Intel Raptor Lake with 24 Cores and 32 Threads Demoed

[Dr_b_]

New Z790 chipset

 

[Deleted member 24505]

Considering they're retaining the same CPU socket, it might not be possible. AMD managed to do it though (from PCIe 3.0 to 4.0), so who knows.

I think meteor is going to be LGA1700 too

https://wccftech.com/intel-13th-gen...ile-tsmc-3nm-gpu-tile-tsmc-n5-n4-soc-lp-tile/

 

[TheLostSwede]

I think meteor is going to be LGA1700 too

https://wccftech.com/intel-13th-gen...ile-tsmc-3nm-gpu-tile-tsmc-n5-n4-soc-lp-tile/

That's the rumour, yes.
Doesn't mean they won't make changes to the platform.
How many LGA-115x platforms are there?

 

[dj-electric]

That's the rumour, yes.
Doesn't mean they won't make changes to the platform.
How many LGA-115x platforms are there?

Expect changes in native PCIE5 support for M.2 off the CPU. Hopefully retrospectively to existing hardware, but i have my doubts

 

[WhoDecidedThat]

Again, from a cooling perspective, that may not be possible.

1 P-core has roughly the size of 4 E-cores, so even in theory you can only have at most12 P-cores and then power efficiency for MT tasks would be thrown out of the window.

4 E cores consume more power than 1 P core and give more performance than 1 P core. Adding 16 E cores (v/s adding 4 P cores) means more heat generation, more power consumption and more Multi Threaded performance. The reason why Intel is going with E cores is to get more performance for the same chip size (better MT performance per transistor).

P cores do not have worse power efficiency. They have worse area efficiency. And area efficiency is very important on EUV nodes as EUV nodes are very expensive.

For the same die area.
I would prefer 14+0 instead of 8+16 as a desktop CPU

We should be get a 8 P core + 0 E core chip too. Just like we have a 6 P + 0 E chip this generation.

 

[john_]

Isn't Alder Lake outperforming Zen 3 though? It's not that it's cheap. It's that it works better in many situations.

Those 8 performance cores outperform AMD's Zen 3 cores. Those E cores. They just don't. E cores are not faster than AMD's Zen 3 cores. And considering that future Intel top models will be just increasing the number of E cores, we will be stuck with 8 P cores for years. The E cores will be there to push the number of cores on the CPU box higher, the easy way and also the most profitable way for Intel. Not the best way for the consumer. If AMD follows the same path, we could get stuck with specific number of P core models in the mainstream platform for the next 10+ years(AMD could move backwards cutting max P cores to 8 and adding Zen4c cores like what Intel is doing). With Intel bringing back the HEDT line, it's obvious that limiting the number of P cores in the mainstream platform will only help Intel's and maybe in the future AMD's profit margins, not the consumer who today can have a 16 P core CPU, with a sub $100 motherboard.

 

[Vayra86]

I really get angry at this part of the news post. The author thinks that this new kind of stagnation, where performance cores are staying the same and Marketing cores are increasing, is something exciting.
It's not.

Thinking that the future will bring configurations with 8+16, then 8+24 and then 8+32 cores, where only those 8 cores are performance cores, for me it's not something exciting. It's marketing. A cheap way for Intel to match the number of cores AMD is offering. Not much different than what AMD did in the past, where it was taking 1,5 core and was marketing it as a full dual core module. In both cases we have MARKETING. Feeling excitting about MARKETING, is not something that I like seeing in a technology site like TechPowerUp.

That's a sign of rose tinted glasses, sorry bud.

See... the problem is, we can bounce that ball back to AMD just as well. Let's rewind - was there ANY news item like this at the time of Piledriver/Bulldozer that said 'you get 8 cores that aren't real cores', versus the Intel quad fest? I don't think so. But AMD did build a far worse performing solution, even the Intel Quads it was marketing against (quite explicitly, too) were often faster, especially in single threaded applications.

These are new technologies - both the chiplet approach and getting a fast interconnect to pull them together, and Intel's approach to higher core counts with some bigLittle thing. The technologies are, right now, competing. It remains to be seen what will 'win' in the end, or maybe we'll even get and keep both for longer periods of time. Maybe one technology works for a consumer segment, while another works for enterprise/datacenter?

