Intel Meteor Lake Desktop CPUs Spotted in Presentation, Leak Indicates Core i3 and i5 Only

[T0@st]

Following on from yesterday's news of Meteor Lake's "Adamantine" L4 cache another leaky bit of information has popped up. A tipster on Twitter, Bionic_squash, has uploaded a slide from a supposedly official Intel presentation document, and it shows a small selection of Meteor Lake-S and Arrow Lake-S desktop CPUs, as well as the refreshed Raptor Lake-S series. The majority of recent leaks have pointed to laptop variants of Intel's fourteenth generation Core lineup, and not much has emerged about a desktop-dedicated range in a while - prompting further murmurs about Team Blue canning that side of things. The Meteor Lake-S family is still in the works according to the leaked chart and industry experts reckon that a product launch is due later in the year.

By looking closely at the chart, it shows that the Meteor Lake-S desktop processors are limited to 35 and 65 W TDPs, meaning that Core i3 and i5 lines are the only offerings within the 14th generation desktop lineup. Performance enthusiasts will need to look at the 15th gen Arrow Lake-S lineup - where the big i7 and i9 CPUs (up to 125 W) sit, or the refreshed Raptor Lake lineup which also offers a wide range of options - from i3 up to i9. Industry experts are a bit puzzled about Meteor Lake's prospects in the desktop processor sector - when considering a (speculated) skew to more entry-level and mid-range minded customers. Will Intel lose out by not offering more powerful variants, or are they working on a refreshed 14th generation product lineup for 2025?



View at TechPowerUp Main Site | Source

 

[Dristun]

How fast at 65W though? I wouldn't mind upgrading if the 6+8 is going to comfortably beat, say, 8+8 13700 at 65W.

 

[docnorth]

I5's are fine for me, but after 2-3 months the new PC must be at the office. So the choice is between Intel RPL i5 and AMD 7600 (or 7700) non-X.

 

[ZoneDymo]

I5's are fine for me, but after 2-3 months the new PC must be at the office. So the choice is between Intel RPL i5 and AMD 7600 (or 7700) non-X.

Go AMD, just cuz its kewl

 

[R0H1T]

Yields must be pretty bad for them to be releasing just till i5, this happened with Broadwell & the infamous Cannonlake which was even worse!

 

[ixi]

How fast at 65W though? I wouldn't mind upgrading if the 6+8 is going to comfortably beat, say, 8+8 13700 at 65W.

Intel and 65w power draw... Is it possible?

Well, lets see performance, but weird to see only up to i5. Maybe by going i5 they do not want to overcome current i7?

 

[Nanochip]

So meteor lake is the i5 and arrow lake is the i7/i9?

What’s going on?

Both lakes have different core generation designs, and process nodes so the question is, will arrow lake-a launch along side meteor lake-s?

 

[trsttte]

Oh wow, stuff really not going Intel's way. They can limp along for many more years to come, but pressure must be mounting on

 

[kondamin]

Clocks must be really bad if that really is the case

 

[TheinsanegamerN]

The fact they have two different VRM requirements makes me think we will end up seeing I3/I5 only boards out of this. Maybe i7/i9 will become sudo HDET again?

 

[Darmok N Jalad]

Yields must be pretty bad for them to be releasing just till i5, this happened with Broadwell & the infamous Cannonlake which was even worse!

Cannon Lake was a disaster. When the first product out the gate is a very delayed dual core with a disabled iGPU, you know it’s bad. I thought Broadwell had a bit more availability eventually, but that was a longer time ago.

 

[N/A]

This time nothing is disabled, 6 core tile is 6 core enabled. and the GPU tile may not even be present, that means i5 14400F for cheap.

But I waited for this long. the angstrom era is just around the corner.

 

[Minus Infinity]

This is exactly what Tom at MLiD posted a few weeks ago from his sources. ML is only for i5 and i3 desktop and i7 and i9 will have to wait for Arrow Lake. Still 14600K should decimate 13600K if the large IPC uplifts on both P and E cores are true.

I just hope we see real progress on performance per watt and massive power reductions in MT applications.

So meteor lake is the i5 and arrow lake is the i7/i9?

What’s going on?

Both lakes have different core generation designs, and process nodes so the question is, will arrow lake-a launch along side meteor lake-s?

Arrow Lake is essentially a refined version of ML, it's not fundammentally a different design or architecure. It leverages a refined process node, and improved P and E cores and probably higher clocks. Meteor Lake is a radical shift in design and is proving difficult to get right. Meteor Lake is a year late. Don't forget ML is not only new process node, it uses chiplets and a Foveros FPGA interposer.

I would wait for Arrow Lake no matter how good Meteor Lake would be. I'm not buying into beta designs.

 

[usiname]

Another disaster not capable to make its way to the desktop, but intel found the solution, 14900h (6+8) in desktop package as 14600k. Lets see how bad will be the new 7nm and how long will took them to fix it. Maybe in the end they will sell 100% of the production of their fabs to 3rd party to fund 100% waffers from TSMC. After all, that is why the tax payers money went to intel, right?

