Intel's response to RyZen.

[Arachnotronic]

Amd could do a 6 ccx Zen for 12 cores with higher speed ddr4 for 210 to 220 mm on 14llp if they want to go that path. That's a 48 core chip in server space. Unless Intel moves away from monolithic soc's it's hard to do a core war with amd/Zen/infinity fabric

Intel is moving to building future dies in pieces connected via EMIB.

 

[Arachnotronic]

Intel will ship 10nm in volume only in 2018 but that's gonna be Core M kind of thing and desktop is more likely early 2019.
What does 10nm give them in desktop though? Same core and slightly higher clocks, if 10nm clocks.

Desktop/high perf mobile should be on 10nm+ in early 2019, using the Ice Lake core -- a new CPU core.

AMD in early 2019 can go 7nm and 12 cores if there are no delays for GloFo's 7nm. Feature size it would be slightly ahead of Intel, not that it matters much.
Look at Zen, they are half a node behind but pack 8MB L3$ in a smaller area.

Provide evidence that GloFo 7nm has smaller feature sizes than Intel 10nm.

Core wise, Intel's core is very very mature, hard to get more out of it. Zen is new, with Zen plus they can boost both IPC and clocks in a significant way and 7nm gives them more room in 2019.
We'll see soon how Zen really performs but AMD's position relative to Intel can only improve before Intel has a new core.

Ice Lake should be a new core.

Ofc Intel's reaction is driven by Wall Street. How do they maximize market share, margins and ASPs? Best way is to do nothing ,just count on marketing and "complicated" deals with OEMs and large retailers.

Lagrange multipliers?

In all seriousness, you maximize margin by improving the competitiveness of your products while at the same time keeping cost structure in check at each price point (i.e. segmentation).

But Intel has never been in this position before. They have 97% revenue share in PC, inflated ASPs and margins and the market is declining. If they lose 15% share and at the same time ASPs decline 10% and the market declines 5%, it's really bad.Ofc that's for the overall PC market not Summit Ridge as Raven Ridge is also Ryzen.
In desktop they might just sell 6 cores+GPU and argue that 6 cores is enough. A few months later, the CEO gets the boot and they start working on something relevant.

Intel is predicting a low-single digit decline thru 2018 in its client business, so clearly they're not expecting to lose the kind of revenue you are thinking (0.85*0.9*0.95 = ~0.73).

 

[Conroe]

How about a free game code? FX has one.

 

[revanchrist]

There are some whispers from Chiphell that Intel will do a price adjustment for current Kabylake and Broadwell-X lineup by 19th of March. And about something something delay in the upcoming 10nm roadmap. No idea about the reliability of this rumor, will check back again by end of March.

 

[imported_jjj]

Desktop/high perf mobile should be on 10nm+ in early 2019, using the Ice Lake core -- a new CPU core.


Provide evidence that GloFo 7nm has smaller feature sizes than Intel 10nm.


Lagrange multipliers?

In all seriousness, you maximize margin by improving the competitiveness of your products while at the same time keeping cost structure in check at each price point (i.e. segmentation).



Intel is predicting a low-single digit decline thru 2018 in its client business, so clearly they're not expecting to lose the kind of revenue you are thinking (0.85*0.9*0.95 = ~0.73).

Ice Lake in early 2019 is pure speculation and ,if it is,, there is no indication that it is faster than the current core as for Intel a smaller low power core is a better fit in PC.
Someone else has provided a slide on the previous page with current expectations for future nodes and , BTW, you can't even provide evidence that Intel will ever have 10nm, it's just not how it works.Not that feature size has much to do with perf or cost nowadays.
Increasing competitiveness with what? Even your extra optimistic claim puts that new core 2 years from now.