Maybe Zen will adopt a big little + Chiplet approach one day?
Maybe Intel will do the same?

We can't tell. So its not far fetched to market cores as cores and distinguish them as E-cores. Intel isn't doing much different: they're saying 'these are the configs: P+E and the result is X cores Y threads'. That's as honest as you're going to get.

So is it really just marketing? And if it was, what was Piledriver and Bulldozer then? Those were cores with shared resources just the same, except implemented in a way nobody had a use for.

 

[Richards]

Should dominate the client market with this

 

[john_]

That's a sign of rose tinted glasses, sorry bud.

See... the problem is, we can bounce that ball back to AMD just as well. Let's rewind - was there ANY news item like this at the time of Piledriver/Bulldozer that said 'you get 8 cores that aren't real cores',

Stopped reading there. Why? Because you start with an accusation and at the same time prove that you didn't bothered reading my post entirely. Read the whole post, remove your rose tinted glasses and try again.

I never bought a Bulldozer CPU, with an exception of a short period having an FM2+ APU(bought it for the integrated GPU, not the CPU part), because those Bulldozers where not real 4-6-8 core CPUs. Just marketing.

 

[eidairaman1]

Isn't Alder Lake outperforming Zen 3 though? It's not that it's cheap. It's that it works better in many situations.

Yeah with inefficiency, intel bulldozer?

That's a sign of rose tinted glasses, sorry bud.

See... the problem is, we can bounce that ball back to AMD just as well. Let's rewind - was there ANY news item like this at the time of Piledriver/Bulldozer that said 'you get 8 cores that aren't real cores', versus the Intel quad fest? I don't think so. But AMD did build a far worse performing solution, even the Intel Quads it was marketing against (quite explicitly, too) were often faster, especially in single threaded applications.

These are new technologies - both the chiplet approach and getting a fast interconnect to pull them together, and Intel's approach to higher core counts with some bigLittle thing. The technologies are, right now, competing. It remains to be seen what will 'win' in the end, or maybe we'll even get and keep both for longer periods of time. Maybe one technology works for a consumer segment, while another works for enterprise/datacenter?

Maybe Zen will adopt a big little + Chiplet approach one day?
Maybe Intel will do the same?

We can't tell. So its not far fetched to market cores as cores and distinguish them as E-cores. Intel isn't doing much different: they're saying 'these are the configs: P+E and the result is X cores Y threads'. That's as honest as you're going to get.

So is it really just marketing? And if it was, what was Piledriver and Bulldozer then? Those were cores with shared resources just the same, except implemented in a way nobody had a use for.

Years later under gaming loads for those with AM3 saw a gain in performance compared to quadcore HT in games, what can it be said, physical cores even if they share resources are better than virtual cores.

 

[Cutechri]

For the same die area.
I would prefer 14+0 instead of 8+16 as a desktop CPU

Good luck cooling that

 

[birdie]

4 E cores consume more power than 1 P core and give more performance than 1 P core. Adding 16 E cores (v/s adding 4 P cores) means more heat generation, more power consumption and more Multi Threaded performance. The reason why Intel is going with E cores is to get more performance for the same chip size (better MT performance per transistor).

P cores do not have worse power efficiency. They have worse area efficiency. And area efficiency is very important on EUV nodes as EUV nodes are very expensive.

This is not the case:

Listed in red, in this test, all 8P+8E cores fully loaded (on DDR5), we get a CPU package power of 259 W. The progression from idle to load is steady, although there is a big jump from idle to single core. When one core is loaded, we go from 7 W to 78 W, which is a big 71 W jump. Because this is package power (the output for core power had some issues), this does include firing up the ring, the L3 cache, and the DRAM controller, but even if that makes 20% of the difference, we’re still looking at ~55-60 W enabled for a single core. By comparison, for our single thread SPEC power testing on Linux, we see a more modest 25-30W per core, which we put down to POV-Ray’s instruction density.

By contrast, in green, the E-cores only jump from 5 W to 15 W when a single core is active, and that is the same number as we see on SPEC power testing. Using all the E-cores, at 3.9 GHz, brings the package power up to 48 W total.

