Interesting that the language used for the 'no comment' on EMIB (product intercept) is the same as used to describe the stop-the-presses addition of hardware security mitigations to Cascade Lake (intercept product flow).
Unfortunately marketers usually hype about things they don't have a clue about. First rule of marketing: Lie, lie, lie in another words sell, sell, sell.
Keep in mind, that EMIB is likely not intended for the server/workstation group currently - but it almost guarantee to be mobile and desktop line in the future.
One thing interesting with server line of CPU's is to use multiple GPU in multi-cpu configuration. It nice to here that scalar able line has unified 2/4/8 cpu configurations
I would be too if I was in Intel's marketing department. I would talk about it once it's on better ground as of right now there is no way to positive spin it.
I'm still wondering if Intel thinks Cascade/Cooper Lake will be competitive with AMD's EPYC once its out on TSMC's 7nm process or if Ice Lake is the competitor a year later.
It's not uncommon for anyone doing their own production (Intel, Toshiba, IBM GF, etc) to not talk about process changes because they are often doing work that is not necessarily parallel with TSMC/ GF proper, experimenting with alternative processes and even substrate compositions that nonlinearly scale different components (i.e. memory traps will be larger than gate spacing, trace thickness different than trace exclusion zone, etc) compared to the traditional methods.
On that note, any chance that Anandtech will try a similar interview format with more low level competitors in the SEMI space? You know, LAM, Applied Materials, KLA Tencor, or even lower level down to the companies that actually make the EFEMs, vacuum pumps, lasers, and robots for those systems? There are some pretty interesting companies large and small that could fill in gaps that Intel and the others won't state directly
She gave more specific information about Ice Lake than what AMD has given about Epyc 2, and according to all the propaganda Epyc 2 is practically on sale right now.
The difference is that according to Intel's initial roadmaps, Ice Lake should already be in consumer's hands. It has been on their roadmaps for a long time.
AMD thus far has had the next Epyc chip penned in for 2019 and well, that is still in the future. It could fall flat on its face in terms fo promise as it obviously hasn't shipped yet but so far it isn't behind schedule.
Indeed. It seems Intel's initial plans for 10 nm products have all taped out in the design phase but are simply waiting for manufacturing to get yields right.
The initial Cannon Lake chips for ultra mobile are at least 18 months behind. Cannon Lake for desktops was killed off roughly two years ago when Coffee Lake emerged on the road maps to fill that gap. What was supposed to come after desktop Cannon Lake was Ice Lake for consumers and that is clearly a 2019 part now.
The 10 nm Knights Hill chip for the Xeon Phi was outright cancelled. Cannon Lake-EP/EX had appeared on early server roadmaps as the chip after Sky Lake-EP/EX. This has since been replaced by Cascade Lake with Cooper Lake being a stop gap between Ice Lake-SP and Cascade Lake.
This interview reads me like they don't know exactly when 10 nm will be in ready for mass production of the large dies for server parts. EMIB is the one thing that can save Intel as it'd permit breaking down the large dies at the high end into several smaller dies that can yield better. The catch is that it increases costs but at this juncture, Intel has nothing to lose.
I guess we will see a release of when AMD was on top. Lot of PR events for the next 2 years. AMD will be on 7nm next year. a full year or more ahead of Intel, given even 2020 is a big question mark for Intel (they had said 2015 for 10nm). I would say, by next year intel will not be competitive in any segment not server, not mobile, not desktop.
Not looking good for Intel but as long at they can compete where most of the volumes of sales are it's hardly a disaster. Look at HEDT as an example where TR2 is very good but Intel can compete with all but the top bin and the market for ~$1,800 HEDT is too small to be a concern. The issue for Intel is that they will more than likely need to drop prices to compete for the higher volume sales SKUs at some point.
Intel so far remains competitive in the ultra lower power area. Their tweaks to power management and their 14 nm process have been rather impressive.
Increase the power budget a bit and Intel quickly loses to AMD mobile chips. On desktop Intel still reigns in single threaded performance and can scale to high clocks, giving them performance leadership until heavy multithreaded workloads come into play. Zen 2 is a big variable here as a modest increase in both clock speeds and IPC throughput can erase Intel's edge. If AMD continues to offer more cores for the money, they'll dominate but the jury is certainly still out on this match up.
