Intel Releasing 10 core 20 thread i9-10900KF for $499 very soon... 5.2 Ghz boost

[Space Lynx]

Right back at ya, with your edited quote(s). Again if it was not for AMD shit be worst for all of us, just a fact that prices would be higher.

it is nice having competition again. intel would have sat on 14nm variants for 5 more years if they had no comeptition

 

[cucker tarlson]

Right back at ya, with your edited quote(s). Again if it was not for AMD shit be worst for all of us, just a fact that prices would be higher.

apparently it works both ways.
not all of us make pc part purchases to make a statement for the corporations.
a thinking customer will buy whatever is good for them,not for either corporation.

 

[biffzinker]

I should of waited for AMD's 5700/XT, and Nvidia's answer. At the time I was looking for more performance than Vega offered to replace a RX 480. I've been happy with the performance the RTX 2060 offers.

 

[londiste]

But there’s nothing to keep AMD from undercutting this the day it launches. I would bet that the 3900X is much cheaper to produce

Why exactly do you think 3900X is much cheaper to produce?
- 3900X consists of 14nm IO die (~125mm^2) and two 7nm CCD dies (76mm^2).
- Both Zen and Zen+ dies are around 209mm^2.
- We do not know exactly how large Intel's 10-core thing is but since Skylake and derivatives have a long history, we know that 4 core die (7700K) is ~125mm^2, 6 core die (8700K) is ~175mm^2 and 8-core die (9900K) is ~200mm^2. The numbers are not exact, a few mm^2 here or there but close enough. Adding two more cores would put a 10-core die at ~225mm^2 assuming no other major changes are made.

Edit:
I messed up the sizes here. 2 additional cores add about 25mm^2 but I somehow went +50mm^2 from 4 to 6 cores. See comment below from @ppn . He correctly estimates 10-core CPU at 200mm^2 and 12-core CPU at 225mm^2.

200-250mm^2 is not a big die yet, it's reasonable in terms of yields and production.
3900X has smaller dies but more die size and 7nm is more expensive.

When it comes to cost also consider in case of Intel CPUs Intel is both CPU architecture designer as well as manufacturer. They can work with both design and production margins. In case of AMD, AMD designs the CPU architecture but dies are manufactured by TSMC. TSMC wants its own profit and its own margins regardless of what AMD does.

I am not saying Intel would be willing to cut into its margins - recent history shows exactly the opposite - but when shit hits the fan, they can.

it is nice having competition again. intel would have sat on 14nm variants for 5 more years if they had no comeptition

Are you serious? History quite clearly shows Intel has had no intention of sitting on an older node if they can produce on a smaller one. Sandy Bridge was at 32nm, Ivy Bridge was at 22nm, Broadwell was at 14nm.
They are on 14nm because they do not have a new process node available for production, pure and simple.

Hell, Intel is sitting on 14nm variants for 5 years even though they do have competition.

 

[TheLostSwede]

3900X has smaller dies but more die size and 7nm is more expensive.

When it comes to cost also consider in case of Intel CPUs Intel is both CPU architecture designer as well as manufacturer. They can work with both design and production margins. In case of AMD, AMD designs the CPU architecture but dies are manufactured by TSMC. TSMC wants its own profit and its own margins regardless of what AMD does.

I am not saying Intel would be willing to cut into its margins - recent history shows exactly the opposite - but when shit hits the fan, they can.

Yes, 7nm wafers are more expensive to make, but this is also why AMD is only making the chiplets at 7nm. Assuming the yields are decent, they should get a lot of chiplets per wafer.
The only number that came out early on, was a 70% yield, which has hopefully improved by now.
One main advantage of chiplets, as you you can see below, is that you're likely to have a much higher yield of good dies, over making larger chips.

Chiplet - WikiChip

A chiplet is an integrated circuit block that has been specifically designed to work with other similar chiplets to form larger more complex chips. In such chips, a system is subdivided into functional circuit blocks, called 'chiplets', that are often made of reusable IP blocks.

en.wikichip.org

You're right that Intel has more control over things though, but, as with so many things, there's also internal pricing, so it's not like Intel's manufacturing costs are free of charge.
Samsung is a great example here as well, where some of its BU's have outsourced chip production, as it would've been more costly to do it in-house, or they simply didn't have the production capability at times to keep up with demand.

