Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

[Tigerick]

As Hot Chips 34 starting this week, Intel will unveil technical information of upcoming Meteor Lake (MTL) and Arrow Lake (ARL), new generation platform after Raptor Lake. Both MTL and ARL represent new direction which Intel will move to multiple chiplets and combine as one SoC platform.

MTL also represents new compute tile that based on Intel 4 process which is based on EUV lithography, a first from Intel. Intel expects to ship MTL mobile SoC in 2023.

ARL will come after MTL so Intel should be shipping it in 2024, that is what Intel roadmap is telling us. ARL compute tile will be manufactured by Intel 20A process, a first from Intel to use GAA transistors called RibbonFET.





Intel Core Ultra 100 - Meteor Lake



As mentioned by Tomshardware, TSMC will manufacture the I/O, SoC, and GPU tiles. That means Intel will manufacture only the CPU and Foveros tiles. (Notably, Intel calls the I/O tile an 'I/O Expander,' hence the IOE moniker.)

 

[Thunder 57]

I mean amd doesn't even have the volume for mobile now, imagine the next few years with even more limited wafers

Maybe that's why they are allegedly using Samsung in the future.

 

[Markfw]

Maybe that's why they are allegedly using Samsung in the future.

And more fabs are being built every day. Its WAY too early to say that AMD is MFG bound. I have never been unable to get the AMD chip I wanted, including server chips.

 

[DrMrLordX]

I mean amd doesn't even have the volume for mobile now, imagine the next few years with even more limited wafers

AMD isn't limited by wafer supply.

 

[Joe NYC]

I mean amd doesn't even have the volume for mobile now, imagine the next few years with even more limited wafers

It seems like you are suggesting that TSMC will sell wafers to Intel at expense of AMD. That would be a silly thing for TSMC to do. Do you think TSMC is run by idiots?

 

[Schmide]

I don't study fabs, but I've read enough to understand that you don't just phone in orders. You go to the factory, set up shop, and run designs through the system in a iterative process. Sure there is competition for space, but if you have a seat at the table, you're going to get scheduled.

 

[Hulk]

There is always a knee jerk reaction to production shortages. It all cycles up and down. There is suddenly a big surge in demand due to whatever reason, perhaps everyone working home with Covid, or a new technology everyone wants like SSD's... Demand will initially be high with low supply so there will of course be high prices. Manufacturers see the "hole" in the sector and gear up to make some money, they overproduce to be "first in" and then prices plummet and they're sitting on stock that has to go before it's outdated (good for us). Look at SSD's now, same with memory, GPU's are also realistically priced. I just bought and ARC 750 LE for $200.

Don't worry about CPU supply. If there is a demand the supply will eventually be there. As always if you must be a first adopter you will pay.

Prices on CPU's always hold better because there are only two suppliers, AMD and Intel (talking x86). Until the last 10 years or so it was really only Intel, which was why we were seeing their Extreme Editions going for $1,000. AMD fixed that right quick with a little competition

There is more competition in the SSD, memory, and perhaps a bit more with the GPU segment with Intel gaining a beach head so prices are more competitive.

 

[dullard]

Prices on CPU's always hold better because there are only two suppliers, AMD and Intel (talking x86). Until the last 10 years or so it was really only Intel, which was why we were seeing their Extreme Editions going for $1,000. AMD fixed that right quick with a little competition

That isn't how competition works. When AMD first beat Intel in performance, what was their price? $1299 in 1000-unit quantities. https://www.computerworld.com/article/2592477/amd-ships-1-ghz-processor.html And how are top HEDT prices now? Multiple times that amount.

Prices are based on what the market will bear. True competition requires dozens or hundreds of suppliers. A duopoly has only a little impact on price, and no impact if they can differentiate themselves (more cores, more GHz, etc). With any differentiation that the consumer cares about, you essentially have two monopolies.

 

[Hulk]

That isn't how competition works. When AMD first beat Intel in performance, what was their price? $1299 in 1000-unit quantities. https://www.computerworld.com/article/2592477/amd-ships-1-ghz-processor.html And how are top HEDT prices now? Multiple times that amount.

Prices are based on what the market will bear. True competition requires dozens or hundreds of suppliers. A duopoly has only a little impact on price, and no impact if they can differentiate themselves (more cores, more GHz, etc). With any differentiation that the consumer cares about, you essentially have two monopolies.

You are partly correct.

I would say that prices are based on what the market will bear AND competition AND how badly the item is required. If you need insulin to survive and there is only one vendor then you will bear a pretty high price, right?

