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.



Comparison of upcoming Intel's U-series CPU: Core Ultra 100U, Lunar Lake and Panther Lake

Model	Code-Name	Date	TDP	Node	Tiles	Main Tile	CPU	LP E-Core	LLC	GPU	Xe-cores
Core Ultra 100U	Meteor Lake	Q4 2023	15 - 57 W	Intel 4 + N5 + N6	4	tCPU	2P + 8E	2	12 MB	Intel Graphics	4
?	Lunar Lake	Q4 2024	17 - 30 W	N3B + N6	2	CPU + GPU & IMC	4P + 4E	0	8 MB	Arc	8
?	Panther Lake	Q1 2026 ?	?	Intel 18A + N3E	3	CPU + MC	4P + 8E	4	?	Arc	12

Comparison of die size of Each Tile of Meteor Lake, Arrow Lake, Lunar Lake and Panther Lake

	Meteor Lake	Arrow Lake (20A)	Arrow Lake (N3B)	Arrow Lake Refresh (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop Only	Desktop & Mobile H&HX	Desktop Only	Mobile U Only	Mobile H
Process Node	Intel 4	Intel 20A	TSMC N3B	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Q1 2025 ?	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2025 ?	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	6P + 8E ?	8P + 16E	8P + 32E	4P + 4E	4P + 8E
LLC	24 MB	24 MB ?	36 MB ?	?	8 MB	?
tCPU	66.48
tGPU	44.45
SoC	96.77
IOE	44.45
Total	252.15

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.)

 

[AMDK11]

Returning to the RedwoodCove core, I believe that the diagram with changes in the x86 P-core in the GraniteRapids presentation and the data from the MeteorLake presentation complement each other, but do not yet reveal everything that has been changed in RedwoodCove. I really have no reason at this point to think that IPC is substantially similar to Golden/RaptorCove. Why all these changes if the IPC is the same? In my opinion, the issue with RedwoodCove is not whether the IPC is higher, but how much higher it is.

Historically, Core 2 (Conroe - 2006) introduced 33% wider x86 decoding from 3-way to 4-way.We have 4-way decoding up to Sunny/CypressCove, i.e. 2020. That's about 14-15 years of generational improvements, leaving 3 ALUs up to IvyBridge as Haswell introduced 4 ALUs.

Only GoldenCove (2021) introduced wider decoding by as much as 50%, i.e. it moved from a 4-way to 6-way x86 instruction decoder and increased the number of ALU units by 25% from 4 to 5.

I think Intel admitted that the transition from GoldenCove to RedwoodCove is not as serious a change as the transition from SunnyCove to GoldenCove, but it does not rule out IPC gains, because as part of 6-way decoding, through appropriate optimization and core expansion,you can still gain a lot, so I'm not eliminating the higher IPC from RedwoodCove.just because it doesn't have 8-way decoding or 6-8 ALUs.I think optimization and core expansion gives the IPC a gain, the only question is how much.

Personally, I think (this is my feeling) that with the transition to 6-way x86 decoding in Zen 5, AMD will break ground for at least a few generations, where the IPC increase will definitely not be just +5%.I have the impression that AMD whether Intel, despite the rumors, will not switch to more than a 6-way decoder in Zen 6-7 or LionCove-PantherCove.8-way x86 decoding is rather a distant future.But I may be wrong.

 

[Doug S]

…except ASUS, Gigabyte, ASRock, and MSI. Why they don’t invest more in smaller form factors is baffling to me. I have a mini PC not much bigger than a Raspberry Pi + case sitting on my desk and it is absolutely fantastic. If I weren’t a gamer, developer, and content creator, it is all I would need. (it can game, just not anything serious)

Selling boards with the CPU soldered on is a risk for them. They can't offer all CPU SKUs so they would have to pick a few, and then hope they sell well enough they don't get stuck with a bunch of boards with a particular offering that proves unpopular. A $100 board without a CPU sitting in inventory can be discounted 20% and that discount costs them $20. A $100 board with a $200 CPU discounted 20% costs them $60 - and if MTL has been replaced with something better they will have to discount a lot more than 20% to sell it. Whereas if the socket is compatible they might merely need a firmware update then that board can be used with the "something better" so that 20% discount will easily sell it.

Maybe they could do it build to order like laptops are (sometimes) made, but motherboard companies have never done that so they'd have to set themselves up to handle direct customer orders and the build to order process in their factory. They would have to believe the market size is big enough to be worth it.

