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.)

 

[MoistOintment]

His actions of tweeting Bible verses publicly as Intel's CEO is worthy of criticism. Which successful CEO does this?

Really think this is irrelevant. Him and his wife are both religious. He cofounded a religious non-profit, donates a lot to charity, helped fund a Christian College, and has been tweeting a Bible verse every Sunday long before he became CEO.

It's really weird to just get so bent out of shape about his religion. I could care less about this.

 

[igor_kavinski]

Really think this is irrelevant. Him and his wife are both religious. He cofounded a religious non-profit, donates a lot to charity, helped fund a Christian College, and has been tweeting a Bible verse every Sunday long before he became CEO.

Could he please step aside and let someone else be CEO of Intel, as an act of charity?

 

[igor_kavinski]

Just seems like Intel's is full of yes men at the top and Pat is listening to them because he doesn't have a good understanding of anything. He didn't take the challenge seriosly on day 1 and now, 3 years later there is still no real progress.

Nailed it!

 

[jpiniero]

It's hard to get a good CEO. Intel tried really hard to get a good one after Krzanich, Swan already was a stopgap. Pat apparently could only be lured with that kind of money.

Pat jumped because he saw the writing on the wall that VMWare is toast & there's nothing you can really do about it.

Come to think of it, Intel is the same way since the board isn't willing to let him move fast enough to unload the Foundry.

 

[igor_kavinski]

It's hard to get a good CEO. Intel tried really hard to get a good one after Krzanich, Swan already was a stopgap. Pat apparently could only be lured with that kind of money.

Intel's problem is they don't like outsiders and don't want to listen to them. Their best bet would've been Jim Keller and he would've become CEO at half of what they are paying Pat and would've at the very least delivered, if not exceeded expectations.

 

[The Hardcard]

That's a really nice job to have. No wonder he's in such high spirits on stage all the time. No need to show any progress until he's filled his coffers with 5+ years of excellent remuneration and a lavish lifestyle.

It is not a question of need, he can’t show progress for 5+ years. Any changes that could show up by 2024 would have to be fairly minor and in line with what was already was already in progress. Tweaks and adjustments can be made in the final months, but it nearly impossible for such changes to change the character of a semiconductor product. What is before us today is by and large what was laid out 2018 to early 2020.

And it is important to remember what happened in that timeframe. AMD, TSMC, Apple, and Nvidia all began executing - based on decisions made between 2013 -2018 - with historic success for each company.

I am not saying Gelsinger should be let off the hook here, but let’s keep it in context. Few, if any have taken the reigns if a company when so many different transformative developments happening simultaneously. And all these transformational developments were years along when he took over.

 

[igor_kavinski]

It is not a question of need, he can’t show progress for 5+ years.

Then he was the wrong person for the job. Intel needed someone who could make quick decisions. Even getting rid of the fabs and going all in on TSMC if that was what he believed was the only way to get Intel back in the game. Not this crazy madness of trying to regain manufacturing process competitiveness. With the same lousy people who struggled with 10nm??? He pulled wool over everyone's eyes. Seems like the deal made with him was, "hey, we can't lose the fabs. It's what makes us Intel. And all these people. They have families. We can't let them go to ruin". So being the kind hearted soul Pat is, he lied for their sake. To investors. To everyone! This is not how corporations are supposed to work.

 

[Hulk]

Just seems like Intel's is full of yes men at the top and Pat is listening to them because he doesn't have a good understanding of anything. He didn't take the challenge seriosly on day 1 and now, 3 years later there is still no real progress.

He probably had a good understanding of microprocessor architecture.

 

[cannedlake240]

Tbh what was he supposed to do with disasters that were Alchemist and PVC that still were a year or two off when he joined? Dismantle everything and start over? Isn't that basically what theyre doing? Not surprising their AI program is in complete shambles lacking consistent vision

 

[cebri1]

He probably had a good understanding of microprocessor architecture.

 

[cannedlake240]

View attachment 110477

This is all a ploy, Pat is a puppet figurehead controlled by the powerful fab lobby at Intel!

 

[desrever]

He probably had a good understanding of microprocessor architecture.