While a hundred competitors are better than two, there is no doubt that 2 is MUCH better than 1. There is a continual price war between AMD and Intel at every performance point.

 

[TESKATLIPOKA]

That isn't how competition works. When AMD first beat Intel in performance, what was their price? $1299 in 1000-unit quantities. https://www.computerworld.com/article/2592477/amd-ships-1-ghz-processor.html And how are top HEDT prices now? Multiple times that amount.

Prices are based on what the market will bear. True competition requires dozens or hundreds of suppliers. A duopoly has only a little impact on price, and no impact if they can differentiate themselves (more cores, more GHz, etc). With any differentiation that the consumer cares about, you essentially have two monopolies.

7980X's price is $4999. The price is almost 4x higher, true, but look at what you get in return. 1.) 64 cores, 128 threads
2.) 3x higher base clock
3.) not sure how many times higher IPC

If you think about It, then that single core 1GHz Athlon was ridiculously overpriced, and I am not even including the inflation after 23 years.

 

[rainy]

7980X's price is $4999. The price is almost 4x higher, true, but look at what you get in return. 1.) 64 cores, 128 threads
2.) 3x higher base clock
3.) not sure how many times higher IPC

If you think about It, then that single core 1GHz Athlon was ridiculously overpriced, and I am not even including the inflation after 23 years.

Yes, his memory is quite selective but maybe it's not that surprising for Intel fan.
I have a much newer example: i7-6950X (10 cores) released by Intel in Q2 2016 for "just" 1723 dollars.
https://www.intel.com/content/www/u...-25m-cache-up-to-3-50-ghz/specifications.html

Btw, when this 1 GHz Athlon was released, AMD had just single fab (FAB 30).

 

[dullard]

While a hundred competitors are better than two, there is no doubt that 2 is MUCH better than 1. There is a continual price war between AMD and Intel at every performance point.

How do you personally define "MUCH"?

There are numerous duopoly pricing models around with various assumptions. One famous model is the Cournot Model that makes simple assumptions (such as no collusions between companies, that the products can reasonably be interchanged by the users, and that the firms act rationally to maximize profits). In the Cournot model, the maximum profit for a monopoly occurs at price [a + c] / 2 and the maximum profit for a duopoly occurs at price [a + 2*c] / 3, where:

• 'a' is the price when the product is scarce. That is the maximum price that someone rational will pay when free to make choices (i.e. not a life/death situation with insulin).
• 'c' is the marginal cost. That is, the cost to make one more item. This excludes all research costs, equipment costs, start up costs etc since that was all factored into the first item produced. If AMD makes 10 million CPUs of a given style, then 'c' is how much more does it cost to make 10 million and one CPUs of that style.

Lets try a few numbers. Suppose a rational person would pay $1000 for an i7 Intel CPU or $1000 for a Ryzen 7 CPU if there was a shortage of i7/Ryzen 7 CPUs. Remember there are other CPUs, there is just a shortage of this line of i7/Ryzen 7 CPU in this example. If the i7/Ryzen 7 CPU cost more than $1000, then they'd just switch to the i9 equivalent. Thus 'a' = $1000. Suppose it costs the companies $50 to make one more CPU than they would otherwise make. Thus, 'c' = $50. Then the monopoly price is $525 and the duopoly price is $367. The difference is a 30% savings.

Now, what about a lower end model. Suppose a rational person would pay $250 for an i3 Intel CPU or a Ryzen 3 CPU if there was a shortage of i3/Ryzen 3 CPUs. Remember there are other CPUs, there is just a shortage of this line. If the i3/Ryzen 3 CPU cost more than $250 then they'd just switch to the i5 equivalent. Thus 'a' = $250. Suppose it costs the companies $50 to make one more CPU than they would otherwise make. Thus, 'c' = $50. Then the monopoly price is $150 and the duopoly price is $117. The difference is a 22% savings.

But, if the companies can differentiate the product sufficiently, this violates the "reasonably be interchanged by the users" assumption. In this case, both can act as monopolies and the difference with 1 or 2 companies could be nothing. In fact, having competition could actually raise prices. Look up "price-increasing competition". Basically it would be more efficient for a single monopoly to make all the CPUs and thus costs "c" decrease in the monopoly case and prices could decrease if one of the two goes out of business.

 

[Hulk]

How do you personally define "MUCH"?