 

[cebri1]

GLC on Intel 7 appears to be as efficient (or marginally less) efficient than Zen 3. If RWC is as efficient as Zen 4, I would be pretty impressed tbh, because Zen 4 increased efficiency from node, but then also increased efficiency through architecture. While I think it's likely RWC has slightly higher architectural perf/watt than GLC as well, I doubt it's anywhere near where Zen 4 got vs Zen 3, and thus if RWC is matching Zen 4 in perf/watt, it's because the node turned out better than expected (or Intel's DTCO team went hard)

Nothing to add really, but always good to remember how efficient an 13900K can be if you let it (at the cost of losing the performance crown).

 

[deasd]

GLC on Intel 7 appears to be as efficient (or marginally less) efficient than Zen 3. If RWC is as efficient as Zen 4, I would be pretty impressed tbh, because Zen 4 increased efficiency from node, but then also increased efficiency through architecture. While I think it's likely RWC has slightly higher architectural perf/watt than GLC as well, I doubt it's anywhere near where Zen 4 got vs Zen 3, and thus if RWC is matching Zen 4 in perf/watt, it's because the node turned out better than expected (or Intel's DTCO team went hard)

Intel4/RWC needs magic to do so.

As someone tested, at 5.2Ghz 1T Zen4 use only 30% of power of a GLC(or RaptorCove) and 2T <50% of power.

Nobody could tell if GLC is bugged or not. If yes RWC has a chance to get fix.

https://twitter.com/x/status/1702629569834938549

 

[Abwx]

Nothing to add really, but always good to remember how efficient an 13900K can be if you let it (at the cost of losing the performance crown).

View attachment 86386

At 88W the 7950X score 11863 pts, that s 134.8 pts/W, so this graph is BS.

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

 

[cebri1]

At 88W the 7950X score 11863 pts, that s 134.8 pts/W, so this graph is BS.

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

No it's not redacted and der8auer is pretty well respected tester. Sorry for the facts.

We have a zero tolerance for profanity in the tech forums.
Please don't do that again.

Iron Woode
Super Moderator

 

[Abwx]

No it's not redacted and der8auer is pretty well respected tester. Sorry for the facts.

The fact is that he did implement 90W TDP wich lead to 121.5W PPT, that s the actual power at wich he did set the 7950X, give us the link, we ll see that his score is quite higher than the one of Computerbase at 88W and amount to a test at said 121.5W.

And btw, Computerbase are known for their precision, you can see on the link that i posted that they do a lot of tests at different powers.

Edit : There s no 90W eco mode on the 7950X, the stock eco mode is at 105W TDP/142W PPT, if one set the TDP at 90W then the CPU will push up to its PPT wich is TDP x 1.35.

 

[AMDK11]

I think I know where MLID got the information that GraniteRapids is RedwoodCove+, which is not true.He probably saw these slides from Intel Innovation 23 before and thought that since there are more changes in the description of RedwoodCove on GraniteRapids and when presenting the new P core, MeteorLake is mentioned only L1 64KB, he misinterpreted it as RedwoodCove in MTL it's basically thesame as Golden/RaptorCove, and since more changes were described in GranitaRapids, he assumed it was RedwoodCove+.But this is not true and the guy is misinterpreting the data from his source, which was has already been demonstrated many times.Moreover, he himself admitted that he is not familiar with technical issues and confuses IPC with efficiency and energy efficiency.

EDIT:
RedwoodCove Core:
- Larger instruction cache: 64KB 16-way (Golden/RaptorCove L1-I 32KB 8-way)
- Improved code prefetching
- Smarter code prefetching engine
- Improved branch prediction accuracy and reduced miss-hit penalty
- Micro-op queue 192 for ST or 2x96 for SMT
Golden/RaptorCove 144 ST/2x 72 SMT
Sunny/CypressCove 70 ST/2x 70 SMT
Skylake 64 ST/2x 64 SMT
Haswell 56 ST/2x 56 SMT
- Greater performance, lower instruction latency FPU with 4-5 to 3 (logic extension)
- AMX supports TF32 and FP16
- New L2 design, more preferred LLC(L3) prefetch engine, higher throughput.

I believe that there are more changes, but if they do not bring an increase in IPC, I will be very disappointed.Meanwhile, I'm waiting for the premiere.

 

[PJVol]

Even assuming these numbers are accurate, we don’t know how much performance scales with power.