Server isn't as gloomy as many make it out to be. Yeah, CPU performance will likely be eclipsed by the next generation of Epyc parts in legacy code but Intel does have promise outside of the CPU area. Intel's on-package fabrics are still going forward (OmniPath) and on-pacakge accelerators for specific workloads are moving forward (FPGA is shipping now, Nervana soon?). Optane memory has some great potential for large dataset workloads. AVX-512 is there if code can leverage the wide SIMD processing. I wouldn't be bullish about Intel in the server space but they enough going for them to be the primary choice in some very big niche uses.
The other wild card are the numerous high profile security flaws that have hit Intel exceptionally hard. It does sound like from the 10 nm delays that Intel has worked in fixes into hardware. Silver lining I guess.
So... I still don't understand. Are the new consumer CPUs (9900k, 9700k, etc.) Cascade lake and they also contain the security mitigations for Spectre/Meltdown, or they are not Cascade lake or they just don't contain the mitigations?
Does that mean that there will be no security mitigations on those? Because if they release new CPUs without the fix that is just like spitting their customers in the face after taking their money... plain insult.
No, it's due to timing. They couldn't implement the changes into the Coffee Lake iteration. Since these new Coffee Lakes are not much different than the older ones. They were however able to fix for Cascade and I assume Intel's desktop 10th gen chips.
"To summarize the problem: essentially, Intel's CPUs ignore their operating system kernel page tables. ... The processor therefore may consult the page tables to convert the app's virtual memory address to the corresponding physical RAM address.
THIS TAKES TIME, and today's Intel CPUs WILL NOT WAIT for a page table walk to complete ..."
The INTENDED designs of INTENTIONALLY not following the specs, the "Foreshadow" triplets and "Meltdown," to gain performance are the INTENDED frauds of cheating for performance in criminal court, aren't they?
The summer is almost over, but the related government organizations are still hibernating with paycheck auto-and-direct deposit?
One simple question to clarify Intel's design mentality whether following the specs or not:
Is "not following the well defined privilege levels" a design bug or "the INTENDED design, no bug"?
AGAIN, not following the specs causes "Foreshadow" triplets!!!
Is there more "not following the specs"??? Or the mentality is just wait to let outsiders peel onions to get them out instead of proactively disclose all "not following the specs" and having them mitigated all at once???
Very nice in formulated Interview for what they can mention.
I found a couple of things that really stuck out
1. Spectra Meltdown fixes in hardware coming before 10nm. Specially in Cascade - it to me shows that Intel is very hard in resolving the issue in Hardware also.
2. That Intel is investigating EMIB on server line.
3. "LS: I think that’s just getting into a mess of naming things that doesn’t serve anyone’s interests." - this topic is not about the naming of line of products but I believe she is referring to 10nm by Intel and 10nm/7nm used by competition - that they can be name that but there not equal. Also 14nm in Cascade Lake - is vastly different in Architecture than earlier 14nm
Overall, nice interview and I think Lisa is very clever in telling what she can not telling too much. The biggest thing I got out - improvement come even if 10nm is not there yet and 10nm is just a name.
1) It seems that Cascade was slightly delayed to get the fixes into hardware. This I feel was the appropriate action to take given how impactful the software fixes were in the server area.
2) They have been investigating this for awhile. The arrival of EMIB has been a question of when and not if for awhile.
3) This is rather important as Intel previously made distinction between 10 nm and 10 nm+ nodes. Desktop Ice Lake is to be made using the second generation 10 nm+ node for example while mobile Cannon Lake is the initial 10 nm node. This reads to me that Intel themselves doesn't really know or what they were going to call '10 nm+' has morphed into the plain 10 nm terminology.
Since the new process seems to be a train-wreck in progress, it doesn't seem like rocket science for Intel to spend some of the time working on the security defects right?
It's hard to keep track of all the 14nm parts. So far we have seen: Broadwell - 14nm version of Haswell Skylake - successor architecture to Haswell Kaby Lake Coffee Lake
And now we are hearing about: Cascade Lake Cooper Lake
So that's five iterations of the Skylake architecture on 14nm. There's also Cannon Lake (the 10nm shrink of Skylake) which is being used in dual core mobile processors only.
Broadwell does contain architecture enhancements over Haswell. They're just not very big. On the other hand Skylake, Kaby Lake, Coffee Lake, Whiskey Lake and Cannon Lake are all the same, CPU wise.
Skylake-X, Cascade and Cooper differ in architecture.
It would be great if you asked the readers for questions in advance. I'm sure we would have some interesting ones. I'd hoped there might be some commentary about the foreshadow security vulnerabilities too.
Lisa Spelman seems unwilling to spill the beans on EMIB, 10nm, and either had limited knowledge, or was not willing to share them.
