mixed components. & it's all about complexity number of steps and masks, patterns. That's why EUV is so important as it reduces number of nedded patterns lowering the number of steps & driving the cost down. In time evaluation period will also go down cutting the design cost. TSMC rushed their first gen 7nm FinFET using lo density lib & it has relatively small advantages over their last 10nm node especially regarding density. They won QC deal thanks to timing. Samsung is currently in lead regarding 7nm and it is "very suitable" as it's at least 30% cheaper per gate to non EUV 7nm and 66% denser than Samsungs own 10nm which is in both cases huge. On the other hand they don't have a timing on their side but they will have a lot of free capacity while TSMC is currently over crowded. GloFo disappointed dropping the 7nm node, while 22nm FD-SOI is good & cutting edge for analog and mixed, relatively cheap and good enough for main stream & lower tier mobile being the first choice for IoT it simply ain't good enough for HPC & beading edge mobile. If GloFo had both it would have become really interesting with unique win win mix of available node's & capabilities. Samsung does own the most IP's & has the largest offering of IP's and licensable IP's for it's nodes. In fact that's how it become the largest semiconductor manufacturer.
Yeah, I think one of the major hurdles right now is that ASML are unable to speed up their deliveries of EUV equipment and the foundries just have to wait in line.
Wasn't samsung 7nm supposed to be full EUV? Did they add EUV to their 8nm to make something similar to TSMC 7nm+? Wonder if the new exynos will use this or go with 8nm (the design cycle could hinder the transition from 7 to 8nm).
Unfortunately we won't see full EUV for at least two more years. Samsungs 8nm FinFET is only the rebranded 10 nm one with denser rooting lib in use & DUV.
If by "full EUV" you mean "using EUV for all layers", no. EUV is used to replace 193nm immersion lithography for layers that would require double or quadruple patterning. For something like metal layer 10 where you're putting doing a 1000 nm trace, you will always use normal 193nm lithography. It is faster/cheaper/better.
That is what I meant (sorry am just an enthusiast, I'm not aware of all the intricacies of silicon manufacturing), I just thought that Samsung said they would go with the "full" EUV for 7nm sometime in 2019 (risk in 2018), using it only for critical layers like TSMC 7+ is news to me and the tone of the article seemed the same (it mentioned how using EUV for only a few layers would alleviate the shortage ASML Twinscan NXE:3400B that would affect a "full" EUV node, and Samsung declared a "silent revolution").
They'll move more layers over it in future generations (which means that they will make them smaller and therefore requiring of EUV), but big things like power planes are never going to be EUV. It doesn't make sense.
From what I remember from Samsung press releases, EUV was always a 'critical layer only' usage for 1st gen 7nm. There were originally more advanced 7nm nodes (i.e. 7LPPS) that would use EUV for more (but not all) masking layers, but I haven't seen an updated road map with those details in a bit.
AFAIK the only node currently on their map with 100% EUV is 5nm
7LPP was supposed to gain 30% area efficiency by reducing tracks. If the tracks are still reduced that way, the design rules are virtually the same as 10nm.
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
42 Comments
Back to Article
shabby - Wednesday, October 17, 2018 - link
In other news Intel starts its new and improved 14nm+++ process...Amandtec - Wednesday, October 17, 2018 - link
Which they will review themselves and cool with a 1000W cooler hidden under the table.Gondalf - Thursday, October 18, 2018 - link
At least it yields, allowing billions of net profit.nevcairiel - Thursday, October 18, 2018 - link
This is a low power process though, big difference the desktop chips Intel makes.bobhumplick - Sunday, October 28, 2018 - link
nevcairiel exactlybobhumplick - Saturday, November 2, 2019 - link
yep. all these clickbait articles and vids going around awhile back about amd outselling intel 2 to 1. but thats at 1 retailer that only sells boxed cpus right after the launch of amd's lineup and months after the launch of intels (when everybody that wanted one already has one).for the price of a boxed i5 9400 a dell factory in asia can probalby build a complete i5 9400 system. and none of the factories makes these machines with amd parts because all amd has are 14nm\12nm apus with only 4 cores. while intel has a complete lineup with bundled nic, chipset, cpu etc that actually means intel machines are cheaper to make.
i would say intel outsells amd 20 to 1 at least. probably 50 to 1. its just you dont see those machines. yet they outnumber people almost everywhere we work
all intel has to do, is release cometlake with a 6\12 i5 for 250 ish and a 8\16 i7 (350 ish). get the k- model i5, get a 100 dollar z490 board and get some ram (my z370 pro4 was 110 at launch). overclock (actually near 5ghz will probably be stock) and you have gaming performance that beats amd's whole lineup.
i know not everybody just games but in desktops gaming is the driving force, the only expanding segment in a shrinking market. if you need a 12 core on mainstream then get amd. but what i just said has 90% of the population covered.
and all those people that are only gaming who said that the difference in performance between an intel or amd system in gaming are all of a sudden buying 500 dollar 12 core cpus....that game worse than 330 dollar 8700ks, and soon to be 250 dollar i5's.
