Intel Shelves Z390 Express As We Knew It, Could Re-brand Z370 as Z390

[las]

Yeah, it's getting increasingly more complicate to compare. And yet, as Hector Ruiz himself once said: "customers aren't buying nanometers".
Does anyone remember today how Vega was supposed to draw less power because it was built on 14nm (as opposed to Pascal's 16nm)? We can never infer anything meaningful from these number, yet somehow we keep trying...

You are right. Maybe we should not care that much about nm..
Sandy Bridge to Ivy Bridge.. 32nm to 22nm.. Didn't change much either..

 

[bug]

You are right. Maybe we should not care that much about nm..
Sandy Bridge to Ivy Bridge.. 32nm to 22nm.. Didn't change much either..

You've completely missed my point. But carry on...

 

[las]

You've completely missed my point. But carry on...

It was a joke.

 

[ensabrenoir]

....what ever they do .....come up with a different naming structure totally cause this 3rd grade one-upper-man-ship they're doing with Amd is tired.

 

[Dave65]

Again, thank you AMD, this has put Intel in a position where they have to be innovative and tech savvy again.. Can only be healthy for consumers and the field of tech.

 

[ppn]

Obviously nanometers mean alot. not so much when the chipset has TDP of 6 watts, but still.

14 samsung and 16nm by tsmc, are very comparable in size and power usage.

22nm and 14nm both by intel are not the same. 22x22/14x14 or the latter representing 40% of the original size. or 60% less power.

 

[efikkan]

I guess 2019 looks to be another year of "spec wars". Intel might have been struggling with 10 nm so far, and some have even declared 2019 the year of AMD, but I wouldn't count them out just yet. Hopefully TSMC 7 nm is good, but we don't know yet, we will probably get some good indications in Q4 2018 as the first production batches are complete. But even if TSMC's 7 nm is more successful than any previous node, it still will take some time to improve yields and production volume. And remember that Intel's 10 nm node will continue to improve throughout 2018 and 2019.

Still, as some have already touched upon, the production node is just a small piece of the picture. As everyone should know by this point, node scale names are just pure marketing. It only denotes the smallest "design feature", not the actual density of a specific design. There are also numerous other factors which impacts the tapeout of a design, and of course yields and the voltage required. So be careful about estimating the success of various nodes before some of then are even in volume production.

 

[Tomorrow]

Intel 14nm = TSMC/GloFlo 10nm
Intel 10nm = TSMC/GloFlo 7nm
Intel 7nm = TSMC/GloFlo 5nm

TSMC/GloFlo always market their processes as more cutting edge than they actually are. So despite surface level nm advantage individual component (pitch) sizes are actually bigger than Intel's. That is why Intel had some much trouble with 14nm at the beginning and even more trouble with 10nm now.

So if TSMC/GloFlo gets 10nm out next year in mass they would have equaled Intel's 14nm down to pitch size.
However Intel superior pitch size does not mean squat if they can't get it into production in mass.

Currently it is expected that Intel's 10nm will enter mass production in 2H 2019. By that time TSMC/GloFlo will have 7nm in production meaning they would have introduced equal pitch size before intel. Even if Intel somehow gets their 7nm fasttracked to 2020/2021 TSMC/GloFlo will have 5nm by then and still atleast equal if not better.

 

[stimpy88]

That is perhaps not the most brilliant theory I've ever heard. A much more likely explanation is that, Intel spied to know that AMD axed their X490, and in great relief, canceled their 14nm Z390, which had already been caught in a great production mess.

Never say never! AMD have done this before, with their very good friends at nVidia, no less.

 

[Final_Fighter]

He's a freelancer because corporations don't want microarchitectures from scratch year after year after year. They iterate on established microarchitectures before pursuing a new one. Keller has no interest in that process.

That said, Keller is apparently senior vice president at Intel which is a cushy executive job, not engineering.

i believe they put him in this position so that hes comfortable and not going around and helping make products for other companies. if he is at intel then hes not over at amd helping make zen3 or something else.

 

[bug]

i believe they put him in this position so that hes comfortable and not going around and helping make products for other companies. if he is at intel then hes not over at amd helping make zen3 or something else.

