Intel to Go Ahead with "Meteor Lake" 6P+16E Processor on the Desktop Platform?

[mama]

Don't be hasty. Let's see what it performs like at release. Assuming of course it releases anytime in the next two years.

 

[sanamarcar]

I'm so confused by people's obsession for p-cores and their disdain for e-cores. The e-cores roughly equivalent to skylake cores while taking up way less energy and die space.

1 P-core takes roughly as much space as 4 E-cores, and the E-cores would perform better in pure multi-threaded scenarios.

Given a choice between 10P + 0E cores or 6P +16E cores, there's a very good case to be made for 6+16 config. Games will see negligible impact with 2 fewer P-cores, and multi-threaded apps that can actually use 6+ cores, will see a massive boost because of the extra performance that comes with additional E-cores.

The caveat to all this is that the 6P+16E part can't be a 14900, it must be a 14600-class part, and be priced as such.

And people forget that the limit is not set by the size of the silicon, it is the W. Why do they want 10P cores if 4P at high frequencies consume 150W easily... Intel does what it has to do. In last SD Soc u have 4 types of cores, and P core is only 1 xD.

 

[evernessince]

I'm so confused by people's obsession for p-cores and their disdain for e-cores. The e-cores roughly equivalent to skylake cores while taking up way less energy and die space.

1 P-core takes roughly as much space as 4 E-cores, and the E-cores would perform better in pure multi-threaded scenarios.

Intel pitched these E-cores as being efficient. They are for sure, but what is the point of 16 E-cores? Efficient or just an excuse to bump performance on the cheap?

Alright so let's just dispel the notion that E-Cores are super efficient:

As you can see from the chart above, even with only E-Cores enabled in a task they are most effective in E-Cores are actually less efficient than if you enabled just the P-Cores. The only purpose of E-Cores is so that Intel can cram more cores into it's processors.

E-Cores are massively slower than P-cores:

The only metric they look good in is performance per mm2.

It's easy to see why people do not want a 6P 16E core CPU.

Given a choice between 10P + 0E cores or 6P +16E cores, there's a very good case to be made for 6+16 config. Games will see negligible impact with 2 fewer P-cores, and multi-threaded apps that can actually use 6+ cores, will see a massive boost because of the extra performance that comes with additional E-cores.

The caveat to all this is that the 6P+16E part can't be a 14900, it must be a 14600-class part, and be priced as such.

According to the article, it is the max they will have that gen:

"Apparently, Intel will brazen it out against AMD with a maximum CPU core-count of just 6 performance cores and 16 efficiency cores possible for "Meteor Lake.""

That's a problem, because this is a product that isn't even released yet and the top SKU is poised to loose some performance in games that already utilize more than 6 cores. By the time this processor releases, there will certainly be even more especially as more titles get RT. Then consider how well those 6 cores will cope with future titles as well for what should be a flagship CPU. Intel is going to need to bring an extremely hefty IPC uplift to offset the loss. It kind of also gives AMD a pass to not increase core counts again as well, given that they've been able to match Intel's higher core counts with higher performing cores.

 

[SL2]

It's amazing how every "ecores bad" comment is here from AMD users who will buy an 6+8 zen 5 and love it

You assume they're the same, or comparable, just based on the fact that they're the smaller cores in respective CPU. That's what's amazing here.
Then you're hinting at some people being hypocritical if they buy a future product that we know nothing about.

Just wow.

Spoiler

The Zen 4C core leaks speaks for themselves. They're closer to Zen 4 cores than Gracemont in terms of performance, while still being much smaller than Zen 4 cores.
If that's any indication of what the smaller cores in Zen 5 CPU's are like it will be an interesting product for sure, but I don't care about that at this point.
It's pointless, and it will be worthless information when it actually launches.

I personally refuse to get amazed in either direction this long before launch. We've seen it before, but we know that we need actual benchmarks before getting amazed.