Intel's predictions are so ridiculous that it is not even funny. Intel is predicting that they keep market share, they push ASPs up and AMD doesn't exist, it's all on a mild decline in the PC market.
But hey, lets do some math ,not that you'll ever accept anything that doesn't paint Intel in a shiny bright light.
Lets say the PC market was 32 billion in 2016 and we know AMD had about 1 billion of that.
Lets say in 2018 the PC market is down in 2018, vs 2016 by 10 % in units. Lets say that a competitive AMD will have 15% revenue share. And lets say that ASPs compress by only 10% due to AMD being back. Added all up Intel's PC revenue would be down by 9 billion.
And that's a mild scenario.So sure lets listen to all the idiotic claims Intel makes and believe in them,like it's a religion. Intel its current strategy won't be able to support the fabs very soon and they might not survive if they don't change leadership.

 

[CHADBOGA]

Intel is predicting a low-single digit decline thru 2018 in its client business, so clearly they're not expecting to lose the kind of revenue you are thinking (0.85*0.9*0.95 = ~0.73).

They might end up being proven wrong on this prediction.

 

[Arachnotronic]

Ice Lake in early 2019 is pure speculation

Intel has said that it's committed to annual product refreshes. Q1 2017 = Kaby Lake, Q1 2018 = Coffee Lake, Q1 2019 = Ice Lake, Q1 2020 = Tiger Lake, and so on.

and ,if it is,, there is no indication that it is faster than the current core as for Intel a smaller low power core is a better fit in PC.

All of Intel's "new cores" are faster than the older ones.

Someone else has provided a slide on the previous page with current expectations for future nodes and , BTW, you can't even provide evidence that Intel will ever have 10nm, it's just not how it works.Not that feature size has much to do with perf or cost nowadays.
Increasing competitiveness with what? Even your extra optimistic claim puts that new core 2 years from now.

Will "ever" have 10nm? They are starting up 10nm factories as we speak and should begin mass production of Cannon Lake-Y (and maybe Cannon Lake-U 2+2) at the end of 2017.

Intel's predictions are so ridiculous that it is not even funny. Intel is predicting that they keep market share, they push ASPs up and AMD doesn't exist, it's all on a mild decline in the PC market.

Maybe Intel has some insight into the PC market that you aren't aware of? Ryzen is nice but it's targeting a niche within a niche in the overall desktop market, and Intel is likely to respond with new products and/or price cuts to try to maintain market share. AMD does not operate in a vacuum.

But hey, lets do some math ,not that you'll ever accept anything that doesn't paint Intel in a shiny bright light.

Now you're just baiting.

Lets say the PC market was 32 billion in 2016 and we know AMD had about 1 billion of that.
Lets say in 2018 the PC market is down in 2018, vs 2016 by 10 % in units. Lets say that a competitive AMD will have 15% revenue share.

It took Intel 5 years to gain 10% worth of market share against an AMD that practically gave up on the low end of the PC market (Intel gained a lot of share in the low end) and gave up iterating on high-end desktop (after Vishera in 2012, AMD released nothing new). Gaining that kind of unit share in such a short amount of time is not likely to happen.

And lets say that ASPs compress by only 10% due to AMD being back. Added all up Intel's PC revenue would be down by 9 billion.

OK, so Intel would cut prices AND lose significant market share? Then what would be the point of the price cuts?

And that's a mild scenario.So sure lets listen to all the idiotic claims Intel makes and believe in them,like it's a religion. Intel its current strategy won't be able to support the fabs very soon and they might not survive if they don't change leadership.

I've bookmarked this post. Let's see how your prediction goes. $9 billlion of Intel revenues in CCG gone by the end of 2017 with AMD more than doubling its sales in 2017 compared to 2016. We'll see.

 

[Arachnotronic]

They might end up being proven wrong on this prediction.

Absolutely a possibility. I would say, though, that when trying to predict these sorts of things, it is always important to assume that all parties will play their hands in an optimal fashion.

I think if Intel plays its hand in a fashion that's even within the ballpark of optimal, a 15% decline in revenue share in one year isn't likely to happen.

EDIT: Let me add to this.

So think about this from a business POV.

Let's say that 40% of the PC market is desktop, 60% is notebook. Right off the bat, 60% of the market is inaccessible to these Summit Ridge Ryzen chips. The remaining part is where Ryzen can theoretically play.

