Coffeelake thread, benchmarks, reviews, input, everything.

[krumme]

Of course you'd need a mask set, but an 8 core CFL will probably generate literally billions in revenue during its life time, $10-20 million for a new mask set is nothing.

Agree its revenue in the B. But where does this 10 to 20M come from?

 

[Arachnotronic]

Agree its revenue in the B. But where does this 10 to 20M come from?

State of the art production mask tapeouts can cost as much as $3M. Chips that ship in the millions must use these production masks to get cost down. Meanwhile, companies that sell chips in low volume at high prices may choose to use one time shuttles (like MOSIS) to produce a batch of chips at a time using a multi project

http://www.adapteva.com/andreas-blog/semiconductor-economics-101/

I'm assuming that Intel masks are significantly more expensive than typical mask sets, and even then, it's peanuts.

 

[PeterScott]

http://www.adapteva.com/andreas-blog/semiconductor-economics-101/

I'm assuming that Intel masks are significantly more expensive than typical mask sets, and even then, it's peanuts.

https://www.anandtech.com/show/11658/the-amd-ryzen-3-1300x-ryzen-3-1200-cpu-review
"... to design a new set of masks on GlobalFoundries 14nm process is many tens of millions of dollars ..."

I can't find the link, but I remember reading it was 30 million for 22nm masks and up to 80 million for 14nm masks. I also remember reading that smaller the process the more masking steps and they up over 50 masking steps now.

 

[krumme]

https://www.anandtech.com/show/11658/the-amd-ryzen-3-1300x-ryzen-3-1200-cpu-review
"... to design a new set of masks on GlobalFoundries 14nm process is many tens of millions of dollars ..."

I can't find the link, but I remember reading it was 30 million for 22nm masks and up to 80 million for 14nm masks. I also remember reading that smaller the process the more masking steps and they up over 50 masking steps now.

Yeaa i remember something about 80-100M too. Not peanuts i would say even for chipzilla but compared to having a uncompetitive product or forcing a new process it surely is a very cheap insurance.
Seems like a super desktop product to me with its high fmax and they can use 10nm for servers and mobile.
Intel have shown thet can manage multiple scenarios at the same time. Haifa and Oregon competing and cooperating.

 

[Arachnotronic]

https://www.anandtech.com/show/11658/the-amd-ryzen-3-1300x-ryzen-3-1200-cpu-review
"... to design a new set of masks on GlobalFoundries 14nm process is many tens of millions of dollars ..."

I can't find the link, but I remember reading it was 30 million for 22nm masks and up to 80 million for 14nm masks. I also remember reading that smaller the process the more masking steps and they up over 50 masking steps now.

$80 million would still be negligible for a product that will generate billions of dollars.

Put another way -- Intel could lose far more than $80 million in CPU sales to AMD if it doesn't have a leading 8 core product.

 

[PeterScott]

$80 million would still be negligible for a product that will generate billions of dollars.

Put another way -- Intel could lose far more than $80 million in CPU sales to AMD if it doesn't have a leading 8 core product.

Again it doesn't show up on any leaked roadmap, and Intel is well positioned with 6C coffee lake for the coming year.

It could happen, but I don't expect it, based on no real evidence.

We know Ice Lake is officially gen 9, and should show up about a year after Gen 8.

IMO Ice Lake will be the next major update for the desktop, not 8C CL.

 

[raghu78]

Again it doesn't show up on any leaked roadmap, and Intel is well positioned with 6C coffee lake for the coming year.

It could happen, but I don't expect it, based on no real evidence.

We know Ice Lake is officially gen 9, and should show up about a year after Gen 8.

IMO Ice Lake will be the next major update for the desktop, not 8C CL.

Given Intel's 10nm woes do not expect Icelake before H1 2019. CFL 8C is the next major CPU from Intel and should probably arrive by mid-early q3 2018. Moreover I think Intel's execution at 10 does not instill a lot of confidence that 10+ will solve all the problems with yield and performance. Icelake might end up clocking much lower than CFL 8c for base, all core turbo and max turbo. So it remains to be seen what is the nett gain in terms of ST and MT over CFL 8c.