We know Intel is having similar problems, partially due to them making cellular modems for Apple, which is why we have a desktop consumer CPU shortage. If you don't have the fab space/production lines to meet demand, then something is going to suffer and in Intel's case, it has been their own products where they're not under contractual obligation to supply a certain amount of chips per month.

On top of that, these companies needs to keep shareholders happy, so they need to make a certain profit per chip, which is why we're now seeing AMD increasing the cost of their products. This was kind of expected, but I doubt people are happy about it.

 

[londiste]

All I am saying is that it is not clear-cut like it is often made out to be - chiplets, so AMD wins because cheap.

At these die sizes and with 14nm being much more mature than 7nm is, I seriously doubt chiplet design has that significant of a benefit. Again, AMD has been competing extremely well in the last 2.5 years with monolithic dies that are within 10% of the size what I assume Intel's 10 core will be.

By the way, AMD has said they only make chiplets on 7nm because IO does not scale down well. From what I understand it is not so much about IO being problematic but about making IO on 7nm being waste because features are large enough that they can be manufactures much more efficiently on 12/14nm.

 

[biffzinker]

14nm IO die (~125mm^2)

The I/O die is 12nm for desktop or 14nm for Threadripper/EPYC is what I had read.

 

[TheLostSwede]

All I am saying is that it is not clear-cut like it is often made out to be - chiplets, so AMD wins because cheap.

At these die sizes and with 14nm being much more mature than 7nm is, I seriously doubt chiplet design has that significant of a benefit. Again, AMD has been competing extremely well in the last 2.5 years with monolithic dies that are within 10% of the size what I assume Intel's 10 core will be.

By the way, AMD has said they only make chiplets on 7nm because IO does not scale down well. From what I understand it is not so much about IO being problematic but about making IO on 7nm being waste because features are large enough that they can be manufactures much more efficiently on 12/14nm.

Indeed. They have a good chance of having good yield on the chiplets, but as a lot of CPUs need two or more of them, they better have good yields or they're screwed.

Well, mature doesn't mean better. TSMC seems to be pretty damn good at cranking out maximum chips per wafer, as if they weren't, the business would've gone elsewhere.
On the other hand, Intel should have as good yields as they'll ever get on 14nm, so there's nothing to improve, whereas AMD and TSMC might still have some improvements coming that will increase yields.

And this will be/might already be a problem for Intel with the move to 10nm. Plenty of things don't scale well and that's why Intel is working hard on EMIB (Embedded Multi-Die Interconnect Bridge), as it'll allow them to put multiple different production nodes inside a single processor. I'm sure we'll see something similar from a wide range of chip makers in the future. Glue is going to get popular...

Source: https://www.tomshardware.com/news/intel-emib-interconnect-fpga-chiplet,35316.html

 

[biffzinker]

TSMC seems to be pretty damn good at cranking out maximum chips per wafer, as if they weren't, the business would've gone elsewhere.

Likely why Intel moved to having Atom, and the lower end chipsets (H310) fabbed at TSMC.

It would seem Intel may of been outsourcing the production of Atom all the way back to 2009.

Why Is Intel Out-Sourcing Atom?

One of the questions Intel has undoubtedly got lots of briefing notes on for this evening's announcement of its outsourcing deal on Atom with TSMC is: Why

www.electronicsweekly.com

 

[londiste]

On the other hand, Intel should have as good yields as they'll ever get on 14nm, so there's nothing to improve, whereas AMD and TSMC might still have some improvements coming that will increase yields.

Pretty sure AMD is moving to N7+ and N6

 

[TheLostSwede]

Pretty sure AMD is moving to N7+ and N6

Which in turn, might mean worse yields, at least for a time.

 

[londiste]

Which in turn, might mean worse yields, at least for a time.

Yes. But also worth it because small improvements are expected in both density and power/frequency. Couple hundred MHz and more transistors go a long way

 

[TheLostSwede]

Oh and with regards to the topic of this thread, not going to be soon, expect it to be announced at Computex next year.

 

[ppn]

Why exactly do you think 3900X is much cheaper to produce?
- 3900X consists of 14nm IO die (~125mm^2) and two 7nm CCD dies (76mm^2).
...