Now you're down to 40% of PCs, or the entirety of the desktop market. But roughly 50% of desktop market is consumer, 50% is corporate, and corporate buyers tend not to be interested in CPUs from AMD, let alone iGPU-less ones. So I'd say the market is pretty much down now to half of the overall desktop market, or 20% of the overall PC market.

Within that half of the desktop market that goes to consumer, how much of that is enthusiast gaming? Let's call it 25% (probably on the high side here).

So now you've got these Ryzen chips going after the enthusiast gaming part of the consumer desktop market. Of your overall PC TAM you're down to:

1*0.4*0.5*0.25 = 5% of the overall PC market. Let's say that revenue share of overall PC market is 2x the unit share because of generally higher ASPs. So, 10% of the total revenue in the PC market for these Ryzen chips.

How much of that market is AMD realistically going to get in a year? 7700K is still arguably the best gaming CPU out there today (and it's at $350), and Intel can cut prices on Broadwell-E, has Skylake-X coming out, and so on. On top of a good set of products (competitiveness can be modulated with price movements), Intel's brand is stronger, relationships with the key system vendors is stronger (and those system vendors buy lots of other chips for their other product lines, so much easier for volume discounts), sales force is more comprehensive, etc.

Let's say AMD gets a full 30% of that enthusiast market (again, optimistic, but this is basically AMD's share of the dGPU market).

That's 30% of 10% of the revenue share in the PC market, or 3% revenue share.

3% of Intel's Platform revenues (i.e. CPU+chipset) in 2016 ($30.7B) would be about $921 million.

And from there, recognize that AMD's share of the gaming/enthusiast CPU market is currently non-zero (a lot of FX chips still get sold, and I'm sure people still buy those Kaveris on FM2+ for this purpose too).

That's why the estimate another poster gave of $9B revenue share loss on Intel's part in one year doesn't really make a lot of sense to me, but of course we will see.

 

[coercitiv]

corporate buyers tend not to be interested in CPUs from AMD, let alone iGPU-less ones.

Within that half of the desktop market that goes to consumer, how much of that is enthusiast gaming? Let's call it 25% (probably on the high side here).

While I generally agree with the conclusion of your post (Intel will not suffer a major revenue loss) I have some objections for the parts above: the lack of iGPU does not directly make the chips unattractive for corporate buyers or consumers other than enthusiast gamers, the resulting price evidently does that. With the prices announced for lower Ryzen SKUs, they will be competitive with Intel i5-i7 parts even when having to buy an entry level dGPU. They will not be competitive if Intel lowers prices, but that clearly affects revenue.

The only parts immune to Ryzen are i3 and Pentiums, although we might have some surprises in the i3 area, where AM4 budget motherboards should be extremely cheap.

 

[JDG1980]

Let's say that 40% of the PC market is desktop, 60% is notebook. Right off the bat, 60% of the market is inaccessible to these Summit Ridge Ryzen chips. The remaining part is where Ryzen can theoretically play.

That's true enough for now. But when Raven Ridge drops (probably by the end of the year), Intel is going to be facing meaningful competition in the laptop market. We will have to wait for the full reviews to find out exactly what perf/watt Zen offers, but unless AMD is flat-out lying about its TDP figure, it clearly at least matches Intel's best HEDT offerings and most likely exceeds them. At lower clock speeds, it should be even more efficient. Intel currently sells dual-core CPUs for most of its laptop offerings, even the so-called i7s; only a handful of quad-core laptop chips are available from them and these are extremely expensive. That gravy train stops with Raven Ridge, which will make 4C/8T mainstream in the laptop market. Sure, Intel can compete, but they'll no longer have the easy margins of selling 2C/4T chips at inflated prices.

Now you're down to 40% of PCs, or the entirety of the desktop market. But roughly 50% of desktop market is consumer, 50% is corporate, and corporate buyers tend not to be interested in CPUs from AMD, let alone iGPU-less ones. So I'd say the market is pretty much down now to half of the overall desktop market, or 20% of the overall PC market.