GB6 numbers (if legit) are probably more telling due to the wide range of test workloads (2502 / 12545 )

Intel Corporation Meteor Lake Client Platform - Geekbench

Benchmark results for an Intel Corporation Meteor Lake Client Platform with an Intel Core Ultra 7 165H processor.

browser.geekbench.com

 

[Geddagod]

I really have no reason at this point to think that IPC is substantially similar to Golden/RaptorCove. Why all these changes if the IPC is the same? In my opinion, the issue with RedwoodCove is not whether the IPC is higher, but how much higher it is.

Intel themselves are saying RWC is esentially just a GLC port in interviews. There's not going to be much of an IPC increase with RWC. Crestmont has a 4-6% IPC increase, and they listed it out as having an IPC increase, and the fact that they didn't bother with RWC means that the IPC increase for that is almost certainly even lower than 4-6%.

 

[inf64]

I think I know where MLID got the information that GraniteRapids is RedwoodCove+, which is not true.He probably saw these slides from Intel Innovation 23 before and thought that since there are more changes in the description of RedwoodCove on GraniteRapids and when presenting the new P core, MeteorLake is mentioned only L1 64KB, he misinterpreted it as RedwoodCove in MTL it's basically thesame as Golden/RaptorCove, and since more changes were described in GranitaRapids, he assumed it was RedwoodCove+.But this is not true and the guy is misinterpreting the data from his source, which was has already been demonstrated many times.Moreover, he himself admitted that he is not familiar with technical issues and confuses IPC with efficiency and energy efficiency.

EDIT:
RedwoodCove Core:
- Larger instruction cache: 64KB 16-way (Golden/RaptorCove L1-I 32KB 8-way)
- Improved code prefetching
- Smarter code prefetching engine
- Improved branch prediction accuracy and reduced miss-hit penalty
- Micro-op queue 192 for ST or 2x96 for SMT
Golden/RaptorCove 144 ST/2x 72 SMT
Sunny/CypressCove 70 ST/2x 70 SMT
Skylake 64 ST/2x 64 SMT
Haswell 56 ST/2x 56 SMT
- Greater performance, lower instruction latency FPU with 4-5 to 3 (logic extension)
- AMX supports TF32 and FP16
- New L2 design, more preferred LLC(L3) prefetch engine, higher throughput.

I believe that there are more changes, but if they do not bring an increase in IPC, I will be very disappointed.Meanwhile, I'm waiting for the premiere.

Be prepared to be disappointed, I guess.

 

[Hulk]

I don’t think anyone made such a claim. Meteor Lake coming to the desktop means literally just that.

I have a small desktop PC sitting on a shelf that runs an AMD Cezanne mobile chip. The system isn’t much bigger than my wallet in terms of size. MTL-P 6+8 in one of those small boxes would be amazing to have.

As I had thought, no Meteor Lake desktop parts. Just mobile chips in AIO and similar.

Intel Core "Meteor Lake" Confirmed NOT Coming to Desktops, Only AIOs and Mini PCs

Intel's next generation Core "Meteor Lake" processor was confirmed by ComputerBase.de to not release on the desktop platform. The processor will not make it to a socketed desktop package such as the upcoming LGA1851. It will see a mobile-only (notebook and tablets only) launch, with select...

www.techpowerup.com

 

[HoveringStyle]

Intel4/RWC needs magic to do so.

As someone tested, at 5.2Ghz 1T Zen4 use only 30% of power of a GLC(or RaptorCove) and 2T <50% of power.

Nobody could tell if GLC is bugged or not. If yes RWC has a chance to get fix.

https://twitter.com/x/status/1702629569834938549

View attachment 86389


View attachment 86390

Is that in watts? they seem incredibly low values considering the 7600X (5.3GHz) uses 30W in Cinebench 1T


The same as the 13600K which also has a 5.3GHz max clock.

 

[Abwx]

Is that in watts? they seem incredibly low values considering the 7600X (5.3GHz) uses 30W in Cinebench 1T
View attachment 86397

The same as the 13600K which also has a 5.3GHz max clock.

That s a whole CPU package power, idle CPU package power is 13.5W and with 1 CB thread it increase to 33W, so a 7600X core use about 19.5W in CB ST.

13600K idling is at 6.5W and 1T loaded at 30.8W, that s 24.3W for the core.