deil - Thursday, October 18, 2018 - link
Even if all they gain from another take on 14nm is better yeld, hey should feel free to do so. Maybe next AMD product will force them to lower the price.bobhumplick - Saturday, November 2, 2019 - link
intel should lower consumer prices (boxed processors etc) but their real customers, oems etc, actually get really good prices, buying in bundles for the billions of office pc's and laptops. a dell factory, making the cheapest entry business i5 9400 system can probalby make that whole system for the price of an i5 9400.and yields on 14nm are awesome compared to anything else.
and remember those clickbait vids\articles about zen 2 yielding 70% and intel 35%. i mean sure the full 28 core die might only yield 35%, but take a core off and its still sellable so that actually yielded something. its not 65% of 28 core dies are thrown in the trash. plus that is including all the mesh etc
geeze what a ridiculous statement. comparing a 70mm2 8 core chiplet (not even factoring in io die etc) against a 700mm2 intel 28 core die
bobhumplick - Saturday, November 2, 2019 - link
honestly....really....without a doubt...amd HAS to have a PR\marketing firm or division that just goes around and posts this same comment on EVERY SINGLE ARTICLE AND VIDEO KNOWN TO MAN, even when IT ISNT EVEN ABOUT INTELisofilm - Wednesday, October 17, 2018 - link
If true, had to start without pellicles (not perfected yet), in order to not be totally shamed by having to ship international Galaxy S10's with 7nm Snapdragon 8150 from competitor TSMC. Already confirmed to ship S10's to all US and other major countries using 7nm Snapdragon 8150 TSMC chips (as usual).Amandtec - Wednesday, October 17, 2018 - link
To be fair : GF and Intel have found 7nm extremely challanging. Samsung it doing pretty well here. TSMC is flying.Gondalf - Thursday, October 18, 2018 - link
Are you sure "pretty well" ? Your statement is vague. What is "pretty well" ?A bunch of wafers at 50% yields (no pellicles at all because not available not even in upcoming years) done only because Samsung "need" to manufacture the next phone SOC instead of Apple will destroy the high end market?
If true your definition of "pretty well" do not fit to Intel or GloFo business structure.
Ps. no mention to yields on this article, only marketing fluff
jospoortvliet - Saturday, October 20, 2018 - link
They are producing... slow, bad or expensive - something beats nothing ;-)bobhumplick - Sunday, October 28, 2018 - link
cpus running from from 65 watts to a few hundred are a whole different story. if all intel had to do was make a 1 watt soc they would have had it out a couple years agot_oven - Sunday, October 21, 2018 - link
Apple always hurt SS one way or another. Directly SS need to compete with A12 7nm, also Apple put a lot of money into TSMC allow them to develop 7nm so quickly and benefit another SS competitor, name Qualcomm Snapdragon.neblogai - Wednesday, October 17, 2018 - link
Regarding the second paragraph: does it offer a 40% area reduction, or '40% more transistors' in the same area?foobaz - Wednesday, October 17, 2018 - link
don't those work out to be the same thing?neblogai - Wednesday, October 17, 2018 - link
Well, 40% area reduction means 66.6% more transistors per area.ZolaIII - Thursday, October 18, 2018 - link
It means same as written approximately 40% area reduction for same complexity for all included logical blocks including complex one's as the transistor. It's probably even 2x for transistor but much less then 40% for SRAM.ajp_anton - Thursday, October 18, 2018 - link
-40% means 0.6x size0.6x size means 1.67x more (1/0.6)
1.67x more means +67%
vbigdeli - Wednesday, October 17, 2018 - link
I hope to see surface 7 with ARM based processorGondalf - Thursday, October 18, 2018 - link
I doubt the SOC manufacturing cost will fit with Microsoft idea of profit.lilmoe - Wednesday, October 17, 2018 - link
I see a monstrous SoC in the making. Don't disappoint me Sammy.Yojimbo - Wednesday, October 17, 2018 - link
Production gate all around transistors in 2020? I didn't know it was that close...msroadkill612 - Wednesday, October 17, 2018 - link
They seem to be using a lot of arm devices to build the factory :)eek2121 - Thursday, October 18, 2018 - link
hah, you think that's a lot, here in Nashville, 'arm' is all over the place! ;)msroadkill612 - Wednesday, October 17, 2018 - link
AFAIK, and if so (as the article implies from the likely apps for the new tech), it seems worth noting, that wonderful as it is, Samsung are in a slightly different line of work to TSMC.For CISC like intel & amd cpuS, tsmc are the go to guys.
ZolaIII - Thursday, October 18, 2018 - link
Samsung has by far the largest portfolio of IP's and actual products that can be paird & quicker. This makes it ahead of anyone else including TSMC and larger than even Intel. LPP ain't something that you would prefer if you are trying to push oscillator frequency that CISC designs are doing for more than a decade. On the other hand it's very suitable for everything else including FPGA's, GPU's, mobile, servers.Gondalf - Thursday, October 18, 2018 - link
Not much ahead, in fact "Sammy" lost Qualcomm on 7nm and in pratics right now Samsung is delivering the new process for internal use only. A big failure for Foundry division.Your "very suitable" is not supported by facts.