It may very well be that some things are not going as well as Intel would like. Certainly, a 2019 or later release of 10nm is 3 years later - perhaps even more if there are even further delays. We did not get a serious answer for example as to why 10nm is so badly delayed.
I think that Lisa must be under tight NDA and cannot tell us much. It was still an interesting interview. Thanks for holding the interview.
If you have a follow up, could you ask her why in the world they redid the Xeon naming scheme? the Bronze/Silver/Gold/Platinum scheme is honestly far less obvious then the old scheme.
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
40 Comments
Back to Article
edzieba - Wednesday, August 15, 2018 - link
Interesting that the language used for the 'no comment' on EMIB (product intercept) is the same as used to describe the stop-the-presses addition of hardware security mitigations to Cascade Lake (intercept product flow).imaheadcase - Wednesday, August 15, 2018 - link
First rule of marketing, if you don't know don't say. What you do know hype it up like its gods gift to men.milkod2001 - Thursday, August 16, 2018 - link
Unfortunately marketers usually hype about things they don't have a clue about. First rule of marketing: Lie, lie, lie in another words sell, sell, sell.HStewart - Wednesday, August 15, 2018 - link
Keep in mind, that EMIB is likely not intended for the server/workstation group currently - but it almost guarantee to be mobile and desktop line in the future.One thing interesting with server line of CPU's is to use multiple GPU in multi-cpu configuration. It nice to here that scalar able line has unified 2/4/8 cpu configurations
HollyDOL - Wednesday, August 15, 2018 - link
It seems like every passing around 10nm caused severely tight lips.FreckledTrout - Wednesday, August 15, 2018 - link
I would be too if I was in Intel's marketing department. I would talk about it once it's on better ground as of right now there is no way to positive spin it.I'm still wondering if Intel thinks Cascade/Cooper Lake will be competitive with AMD's EPYC once its out on TSMC's 7nm process or if Ice Lake is the competitor a year later.
basroil - Wednesday, August 15, 2018 - link
It's not uncommon for anyone doing their own production (Intel, Toshiba, IBM GF, etc) to not talk about process changes because they are often doing work that is not necessarily parallel with TSMC/ GF proper, experimenting with alternative processes and even substrate compositions that nonlinearly scale different components (i.e. memory traps will be larger than gate spacing, trace thickness different than trace exclusion zone, etc) compared to the traditional methods.On that note, any chance that Anandtech will try a similar interview format with more low level competitors in the SEMI space? You know, LAM, Applied Materials, KLA Tencor, or even lower level down to the companies that actually make the EFEMs, vacuum pumps, lasers, and robots for those systems? There are some pretty interesting companies large and small that could fill in gaps that Intel and the others won't state directly
quadrivial - Wednesday, August 15, 2018 - link
They sure went icy about Ice Lake. I guess nobody at Intel wants to talk about 10nm anything.CajunArson - Wednesday, August 15, 2018 - link
She gave more specific information about Ice Lake than what AMD has given about Epyc 2, and according to all the propaganda Epyc 2 is practically on sale right now.DigitalFreak - Wednesday, August 15, 2018 - link
U mad, bro?Kevin G - Wednesday, August 15, 2018 - link
The difference is that according to Intel's initial roadmaps, Ice Lake should already be in consumer's hands. It has been on their roadmaps for a long time.AMD thus far has had the next Epyc chip penned in for 2019 and well, that is still in the future. It could fall flat on its face in terms fo promise as it obviously hasn't shipped yet but so far it isn't behind schedule.
WinterCharm - Friday, August 24, 2018 - link
And here we see the disgruntled Intel Shill, his natural habitat disturbed by AMD recently becoming competitive again.Kevin G - Wednesday, August 15, 2018 - link
Indeed. It seems Intel's initial plans for 10 nm products have all taped out in the design phase but are simply waiting for manufacturing to get yields right.The initial Cannon Lake chips for ultra mobile are at least 18 months behind. Cannon Lake for desktops was killed off roughly two years ago when Coffee Lake emerged on the road maps to fill that gap. What was supposed to come after desktop Cannon Lake was Ice Lake for consumers and that is clearly a 2019 part now.
The 10 nm Knights Hill chip for the Xeon Phi was outright cancelled. Cannon Lake-EP/EX had appeared on early server roadmaps as the chip after Sky Lake-EP/EX. This has since been replaced by Cascade Lake with Cooper Lake being a stop gap between Ice Lake-SP and Cascade Lake.
