Intel Comet Lake-S Platform Overview / Desktop Platform Consumer Roadmap (Updated)

[Taf the Ghost]

9 hours ago, Waffles13 said:

The question is which node Intel will be competing with by the time their 7nm rolls around. I have no doubt that their 7nm will be better than TSMC's 7nm, but there's a high likelihood that by the time Intel 7nm is actually on the market, TSMC will have their 5nm out and then we are in the exact same situation that we are now. That's why I hate these semantic arguments about which node is better; I'll take a worse node if it's actually on the market in products over a theoretically better node that is vaporware. 

Per TSMC's earnings call, 5nm is in Risk Production right now (so companies are starting to get back early silicon from their designs) and it enters volume production in H1 2020. AMD will be on 5nm by the time Intel is on 7nm, which should be functionally the same size of transistors. 

[Kurupt]

Interesting.

 

I am still very happy with my choice of the i7 9700k.  I will never more than likely use my PC for more than "gaming primary" along with media, movies, music, and web browsing.

[00100000]

I'd be satisfied with an Intel PC, but AMD is clearly better price to performance

[ravenshrike]

2 hours ago, Taf the Ghost said:

Per TSMC's earnings call, 5nm is in Risk Production right now (so companies are starting to get back early silicon from their designs) and it enters volume production in H1 2020. AMD will be on 5nm by the time Intel is on 7nm, which should be functionally the same size of transistors. 

That's when things should start to get really interesting. 4th or 5th gen IF and Ryzen vs 1st gen Fervoros.

[dgsddfgdfhgs]

Seriously!??

Still keeping the K /non-K for overclocking 

 

I dont trust those "reasonable" prizes, this is intel after all

or we shall see $300 on base line mobo lol

 

[Taf the Ghost]

4 minutes ago, ravenshrike said:

That's when things should start to get really interesting. 4th or 5th gen IF and Ryzen vs 1st gen Fervoros.

It's going to be a while before Intel's stacking approach shows up on normal devices. As of right now, it's very low powered devices. The first real products with 3D stacking should either be Chiplets set within a larger die for activate interposer (basically, it's a Network on Chip) or as L4 Cache. A "space saving" L4 will probably be the most effective performance bump over the next few years.

[mr moose]

9 hours ago, Waffles13 said:

Because the 10nm that Intel originally talked about, and the one that many use to defend Intel's node superiority, doesn't exist. The few 10nm parts that came out were absurdly small mobile chips with no GPU, and by all accounts the performance was a downgrade from 14nm. Plus there were rumors a year or two ago (on phone, can't easily check) that Intel themselves had downgraded the 10nm that they were actually producing from the original design, so that even that limited run likely wasn't the same node that they originally announced. 

 

Say I come out tomorrow and announce that I have invented a new type of engine that runs entirely on water, and I'll be showing it off and putting it up for sale next year. Then I spend the next 5 years trying and failing to to produce a working model. Does that mean my original design was any good? Even if it hypothetically works and is super efficient, if it's incapable of being built then it is useless. Maybe it's worth looking at as a study of what not to do in the future, but certainly no one should be looking at me as the leader of innovation if I can't actually make the thing that I designed. Like I said, it's vaporware. 

 

I hope that they get 7nm out as intended. At the end of the day all I want is rad computer shit. I just have next to no faith in Intel's manufacturing after these last few years, and it's baffling to me how often people still defend "Intel 10nm" as being superior/equal to competitors when by all accounts it appears to be impossible to produce. 

 

Sorry, but all I got out of this post was you are allowed to draw conclusion on Intel's 10nm but no one else is.   It's out, it works, it may not be in desktop processors en masse right now, but it is still a valid node and valid technology that people can discuss if they want.

[colonel_mortis]

@BiG StroOnZ I've updated the title to match the apparent consensus in the topic, and in some of the sources, that this is not genuine. Feel free to change the presentation of it as you see fit, but if you'd like to revert the change them please PM me and/or another mod first.

 

Normally we don't do this, but the topic has been going on for quite a while and many people only read the title+first post.

[RejZoR]

I don't think fab node is the main reason for Intel's problems. Clearly, they have architecture capable of reaching high clocks. What I think is screwing them up the most is the fact that they are doing everything in monolithic form and they are having hard time reaching such core counts packed in single chip using 14nm no matter how mature it is. They just don't have architecture and manufacture designed to do the AMD's chiplet thing. So they are waiting for fab process to get smaller so they can make a new monolithic big CPU. But they can't, coz they can't get the fab down to 10nm or 7nm which would allow them to do that much easier. They also can't just do it now because they don't have the logic to connect cores this way. AMD planned that years ahead, Intel apparently hasn't and has relied on projections of node shrink.