Coffee Lake
4-Core 126 mm² die size
6-Core ~149.6 mm² die size
8-Core ~174 mm² die size
10-Core ~199 mm² die size Comet Lake

(theoretical) Rocket Lake 14nm ++++
12 14 and 16 Core ~225 mm² die size w/o Integrated Graphics GFX whatsoever, Xe GFX on separate chip

For instance Quad-core 32nm Sandy Bridge is 216 mm² die size, so 225 mm² 16-core GFX-less is not that much bigger. We are still safe. It will hold the line until Ice lake -S for Desktop.

 

[londiste]

@ppn thanks for the correction. I don't know how I messed that lineup up and now added a comment in my original post about your correction. Thanks!

Intel is not really safe with this though. Die size is only one of the factors. The big one that will bite them is power, especially over many-core load.

That 5.2 GHz for i9-10900KF in the original news blurb might be true but it is very likely that Intel will follow AMD's lead and send high voltages into single or a few cores for these high boost speeds. The same 1.4V that AMD uses for boost clocks in Ryzen 3000 series should allow Intel to boost to 5.2-5.3GHz. They can go further for more. Single core at 1.5V might be doable and that would win them another couple hundred MHz while efficiency goes out the window. Whether the die can take something like that, we simply don't know.

 

[EarthDog]

Intel is not really safe with this though. Die size is only one of the factors. The big one that will bite them is power, especially over many-core load.

Its weird to me when the shoe (power) is on the other foot it is an issue. People are fickle.

 

[Darmok N Jalad]

Why exactly do you think 3900X is much cheaper to produce?
- 3900X consists of 14nm IO die (~125mm^2) and two 7nm CCD dies (76mm^2).
- Both Zen and Zen+ dies are around 209mm^2.
- We do not know exactly how large Intel's 10-core thing is but since Skylake and derivatives have a long history, we know that 4 core die (7700K) is ~125mm^2, 6 core die (8700K) is ~175mm^2 and 8-core die (9900K) is ~200mm^2. The numbers are not exact, a few mm^2 here or there but close enough. Adding two more cores would put a 10-core die at ~225mm^2 assuming no other major changes are made.

Edit:
I messed up the sizes here. 2 additional cores add about 25mm^2 but I somehow went +50mm^2 from 4 to 6 cores. See comment below from @ppn . He correctly estimates 10-core CPU at 200mm^2 and 12-core CPU at 225mm^2.

200-250mm^2 is not a big die yet, it's reasonable in terms of yields and production.
3900X has smaller dies but more die size and 7nm is more expensive.

When it comes to cost also consider in case of Intel CPUs Intel is both CPU architecture designer as well as manufacturer. They can work with both design and production margins. In case of AMD, AMD designs the CPU architecture but dies are manufactured by TSMC. TSMC wants its own profit and its own margins regardless of what AMD does.

I am not saying Intel would be willing to cut into its margins - recent history shows exactly the opposite - but when shit hits the fan, they can.

Are you serious? History quite clearly shows Intel has had no intention of sitting on an older node if they can produce on a smaller one. Sandy Bridge was at 32nm, Ivy Bridge was at 22nm, Broadwell was at 14nm.
They are on 14nm because they do not have a new process node available for production, pure and simple.

Hell, Intel is sitting on 14nm variants for 5 years even though they do have competition.

Simply looking at history, AMD has been able to go cheaper than Intel. AMD has tons less overhead than Intel. It’s not about silicon sizes and substrates—it’s margins. AMD can sell for less because they don’t have to pay as many employees or keep as many buildings running. Intel has fabs scattered across the globe, and Intel depends on those to keep die sizes down to meet those margins. Being stuck at 14++ is no help. AMD can cease production with TSMC at any time with far less risk. This situation is probably close to Intel’s worst nightmare—to make a large (for them) CPU on a HDET platform and sell it for low margins. You can’t just compare two dice and assume all other things are equal.

 

[kapone32]

It would seem that this release is focused on competing with AMD's 12 and 16 core offerings. The 9900K was the answer to the 1700/1800/2700 and the KS was the response to the 3700/3800. As they had nothing to respond to the 3900X and 3950X they have probably been working on this since before the official announcement of the 3900/3950. AMD right now is leading in the CPU space. At every pricing level they have a compelling option. The prices of the 3900/3950 have been a bit controversial as they are more than AMD has ever sold a Desktop CPU for. If indeed Intel does do this though it would force down the price of the 9900K which is already more than $499 and cause greater competition for the 3700/3800 series. AMD would potentially have to lower the price of the 3900/3950 if reviews and other data show that a 10 core hyper threaded chip does bring significant gains for Intel having the extra 2 cores for multi threaded applications and may be the sweet spot for upcoming games.