 

[hs4]

OneRaichu tweeted the plan for MTL-S, which diverts the 6+8 die of MTL-P. However, many tech blogs misread it and interpreted it in a funny way, saying that the cancelled design of MTL-S was diverted to make MTL-P.

What can be predicted with some certainty about Meteor lake:
- Power consumption at the same clock will be 2/3 compared to Intel 7 (data also shown at IEDM2022).
- The maximum clock is about the same or slightly lower than RPL.
- Technologies that reduce power consumption by 20% at low to medium loads, such as DLVR, will be introduced.
- IPC is only slightly higher than RPL.

Simply put, '14600' (best-binned MTL 6+8 die) will be '13600K with 2/3 the power consumption'. As a result, it will have
- ST performance similar to 13900K, 7800X3D +20%.
- MT performance similar to 13600K (PL2 unlimited), 7800X3D +40%.
- Gaming performance similar to 13900K and 7800X3D, more better if ADM cache works
- Maximum power consumption ranges 105-120W
- Power efficiency similar to 7800X3D, 10W lower idle

 

[DavidC1]

I just hope we see real progress on performance per watt and massive power reductions in MT applications.

MLID is wrong about Meteorlake cores having big 15-20% gains per clock. Meteorlake is lucky to get few single digit percentages(4-6%). The reliable leakers like OneRaichu says that. Official Intel presentations say Redwood Cove and Intel 4 offers 20% perf/watt gains, which means in power constrained environments it can perform better than Raptorlake.

This is why Meteorlake-S is low end.

Arrow Lake is essentially a refined version of ML, it's not fundammentally a different design or architecure. It leverages a refined process node, and improved P and E cores and probably higher clocks. Meteor Lake is a radical shift in design and is proving difficult to get right. Meteor Lake is a year late. Don't forget ML is not only new process node, it uses chiplets and a Foveros FPGA interposer.

I would wait for Arrow Lake no matter how good Meteor Lake would be. I'm not buying into beta designs.

No, in terms of performance gains and architecture, Arrowlake is the big one. While the Redwood Cove cores in Meteorlake get maybe 5% per clock, Lion Cove cores in Arrowlake is expanding it greatly. Going from 6 to 8 decoders, and 700+ Reorder buffers.

Simply put, '14600' will be 13600K with 2/3 the power consumption. The result,
- ST performance will be similar to 13900K, 7800X3D +20%.
- MT performance similar to 13600K (PL2 unlimited), 7800X3D +40%.
- Gaming performance similar to 13900K and 7800X3D, more better if ADM cache works
- Maximum power consumption ranges 105-120W
- Power efficiency similar to 7800X3D, 10W lower idle

You'll be very disappointed if that's what you are expecting for Meteorlake. Rather than just parroting tech sites that you criticize(and you are right), go and actually read Intel 4 coverage. The gains in perf/clock and perf/watt drops when the clock frequency goes near 4GHz. 20% clock gain at same power about 3.5GHz drops to only 10-12% at 4GHz.

At 5+GHz the desktop parts play at, you are essentially seeing no gain. And how can ST be 20% faster when the uarch is a minor update and it's not power limited anyway?

5% ST gain, 10% MT gain, in the optimistic scenario. Arrowlake is the big one, but Meteorlake is only slightly better than a refresh.

 

[hs4]

This is exactly what Tom at MLiD posted a few weeks ago from his sources. ML is only for i5 and i3 desktop and i7 and i9 will have to wait for Arrow Lake. Still 14600K should decimate 13600K if the large IPC uplifts on both P and E cores are true.

I just hope we see real progress on performance per watt and massive power reductions in MT applications.


Arrow Lake is essentially a refined version of ML, it's not fundammentally a different design or architecure. It leverages a refined process node, and improved P and E cores and probably higher clocks. Meteor Lake is a radical shift in design and is proving difficult to get right. Meteor Lake is a year late. Don't forget ML is not only new process node, it uses chiplets and a Foveros FPGA interposer.

I would wait for Arrow Lake no matter how good Meteor Lake would be. I'm not buying into beta designs.

Meteor lake and Arrow lake (if 20A) will be very different.
- If Semianalysis' "Meteor Lake Die Shot and Architecture Analysis" is correct, there is little difference between MTL and RPL in IPC.
- If the story posted by AnandTech in summer 2021 is correct, the E-core of Arrow lake will be "Skymont", and according to OneRaichu, a certain IPC improvement (around 20%) can be expected with Skymont.
- Intel 7 and Intel 4, both FinFET, have similar design limitations and can be easily adapted. On the other hand, 20A will introduce RibbonFET and PowerVIA, which will significantly change the design rules, making some changes to the microarchitecture inevitable.
- RibbonFETs enable faster switching and PowerVIA have better high-frequency tolerance; 20A is expected to have a clock improvement of about 15% at the same power consumption.

 

[DavidC1]

- RibbonFETs enable faster switching and PowerVIA have better high-frequency tolerance; 20A is expected to have a clock improvement of about 15% at the same power consumption.