 

Eventually, even Intel will have to go the chiplet route imo. They are just sort of hanging somewhere in the middle, waiting to shrink the node to make new monolithic CPU or make the interconnect logic to do the chiplet thing. Problem is, they have neither. Last time they were gluing CPU's together was Pentium 4 era iirc.

[Taf the Ghost]

2 hours ago, RejZoR said:

I don't think fab node is the main reason for Intel's problems. Clearly, they have architecture capable of reaching high clocks. What I think is screwing them up the most is the fact that they are doing everything in monolithic form and they are having hard time reaching such core counts packed in single chip using 14nm no matter how mature it is. They just don't have architecture and manufacture designed to do the AMD's chiplet thing. So they are waiting for fab process to get smaller so they can make a new monolithic big CPU. But they can't, coz they can't get the fab down to 10nm or 7nm which would allow them to do that much easier. They also can't just do it now because they don't have the logic to connect cores this way. AMD planned that years ahead, Intel apparently hasn't and has relied on projections of node shrink.

 

Eventually, even Intel will have to go the chiplet route imo. They are just sort of hanging somewhere in the middle, waiting to shrink the node to make new monolithic CPU or make the interconnect logic to do the chiplet thing. Problem is, they have neither. Last time they were gluing CPU's together was Pentium 4 era iirc.

For 10nm, Intel stretched the DUV lithography tech too far. So they needed SAQP along with several other physical changes that seems to have not worked. (It appears cobalt got removed from the node all together.) But the real issue was Management. 10nm should have been in good shape in 2014, but they weren't willing to make the hard decisions on the node until the CEO got fired in 2018 and the new CEO was already running things. (This caused the whole "10nm is dead", as they canceled 3 of the 4 Fabs and moved them over to 7nm build outs.)

 

Intel will eventually be on chiplets, but it'll be a few more years. Though Intel's biggest issue is that AMD has a "strategic design" advantage for a while. It's pretty clear how AMD is going to push their own tech, while Intel has to maintain performance in the areas it has it right now. Intel's next big design should be Granite Rapids in 2022/2023 range, which is really the next place Intel will really compete outside of their current strength of high-clocks and laptop.

[Unixsystem]

10 hours ago, mr moose said:

 

Sorry, but all I got out of this post was you are allowed to draw conclusion on Intel's 10nm but no one else is.   It's out, it works, it may not be in desktop processors en masse right now, but it is still a valid node and valid technology that people can discuss if they want.

The "10nm" node that is out is not the 10nm node that Intel talked about years ago. The node that is out provides worse performance than their original 14nm and seemingly only works for absolutely tiny chips due to yields. 

 

If people want to discuss the 10nm that actually exists, the go for it, but in that case Intel is nowhere near fabrication leadership. People are still conflating the original intent of Intel 10nm with the gimped version that we have, and they are completely different things. 

[Tieox]

I think Intel should send Linus a sample of their next CPU but have the processor lid RGB, purely to make his head explode with "WHYYYYY"

 

 

[Unixsystem]

19 hours ago, Results45 said:

 

True. Intel has been saying that they will have 10nm out very very soon for a few years now when they should have jumped straight to 7nm. 

 

While there are already PCs with 10nm Core i3s out in the wild, I guess Intel has decided that it's not worth it to give-up on 10nm despite the "secret issues" it's having to get yields as close to 100% as possible.

 

Also, 10nm chip are in mass production (and are shipping to OEMs & partners now). We can't buy them until the end of the year. 

I'll concede that they may in fact be producing more than I was aware of. I knew about the chinese laptop CPUs but hadn't seen the more recent story. 

 

Still, I think the point still stands in terms of which "10nm" node we are actually getting. The initial numbers we heard for both power efficiency and performance uplift were huge, and while you can assume a certain amount of widdle room with regards to design versus execution, those mobile chips don't provide a lot of hope for what Ice Lake is actually going to manage, at least in terms of the node. If performance is way down from the initial reveal years and years ago, then I'd still say that my point of the original "Intel 10nm" being an unrealistic and poorly thought out node is still valid. 

 

Again, I hope I'm wrong though. Good CPUs are better for everyone. 

[BiG StroOnZ]

27 minutes ago, Waffles13 said:

I'll concede that they may in fact be producing more than I was aware of. I knew about the chinese laptop CPUs but hadn't seen the more recent story. 

 

Still, I think the point still stands in terms of which "10nm" node we are actually getting. The initial numbers we heard for both power efficiency and performance uplift were huge, and while you can assume a certain amount of widdle room with regards to design versus execution, those mobile chips don't provide a lot of hope for what Ice Lake is actually going to manage, at least in terms of the node. If performance is way down from the initial reveal years and years ago, then I'd still say that my point of the original "Intel 10nm" being an unrealistic and poorly thought out node is still valid. 

 

Again, I hope I'm wrong though. Good CPUs are better for everyone. 