The mistake you're making is to assume all these sales are equally valuable. A HEDT CPU can cost 5x-10x or more what a bargain-basement OEM desktop chip costs. Even if the percentage margin is exactly the same, that means one HEDT sale is as good as 5 to 10 entry-level sales.

Corporate sales for Summit Ridge will be marginal, but some of these chips will end up in workstations. Raven Ridge will be able to fully compete in the OEM market.

Intel can cut prices on Broadwell-E, has Skylake-X coming out, and so on.

Intel doesn't cut prices on existing products, not even when they're being discontinued. They may have to rethink Skylake-X pricing or reshuffle the lineup to make it competitive, though. Original rumors was that Skylake-X only went up to 10C/20T, just like Broadwell-E. If that's true, they may have to bring over one of the big-die workstation chips (E5-2687W v5?) to cover the $1723 price point, while everything else gets pushed below $999. I don't think they want to surrender the high price, but if they only offer 10 cores there, they'll get slaughtered.

That's why the estimate another poster gave of $9B revenue share loss on Intel's part in one year doesn't really make a lot of sense to me, but of course we will see.

Probably an exaggeration for this year. But things are going to get worse for Intel - they will likely never again have the kind of massive lead on AMD that they did from 10/2011 to 02/2017. Can they compete? Sure - but that's the thing, for the past 5+ years, they've been used to NOT competing but instead having a cozy monopoly where they can segment 100 different ways and dribble out 5% performance gains. Now they actually have to make an effort.

 

[coercitiv]

The mistake you're making is to assume all these sales are equally valuable. A HEDT CPU can cost 5x-10x or more what a bargain-basement OEM desktop chip costs. Even if the percentage margin is exactly the same, that means one HEDT sale is as good as 5 to 10 entry-level sales.

What's the current average selling price for a CPU in the consumer market?

 

[JDG1980]

Intel is moving to building future dies in pieces connected via EMIB.

I wonder if this ties in with the rumors that Intel will be using AMD graphics in conjunction with their future CPUs. With large-sized dies becoming more and more problematic on smaller process nodes, putting big ASICs together out of "building blocks" makes a lot of sense for the future, and we're clearly seeing both Intel (EMIB) and AMD (Infinity Fabric) going in this direction. Perhaps with Icelake or its successor, Intel will do a 6- to 8-core CPU die which could be linked with a separate Radeon GPU die for consumer-grade chips, and linked two or four together for HEDT/server. (Could a bigger L3 cache be put on a separate die with EMIB? If not, maybe EDRAM could take its place, or even something like HMC, Optane, or HBM2?)

 

[arandomguy]

OK, thanks.

Could you (or anyone) point me in the direction of any reviews that would compare say, the 6900K to the 7700K?

I think its better you search for them yourself because any singular review anyone posts may be (mis)leading even if not intentionally so. The key is to get as much data as possible and be critical of it. Most reviews would use the lower 6 core (6800k, 5820k) or the 8 cores though. I think its best just to actually wait for the Ryzen reviews at this point.

I have issues with how people do CPU reviews for gaming. Like Fallout 4, we know that FO4 is extremely demanding in locations like Corvega and Diamond City in which CPU (as well as memory) may come into play but how many reviewers test these locations? This is why you might get the impression that all CPUs are almost equal for this game but it really isn't.

Something to also keep in mind is some of the performance differences in reviews could be attributable to other areas such as available cache access and as such may not be completely indicative of what the differences would be of the 1700 vs 7700k. But at the same time the larger cache available to the 1700 might carry the same advantage as well in a lot of these games.

Also given the context of this thread I think people should not be so dismissive of stock vs stock comparisons as that is still very valuable to the gaming market as whole. Given we don't actually know what turbo speeds the 1700 will actually maintain in a gaming scenario the 7700k for all we know may hold a 50% ST performance advantage by virtue of clock speed. That would mean you'd need perfect scaling beyond 6/12 in a game just to pull even for the 1700.