 

[CrAsHnBuRnXp]

10 series CPU names dont roll off the tongue easily. We need a new series with new names.

 

[ratirt]

The prices of the 3900/3950 have been a bit controversial as they are more than AMD has ever sold a Desktop CPU for.

It's weird you say that though. Just because AMD never charged that much these prices must be controversial? These processors have much more to offer than Intel's CPUs so the price is higher. It is not about controversial but if it is justified to price these that much. I think these are very good prices for the 3900/3950 AMD CPUs. So I'm trying to say is, look at a product and price to performance or value or anything that processors offers not a company and what it used to sell their products for. Cause that's fools errand to me.
We've moved to 16c from 4c in the desktop segment to what has been server market for fraction of the price.

 

[EarthDog]

We've moved to 16c from 4c in the desktop segment to what has been server market for fraction of the price.

if only a general consumer and enthusiast can utilize more than half that....

 

[kapone32]

It's weird you say that though. Just because AMD never charged that much these prices must be controversial? These processors have much more to offer than Intel's CPUs so the price is higher. It is not about controversial but if it is justified to price these that much. I think these are very good prices for the 3900/3950 AMD CPUs. So I'm trying to say is, look at a product and price to performance or value or anything that processors offers not a company and what it used to sell their products for. Cause that's fools errand to me.
We've moved to 16c from 4c in the desktop segment to what has been server market for fraction of the price.

I am not one of those people that feel the price for those CDPUs are too high. I was just going based on comments I have seen here and other places about people lamenting on how AMD used to be the value king and now isn't. I do not disagree that if AMD had never gone past 4 or 8 cores we would not ever have seen anything like this from Intel for the price being quoted.

 

[AsRock]

apparently it works both ways.
not all of us make pc part purchases to make a statement for the corporations.
a thinking customer will buy whatever is good for them,not for either corporation.

True, which is why i own mainly intel systems, as AMD were not producing any thing near Intels performance back then. Clearly not the case now.

 

[cucker tarlson]

True, which is why i own mainly intel systems, as AMD were not producing any thing near Intels performance back then. Clearly not the case now.

I'm playing the waiting game.
I want either 3700X to drop to former 2700X price (around 1250pln,it's 1500 now) and I'll get it with x570 then (800 for the asus prime) or 9700k drops to current 3700x price (it's 1700 now) and I'll get it with a z390 gaming x (550).

 

[londiste]

Simply looking at history, AMD has been able to go cheaper than Intel. AMD has tons less overhead than Intel. It’s not about silicon sizes and substrates—it’s margins. AMD can sell for less because they don’t have to pay as many employees or keep as many buildings running. Intel has fabs scattered across the globe, and Intel depends on those to keep die sizes down to meet those margins. Being stuck at 14++ is no help. AMD can cease production with TSMC at any time with far less risk. This situation is probably close to Intel’s worst nightmare—to make a large (for them) CPU on a HDET platform and sell it for low margins. You can’t just compare two dice and assume all other things are equal.

- History shows Intel getting stupid high margins. AMD has gone lower in price but usually not because they wanted to.
- Intel has larger company but they also compete in many markets. Foundry is very capital-heavy one at that.
- AMD cannot cease production with TSMC. This is their risk. If something should happen in/to TSMC that sets them back, AMD would be quite out of luck.
- Situation is Intel's nightmare alright but what are you bringing HDET into it? Desktop CPUs were the topic here. Intel's manycore Xeons are screwed in most places, HDET is a niche market and has a lot more to do with marketing rather than profit.
- When it comes to production and cost, dies can be compared fairly well. When it comes to that, having multi-die packages for CPUs is more complex, more costly and more likely to bottleneck somewhere.

The prices of the 3900/3950 have been a bit controversial as they are more than AMD has ever sold a Desktop CPU for.

Sure they have. Athlon 64 FX-53 was over $700, FX-55 was over $800 and FX-57 was little over $1000.