 

[Dr. Dro]

It might shape up to be like BDW-S. ADL and RPL parts are already *plenty* fast, and a regression in core count may not be taken nicely by the market.

 

[Chaitanya]

Yeah so what, 6+8 is very different than 6+16/16+6. I don't mind some e-cores, i'd prefer to have everything full power but if they're able to leverage better efficiency and price (by using smaller die) sure, give me a couple e-cores that can run things in the background. I'm into virtualization so I can have for example the e-cores running less demanding stuff and/or the hypervisor while the p-cores do the demanding stuff like run a game or whatever.

But 16 e-cores!? What am I (or anyone) supposed to do with that? Servers will probably eat up this increased reliance on e-cores, it seems like it's the only response Intel has to massive epyc/threadripper core counts, but for client? It just doesn't compute for me

I dont know that dumb-dumb is suffing in rearside, but I know AMD hasnt made Zen 5 details(use of Assymetric CCXs into single CPU) out to public just yet other than some basic claims.

 

[phanbuey]

You assume they're the same, or comparable, just based on the fact that they're the smaller cores in respective CPU. That's what's amazing here.
Then you're hinting at some people being hypocritical if they buy a future product that we know nothing about.

Just wow.

Spoiler

The Zen 4C core leaks speaks for themselves. They're closer to Zen 4 cores than Gracemont in terms of performance, while still being much smaller than Zen 4 cores.
If that's any indication of what the smaller cores in Zen 5 CPU's are like it will be an interesting product for sure, but I don't care about that at this point.
It's pointless, and it will be worthless information when it actually launches.

I personally refuse to get amazed in either direction this long before launch. We've seen it before, but we know that we need actual benchmarks before getting amazed.

I’m not assuming anything - when did I assume anything?

My point was disparate cores are the future - the current tech trends point to that, it’s already in everyone’s roadmaps to release that. HD cores aren’t going to get the same fps as full cores with hbm/3dvcache.

They’re not hypocritical- they just haven’t used the tech - but they will. That’s not an assumption.

 

[ixi]

If 6 cores can compete with 8, 10 and 12 cores from amd then good, if no. Hell no, muņē waste.

 

[SL2]

Alright so let's just dispel the notion that E-Cores are super efficient:

It's not that simple, you conveniently left out the single thread pic. Why isn't the P-core the most efficient also when running single thread?

What kind of limitation does the E-cores have that causes the efficiency to plummet when using multiple cores?

 

[hs4]

E-cores are pretty much useless in my use cases.
I would treat this as a 6 core CPU
Now we are back to 2017 where 8700k just came out ?

Many reviewers conducted E-core only tests using Affinity adjustment or Alder lake-N, including the article "Intel Core i9-12900K E-Cores Only Performance Review" on this site. According to them, when power is supplied as much as possible

The 12900K's 8 E-cores perform
Cinebench ST: 1050, MT:7700 @ 55 W

The 13900K's 16 E-cores with +10% clock performe
Cinebench ST: 1200, MT:18000 @ 120 W

In other words, the 16 E-cores are comparable to the Ryzen 9 3900X in both ST, MT, and has almost same instruction set. If we follow your definition, the 3900X has 0 cores.

The next gen Crestmont is said to have a 10% IPC improvement or, according to MLID, a 25% or more IPC improvement. If MLID's claims are true (which I don't believe), then Meteor lake's 16 E-cores will be Cinebench ST: 1500, MT: 22000, comparable to 5900X. According to your claim, the 5900X would have 0 cores.

 

[SL2]

I’m not assuming anything - when did I assume anything?

You called out AMD users bashing Intel E, and that they will be buying AMD "E" in the future = "amazing".

Only way this is amazing (and truly hypocritical from AMD users IMO) is if they cores are comparable, otherwise it's apples to oranges.

In a similar way, I could call Intel Arc A770 crap today, but if I buy a newer Intel GPU next year has nothing to do with it. Nothing amazing/weird/hypocritical going on.