He had that but things changed so much since he was at Intel.

 

[OneEng2]

When they first showed off those slides about getting rid of HT for better ST uplift, we all gained hopes that Lion Cove would be an ST screamer. It isn't. So what the heck were those slides for? To cover up their shame for being unable to get HT ready for prime time. I'm seriously tired of everyone associated with Intel who thinks it's completely OK to lie because they have been doing things this way for so long. Would've been a simpler thing for them to just admit that due to a myriad of complications, HT won't be available this year but they will try their best to sneak it into the refresh next year. I wouldn't be complaining as much then.

If Lion Cove or Skymont are not designed for SMT, I don't think it is something you can just tack on. It is my thought that everything from decode to dispatch and everything in between would be effected by SMT.

I think it is easy enough to "turn off", since doing so simply leaves that workflow path idle; however, I think it requires some serious design thought to make a core that uses it.

While Lion Cove may well have the resources built in to accomplish SMT (it is certainly large enough), I suspect there is no way for Skymont to achieve SMT without becoming a good deal larger than it is.

Almost 17 million in compensation for Pat last year. Most in INTC so, you know, their/his future at stake with that compensation.

Yea, how is the poor guy supposed to pay for groceries?

Pretty crazy when 0.01% of top earners make more than the bottom 50%.

 

[MoistOintment]

Then he was the wrong person for the job. Intel needed someone who could make quick decisions. Even getting rid of the fabs and going all in on TSMC if that was what he believed was the only way to get Intel back in the game. Not this crazy madness of trying to regain manufacturing process competitiveness. With the same lousy people who struggled with 10nm??? He pulled wool over everyone's eyes. Seems like the deal made with him was, "hey, we can't lose the fabs. It's what makes us Intel. And all these people. They have families. We can't let them go to ruin". So being the kind hearted soul Pat is, he lied for their sake. To investors. To everyone! This is not how corporations are supposed to work.

Or Pat sees the writing on the wall that x86, *long term* does not have great prospects. So they can either

1) double down on being an x86 designing and just hold off in PC and datacenter against inroads from ARM in a shrinking market space, eventually trying to pivot to being an ARM designer amongst a sea of other ARM designers.

Or 2) double down on being a Fab, where the TAM is growing and the competitors are shrinking. Where Samsung is struggling too and there's an opening from a world that doesn't want the entirety of advanced semi manufacturing being consolidated into a single Taiwanese psuedo-state enterprise. Where even if x86 dies, fab competes for those contracts. Even if Nvidia kills their dGPU efforts, they can try and fab Nvidia dies.

Like, looking forward not 1 year, not 3, not even 8. What does the competitive landscape look like way down the road? Which path laid out guarantees Intel' *long* term future? Why should we want them to cancel fabs and ensure a permanent TSMC monopoly? Because maybe we might get some better gaming CPUs out of it in the short term?

The hand off from design to fab first is easily a very difficult, potentially decade+ long transformation. But zoom out big picture and don't miss the forest for the trees. Look beyond 3 year time scales.

 

[poke01]

1) double down on being an x86 designing and just hold off in PC and datacenter against inroads from ARM in a shrinking market space, eventually trying to pivot to being an ARM designer amongst a sea of other ARM designers.

RISC-V looks more threat than ARM because it requires no license. That is what all companies want, an ISA with no strings attached like licensing. Open and free, this is the biggest threat to x86. It will be huge in China when it matures.

Look at all new CPU startups, all RISC-V based.

 

[Hulk]

ARL is a good start. There. I said it. Lags for some software but still a fast CPU, and does really well with MT productivity.
If Intel could pull 10% IPC out of Lion Cove and better consistency across a broad range of software it would be very strong indeed. I mean next gen. They could actually leave everything alone on the compute tile and just work on LNC. Put that tiles make improvements faster talk into action.

As it is it's only a matter of time before I cave and buy one. Although I'm seriously considering the 9950X, which would be my first AMD system. The 9950X is just so strong everywhere. The allure with ARL for me is all of the fun testing and tweaking I can do with the P's and E's. Because honestly both are more than fast enough for me.