 

You might be interested in this Ice Lake "leak" that just came out, although I'm not sure whether it warrants its own thread - 

Quote

Leaked Core i7-1065G7 SC benchmarks beat Ryzen 9 3900X: 

 

 

Recently, benchmark results leaked for Intel’s 10th generation Ice Lake i7-1065G7 processor, which is designed for use in laptops. Despite it being a laptop chip – so there are power and heat implications – as the Techquila website notes, the benchmarks suggest that it will beat AMD’s new Ryzen 9 3900X. What’s particularly interesting is that the Ryzen 9 3900X is a desktop processor, which means it consumes more power (105W vs the i7-1065G7’s 15W) and runs at higher frequencies. In the leaked Geekbench scores, the Intel Ice Lake Core i7-1065G7, running in a HP Spectre x360 laptop, manages a single-core score of 5,691. Meanwhile, AMD’s flagship Ryzen 9 3900X flagship scores slightly lower at 5,624. While that isn’t a huge difference, the fact that Intel’s new mobile processor appears to edge out AMD’s latest desktop flagship when it comes to single-core performance means AMD could struggle to produce mobile processors that rival Intel's. It’s worth noting that these benchmarks are not confirmed, but if they are real it tells us a few things. First of all, it seems like Intel could be getting its mojo back with laptop processors – with the Core i7-1065G7 producing some excellent results considering it’s a mobile processor.

 

Source: https://www.techradar.com/sg/news/amd-will-struggle-to-compete-with-intel-when-it-comes-to-laptop-processors

Source 2: https://www.digitaltrends.com/computing/intel-ice-lake-i7-benchmark-leak-beats-amd-ryzen-9-3900x/

Source 3: https://browser.geekbench.com/v4/cpu/13954547

[mr moose]

2 hours ago, Waffles13 said:

The "10nm" node that is out is not the 10nm node that Intel talked about years ago. The node that is out provides worse performance than their original 14nm and seemingly only works for absolutely tiny chips due to yields. 

 

If people want to discuss the 10nm that actually exists, the go for it, but in that case Intel is nowhere near fabrication leadership. People are still conflating the original intent of Intel 10nm with the gimped version that we have, and they are completely different things. 

 

So we are only allowed to discuss nodes that are in current leadership production and about their current state.

[thorhammerz]

33 minutes ago, BiG StroOnZ said:

You might be interested in this Ice Lake "leak" that just came out, although I'm not sure whether it warrants its own thread - 

Source: https://www.techradar.com/sg/news/amd-will-struggle-to-compete-with-intel-when-it-comes-to-laptop-processors

Source 2: https://www.digitaltrends.com/computing/intel-ice-lake-i7-benchmark-leak-beats-amd-ryzen-9-3900x/

Source 3: https://browser.geekbench.com/v4/cpu/13954547

It would also be interesting to consider the specs of the memory used in that benchmark - IceLake purportedly supports up to LP4/X-3733 / DDR4-3200, but the prospect of laptop OEMs using fast memory (let alone timings that aren't abjectly atrocious, as is the norm) in laptops is rather lacking in historical precedence.

[Unixsystem]

1 hour ago, mr moose said:

 

So we are only allowed to discuss nodes that are in current leadership production and about their current state.

Conversing with you is like talking to a brick wall. 

 

A brick wall that goes into every thread and deliberately misrepresents everyone else's arguments. 

[mr moose]

1 hour ago, Waffles13 said:

Conversing with you is like talking to a brick wall. 

 

A brick wall that goes into every thread and deliberately misrepresents everyone else's arguments. 

 

The reason it seems like that is because I will continue to discuss the issue at hand rather than dictate to the other person what parts of the discussion are appropriate. 

 

I happen to think any discussion of node size is relevant regardless of it's current application or status,  I don't hold the opinion that node size is relevant to purchasing or purchasing advice, consequently I will discuss both topics all day (I will always argue my perspective just like everyone else)  and why I won't tell you or anyone else to stop voicing their opinion unless that opinion is demonstrably wrong.  

 

 

[Results45]

Remember the whole debate about whether or not the 10-core CPU "specs" were a hoax?

Spoiler

 

From HERE: https://www.tomshardware.com/news/intel-comet-lake-features-specifications-availability-launch,39848.html

 

Well, I found a more reliable source to back it up (the 13th listing)!

Comet Lake-S/H/U CPU Support Listings in Linux CoreBoot:

 

Cheers!  

[AlexOak]

https://arstechnica.com/gadgets/2019/08/intel-reveals-final-details-on-ice-lake-mobile-cpus/ Im not sure if this falls under this but whatever. 

[BiG StroOnZ]

3 hours ago, AlexOak said:

https://arstechnica.com/gadgets/2019/08/intel-reveals-final-details-on-ice-lake-mobile-cpus/ Im not sure if this falls under this but whatever. 

 

 

[BiG StroOnZ]

Confirmation of the upcoming 10-core? - 

 

 

Source 6