 

[hs4]

E-core efficiency I have measured rigorously, but at least 1xP and 2xE cores are both Cinebench MT about 500 points/GHz, and increasing the clock increases power samely. In other words, E-core and P-core have equal power efficiency as long as they are clocked the same.

On the other hand, the 2xE core occupies only about 65% of the area of the 1xP core, so more cores in the same area and lower clocks can achieve the same MT performance at lower power (this is the technique used in mobile GPUs). The table below shows the cases where E-cores can contribute to power efficiency.

 

[Karti]

Please, give i3 segment e-cores aswell

imagine something like

14th gen i-core series

i3- 4P + 2E = 10T
or
4P + 4E = 12T

i5 - just as it is with 13th gen, 6P + 4E = 16T
or
go for 6P +6E = 18T
i7 - 8P + 8E (just like 13th gen)
i9 - 8P + 16E (just like 13th gen) or at least increase it to like 10P + 16E = 36T

could be a nice generation if you ask me

 

[OneMoar]

Alright so let's just dispel the notion that E-Cores are super efficient:

View attachment 285640

As you can see from the chart above, even with only E-Cores enabled in a task they are most effective in E-Cores are actually less efficient than if you enabled just the P-Cores. The only purpose of E-Cores is so that Intel can cram more cores into it's processors.

E-Cores are massively slower than P-cores:

View attachment 285641

The only metric they look good in is performance per mm2.

It's easy to see why people do not want a 6P 16E core CPU.



According to the article, it is the max they will have that gen:

"Apparently, Intel will brazen it out against AMD with a maximum CPU core-count of just 6 performance cores and 16 efficiency cores possible for "Meteor Lake.""

That's a problem, because this is a product that isn't even released yet and the top SKU is poised to loose some performance in games that already utilize more than 6 cores. By the time this processor releases, there will certainly be even more especially as more titles get RT. Then consider how well those 6 cores will cope with future titles as well for what should be a flagship CPU. Intel is going to need to bring an extremely hefty IPC uplift to offset the loss. It kind of also gives AMD a pass to not increase core counts again as well, given that they've been able to match Intel's higher core counts with higher performing cores.

yes because they need to stay in there p0 state longer to perform the same task
this ONLY really applies to benchmark workloads and ONLY when you are benching on ONLY the E-cores
also the E-core take up a lot less die space which means you can have many more of them

all that being said I personally think hybrid architectures are a fad and a stop gap solution

 

[Vayra86]

E-cores are pretty much useless in my use cases.
I would treat this as a 6 core CPU
Now we are back to 2017 where 8700k just came out ?

Sweet! I've always said life's cheap if you have patience, I'm golden then !

Please, give i3 segment e-cores aswell

imagine something like

14th gen i-core series

i3- 4P + 2E = 10T
or
4P + 4E = 12T

i5 - just as it is with 13th gen, 6P + 4E = 16T
or
go for 6P +6E = 18T
i7 - 8P + 8E (just like 13th gen)
i9 - 8P + 16E (just like 13th gen) or at least increase it to like 10P + 16E = 36T

could be a nice generation if you ask me

Yeah awesome, the return of the quadcore in 2023. Intel selling you regression at premium, where can I sign up.

E-core efficiency I have measured rigorously, but at least 1xP and 2xE cores are both Cinebench MT about 500 points/GHz, and increasing the clock increases power samely. In other words, E-core and P-core have equal power efficiency as long as they are clocked the same.

On the other hand, the 2xE core occupies only about 65% of the area of the 1xP core, so more cores in the same area and lower clocks can achieve the same MT performance at lower power (this is the technique used in mobile GPUs). The table below shows the cases where E-cores can contribute to power efficiency.
View attachment 285646

Right, so Intel 12 threads equal AMD 6 thread CPUs, glad we finally figured that out
And then Intel proceeds to destroy its wattage lead by powering those P cores up to 11 and they're back at square one, but people can at least believe they bought something with a bigger number on it.