Nobody put him in any position. Just think how the industry works. AMD's K7 or Intel's Core were launched once and they have been with us for 5 years or more. Even Netburst and Bulldozer had many years on the market. Because once you out a new architecture, you just have to make the investment back. Now out yourself in Keller's shoes: what do you do between architectures? You have your pick between tweaking something you already know inside-out or going elsewhere and building, I don't know, Apple's mobile business.
CPU building industry isn't even the only industry where consultants exist. Hell, I'd be a consultant for software businesses. Except nobody cares for quality and thinks foregoing best practices and going for quantity is how it's done.

 

[Metroid]

Intel 14nm = TSMC/GloFlo 10nm
Intel 10nm = TSMC/GloFlo 7nm
Intel 7nm = TSMC/GloFlo 5nm

You know what that means right? It means that ryzen 2 using 12nm is like intel 16nm and that means ryzen might lead if process node is equal on both sides. Ryzen's 2 single thread performance is 6% lower cause it probably means it has less transistors, with the node smaller, means more transistors which means more performance for amd ryzen. If both sides are on equal footing on process node, I already can see amd having the lead but that is me.

 

[TheGuruStud]

If Intel ever gets that 10nm process to work as intended AMD's in a for a bumpy ride. At this point the playing field is starting to level, let's just hope that AMD can keep up and not go into the comfort zone. Intel is far from irrelevant best to remember that, they have enough R&D budget to come up with something nice, it won't be this year though, the only thing that might be a saving grace would be the 8/16 SKU that's taping out.... if it tapes out this year .... this is from the consumer's POV. As for the other sectors, until the contracts run out Intel's gonna be fine, it's going to be fun to see what happens when said contracts are over.

It's fundamentally broken and will be 3 yrs late AT BEST (2020 mass production...maybe). And they're still scrambling to build a new architecture, so 2022 for IPC lift?

 

[las]

It's fundamentally broken and will be 3 yrs late AT BEST (2020 mass production...maybe). And they're still scrambling to build a new architecture, so 2022 for IPC lift?

We have an insider here. Please tell me more.

 

[TheGuruStud]

We have an insider here. Please tell me more.

Intel already admitted it. I know, reading is hard for the mentally challenged.

 

[las]

Intel already admitted it. I know, reading is hard for the mentally challenged.

Suuuure they did.

 

[Vayra86]

Intel already admitted it. I know, reading is hard for the mentally challenged.

Care to provide a source for that Intel statement? Wccftech articles dont count

 

[bug]

Care to provide a source for that Intel statement? Wccftech articles dont count

The manufacturing process cannot be fundamentally flawed because it's not a new process. It's still multi-patterning, as it was before, but Intel went for quad patterns instead of triple. See https://en.wikipedia.org/wiki/Multiple_patterning#Multi-Patterning_Practices

 

[efikkan]

As a reminder, Ice Lake was taped out on 2017-06-08 and is ready to start mass production right now. It all comes down to when Intel consider the node to be mature. When they push the button on a completely finished design it takes ~6 months to market.

 

[Tomorrow]

As a reminder, Ice Lake was taped out on 2017-06-08 and is ready to start mass production right now. It all comes down to when Intel consider the node to be mature. When they push the button on a completely finished design it takes ~6 months to market.

I have a hard time believing it considering CannonLake only exists as a measly dualcore with IGP disabled on some low end notebooks in Asia. IceLake is nowhere to be seen despite supposedly being taped out ~12 months ago.

 

[bug]

I have a hard time believing it considering CannonLake only exists as a measly dualcore with IGP disabled on some low end notebooks in Asia. IceLake is nowhere to be seen despite supposedly being taped out ~12 months ago.

Well, yields are still an issue so (extremely) limited availability is exactly what you'd expect. The Linux kernel has also received numerous patched for IceLake already, so somewhere, something exists. Irrelevant to me, since I can't buy one, but there are signs it's there.

 

[efikkan]

I have a hard time believing it considering CannonLake only exists as a measly dualcore with IGP disabled on some low end notebooks in Asia. IceLake is nowhere to be seen despite supposedly being taped out ~12 months ago.

Cannonlake was intended to be released in parallel with Coffee Lake, and as far as I know it was inteded as a mobile/low end only chip, kind of similar to Broadwell.