There are numerous duopoly pricing models around with various assumptions. One famous model is the Cournot Model that makes simple assumptions (such as no collusions between companies, that the products can reasonably be interchanged by the users, and that the firms act rationally to maximize profits). In the Cournot model, the maximum profit for a monopoly occurs at price [a + c] / 2 and the maximum profit for a duopoly occurs at price [a + 2*c] / 3, where:

• 'a' is the price when the product is scarce. That is the maximum price that someone rational will pay when free to make choices (i.e. not a life/death situation with insulin).
• 'c' is the marginal cost. That is, the cost to make one more item. This excludes all research costs, equipment costs, start up costs etc since that was all factored into the first item produced. If AMD makes 10 million CPUs of a given style, then 'c' is how much more does it cost to make 10 million and one CPUs of that style.

Lets try a few numbers. Suppose a rational person would pay $1000 for an i7 Intel CPU or $1000 for a Ryzen 7 CPU if there was a shortage of i7/Ryzen 7 CPUs. Remember there are other CPUs, there is just a shortage of this line of i7/Ryzen 7 CPU in this example. If the i7/Ryzen 7 CPU cost more than $1000, then they'd just switch to the i9 equivalent. Thus 'a' = $1000. Suppose it costs the companies $50 to make one more CPU than they would otherwise make. Thus, 'c' = $50. Then the monopoly price is $525 and the duopoly price is $367. The difference is a 30% savings.

Now, what about a lower end model. Suppose a rational person would pay $250 for an i3 Intel CPU or a Ryzen 3 CPU if there was a shortage of i3/Ryzen 3 CPUs. Remember there are other CPUs, there is just a shortage of this line. If the i3/Ryzen 3 CPU cost more than $250 then they'd just switch to the i5 equivalent. Thus 'a' = $250. Suppose it costs the companies $50 to make one more CPU than they would otherwise make. Thus, 'c' = $50. Then the monopoly price is $150 and the duopoly price is $117. The difference is a 22% savings.

But, if the companies can differentiate the product sufficiently, this violates the "reasonably be interchanged by the users" assumption. In this case, both can act as monopolies and the difference with 1 or 2 companies could be nothing. In fact, having competition could actually raise prices. Look up "price-increasing competition". Basically it would be more efficient for a single monopoly to make all the CPUs and thus costs "c" decrease in the monopoly case and prices could decrease if one of the two goes out of business.

In this context I define "much" as making the word "better" even "more" better.
Also, you blinded me with science!

 

[dullard]

In this context I define "much" as making the word "better" even "more" better.
Also, you blinded me with science!

Ultimately, I get triggered when people call 2 companies "competition". When that word should be limited to say 20 or more companies. Also, I took enough economics classes to have a minor in it if my college allowed minors for engineers. I like to let it spill out now and then. Two companies usually lead to lower prices, but not quite to the extent that I thought your posts were implying.

If anyone knows the actual marginal cost for a desktop CPU (including the silicon, labor, quality checks, binning, packaging, shipping, etc.) please feel free to update my assumed $50 number.

 

[Hulk]

Ultimately, I get triggered when people call 2 companies "competition". When that word should be limited to say 20 or more companies. Also, I took enough economics classes to have a minor in it if my college allowed minors for engineers. I like to let it spill out now and then.

I had one economics class as an ME at Rutgers... engineering economics.
I know I feel a heck of a lot more competition when running with someone than when running alone, but then again I was more comfortable in my fluid mechanics lecture than Engineering Economics.

Does the math say that if either Intel or AMD didn't exist prices would be about 22% higher than they are today?

 

[dullard]

Does the math say that if either Intel or AMD didn't exist prices would be about 22% higher than they are today?

Roughly, yes, especially for desktop CPUs. But it does depend on a lot of factors. For example, I don't think Intel has any product that seriously competes with AMD's Threadripper. So, I don't think there is any Threadripper price reduction by having two companies. You could say a similar thing about the upcoming Meteor Lake CPU (trying to make this relevant to this thread). There are so many unique features to Meteor Lake that I don't think the existence of AMD's laptop chips will lower the Meteor Lake prices at all.

 

[Thunder 57]

Ultimately, I get triggered when people call 2 companies "competition". When that word should be limited to say 20 or more companies. Also, I took enough economics classes to have a minor in it if my college allowed minors for engineers. I like to let it spill out now and then. Two companies usually lead to lower prices, but not quite to the extent that I thought your posts were implying.

If anyone knows the actual marginal cost for a desktop CPU (including the silicon, labor, quality checks, binning, packaging, shipping, etc.) please feel free to update my assumed $50 number.

There used to be more competition. Turns out, making CPU's is about the hardest thing to do (that we know of). Other companies for one reason or another couldn't keep up and went under. Same thing happened with fabs.