I think this all sums up to a very clear conclusion for CPU land. 'We're done'. Remarkably similar to what's been going with raster perf on GPUs, isn't it. You only get meaningful performance wins if the silicon is tailored to meet those specific needs; ie, V cache for gaming; RT for pointless epeen.

 

[JustBenching]

Alright so let's just dispel the notion that E-Cores are super efficient:

View attachment 285640

As you can see from the chart above, even with only E-Cores enabled in a task they are most effective in E-Cores are actually less efficient than if you enabled just the P-Cores. The only purpose of E-Cores is so that Intel can cram more cores into it's processors.

E-Cores are massively slower than P-cores:

View attachment 285641

The only metric they look good in is performance per mm2.

It's easy to see why people do not want a 6P 16E core CPU.



According to the article, it is the max they will have that gen:

"Apparently, Intel will brazen it out against AMD with a maximum CPU core-count of just 6 performance cores and 16 efficiency cores possible for "Meteor Lake.""

That's a problem, because this is a product that isn't even released yet and the top SKU is poised to loose some performance in games that already utilize more than 6 cores. By the time this processor releases, there will certainly be even more especially as more titles get RT. Then consider how well those 6 cores will cope with future titles as well for what should be a flagship CPU. Intel is going to need to bring an extremely hefty IPC uplift to offset the loss. It kind of also gives AMD a pass to not increase core counts again as well, given that they've been able to match Intel's higher core counts with higher performing cores.

Although I mostly agree with you, ecores are also efficient in terms of performance per watt per mm2. Because sure 1 ecore is vastly less efficient than 1 pcore, but since 1P core takes the same die space as 4e cores, it would make more sense to compare 4e to 1p

 

[hs4]

Right, so Intel 12 threads equal AMD 6 thread CPUs, glad we finally figured that out
And then Intel proceeds to destroy its wattage lead by powering those P cores up to 11 and they're back at square one, but people can at least believe they bought something with a bigger number on it.

I think this all sums up to a very clear conclusion for CPU land. 'We're done'. Remarkably similar to what's been going with raster perf on GPUs, isn't it. You only get meaningful performance wins if the silicon is tailored to meet those specific needs; ie, V cache for gaming; RT for pointless epeen.

I don't know if you misspoke or don't understand even simple arithmetic, but the 6C-12T of Zen3 and the 12T of Alder lake (in any core configuration) are tuned to be basically equal.

For example, in Cinebench, Geekbench, Blender Benchmark, and 7-zip benchmarks, the 1235U (10C12T) and 5600U (6C12T), strictly limited to 15W, shows almost the same MT score, while the 12500H (12C16T) and 5800H (8C16T) show almost the same MT.

P.S.: 12500H can score comparable to 5600X even when turned down to 45W.

Each cases are based on actual report of CPU Package Power:
12500H (12C16T) @ 45W 11124 (GIGABYTE G5 entertainment mode)
12500H (12C16T) @ 95W 14435 (HP OMEN 16 performance mode)
5700X (8C16T) @ 76W 13802 (TDP65W, PPT76W)
5800X (8C16T) @ 130W 15228 (TDP105W, PPT142W)

 

[usiname]

It's not that simple, you conveniently left out the single thread pic. Why isn't the P-core the most efficient also when running single thread?

What kind of limitation does the E-cores have that causes the efficiency to plummet when using multiple cores?

View attachment 285643

E cores are more efficient in 1 thread because the clock speеd is same or close to the frequence when all E cores are under load, while the P core is boosting much higher in single core tasks. Now take into account that the P cores have HT that boost the perfroamcne with ~30% for minimal power conssumption increase. That show very well what trash are the E cores

 

[InVasMani]

Alright so let's just dispel the notion that E-Cores are super efficient:

View attachment 285640

As you can see from the chart above, even with only E-Cores enabled in a task they are most effective in E-Cores are actually less efficient than if you enabled just the P-Cores. The only purpose of E-Cores is so that Intel can cram more cores into it's processors.