That's only true at lower clocks. At higher clocks that gain is significantly reduced. Again, look at the Intel 4 graph.

E-cores on the other hand I expect different. Skymont is probably 30% per clock faster than Crestmont(Meteorlake).

 

[hs4]

Another disaster not capable to make its way to the desktop, but intel found the solution, 14900h (6+8) in desktop package as 14600k. Lets see how bad will be the new 7nm and how long will took them to fix it. Maybe in the end they will sell 100% of the production of their fabs to 3rd party to fund 100% waffers from TSMC. After all, that is why the tax payers money went to intel, right?

One AnandTech post estimates the Intel 7 yield based on the percentage of 'F' SKUs shipped. It says that nearly 90% of the B0 die being able to ship as non-F 13900K/KS. As far as I can tell, Intel 4 yields are also high enough. On the other hand, TSMC's N3 virtually delays 2 years, and GAA/BPD is expected to be introduced later than Intel's, so the situation is reversing in process node development.

The Sapphire Rapids development team was interviewed by the New York Times in January of this year, and it seems that there is an awareness within the company that the new design challenges associated with tiling are increasing. They said that if the product lineup is not narrowed down, the design capacity will be exceeded. This is probably the reason why MTL-S was reduced to 6+8 dies in common with MTL-P.

 

[londiste]

Yields must be pretty bad for them to be releasing just till i5, this happened with Broadwell & the infamous Cannonlake which was even worse!

Probably more to do with the first proper desktop chiplet design from Intel.

 

[JustBenching]

How fast at 65W though? I wouldn't mind upgrading if the 6+8 is going to comfortably beat, say, 8+8 13700 at 65W.

You mean the 6+8 i5 at 65w beating a stock 13700k? That's not happening. No way. And contrary to popular belief, Intel CPUs are actually very efficient, power limit your 13700k to 125w and youd be surprised as hell

 

[Broken Processor]

The fact they have two different VRM requirements makes me think we will end up seeing I3/I5 only boards out of this. Maybe i7/i9 will become sudo HDET again?

If they did that AMD would eat their lunch.

 

[Vayra86]

Oh wow, stuff really not going Intel's way. They can limp along for many more years to come, but pressure must be mounting on

Yep they're still on a slow, painful monolithic death trail

Hell must freeze over before I buy anything Intel high end these days. I totally get why they stop at i5. Anything over it is a total waste of time and most of all energy.
On the gaming front they have absolutely nothing to combat X3D either. At best they reach performance parity while using triple the energy, but most of the time even that isn't possible.

You mean the 6+8 i5 at 65w beating a stock 13700k? That's not happening. No way. And contrary to popular belief, Intel CPUs are actually very efficient, power limit your 13700k to 125w and youd be surprised as hell

They are efficient but the performance ceiling at which they're still efficient is not nearly as close to topping charts as they would like.

 

[hs4]

MLID is wrong about Meteorlake cores having big 15-20% gains per clock. Meteorlake is lucky to get few single digit percentages(4-6%). The reliable leakers like OneRaichu says that. Official Intel presentations say Redwood Cove and Intel 4 offers 20% perf/watt gains, which means in power constrained environments it can perform better than Raptorlake.

This is why Meteorlake-S is low end.

No, in terms of performance gains and architecture, Arrowlake is the big one. While the Redwood Cove cores in Meteorlake get maybe 5% per clock, Lion Cove cores in Arrowlake is expanding it greatly. Going from 6 to 8 decoders, and 700+ Reorder buffers.


You'll be very disappointed if that's what you are expecting for Meteorlake. Rather than just parroting tech sites that you criticize(and you are right), go and actually read Intel 4 coverage. The gains in perf/clock and perf/watt drops when the clock frequency goes near 4GHz. 20% clock gain at same power about 3.5GHz drops to only 10-12% at 4GHz.

At 5+GHz the desktop parts play at, you are essentially seeing no gain. And how can ST be 20% faster when the uarch is a minor update and it's not power limited anyway?

5% ST gain, 10% MT gain, in the optimistic scenario. Arrowlake is the big one, but Meteorlake is only slightly better than a refresh.

The figure below was presented at VLSI Symposium 2022, but the 'industry standard core' is probably not Redwood Cove. At least the Power-Clock curve in Intel 7 is not similar to that of Golden cove. The absolute values of that curves should not be of much concern, as the Power-Clock curve should be different depending on the choice of cell libraries and pipeline structure.

From this figure, it is sufficient to know that power consumption at the same frequency decreases by about 50-30% in most ranges. This is the only official data I have some confidence in. If you have other sources, please let me know.

how can ST be 20% faster when the uarch is a minor update and it's not power limited anyway?

I apologize for the confusing writing style here. I wanted to write, "ST perf will be the same as the 13900K, which means it will be 20% faster than the 7800X3D."

 

[R0H1T]

Unless it clocks to the moon 20% over & above 7800x3d seems highly unlikely.