 

[eek2121]

Part of the reason newer mobile chips are so hard to come by is that older chips still flood the market. As long as that continues to be the case, AMD won’t overcommit to releasing the newer stuff for fear of driving down ASPs. On top of that, those older parts tend to be cheaper to make in some cases.

I am actually going to be keeping an eye on both Meteor Lake and the Phoenix refresh. Also Zen 5 if it launches before I decide to upgrade. All 3 look promising.

 

[Markfw]

Part of the reason newer mobile chips are so hard to come by is that older chips still flood the market. As long as that continues to be the case, AMD won’t overcommit to releasing the newer stuff for fear of driving down ASPs. On top of that, those older parts tend to be cheaper to make in some cases.

I am actually going to be keeping an eye on both Meteor Lake and the Phoenix refresh. Also Zen 5 if it launches before I decide to upgrade. All 3 look promising.

Actually, I have to agree with this. While my primary boxes are desktop, I have to have a recent and good quality laptop. The one I got back in 2016 is still working fine, except it only works on AC (which I hardly ever use battery). So recently I got a new one. Thats one every 7 years, while my desktops are about every 3 years.

 

[Hulk]

Roughly, yes, especially for desktop CPUs. But it does depend on a lot of factors. For example, I don't think Intel has any product that seriously competes with AMD's Threadripper. So, I don't think there is any Threadripper price reduction by having two companies. You could say a similar thing about the upcoming Meteor Lake CPU (trying to make this relevant to this thread). There are so many unique features to Meteor Lake that I don't think the existence of AMD's laptop chips will lower the Meteor Lake prices at all.

Wouldn't the Xeon parts or AMD server parts be competing to some extent with Threadripper? I mean if TD didn't exist then the only option would be the super expensive parts if you need 50+ cores, right?
At the end of the day for mobile I think battery life is going to be the deciding factor for many people, meaning efficiency. Seems like more choices on the shelves would mean more competition.

 

[SiliconFly]

A new theory (a.k.a the rear-view mirror theory)

Since Intel is placing massive orders with TSMC N3B for LNL, and Intel 20A is almost ready for manufacturing, I think Intel has very high confidence in LNC. And LNC being a grounds up power-efficient design without hyper-threading, I'm starting to think Intel has modeled LNC around ARM designs or more specifically Apple M-series designs.

In which case, LNC might have significantly higher IPC at a given clock compared to RWC (say 3 GHz). And LNC's max clock will significantly be lower at less then 5 GHz. Maybe even just 4 GHz. But even at these lower max clocks, it may still edge out RWC in performance.

If true. it signals Intel's complete departure from decades of power hungry higher-clocked designs. Is LNC the "rear-view mirror" Pat was talking about?

 

[Joe NYC]

A new theory (a.k.a the rear-view mirror theory)

Since Intel is placing massive orders with TSMC N3B for LNL, and Intel 20A is almost ready for manufacturing, I think Intel has very high confidence in LNC. And LNC being a grounds up power-efficient design without hyper-threading, I'm starting to think Intel has modeled LNC around ARM designs or more specifically Apple M-series designs.

In which case, LNC might have significantly higher IPC at a given clock compared to RWC (say 3 GHz). And LNC's max clock will significantly be lower at less then 5 GHz. Maybe even just 4 GHz. But even at these lower max clocks, it may still edge out RWC in performance.

If true. it signals Intel's complete departure from decades of power hungry higher-clocked designs. Is LNC the "rear-view mirror" Pat was talking about?

He was talking about Alder Lake vs. Zen 3.

 

[SiliconFly]

He was talking about Alder Lake vs. Zen 3.

Who?

(My post isn't related to those, jfyi)

 

[SiliconFly]

A very interesting piece of information from MLID's latest video. AMD is all set to adopt Intel's MTL design in Zen 6. Zen 6 will have a NOC (Network-on-Chip) like MTL, and will also feature LP cores. Possible in mid 2026. Thats almost 3 years from now. 🙄

 

[Joe NYC]

Who?

(My post isn't related to those, jfyi)

Pat Gelsinger:

Intel CEO Gelsinger Says AMD Is "In the Rear-View Mirror" After Alder Lake

The war of words escalates.

www.tomshardware.com

 

[Joe NYC]

A very interesting piece of information from MLID's latest video. AMD is all set to adopt Intel's MTL design in Zen 6. Zen 6 will have a NOC (Network-on-Chip) like MTL, and will also feature LP cores. Possible in mid 2026. Thats almost 3 years from now. 🙄

That Network on the chip of Zen 6 will be a server grade variant, based on the Pensando datacenter DPU cards.