E-Cores are massively slower than P-cores:

View attachment 285641

The only metric they look good in is performance per mm2.

It's easy to see why people do not want a 6P 16E core CPU.



According to the article, it is the max they will have that gen:

"Apparently, Intel will brazen it out against AMD with a maximum CPU core-count of just 6 performance cores and 16 efficiency cores possible for "Meteor Lake.""

That's a problem, because this is a product that isn't even released yet and the top SKU is poised to loose some performance in games that already utilize more than 6 cores. By the time this processor releases, there will certainly be even more especially as more titles get RT. Then consider how well those 6 cores will cope with future titles as well for what should be a flagship CPU. Intel is going to need to bring an extremely hefty IPC uplift to offset the loss. It kind of also gives AMD a pass to not increase core counts again as well, given that they've been able to match Intel's higher core counts with higher performing cores.

If you're going to make a comparison about E cores using a chart with the 1st generation Alder Lake E cores rather than Raptor Lake E cores really isn't the best place to start. That's even more true when talking about 3rd generation E cores making even further refinements to E cores designing. It's a bad comparison choosing to look at the older design especially when there were big changes to the design with Raptor Lake for the better like individual E core multipliers and voltage controls instead of clusters of four. Effectively it's higher resolution control over the multiplier setting and voltages which leads to better sloping of parameter settings for efficiency and performance.

Intel made the right changes to E cores with Raptor Lake as a whole it was good step forward. If anything they just pushed P core frequency too damn far at peak. It's been pointed out that it's fairly easy to correct most of the efficiency problem w/o sacrificing overly heavily. To be fair the AM5 7800 vs AM5 7800X is quite a difference as well and shows AMD pushed the 7800X a lot further and could've had a lot better efficiency as well from the get go. Neither company were perfect on that end.

I'm so confused by people's obsession for p-cores and their disdain for e-cores. The e-cores roughly equivalent to skylake cores while taking up way less energy and die space.

1 P-core takes roughly as much space as 4 E-cores, and the E-cores would perform better in pure multi-threaded scenarios.

Given a choice between 10P + 0E cores or 6P +16E cores, there's a very good case to be made for 6+16 config. Games will see negligible impact with 2 fewer P-cores, and multi-threaded apps that can actually use 6+ cores, will see a massive boost because of the extra performance that comes with additional E-cores.

The caveat to all this is that the 6P+16E part can't be a 14900, it must be a 14600-class part, and be priced as such.

I hear you on that people have a unhealthy obsession with the P cores thinking it's going to automatically give them more frame rates more than anything else. Most of the gains in that area are from IPC and additional frequency and Intel's already pushed frequency against the wall repeatedly. There isn't too much further than push it w/o a more serious node shrink and/or other newer fabrication techniques entirely away from silicone. There is one area I think Intel could do with P cores and that's a shared L2 cache between them same as the E cores does currently.

They could've gone as far as a 4 P cores with 20 E cores if they wanted and it still would've made sense. The E core clusters have more L2 cache that they share than a individual P core in-spite of similar die space area. Something Intel could possibly do with P cores Intel is have a shared HT by putting pairs of P cores in clusters. They could also reintroduce AVX-512 on P cores for clusters for every other P core or 1 per cluster. The P cores could likewise have a shared L2 cache. I'm fine with with leveraging more E cores up to a point. Below about 4P cores it starts getting more questionable if it's worth going lower though even 2P cores from Skylake generation are pretty robust at gaming.

 

[BoboOOZ]

If the IPC gain is great, even 6 core Meteor lake could beat the 8 cores Raptor Lake in gaming. If they release it on desktop, it's probably because they think it's worth it. Else why bother?

What worries me is that they also thought worth launching the 11900 & co, what some reviewers deemed like a waste of sand...