The numbers are here on the third graph :

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

 

[HoveringStyle]

Taking idle=uncore, the raptor lake SPEC power usage seems in line CB but Zen 4 is roughly 10W lower. What gives ¯\_(ツ)_/¯

 

[Abwx]

CB is FP, this seems to use more power than SPECint, wich is an INT code test, should compare with SPECfp to have an idea, but surely that the latter use more power and will mimick CB power wise.

 

[HoveringStyle]

The 13900K was also tested in SPECint and its power values are still in line with CB but it's entirely possible Zen 4 is just that different in that aspect.

 

[Abwx]

The 13900K was also tested in SPECint and its power values are still in line with CB but it's entirely possible Zen 4 is just that different in that aspect.

Zen 4 seems to be much more efficient than RPL P core for anything Integer, even when accounting for node difference, so that s likely due to uarchitectural advantage for such code.

FI in Handbrake H264 the 13900K consume 249W and the 7950X 189W while still being 13% faster, even when limited to 142W it still manage to be ahead by 9%

So at equal score it would use something like 110W wich is 2.26x more efficient, that s way more than the advantage provided by TSMC s N5 over Intel 7.

The numbers i quoted are in those tests :

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

 

[Geddagod]

Zen 4 seems to be much more efficient than RPL P core for anything Integer, even when accounting for node difference, so that s likely due to uarchitectural advantage for such code.

The difference between the spec testing vs the CB testing is way higher than one would expect, even if Zen 4 is more efficient in INT workloads.

FI in Handbrake H264 the 13900K consume 249W and the 7950X 189W while still being 13% faster, even when limited to 142W it still manage to be ahead by 9%

So at equal score it would use something like 110W wich is 2.26x more efficient, that s way more than the advantage provided by TSMC s N5 over Intel 7.

The numbers i quoted are in those tests :

You would want to compare 8+16 with a 12 core AMD processor, not a 16 core one, if you want to try to equalize "core counts"
But honestly, we don't even have to do that. Why bother with the little cores ruining GLC vs Zen 3/Zen 4 perf/watt testing? Just compare a 12400 vs 7600x. Or RPL with e-cores disabled vs a 7600x. Though that perf data is hard to come by from what I've seen.

 

[Abwx]

The difference between the spec testing vs the CB testing is way higher than one would expect, even if Zen 4 is more efficient in INT workloads.

In CB ST a Zen 4 core use about 37W while a single RPL core use 47W, in this case that s the process advantage wich is at work, but in the case of Handbrake it need only 115W to match the stock 13900K, that s comparable to the difference in SPECint, wich point to much better uarch efficency for INT.

You would want to compare 8+16 with a 12 core AMD processor, not a 16 core one, if you want to try to equalize "core counts"
But honestly, we don't even have to do that. Why bother with the little cores ruining GLC vs Zen 3/Zen 4 perf/watt testing? Just compare a 12400 vs 7600x. Or RPL with e-cores disabled vs a 7600x. Though that perf data is hard to come by from what I've seen.

Since 2 e cores are supposed to have as much throughput than 1 P core using SMT it s straightforward that 16 e cores are the equivalent of 8 P cores, so 13900K vs 7950X is the adequate comparison.

Now if you want a core/core comparison in Handbrake a 7700X@65W is 2% faster than a 12900K with e cores disabled according to computerbase measurements, knowing that those 8 P cores run at something like 180W.

Granted ADL has a less efficient process than RPL, but that would reduce its power to 130-140W, still far from 65W, we fall again in the 2x better perf/Watt of the 7950X vs 13900K.

 

[Geddagod]

In CB ST a Zen 4 core use about 37W while a single RPL core use 47W, in this case that s the process advantage wich is at work, but in the case of Handbrake it need only 115W to match the stock 13900K, that s comparable to the difference in SPECint, wich point to much better uarch efficency for INT.

Uh no. A) Why again are we using handbrake? B) Why again are we comparing a 13900k vs a 7950x, which is muddled because the core count inequality but also the fact we can't even compare GLC vs Zen 4 due to Gracemont... you know.. existing
The logic here does not track, at all, not even in the slightest.

Since 2 e cores are supposed to have as much throughput than 1 P core using SMT it s straightforward that 16 e cores are the equivalent of 8 P cores, so 13900K vs 7950X is the adequate comparison.