 

[A///]

To be honest, I suspect that 4 cores will be enough for gaming- right now, anyway. Too many games don't scale well past 4 cores, and the 7700K still has a big single-thread advantage- 4.2GHz vs 3.6GHz, with significantly better IPC. I think it will still be the gaming king.

I don't do much gaming, but I was under the impression that the K i5 was the gaming king, and the i7 was also if you needed the ability to run things in the background and you were a media creator, or have things changed in the last 5 years?

 

[arandomguy]

I don't do much gaming, but I was under the impression that the K i5 was the gaming king, and the i7 was also if you needed the ability to run things in the background and you were a media creator, or have things changed in the last 5 years?

Unless you subscribe to the HT has drawbacks opinion the mainstream i7 has always been faster regardless since it is clocked higher along with more cache. It's always been a price consideration to not go with it.

But many games are starting to show higher differential between the i7 and i5 K CPUs even without factoring in clock speed.

 

[CHADBOGA]

Absolutely a possibility. I would say, though, that when trying to predict these sorts of things, it is always important to assume that all parties will play their hands in an optimal fashion.

I think if Intel plays its hand in a fashion that's even within the ballpark of optimal, a 15% decline in revenue share in one year isn't likely to happen.

EDIT: Let me add to this.

So think about this from a business POV.

When Intel predicts a low-single digit decline thru 2018 in its client business, is that end date:

31/12/18 or as they say in the USA, 12/31/18
30/6/18 or as they say in the USA, 6/30/18
31/12/17 or as they say in the USA, 12/31/17

Or some other date?

I assumed it meant 31/12/18 or as they say in the USA, 12/31/18

 

[lobz]

Intel is moving to building future dies in pieces connected via EMIB.

but that's still a _very_ long way down the road, in a sense that the competition's effective mcm solution is right on the doorstep. The emib might turn out more efficient but that's a story for at least 2 years later.

 

[NTMBK]

I don't do much gaming, but I was under the impression that the K i5 was the gaming king, and the i7 was also if you needed the ability to run things in the background and you were a media creator, or have things changed in the last 5 years?

The i5 is almost as good, but the i7 has a slight edge thanks to extra cache and hyperthreading (as well as higher stock clocks).

 

[lobz]

When Intel predicts a low-single digit decline thru 2018 in its client business, is that end date:

31/12/18 or as they say in the USA, 12/31/18
30/6/18 or as they say in the USA, 6/30/18
31/12/17 or as they say in the USA, 12/31/17

Or some other date?

I assumed it meant 31/12/18 or as they say in the USA, 12/31/18

wat

 

[unseenmorbidity]

I don't do much gaming, but I was under the impression that the K i5 was the gaming king, and the i7 was also if you needed the ability to run things in the background and you were a media creator, or have things changed in the last 5 years?

That's how things used to work, yes. Now, things are changing for gaming. The i3s are bordering on obsolete, 4c/4t are the budget system, 4c/8t are the "good enough" system, and anything higher is where the real money is.

 

[Arachnotronic]

When Intel predicts a low-single digit decline thru 2018 in its client business, is that end date:

31/12/18 or as they say in the USA, 12/31/18
30/6/18 or as they say in the USA, 6/30/18
31/12/17 or as they say in the USA, 12/31/17

Or some other date?

I assumed it meant 31/12/18 or as they say in the USA, 12/31/18

I think they are saying for the next three year, so the average annual decline in CCG revenue will be be between 1-4% through 2019 -- at least this is my understanding.

They are predicting a mid-single digit decline this year for 2017 compared to 2016.

 

[Arachnotronic]

but that's still a _very_ long way down the road, in a sense that the competition's effective mcm solution is right on the doorstep. The emib might turn out more efficient but that's a story for at least 2 years later.

The MCM solution isn't really ideal, otherwise Intel would have just done that and have been finished. The EMIB solution is supposed to make the fact that it's multiple dies as "transparent" as possible.