 

[Crackong]

Many reviewers conducted E-core only tests using Affinity adjustment or Alder lake-N, including the article "Intel Core i9-12900K E-Cores Only Performance Review" on this site. According to them, when power is supplied as much as possible

The 12900K's 8 E-cores perform
Cinebench ST: 1050, MT:7700 @ 55 W

The 13900K's 16 E-cores with +10% clock performe
Cinebench ST: 1200, MT:18000 @ 120 W

In other words, the 16 E-cores are comparable to the Ryzen 9 3900X in both ST, MT, and has almost same instruction set. If we follow your definition, the 3900X has 0 cores.

The next gen Crestmont is said to have a 10% IPC improvement or, according to MLID, a 25% or more IPC improvement. If MLID's claims are true (which I don't believe), then Meteor lake's 16 E-cores will be Cinebench ST: 1500, MT: 22000, comparable to 5900X. According to your claim, the 5900X would have 0 cores.

Maybe you misunderstood my statement.

I said "E-core are useless in my use cases"
That said.
My use case isn't running Cinebench 24/7

My gaming rig doesn't need E-cores, since everybody and their dog knows E-cores doesn't help in gaming, I rather have 1 or 2 extra P cores to handle so-called "Background tasks".

My productivity rig doesn't need E-cores too, since my productivity rig is a VM bare metal running 40+VMs doing software development, all I need is many and equal cores so I could distribute them across the VMs. The last thing I want are some cut down cores with incomplete instruction set and troubles me in virtualization.

Maybe you do Cinebench 24/7 and E-core meant a lot to you.
But I don't.

I could accept pure P-core or pure E-core CPUs, but not the Hybrids.
Hybrid CPUs in my use cases are just waste of sand.

 

[JustBenching]

My gaming rig doesn't need E-cores, since everybody and their dog knows E-cores doesn't help in gaming

Τhen everybody and their dog is wrong. They do in fact help in gaming.

 

[Assimilator]

since everybody and their dog knows E-cores doesn't help in gaming

Maybe you should stop taking advice from dogs.

 

[Broken Processor]

I honestly don't get this. Are they trying to make room for an extreme range next gen? For gamers this will be considered a 6 core part but doubt it will be priced as such.

 

[hs4]

Maybe you misunderstood my statement.

I said "E-core are useless in my use cases"
That said.
My use case isn't running Cinebench 24/7

My gaming rig doesn't need E-cores, since everybody and their dog knows E-cores doesn't help in gaming, I rather have 1 or 2 extra P cores to handle so-called "Background tasks".

My productivity rig doesn't need E-cores too, since my productivity rig is a VM bare metal running 40+VMs doing software development, all I need is many and equal cores so I could distribute them across the VMs. The last thing I want are some cut down cores with incomplete instruction set and troubles me in virtualization.

Maybe you do Cinebench 24/7 and E-core meant a lot to you.
But I don't.

I could accept pure P-core or pure E-core CPUs, but not the Hybrids.
Hybrid CPUs in my use cases are just waste of sand.

I also write codes that split the whole process into multiple threads for image processing and statistical bootstrapping, from eternal-4-cores era.

Even if all cores are homogeneous, the actual execution speed is not homogeneous because of competing access to shared resources (typically two logical cores sharing a physical core, or memory). The situation is surprisingly serious with respect to memory, to the extent that it is necessary to devise ways to prevent multiple threads from accessing memory addresses in close proximity at the same time.

For this reason, in multi-threaded programming, the process is divided into far more pieces than the number of logical cores, and the smaller pieces are processed sequentially when a logical core becomes free (the pieces in the familiar Cinebench image). This is the kind of information that Google doesn't give you very often, but it's a know-how that anyone with experience knows.

If you think that all P-cores work equally well, you lack experience in dealing with many-cores. Besides, as you can see if you try it, under high load, the clocks of P-cores and E-cores are almost equal, resulting in each thread on P-core 1C2T and E-core 2C2T being almost the same speed. If you need commercially equal cores, buy an EPYC or Xeon.