 

[Qwertilot]

How would they make 8c CFL work? Even 6c needed some quite real perf/watt improvements to make true sense.

 

[epsilon84]

How would they make 8c CFL work? Even 6c needed some quite real perf/watt improvements to make true sense.

The same way AMD makes it work I guess - by using lower clockspeeds (and voltages) which in turn lowers power consumption.

Hypothetically speaking, a ~4GHz 8C CFL chip should be very competitive even against Ryzen 2. Intel and AMD will somewhat converge, both addressing their respective weaknesses (core count for Intel, clockspeed/IPC for AMD)

 

[jpiniero]

How would they make 8c CFL work? Even 6c needed some quite real perf/watt improvements to make true sense.

Lower clocks obviously.

 

[PeterScott]

Given Intel's 10nm woes do not expect Icelake before H1 2019. CFL 8C is the next major CPU from Intel and should probably arrive by mid-early q3 2018. Moreover I think Intel's execution at 10 does not instill a lot of confidence that 10+ will solve all the problems with yield and performance. Icelake might end up clocking much lower than CFL 8c for base, all core turbo and max turbo. So it remains to be seen what is the nett gain in terms of ST and MT over CFL 8c.

Based on what? Wishing it so? Wishful thinking doesn't make CPU's materialize.

ICE Lake is on the official Intel roadmap. 8C CL is not.

I expect ICE Lake will be released in 18-Q4, Volume in 19-Q1. Approximately One year after a similar release for Coffee Lake.

Which lines up with latest process projections:

 

[Qwertilot]

Lower clocks obviously.

Hardly much of an upgrade then is it? They need to maintain/ideally increase single thread performance for their top mainstream chips. Still very important for a lot of their uses.

 

[Dribble]

Lower clocks obviously.

Mainstream that wouldn't work. Single thread performance is more important going from 6->8 cores. No one is going to buy it if it does a lot of mainstream tasks slower. If you really have a use for 8+ cores you'd buy HEDT with it's quad channel memory and other high core count features.

 

[LTC8K6]

How would they make 8c CFL work? Even 6c needed some quite real perf/watt improvements to make true sense.

Same clocks as the 8700 should work okay, it's only 65W tdp. 4.2 for 8 cores and then match the 8700 for the rest of the speeds.

 

[IEC]

Mainstream that wouldn't work. Single thread performance is more important going from 6->8 cores. No one is going to buy it if it does a lot of mainstream tasks slower. If you really have a use for 8+ cores you'd buy HEDT with it's quad channel memory and other high core count features.

ST turbo can be the same or even slightly higher with no impact on TDP.

All-core turbo will by necessity be slightly lower, but the MT performance can still be better overall thanks to 33% more cores vs a hexacore.

 

[Zucker2k]

Hardly much of an upgrade then is it? They need to maintain/ideally increase single thread performance for their top mainstream chips. Still very important for a lot of their uses.

4.0 would be the all core turbo. Single thread turbo would be 4.7GHz or higher.

 

[Arachnotronic]

Based on what? Wishing it so? Wishful thinking doesn't make CPU's materialize.

ICE Lake is on the official Intel roadmap. 8C CL is not.

I expect ICE Lake will be released in 18-Q4, Volume in 19-Q1. Approximately One year after a similar release for Coffee Lake.

Which lines up with latest process projections:

I don't see Ice Lake here:

Looks to me that Z390 will begin shipping at the end of Q2. 8 core CFL introduced alongside Z390 would be my bet.

 

[jpiniero]

Mainstream that wouldn't work. Single thread performance is more important going from 6->8 cores. No one is going to buy it if it does a lot of mainstream tasks slower. If you really have a use for 8+ cores you'd buy HEDT with it's quad channel memory and other high core count features.

The SCT (at least with 8C CFL) doesn't need to be lower, esp if Intel were to no longer require that all cores hit the top turbo frequency. A theoretical Icelake desktop processor is another story since we can be confident it would clock lower.

 

[PeterScott]

I don't see Ice Lake here:

Looks to me that Z390 will begin shipping at the end of Q2. 8 core CFL introduced alongside Z390 would be my bet.