Looks like the entire Foundry world is in panic, they see the bulk of customers on 14nm or 12nm for yields reasons and they try to move something with absurd announcements about 5nm or 3nm. Pretty ridicule i have to say, knowing the very bad shape of EUV development.
They are at the point that the operative profit forecast is smaller that R&D costs and equipment costs.
IMO GloFo was right, waiting for a really suitable litographic solution for smaller geometries.
FullmetalTitan - Thursday, October 18, 2018 - link
I don't know where you found your magic crystal ball that gives you insight into Foundry client acquisitions for Samsung. The foundry world seems to be doing just fine judging by the expansion plans and picking up all of Global Foundary's volume for leading edge devices.ZolaIII - Friday, October 19, 2018 - link
The cost of development & testing increased significantly and FinFET structures aren't suitable for analog or mixed components. & it's all about complexity number of steps and masks, patterns. That's why EUV is so important as it reduces number of nedded patterns lowering the number of steps & driving the cost down. In time evaluation period will also go down cutting the design cost. TSMC rushed their first gen 7nm FinFET using lo density lib & it has relatively small advantages over their last 10nm node especially regarding density. They won QC deal thanks to timing. Samsung is currently in lead regarding 7nm and it is "very suitable" as it's at least 30% cheaper per gate to non EUV 7nm and 66% denser than Samsungs own 10nm which is in both cases huge. On the other hand they don't have a timing on their side but they will have a lot of free capacity while TSMC is currently over crowded. GloFo disappointed dropping the 7nm node, while 22nm FD-SOI is good & cutting edge for analog and mixed, relatively cheap and good enough for main stream & lower tier mobile being the first choice for IoT it simply ain't good enough for HPC & beading edge mobile. If GloFo had both it would have become really interesting with unique win win mix of available node's & capabilities. Samsung does own the most IP's & has the largest offering of IP's and licensable IP's for it's nodes. In fact that's how it become the largest semiconductor manufacturer.Zoolook - Friday, October 19, 2018 - link
Yeah, I think one of the major hurdles right now is that ASML are unable to speed up their deliveries of EUV equipment and the foundries just have to wait in line.porcupineLTD - Thursday, October 18, 2018 - link
Wasn't samsung 7nm supposed to be full EUV? Did they add EUV to their 8nm to make something similar to TSMC 7nm+? Wonder if the new exynos will use this or go with 8nm (the design cycle could hinder the transition from 7 to 8nm).ZolaIII - Thursday, October 18, 2018 - link
Unfortunately we won't see full EUV for at least two more years. Samsungs 8nm FinFET is only the rebranded 10 nm one with denser rooting lib in use & DUV.saratoga4 - Thursday, October 18, 2018 - link
>Wasn't samsung 7nm supposed to be full EUV?If by "full EUV" you mean "using EUV for all layers", no. EUV is used to replace 193nm immersion lithography for layers that would require double or quadruple patterning. For something like metal layer 10 where you're putting doing a 1000 nm trace, you will always use normal 193nm lithography. It is faster/cheaper/better.
porcupineLTD - Thursday, October 18, 2018 - link
That is what I meant (sorry am just an enthusiast, I'm not aware of all the intricacies of silicon manufacturing), I just thought that Samsung said they would go with the "full" EUV for 7nm sometime in 2019 (risk in 2018), using it only for critical layers like TSMC 7+ is news to me and the tone of the article seemed the same (it mentioned how using EUV for only a few layers would alleviate the shortage ASML Twinscan NXE:3400B that would affect a "full" EUV node, and Samsung declared a "silent revolution").saratoga4 - Thursday, October 18, 2018 - link
They'll move more layers over it in future generations (which means that they will make them smaller and therefore requiring of EUV), but big things like power planes are never going to be EUV. It doesn't make sense.FullmetalTitan - Thursday, October 18, 2018 - link
From what I remember from Samsung press releases, EUV was always a 'critical layer only' usage for 1st gen 7nm. There were originally more advanced 7nm nodes (i.e. 7LPPS) that would use EUV for more (but not all) masking layers, but I haven't seen an updated road map with those details in a bit.AFAIK the only node currently on their map with 100% EUV is 5nm
porcupineLTD - Thursday, October 18, 2018 - link
Thanks for the info, from indirect sources (tech sites) I was always under the impression that Samsung intended to go "full" EUV with 7nm.Anymoore - Thursday, October 18, 2018 - link
7LPP was supposed to gain 30% area efficiency by reducing tracks. If the tracks are still reduced that way, the design rules are virtually the same as 10nm.SharpEars - Friday, October 19, 2018 - link
The big question is: Will Samsung manufacture chips for other companies using its fabs like TSMC, so that there is competition in the 7 nm space?thetuna - Friday, October 19, 2018 - link
Grains of SANs??