This interview reads me like they don't know exactly when 10 nm will be in ready for mass production of the large dies for server parts. EMIB is the one thing that can save Intel as it'd permit breaking down the large dies at the high end into several smaller dies that can yield better. The catch is that it increases costs but at this juncture, Intel has nothing to lose.
sharath.naik - Wednesday, August 15, 2018 - link
I guess we will see a release of when AMD was on top. Lot of PR events for the next 2 years. AMD will be on 7nm next year. a full year or more ahead of Intel, given even 2020 is a big question mark for Intel (they had said 2015 for 10nm). I would say, by next year intel will not be competitive in any segment not server, not mobile, not desktop.smilingcrow - Wednesday, August 15, 2018 - link
Not looking good for Intel but as long at they can compete where most of the volumes of sales are it's hardly a disaster.Look at HEDT as an example where TR2 is very good but Intel can compete with all but the top bin and the market for ~$1,800 HEDT is too small to be a concern.
The issue for Intel is that they will more than likely need to drop prices to compete for the higher volume sales SKUs at some point.
Kevin G - Wednesday, August 15, 2018 - link
Intel so far remains competitive in the ultra lower power area. Their tweaks to power management and their 14 nm process have been rather impressive.Increase the power budget a bit and Intel quickly loses to AMD mobile chips. On desktop Intel still reigns in single threaded performance and can scale to high clocks, giving them performance leadership until heavy multithreaded workloads come into play. Zen 2 is a big variable here as a modest increase in both clock speeds and IPC throughput can erase Intel's edge. If AMD continues to offer more cores for the money, they'll dominate but the jury is certainly still out on this match up.
Server isn't as gloomy as many make it out to be. Yeah, CPU performance will likely be eclipsed by the next generation of Epyc parts in legacy code but Intel does have promise outside of the CPU area. Intel's on-package fabrics are still going forward (OmniPath) and on-pacakge accelerators for specific workloads are moving forward (FPGA is shipping now, Nervana soon?). Optane memory has some great potential for large dataset workloads. AVX-512 is there if code can leverage the wide SIMD processing. I wouldn't be bullish about Intel in the server space but they enough going for them to be the primary choice in some very big niche uses.
The other wild card are the numerous high profile security flaws that have hit Intel exceptionally hard. It does sound like from the 10 nm delays that Intel has worked in fixes into hardware. Silver lining I guess.
realslipk - Wednesday, August 15, 2018 - link
So... I still don't understand. Are the new consumer CPUs (9900k, 9700k, etc.) Cascade lake and they also contain the security mitigations for Spectre/Meltdown, or they are not Cascade lake or they just don't contain the mitigations?DigitalFreak - Wednesday, August 15, 2018 - link
The 9xxx series processors are still Coffee Lake.realslipk - Wednesday, August 15, 2018 - link
Does that mean that there will be no security mitigations on those? Because if they release new CPUs without the fix that is just like spitting their customers in the face after taking their money... plain insult.SonicKrunch - Wednesday, August 15, 2018 - link
No, it's due to timing. They couldn't implement the changes into the Coffee Lake iteration. Since these new Coffee Lakes are not much different than the older ones. They were however able to fix for Cascade and I assume Intel's desktop 10th gen chips.MrSpadge - Thursday, August 16, 2018 - link
A speculation is that that's the reason hardly any 9xxx CPUs have HT.maroon1 - Saturday, August 18, 2018 - link
What has HT anything to do with Spectre/Meltdown ?! Just askingwow&wow - Wednesday, August 15, 2018 - link
"To summarize the problem: essentially, Intel's CPUs ignore their operating system kernel page tables....
The processor therefore may consult the page tables to convert the app's virtual memory address to the corresponding physical RAM address.
THIS TAKES TIME, and today's Intel CPUs WILL NOT WAIT for a page table walk to complete ..."
The INTENDED designs of INTENTIONALLY not following the specs, the "Foreshadow" triplets and "Meltdown," to gain performance are the INTENDED frauds of cheating for performance in criminal court, aren't they?
The summer is almost over, but the related government organizations are still hibernating with paycheck auto-and-direct deposit?
smilingcrow - Wednesday, August 15, 2018 - link
Don't drive drunk or at least until you gets your eyes checked out.wow&wow - Wednesday, August 15, 2018 - link
One simple question to clarify Intel's design mentality whether following the specs or not:Is "not following the well defined privilege levels" a design bug or "the INTENDED design, no bug"?
AGAIN, not following the specs causes "Foreshadow" triplets!!!
Is there more "not following the specs"??? Or the mentality is just wait to let outsiders peel onions to get them out instead of proactively disclose all "not following the specs" and having them mitigated all at once???