 

[Arachnotronic]

What's the current average selling price for a CPU in the consumer market?

About $120 for CPU + PCH.

 

[Arachnotronic]

We will have to wait for the full reviews to find out exactly what perf/watt Zen offers, but unless AMD is flat-out lying about its TDP figure, it clearly at least matches Intel's best HEDT offerings and most likely exceeds them.

Intel's best HEDT offerings are based on Broadwell arch and first gen 14nm process. Raven Ridge will have to compete with Kaby Lake Refresh-U/Coffee Lake built on a third generation 14nm process. 14nm+ already gave huge improvements in notebooks, and I think 14nm++ will continue that.

At lower clock speeds, it should be even more efficient. Intel currently sells dual-core CPUs for most of its laptop offerings, even the so-called i7s; only a handful of quad-core laptop chips are available from them and these are extremely expensive. That gravy train stops with Raven Ridge, which will make 4C/8T mainstream in the laptop market. Sure, Intel can compete, but they'll no longer have the easy margins of selling 2C/4T chips at inflated prices.

Intel's Kaby Lake Refresh-U will be a quad core part and the dies just aren't that much bigger than the current 2C/4T parts. And 14nm process should be even cheaper/better yielding by the time KBL-R U/CFL hit.

The mistake you're making is to assume all these sales are equally valuable. A HEDT CPU can cost 5x-10x or more what a bargain-basement OEM desktop chip costs. Even if the percentage margin is exactly the same, that means one HEDT sale is as good as 5 to 10 entry-level sales.

I'm not making that mistake, I explicitly assumed in the post that you are responding to that gaming desktops drive 2x the ASPs of overall desktop CPU sales.

Corporate sales for Summit Ridge will be marginal, but some of these chips will end up in workstations. Raven Ridge will be able to fully compete in the OEM market.

If you are building a workstation and your time is money, you are not messing with a Summit Ridge. You are going to buy something like a Xeon E5 2687W v4, probably in a 2-socket config.

Intel doesn't cut prices on existing products, not even when they're being discontinued.

That's been true lately, but go back in time during periods when Intel faced greater competitive pressure than it has in the Bulldozer era, and you will see they were perfectly willing to cut prices, keep older CPUs/platforms in the market to address lower price points, etc.

They may have to rethink Skylake-X pricing or reshuffle the lineup to make it competitive, though. Original rumors was that Skylake-X only went up to 10C/20T, just like Broadwell-E. If that's true, they may have to bring over one of the big-die workstation chips (E5-2687W v5?) to cover the $1723 price point, while everything else gets pushed below $999. I don't think they want to surrender the high price, but if they only offer 10 cores there, they'll get slaughtered.

And we finally agree. The people who buy 6950X are people really just looking for unlocked Xeons anyway, so I think that's exactly what they'll do -- 12c/24t SKX-X at $1700, 10c/20t at $999-$1299, and so on. I could even see them offering the 6 and 8 core models in both 28 PCIe lane configs and 44 lane PCIe configs (full 44 lanes only for 10 core and 12 core).

Probably an exaggeration for this year. But things are going to get worse for Intel - they will likely never again have the kind of massive lead on AMD that they did from 10/2011 to 02/2017. Can they compete? Sure - but that's the thing, for the past 5+ years, they've been used to NOT competing but instead having a cozy monopoly where they can segment 100 different ways and dribble out 5% performance gains. Now they actually have to make an effort.

Intel's practise with effective segmentation is going to be an asset in competing with AMD, not a liability. See preceding paragraph.

 

[LTC8K6]

Intel's 14nm+ actually gave no discernible ipc improvements, mobile or desktop.
The sole reason the mobile chips do better on benches is their ability to stay within their TDP envelope while boosting to higher clocks for longer periods.

It doesn't look like there is much more clock headroom for Intel desktop chips with the 14nm process.

KL-X is not expected to have much of a clock boost, and it's TDP at 112W would seem to be a further indication that it's at the limit.