That's a chipset. No sign of 8C CL.

I do see Ice Lake here:
https://www.intel.com/content/www/u...rocessors-and-chipsets/ice-lake/overview.html
"The Ice Lake processor family is the next generation Intel® Core™ processor family. These processors utilize Intel’s industry-leading 10 nm+ process technology."

That is straight public announcement from Intel. This is Intels official Gen 9 CPU, it should appear about year after Gen 8. It will be built on 10nm+, which has a process readiness of 18-H2. 10nm+ is the perfect opportunity to up core counts again, as it has much improved perf/watt.

Basically we have some varying expectations of publicly confirmed product (Ice-Lake) and of a totally unconfirmed/unsubstantied one (8C CL) dropping in a similar time frame.

I am lot more likely to believe the one Intel Publicly confirmed, than the other which is largely imagined.

 

[TheF34RChannel]

That's a chipset. No sign of 8C CL.

I do see Ice Lake here:
https://www.intel.com/content/www/u...rocessors-and-chipsets/ice-lake/overview.html
"The Ice Lake processor family is the next generation Intel® Core™ processor family. These processors utilize Intel’s industry-leading 10 nm+ process technology."

That is straight public announcement from Intel. This is Intels official Gen 9 CPU, it should appear about year after Gen 8. It will be built on 10nm+, which has a process readiness of 18-H2. 10nm+ is the perfect opportunity to up core counts again, as it has much improved perf/watt.

Basically we have some varying expectations of publicly confirmed product (Ice-Lake) and of a totally unconfirmed/unsubstantied one (8C CL) dropping in a similar time frame.

I am lot more likely to believe the one Intel Publicly confirmed, than the other which is largely imagined.

Bold: that's my issue with this whole thing as well. I believe the 8C CFL originated at a support guy (I highly doubt support staff have in the ins and outs on unannounced and unreleased future products)? It then spread rapidly across the internet hype train simply because we want an 8C (which we will get, confirmed, with ICL). Recently I too said to be convinced one was coming but having thought it over more clearly it doesn't make a whole lot of sense in light of a by then nearby 8C ICL release. Time will tell, but I don't see it [8C CFL].

 

[jpiniero]

Recently I too said to be convinced one was coming but having thought it over more clearly it doesn't make a whole lot of sense.

That's sort of what I've been saying; that the 8C CFL only makes sense if Icelake desktop was cancelled due to 10 nm yield.

 

[Lodix]

That's a chipset. No sign of 8C CL.

I do see Ice Lake here:
https://www.intel.com/content/www/u...rocessors-and-chipsets/ice-lake/overview.html
"The Ice Lake processor family is the next generation Intel Core processor family. These processors utilize Intel’s industry-leading 10 nm+ process technology."

That is straight public announcement from Intel. This is Intels official Gen 9 CPU, it should appear about year after Gen 8. It will be built on 10nm+, which has a process readiness of 18-H2. 10nm+ is the perfect opportunity to up core counts again, as it has much improved perf/watt.

Basically we have some varying expectations of publicly confirmed product (Ice-Lake) and of a totally unconfirmed/unsubstantied one (8C CL) dropping in a similar time frame.

I am lot more likely to believe the one Intel Publicly confirmed, than the other which is largely imagined.

And per intel we should have had alredy 10nm cpus by now. If they can't have 10nm ready now I hardly doubt they will be able to release mass production of 10nm+ in a meaningful volumen in 2018. Which is a bad thing because then they will face 7nm products from AMD.

 

[PeterScott]

And per intel we should have had alredy 10nm cpus by now. If they can't have 10nm ready now I hardly doubt they will be able to release mass production of 10nm+ in a meaningful volumen in 2018. Which is a bad thing because then they will face 7nm products from AMD.

I really don't think you are going to see 7nm volume from AMD in 2018.

 

[Lodix]

I really don't think you are going to see 7nm volume from AMD in 2018.

Neither do I, but I was talking about 2019.

 

[jpiniero]

Yeah. The 8C CFL would be a Q3/4 2018 - Q3/4 2019 type deal. So in theory any 7 nm AMD product would be out before Intel refreshes.