 

[poke01]

As it is it's only a matter of time before I cave and buy one. Although I'm seriously considering the 9950X, which would be my first AMD system. The 9950X is just so strong everywhere. The allure with ARL for me is all of the fun testing and tweaking I can do with the P's and E's. Because honestly both are more than fast enough for me.

Try the AMD platform, Hulk. One shouldn’t reward mediocrity and that is what ARL is as a whole.
AMD has a better upgrade path as well. The 9950X3D next year is one of those paths.

You can always go back to Intel when Nova lake comes and you will also have an AMD board that is likely capable of Zen6.

 

[Hulk]

Try the AMD platform, Hulk. One shouldn’t reward mediocrity and that is what ARL is as a whole.
AMD has a better upgrade path as well. The 9950X3D next year is one of those paths.

You can always go back to Intel when Nova lake comes and you will also have an AMD board that is likely capable of Zen6.

Good points.

 

[OneEng2]

LOL. Monday morning CEO'ing.

Ok, I'll bite.

x86 isn't dead and wont be for the foreseeable future. While ARM is a fine ISA, it has spent most of its life being designed for inexpensive, low power embedded application. IIRC, the first part of an x86 decode is to break the CISC into RISC like equal length instruction and data for pipelining. These modern x86 processors have way more in common with RISC than they do with the original CISC processors of old IMO.

x86 is still dominant in DC and HPC as well as laptops and desktops. In the data center in specific, many of the software licenses are based on core count. Making larger clusters of ARM cores to combat higher performing x86 cores will be difficult to overcome in these instances.

Besides, it is crazy to believe that AMD and Intel couldn't produce a pretty darned good ARM processor if they wanted to.

I can see ARM being a challenge in the thin-and-light laptop market and maybe even desktop replacement market.

As for the fabs, Intel is deeply ingrained in its fab technology. I would argue the history of Intel's successful CPU's overshadows its excellence in fab technology, but in many (most?) cases the CPU's were successful BECAUSE of its excellence in fab technology. A decade or so ago, I used to say "Of course Intel can make a faster processor than AMD with twice the transistor budget and twice the power budget". Back in the day, Intel was 1 to 2 die shrinks ahead of everyone on the planet it seemed.

So you either save the fab part of Intel, or sell off the company.

I wonder though if Intel has the processes, experience, tools, and technical know-how to on-board an external design. All of these things are needed desperately in order to make someone else's chip a success.

Will Pat G succeed? I figure he is right. If 18A doesn't fly like a bird, he is in trouble. Now, I don't know if that is fair or not since turning around a company as big as Intel that has dug itself into a hole as deep as they have is something that takes more like 10 years vs 4 or 5. Intel didn't get into this mess in 4 or 5 years. It is crazy to expect they can dig out in that amount of time IMO. FWIW, based on my experience in executive management, I seriously doubt the board, the market, or the investors will give him the time it will take to right the ship and bring the company back into the profitability of yester-years.

The C suite is a strange place. You can't tell the board and investors the truth about the timing or money, or they will never let you do what needs done to fix things in the first place.... so you set as low of an expectation as you can get away with at the board level, and then start making plans on how to stretch that out with a string of interim achievements that keeps things moving (and provides positive news to the investors) until you can get it done for real.

Of course, if you make 17M / Year, you should be able to fart rainbows on cloudy days and produce technicolor unicorns at will .

 

[OneEng2]

Try the AMD platform, Hulk. One shouldn’t reward mediocrity and that is what ARL is as a whole.
AMD has a better upgrade path as well. The 9950X3D next year is one of those paths.

You can always go back to Intel when Nova lake comes and you will also have an AMD board that is likely capable of Zen6.

I have only built one Intel system in the past 40 years (first one was an 8088 ), and that was a Core 2 Duo. The Core 2 Quad never came down out of the clouds forever although I did finally get one for under $250.