OMG. Idk how many times I have to tell you this. It is asinine to compare 2 E-cores as a big core unless you want to try making Intel seem worse than they really are. If you look at a die shot, which I have sent you plenty, you can see, physically on the die, if Intel didn't develop any E-cores at all, they would have slotted in 1 P-core in the place of a 4-e core cluster. This is the third time I have pointed this out to you. No one compares 2 E-cores to 1 P core, you aren't being weird and quirky, you're not fighting in the face of bias, by equalizing them like that. Everyone uses the 4 E-cores = 1 "P" core measurement because that's how Intel literally developed the E-cores- a 4core cluster all but slots into the place where they would have had a p core on the die.

Now if you want a core/core comparison in Handbrake a 7700X@65W is 2% faster than a 12900K with e cores disabled according to computerbase measurements, knowing that those 8 P cores run at something like 180W.

Granted ADL has a less efficient process than RPL, but that would reduce its power to 130-140W, still far from 65W, we fall again in the 2x better perf/Watt of the 7950X vs 13900K.

This entire statement is something else, man.
First of all, we are going to circle back to what perf/watt means. Again, we had this discussion before. Remember, you equalize the denominator, not the numerator. You're talking about watt/perf. This is an important distinction to make, because saying Intel is 2x behind in perf/watt is just not true and makes them seem much further behind then they really are.
But anyway, this is what I see on computerbase:

At 88 watts, the 7700x has ~35% more perf/watt than a 12900k with only 8 P-cores.
This isn't really ideal, considering I doubt RWC brings more than a 20% increase in perf/watt, meaning Intel would be behind (big core vs big core) by ~10% in perf/watt, but whatever.

 

[H433x0n]

Uh no. A) Why again are we using handbrake? B) Why again are we comparing a 13900k vs a 7950x, which is muddled because the core count inequality but also the fact we can't even compare GLC vs Zen 4 due to Gracemont... you know.. existing
The logic here does not track, at all, not even in the slightest.

OMG. Idk how many times I have to tell you this. It is asinine to compare 2 E-cores as a big core unless you want to try making Intel seem worse than they really are. If you look at a die shot, which I have sent you plenty, you can see, physically on the die, if Intel didn't develop any E-cores at all, they would have slotted in 1 P-core in the place of a 4-e core cluster. This is the third time I have pointed this out to you. No one compares 2 E-cores to 1 P core, you aren't being weird and quirky, you're not fighting in the face of bias, by equalizing them like that. Everyone uses the 4 E-cores = 1 "P" core measurement because that's how Intel literally developed the E-cores- a 4core cluster all but slots into the place where they would have had a p core on the die.

This entire statement is something else, man.
First of all, we are going to circle back to what perf/watt means. Again, we had this discussion before. Remember, you equalize the denominator, not the numerator. You're talking about watt/perf. This is an important distinction to make, because saying Intel is 2x behind in perf/watt is just not true and makes them seem much further behind then they really are.
But anyway, this is what I see on computerbase:
View attachment 86411
At 88 watts, the 7700x has ~35% more perf/watt than a 12900k with only 8 P-cores.
This isn't really ideal, considering I doubt RWC brings more than a 20% increase in perf/watt, meaning Intel would be behind (big core vs big core) by ~10% in perf/watt, but whatever.

There’s no way he’s acting in good faith. It’s a waste of time to engage with him.

 

[deasd]

Is that in watts? they seem incredibly low values considering the 7600X (5.3GHz) uses 30W in Cinebench 1T


The same as the 13600K which also has a 5.3GHz max clock.

The author test cpu Vcore VRM only, which means SoC(IODIE) is excluded which is a huge variable. Also cinebench is different.

 

[igor_kavinski]

GB6 numbers (if legit) are probably more telling due to the wide range of test workloads (2502 / 12545 )

Intel Corporation Meteor Lake Client Platform - Geekbench

Benchmark results for an Intel Corporation Meteor Lake Client Platform with an Intel Core Ultra 7 165H processor.

browser.geekbench.com

ASUSTeK COMPUTER INC. ROG Flow Z13 GZ301VV_GZ301VV vs Intel Corporation Meteor Lake Client Platform - Geekbench

MTL is somehow very potent in multicore file compression and object detection tests. Could be higher frequency or the increased L1/L2 or much higher bandwidth RAM.

 

[Henry swagger]

Intel Arrow Lake-H CPUs Feature Lion Cove P-Cores, Skymont E-Cores & Possibly Crestmont LP E-Cores

As the Intel Meteor Lake launch nears, rumors on the next-gen Arrow Lake & Lunar Lake CPUs have already started floating on the internet.

wccftech.com

Arrow will have soc cores too for desktop ? 🤔