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.)

 

[9949asd]

No, AVX512 has basically no effect on Zen GB6 scores.

Page 609 - Discussion - Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

Page 609 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

it will bro, look at the AES score, It’s useless .

 

[Josh128]

I believe the 3450/23K scores are probably whats coming with retail. 3186 is lower than a 9600X, I dont believe that. Geekbench really is terrible to judge CPUs by. Even on the same SKU, theres often a 100+ point variance in ST.

 

[Hitman928]

it will bro, look at the AES score, It’s useless .

AES score is GB5, not GB6. Did you even look at the post I linked to?

 

[OneEng2]

I guess my overall thought process is still the following:

1. Arrow Lake meets, or slightly exceeds AMD's 9550 in single threaded benchmarks (real world apps .... we will see)
2. Arrow Lake exceeds AMD's 9550 by around 20% in multi-threaded benchmarks.
3. In order to achieve this level of performance, Arrow Lake is using a process which gives it better performance or lower power AND gives Intel a larger transistor budget.
4. Intel is likely paying more for each Arrow Lake die than AMD pays for each Zen 5 die.
5. AMD's Turin (Zen 5 Server) Doesn't really have a competing product from Intel at this time. This will provide AMD with growing market and margin share moving forward.

Still, I have been impressed with Intel's ability in the past to overcome gross mistakes (like P4 and RAMBUS), and still retain market share and margin over AMD. Of course, it did so by being 1 to 2 die shrinks ahead of AMD at all times.

I guess I am wondering if Intel can pull the same financial magic as it has in the past when it must do so with a more expensive process, and without the huge profit from their server product to hold up the overall margins.

 

[9949asd]

AES score is GB5, not GB6. Did you even look at the post I linked to?

When you see the r23 r24, you will see there are a big gap one is 2300 one is 2500.

 

[Doug S]

More nuts than Apple?

That's something I've been wondering about for some time. In annual units (i.e. chips) Apple easily beats Intel, but Intel's server CPUs mean their average size per chip is larger. But recently AMD has really cut into their high end (i.e. very large) server CPUs.

So best guess Apple and Intel's annual wafer volume is probably roughly comparable. I wouldn't be surprised if we found out either did more, but no way the difference is more than 40-50% and probably more like 10-20%. Intel has a lot more different designs so they'd have several times the number of mask sets etc. so they'd have more "volume" from that - which is additional load on a foundry in a different way than simple number of wafers.

 

[Hitman928]

When you see the r23 r24, you will see there are a big gap one is 2300 one is 2500.

We were talking about GB6, but ok. . .

 

[9949asd]

We were talking about GB6, but ok. . .

It’s ok, you didn’t get my point.

 

[511]

I guess my overall thought process is still the following:

1. Arrow Lake meets, or slightly exceeds AMD's 9550 in single threaded benchmarks (real world apps .... we will see)
2. Arrow Lake exceeds AMD's 9550 by around 20% in multi-threaded benchmarks.
3. In order to achieve this level of performance, Arrow Lake is using a process which gives it better performance or lower power AND gives Intel a larger transistor budget.
4. Intel is likely paying more for each Arrow Lake die than AMD pays for each Zen 5 die.
5. AMD's Turin (Zen 5 Server) Doesn't really have a competing product from Intel at this time. This will provide AMD with growing market and margin share moving forward.

This Turin Part is where you are wrong AMD will no longer have Insane Advantage vs Intel due to GNR Both are 128 Cores CPU with AMD having higher Performance per core Intel will have bunch of accelerators previously it was 64C vs 96C so the core counts are evenly matched Intel having better SW and having better horsepower but it will be lot closer than past

Still, I have been impressed with Intel's ability in the past to overcome gross mistakes (like P4 and RAMBUS), and still retain market share and margin over AMD. Of course, it did so by being 1 to 2 die shrinks ahead of AMD at all times.

I guess I am wondering if Intel can pull the same financial magic as it has in the past when it must do so with a more expensive process, and without the huge profit from their server product to hold up the overall margins.

AMDs Server Supremacy is due to Being a Node ahead just like Intel on comparable Node we will see

 

[Hitman928]

It’s ok, you didn’t get my point.

I got it, it was just a tangent to the discussion.

 

[9949asd]

I got it, it was just a tangent to the discussion.

Like I said, you didn’t get the point or refuse to get the point.

 

[Hitman928]

Like I said, you didn’t get the point or refuse to get the point.

Ok, maybe so. Why not explain it for me as it relates to the GB6 discussion. . .

 

[OriAr]

This looks like maybe 10% ST and 5% better MT than a 14900K?

View attachment 107673

GB multithreaded benchmarks are beyond useless, better not to take them seriously. (Their ST benchmarks are decent, their MT benches don't scale that well at all.

 

[511]

Big Win for Intel

Intel and AWS Expand Strategic Collaboration, Helping Advance U.S.-Based Chip Manufacturing

www.intc.com

 

[9949asd]

Ok, maybe so. Why not explain it for me as it relates to the GB6 discussion. . .