4 E-cores are not only significantly faster in MT workloads vs 1 P core, they also consume less power. It was all discussed at least three months ago and people still continue to misunderstand ADL.

 

[Deleted member 24505]

This is not the case:

Listed in red, in this test, all 8P+8E cores fully loaded (on DDR5), we get a CPU package power of 259 W. The progression from idle to load is steady, although there is a big jump from idle to single core. When one core is loaded, we go from 7 W to 78 W, which is a big 71 W jump. Because this is package power (the output for core power had some issues), this does include firing up the ring, the L3 cache, and the DRAM controller, but even if that makes 20% of the difference, we’re still looking at ~55-60 W enabled for a single core. By comparison, for our single thread SPEC power testing on Linux, we see a more modest 25-30W per core, which we put down to POV-Ray’s instruction density.

By contrast, in green, the E-cores only jump from 5 W to 15 W when a single core is active, and that is the same number as we see on SPEC power testing. Using all the E-cores, at 3.9 GHz, brings the package power up to 48 W total.

4 E-cores are not only significantly faster in MT workloads vs 1 P core, they also consume less power. It was all discussed at least three months ago and people still continue to misunderstand ADL.

Or say E cores are useless, fidiots

 

[lexluthermiester]

True, e-core are almost useless for gamers and home office users.

Not true at all. Ecores can be used to do behind the scenes workloads in a game and Office programs can run well exclusively on Ecores. A lot of them together can make for a very powerful platform.

The question is about the sideeffect: half of the die will be e-cores, so cooling might be a problem.

Clearly you understand how Ecores work... /s

 

[Cutechri]

Not true at all. Ecores can be used to do behind the scenes workloads in a game and Office programs can run well exclusively of Ecores. A lot of them together can make for a very powerful platform.

Wow, someone with sense? A rarity nowadays

 

[TheinsanegamerN]

Intel tick-tock model always allow one generational upgrade for tick, no upgrade for tock, this is just last decade on repeat.

If intel suddenly say 14th gen won't require mobo upgrade I will start praising her majesty the Queen Lisa Su though.

No no no shhhh intel always required a different motherboard for every CPU ever made until mommy su blessed us with her presence!

Not true at all. Ecores can be used to do behind the scenes workloads in a game and Office programs can run well exclusively of Ecores. A lot of them together can make for a very powerful platform.

Clearly you understand how Ecores work... /s

The same thing can be done with P cores, and a lot more work can be done on a P core then an E core.

 

[Deleted member 24505]

No no no shhhh intel always required a different motherboard for every CPU ever made until mommy su blessed us with her presence!


The same thing can be done with P cores, and a lot more work can be done on a P core then an E core.

The point is, why run say discord on a P core when it can be run on a low power E core in the background.

 

[WhoDecidedThat]

This is not the case:

Listed in red, in this test, all 8P+8E cores fully loaded (on DDR5), we get a CPU package power of 259 W. The progression from idle to load is steady, although there is a big jump from idle to single core. When one core is loaded, we go from 7 W to 78 W, which is a big 71 W jump. Because this is package power (the output for core power had some issues), this does include firing up the ring, the L3 cache, and the DRAM controller, but even if that makes 20% of the difference, we’re still looking at ~55-60 W enabled for a single core. By comparison, for our single thread SPEC power testing on Linux, we see a more modest 25-30W per core, which we put down to POV-Ray’s instruction density.

By contrast, in green, the E-cores only jump from 5 W to 15 W when a single core is active, and that is the same number as we see on SPEC power testing. Using all the E-cores, at 3.9 GHz, brings the package power up to 48 W total.

4 E-cores are not only significantly faster in MT workloads vs 1 P core, they also consume less power. It was all discussed at least three months ago and people still continue to misunderstand ADL.

I stand corrected. Thanks for the article. I did some additional napkin math.

8 P cores = 239 watts, 1 P core = 78 watts => Difference is 161 watts -> Divide this by 7 = 23 watts per core
8 E cores = 48 watts, 1 E core = 15 watts => Difference is 33 watts -> Divide by 7 => 4.7 watts per core

That is a massive difference in terms of power consumption.