AMD has been rock solid on sticking with a socket for multiple processor releases. I am on AM4 now and will likely stay there as long as my machine keeps working (I have a 5800x (2022) that I upgraded from a 3200g (2019)). Next upgrade Zen 7 on AM6 (2028) unless I am forced to do Zen 6 on AM5 .... but I have a feeling that Zen 7 will be on AM6 making the AM5 platform EOL for future upgrades

 

[OneEng2]

ARL is a good start. There. I said it. Lags for some software but still a fast CPU, and does really well with MT productivity.
If Intel could pull 10% IPC out of Lion Cove and better consistency across a broad range of software it would be very strong indeed. I mean next gen. They could actually leave everything alone on the compute tile and just work on LNC. Put that tiles make improvements faster talk into action.

As it is it's only a matter of time before I cave and buy one. Although I'm seriously considering the 9950X, which would be my first AMD system. The 9950X is just so strong everywhere. The allure with ARL for me is all of the fun testing and tweaking I can do with the P's and E's. Because honestly both are more than fast enough for me.

I think the follow-on for Arrow Lake is going to do just fine. My gut feeling is that they need to get around a few bottlenecks with the ring bus likely associated with their new packaging with tiles, and perhaps tweak Lion Cove a bit as well as get a few scheduling issues ironed out. I wouldn't be surprised if they can get 20-30% out of the old girl . As many have observed, Lion Cove is AWFUL big for as little performance as it is getting on N3B.

Of course, for gaming, I think we are about to see "Game, Set and Match" for AMD with the 9800X3D. Until Intel can manage a similar technology with as good of latency as AMD has managed with their stacked L3, I don't see gamers buying Intel. It just doesn't make any sense.

 

[Hulk]

Is the issue with LNC L3 latency and/or bandwidth and or latency/bandwidth to main memory?

 

[adroc_thurston]

Did AMD have over 100K employees to feed?

No. but they also had a lot less revenue to work with.

 

[511]

By what metric? Intel is just as bad now if not worse off than under any previous CEO. Their future is also worse than ever. Pat is the worst CEO in Intel history at this point since he is driving the company to bankruptcy. Intel stock in down >60% since he took over.

Worse than Kranzich i doubt he is the most decent CEO after Craig Barrett he is doing what previous CEO should have done invest in freaking company not invest in shareholders company is.

He corrected many things that were wrong With Intel he could have done a better job though i will give you that

Because the CEO is responsible for the company and its in their job description? Why does anyone need to be an "armchair CEO" to give criticism?

It is fine to criticize one for their mistakes but criticizing him for someone else's mistake not good

 

[MoistOintment]

x86 isn't dead and wont be for the foreseeable future. While ARM is a fine ISA, it has spent most of its life being designed for inexpensive, low power embedded application. IIRC, the first part of an x86 decode is to break the CISC into RISC like equal length instruction and data for pipelining. These modern x86 processors have way more in common with RISC than they do with the original CISC processors of old IMO.

x86 is still dominant in DC and HPC as well as laptops and desktops. In the data center in specific, many of the software licenses are based on core count. Making larger clusters of ARM cores to combat higher performing x86 cores will be difficult to overcome in these instances.

Dead? No. But it doesn't take a fortune teller to see market trends. ARM is making inroads into DC. It's making inroads into client. AMD and Intel will milk x86 as long as possible because its duopoly / exclusion is to their advantage.

But where do you see x86/ARM marketshare split in client and DC in 2030? 2040? And when AMD Intel eventually have to make their own ARM chips, now they don't have an exclusive ISA to give them an advantage. The number of direct competitors increase, and how does that marketshare split look? And the history of ARM doesn't really matter or the inner workings of how x86 breaks down instructions into more RISC like bites: What matters is what software is compiled to run on.

What does Graviton signal for the industry? Microsoft making a custom ARM chip? If a lot of the big hyperscalers are going to custom ARM chips designed in house, or see CPUs as nothing more than a necessary middleman to run large Nvidia dGPU / accelerator clusters, where does that leave Intel/AMD in 10+ years?

Point is that the overall market is changing. Foundry affords more growth opportunities than Intel trying to claw back their previously lost marketshare - and that's never happening. Intel won't ever be returning to the previous marketshares they held in client in DC, so I don't see why dropping fabs to chase that would be in their best interest (what others are saying, not you)