The avx512 on zen5 will scale 2-3% more score in gb6. When Zen5 and RPL doing 2300 in r23, they will do around 3500 and 3350 in gb6. Don’t tell me you don’t know that?

 

[Hitman928]

The avx512 on zen5 will scale 2-3% more score in gb6. When Zen5 and RPL doing 2300 in r23, they will do around 3500 and 3350 in gb6. Don’t tell me you don’t know that?

So you are linking AVX-512 enabled versus disabled scores in GB6 to specific scores in Cinebench r23? Do you have proof for any of this and how is it related/relevant?

 

[9949asd]

Big Win for Intel

Intel and AWS Expand Strategic Collaboration, Helping Advance U.S.-Based Chip Manufacturing

www.intc.com

🤣intel keeping the fab will not be a good idea.

 

[511]

🤣intel keeping the fab will not be a good idea.

About that it will be seperate entity under Intel

A message from Intel CEO Pat Gelsinger to employees regarding the next phase of Intel's transformation

www.intc.com

 

[AcrosTinus]

🤣intel keeping the fab will not be a good idea.

Intel keeping the fab is the only way to not turn into a commodity.
Chips are necessary but consumers don't know or appreciate what type of computing power they have on their hands, it should just work, show insta and play games.

In order to stay necessary to society they cannot become a spoon or a fork like Apple. The fab guarantees independence of the west and I fear the next wars are supply chain wars...

 

[OneEng2]

This Turin Part is where you are wrong AMD will no longer have Insane Advantage vs Intel due to GNR Both are 128 Cores CPU with AMD having higher Performance per core Intel will have bunch of accelerators previously it was 64C vs 96C so the core counts are evenly matched Intel having better SW and having better horsepower but it will be lot closer than past

Not that I have been following this as closely as I used to, but a little research dug up this:

AMD Turin - 192 cores Zen 5c and Zen 5 variants (I think) - release Q4 2024
Intel Sierra Forest - 144 E cores - Release June 2024
Intel Sierra Forest - 288 E cores - Release Q4 2024 (Intel hasn't been very good about meeting their server roadmap lately though)
Intel Granite Rapids - 2025

Keeping in mind that AMD E Cores still have SMT that generally provides about a 30% boost in multi-threaded workloads, it still looks to me like AMD Turin variants will be quite competitive to even the 244 core variants (that currently don't exist) in terms of performance.

I haven't seen an analysis yet of the Sierra Forest version of E cores used (surely not Arrow Lake class, but rather the previous gen right?) vs AMD 5C E core, but I am guessing it may be that the Zen 5c is a more performant core as well.

I guess what I am guessing is that Intel is still going to be in a losing position against Turin (or at best break even) throughout 2025.

In 2026 AMD will double their core count with 32 core CCX's providing a 512 core EPYC on Zen 6 on TSMC 2nm.

Now I can't see that far into the future; however, it doesn't look to me like Intel is going to be scarfing up the server profits as they did prior to AMD Zen in the next few years. The best they can hope for is to stop the market share bleeding IMO, and even this will come at a greatly decreased profit margin compared to what they once enjoyed.

Which still means that Intel is playing catchup.

Furthermore, their late entry into AI compute is going to be a big drag as well. I think even AMD got caught here with their pants down while NVIDIA ran away with the market. Still, AMD is on its 3rd gen of AI processors (I think) and Intel is just getting started. Makes me think of how AMD leapt ahead of Intel in the integrated graphics market. Again, Intel got caught sleeping at the wheel (seems like their newest integrated graphics is caught up though).

I feel like Intel has gotten by in the past with a combination of market supremacy (and the market power they were able to deploy) and a crazy big lead in process node technology.

I wonder how Intel will fare when it doesn't have cash flowing like rivers (which it doesn't anymore) and it has to compete on the same process node as AMD (which it still really isn't doing. It's paying for a much more expensive node in order to remain competitive ... except for the server dense market where AMD is using N3X so things are about even there with regards to process anyway).

All the bad financial misses have been putting pressure on its money as well. Turns out the market doesn't like it much when you repeatedly miss profit and revenue targets. Last thing I read was no dividends either. That has to make a few very wealthy people upset.

Anyway. Just eating popcorn and enjoying the show as this all plays out.

 

[OneEng2]

Intel keeping the fab is the only way to not turn into a commodity.
Chips are necessary but consumers don't know or appreciate what type of computing power they have on their hands, it should just work, show insta and play games.

In order to stay necessary to society they cannot become a spoon or a fork like Apple. The fab guarantees independence of the west and I fear the next wars are supply chain wars...

Not sure I agree. AMD no longer has a fab. NVIDIA doesn't have a fab. Apple doesn't have a fab. They all seem to be doing just fine. I actually think the future is fabless CPU design. It is horridly difficult to compete with a fab that does NOTHING but do fab well, and does it at a volume that can pay off the silly high NRE cost of new node development. Intel was able to do this when they produced more chips that all others in the industry by many times over. Now that tiled processors, mix and match logic, etc is around, and the very high performance chips are across many markets and many companies, I don't think this model works well anymore.

I do agree with you on the geopolitical issue though. Arizona TSMC factory can't get up and running fast enough for my taste.