 

[lexluthermiester]

The same thing can be done with P cores, and a lot more work can be done on a P core then an E core.

While true, Pcores use WAY more more electricity and even mutlti-threaded, Pcores can, in some workloads, get less done than a bunch of non-multi-threaded Ecores.

I stand corrected. Thanks for the article. I did some additional napkin math.

8 P cores = 239 watts, 1 P core = 78 watts => Difference is 161 watts -> Divide this by 7 = 23 watts per core
8 E cores = 48 watts, 1 E core = 15 watts => Difference is 33 watts -> Divide by 7 => 4.7 watts per core

That is a massive difference in terms of power consumption.

Exactly.

 

[Minus Infinity]

Maybe Zen will adopt a big little + Chiplet approach one day?
Maybe Intel will do the same?

What do mean maybe. Zen 5 is confirmed as big,little design. The little cores will be Zen 4c cores at this stage and they will destroy Gracemont+++ cores. So come late 2023 early 2024 Zen 5 vs Meteor Lake will be very interesting. AMD is also possibly releasing all Zen 4c core CPU's with higher core counts for those that require huge MT performance. 4c cores will strip away cache and not have the IP uplifts of regular 4 cores, but should still be stronger than Zen 3 cores. A 24/32 core Zen 4c will be a MT beast.

 

[thestryker6]

I'm not sure what people who are complaining about E-cores are doing with their desktop systems that ADL isn't good for, but HEDT is too expensive for. Ideally Intel will start HEDT processors at 10/12c+ which will allow HEDT to be affordable again rather than blurring the lines and screwing over anyone who needs PCIe lanes like AMD has done.

 

[trieste15]

I really get angry at this part of the news post. The author thinks that this new kind of stagnation, where performance cores are staying the same and Marketing cores are increasing, is something exciting.
It's not.

Thinking that the future will bring configurations with 8+16, then 8+24 and then 8+32 cores, where only those 8 cores are performance cores, for me it's not something exciting. It's marketing. A cheap way for Intel to match the number of cores AMD is offering. Not much different than what AMD did in the past, where it was taking 1,5 core and was marketing it as a full dual core module. In both cases we have MARKETING. Feeling excitting about MARKETING, is not something that I like seeing in a technology site like TechPowerUp.

This is American Capitalism at its worst.

 

[lexluthermiester]

This is American Capitalism at its worst.

Opinion not supported by merit.

 

[Jhart1228]

This is American Capitalism at its worst.

It's not quite creating Banana Republics or funding the Nazi war machine, but creating E cores comes close.

 

[john_]

I'm not sure what people who are complaining about E-cores are doing with their desktop systems that ADL isn't good for, but HEDT is too expensive for. Ideally Intel will start HEDT processors at 10/12c+ which will allow HEDT to be affordable again rather than blurring the lines and screwing over anyone who needs PCIe lanes like AMD has done.

Really? Because Intel's HEDT line was cheap before the first series of Ryzen processors from AMD, right? And I am also sure that Intel will start from 10/12c+ models, because there where never quad core HEDT models on Intel's HEDT platform, correct? Also Intel never uses CPU features for market segmentation. They where the good guys before AMD, who "screw over" people, by starting offering them up to 64 cores and 128 PCIe lanes in the HEDT line, with all features enabled from top to the bottom model.
No we have now to be thankful to Intel for blurring the lines between Atom CPUs and what we knew as Core CPUs until recently. We should be thankful to Intel because in the future we will be getting CPUs with more E cores than P cores. We shouldn't be complaining because what the hell are we doing with our desktops anyway? Why pay $500-700 for 16 P cores when we can pay the same and get 8 P and 8 E cores? Why pay today $800-$1000 for 24 P cores, when we can pay in the future the same for 8 P and 16 E cores? Why pay $1500 for 32 P cores, when we can pay the same in the future for 8 P and 24 E cores?

Intel, AMD, Nvidia, they are not football teams. Their gains are not always something to cheer for. Finding ways to maximize their profits, it's good, but not always something to make us happy. AMD will follow with Zen4c cores and we will end up paying for performance cores and only getting a percentage of performance cores in the final product.