HStewart - Wednesday, August 15, 2018 - link
Ian,Very nice in formulated Interview for what they can mention.
I found a couple of things that really stuck out
1. Spectra Meltdown fixes in hardware coming before 10nm. Specially in Cascade - it to me shows that Intel is very hard in resolving the issue in Hardware also.
2. That Intel is investigating EMIB on server line.
3. "LS: I think that’s just getting into a mess of naming things that doesn’t serve anyone’s interests." - this topic is not about the naming of line of products but I believe she is referring to 10nm by Intel and 10nm/7nm used by competition - that they can be name that but there not equal. Also 14nm in Cascade Lake - is vastly different in Architecture than earlier 14nm
Overall, nice interview and I think Lisa is very clever in telling what she can not telling too much. The biggest thing I got out - improvement come even if 10nm is not there yet and 10nm is just a name.
smilingcrow - Wednesday, August 15, 2018 - link
"10nm is just a name."Maybe that can be Intel's next marketing line?
"We are not a number, we are Intel".
Kevin G - Wednesday, August 15, 2018 - link
1) It seems that Cascade was slightly delayed to get the fixes into hardware. This I feel was the appropriate action to take given how impactful the software fixes were in the server area.2) They have been investigating this for awhile. The arrival of EMIB has been a question of when and not if for awhile.
3) This is rather important as Intel previously made distinction between 10 nm and 10 nm+ nodes. Desktop Ice Lake is to be made using the second generation 10 nm+ node for example while mobile Cannon Lake is the initial 10 nm node. This reads to me that Intel themselves doesn't really know or what they were going to call '10 nm+' has morphed into the plain 10 nm terminology.
HStewart - Wednesday, August 15, 2018 - link
I am talking about more specific 14nm and 10nm - and also more importantly competition's 10nm and 7nm.dtack - Wednesday, August 15, 2018 - link
Since the new process seems to be a train-wreck in progress, it doesn't seem like rocket science for Intel to spend some of the time working on the security defects right?KAlmquist - Wednesday, August 15, 2018 - link
It's hard to keep track of all the 14nm parts. So far we have seen:Broadwell - 14nm version of Haswell
Skylake - successor architecture to Haswell
Kaby Lake
Coffee Lake
And now we are hearing about:
Cascade Lake
Cooper Lake
So that's five iterations of the Skylake architecture on 14nm. There's also Cannon Lake (the 10nm shrink of Skylake) which is being used in dual core mobile processors only.
Kevin G - Thursday, August 16, 2018 - link
There is also the 14 mm Whiskey Lake.Also Cannon Lake originally had desktop parts but those were cancelled two years ago.
MrSpadge - Thursday, August 16, 2018 - link
Broadwell does contain architecture enhancements over Haswell. They're just not very big. On the other hand Skylake, Kaby Lake, Coffee Lake, Whiskey Lake and Cannon Lake are all the same, CPU wise.Skylake-X, Cascade and Cooper differ in architecture.
dtack - Wednesday, August 15, 2018 - link
It would be great if you asked the readers for questions in advance. I'm sure we would have some interesting ones. I'd hoped there might be some commentary about the foreshadow security vulnerabilities too.CrazyElf - Wednesday, August 15, 2018 - link
Lisa Spelman seems unwilling to spill the beans on EMIB, 10nm, and either had limited knowledge, or was not willing to share them.It may very well be that some things are not going as well as Intel would like. Certainly, a 2019 or later release of 10nm is 3 years later - perhaps even more if there are even further delays. We did not get a serious answer for example as to why 10nm is so badly delayed.
I think that Lisa must be under tight NDA and cannot tell us much. It was still an interesting interview. Thanks for holding the interview.
HollyDOL - Thursday, August 16, 2018 - link
I think so as well... And after all no answer at all is also an answer...Mr Perfect - Thursday, August 16, 2018 - link
If you have a follow up, could you ask her why in the world they redid the Xeon naming scheme? the Bronze/Silver/Gold/Platinum scheme is honestly far less obvious then the old scheme.WasHopingForAnHonestReview - Friday, August 17, 2018 - link
"we said that a bunch of PhDs all changed their theses for it"Ive learned a new word. Theses. Plural of thesis. I dont like this word.
Robertim - Saturday, August 18, 2018 - link
Can't wait to see Intel behind AMD in every aspect next year (price, performance, efficiency). They just can't fight the 7nm with their 14nm++++++*[email protected] - Sunday, August 19, 2018 - link
I usually enjoy reading interviews here. This gal was not enjoyable, she had nothing to say.