 

[cannedlake240]

Granite Rapids - 2025

It launches this September, Intel's words.

512 core EPYC

Nope... Also Diamond Rapids exists. Not so sure about the Forest line tho, supposedly Intel canned it to fund AI accelerator development ??

 

[AcrosTinus]

Not sure I agree. AMD no longer has a fab. NVIDIA doesn't have a fab. Apple doesn't have a fab. They all seem to be doing just fine. I actually think the future is fabless CPU design. It is horridly difficult to compete with a fab that does NOTHING but do fab well, and does it at a volume that can pay off the silly high NRE cost of new node development. Intel was able to do this when they produced more chips that all others in the industry by many times over. Now that tiled processors, mix and match logic, etc is around, and the very high performance chips are across many markets and many companies, I don't think this model works well anymore.

I do agree with you on the geopolitical issue though. Arizona TSMC factory can't get up and running fast enough for my taste.

They seem to be doing fine due to supply chains not being impacted currently, a sanction here, a flood there some wars sprinkled on top and AMD is out of the race. Even Nvidia is waiting to see how the Intel node will pan out before making orders.
Intel must never let go of the fab, x86 is a commodity, no one really cares what is in their system but being the backbone on the entire modern society is on a different level.

No "average" consumer is really starved for computing power, even Intel realized that, look at Lunar Lake.

 

[AcrosTinus]

Not that I have been following this as closely as I used to, but a little research dug up this:

AMD Turin - 192 cores Zen 5c and Zen 5 variants (I think) - release Q4 2024
Intel Sierra Forest - 144 E cores - Release June 2024
Intel Sierra Forest - 288 E cores - Release Q4 2024 (Intel hasn't been very good about meeting their server roadmap lately though)
Intel Granite Rapids - 2025

Keeping in mind that AMD E Cores still have SMT that generally provides about a 30% boost in multi-threaded workloads, it still looks to me like AMD Turin variants will be quite competitive to even the 244 core variants (that currently don't exist) in terms of performance.

I haven't seen an analysis yet of the Sierra Forest version of E cores used (surely not Arrow Lake class, but rather the previous gen right?) vs AMD 5C E core, but I am guessing it may be that the Zen 5c is a more performant core as well.

I guess what I am guessing is that Intel is still going to be in a losing position against Turin (or at best break even) throughout 2025.

In 2026 AMD will double their core count with 32 core CCX's providing a 512 core EPYC on Zen 6 on TSMC 2nm.

Now I can't see that far into the future; however, it doesn't look to me like Intel is going to be scarfing up the server profits as they did prior to AMD Zen in the next few years. The best they can hope for is to stop the market share bleeding IMO, and even this will come at a greatly decreased profit margin compared to what they once enjoyed.

Which still means that Intel is playing catchup.

Furthermore, their late entry into AI compute is going to be a big drag as well. I think even AMD got caught here with their pants down while NVIDIA ran away with the market. Still, AMD is on its 3rd gen of AI processors (I think) and Intel is just getting started. Makes me think of how AMD leapt ahead of Intel in the integrated graphics market. Again, Intel got caught sleeping at the wheel (seems like their newest integrated graphics is caught up though).

I feel like Intel has gotten by in the past with a combination of market supremacy (and the market power they were able to deploy) and a crazy big lead in process node technology.

I wonder how Intel will fare when it doesn't have cash flowing like rivers (which it doesn't anymore) and it has to compete on the same process node as AMD (which it still really isn't doing. It's paying for a much more expensive node in order to remain competitive ... except for the server dense market where AMD is using N3X so things are about even there with regards to process anyway).

All the bad financial misses have been putting pressure on its money as well. Turns out the market doesn't like it much when you repeatedly miss profit and revenue targets. Last thing I read was no dividends either. That has to make a few very wealthy people upset.

Anyway. Just eating popcorn and enjoying the show as this all plays out.

Don't be so doom and gloom.
AMD had its decade of shame and revived, so can Intel.
We are talking about high tech products where basically no one can break into the market.

This is not AMD Bulldozer against Core, this is Intel with competitive products against another competitor, no product really outshines the other.

I still think though that Intel is significantly better than AMD but most people will never get into the niche conditions where Intel just has more polish.
(IOMMU, Virtualization performance, Intel math library, Intel Network stack performance, Intel storage performance, Intel RND 4K performance, Intel memory controller performance, Intel Media Encoders).

All these things make Intel on desktop or server the first choice for me. I even tried connecting my Optane P5810X to my AMD 9950X server it has worse storage performance than a Raptor Lake i9...

 

[Markfw]

They seem to be doing fine due to supply chains not being impacted currently, a sanction here, a flood there some wars sprinkled on top and AMD is out of the race. Even Nvidia is waiting to see how the Intel node will pan out before making orders.
Intel must never let go of the fab, x86 is a commodity, no one really cares what is in their system but being the backbone on the entire modern society is on a different level.

No consumer is really starved for computing power, even Intel realized that, look at Lunar Lake.

This is not correct. Statistically, many consumers are not starved, but there are quite a few like me, and many others who really could use threadripper in an affordable package, but can not get.