News Leaked Intel Arrow Lake chipset diagram show more PCIe lanes, no support for DDR4 — new chipset boasts two M.2 SSD ports connected directly to CPU

This is not the first time an Arrow Lake-S chipset diagram, or any platform detail has been shared between distributors, partners, AIB members though. There have been discussions going on before as well, but the info wasn't made public unlike this one.


But anyway, as an example, here is one slide taken from MSI's recent Dragon Shield "Wuhan" event in Asia. Arrow Lake-S "Core Ultra 200" Desktop platform was also teased. The series could feature a brand new DDR5 memory controller, with an updated IMC.


cERlgBx.jpeg




Well, since this article is about "Arrow Lake" CPU lineup, let me point this out here itself. This leak also applies to the upcoming LUNAR LAKE CPU series as well. This is NOT any rumor. But official specs confirmation directly from INTEL.

Anyway, as we know the upcoming 'Arrow and 'Lunar Lake' CPU lineup would be sporting new "Lion Cove" P-core, and "Skymont" E-core architectures, a recent Intel presentation targeted at PC OEMs highlighted one important aspect of the "Skymont" E-cores.

The slides are too small and blurry, and the original media post also got deleted, but we have few details.

"Skymont" E-core is said to offer a double-digit IPC gain over the "Crestmont" E-core powering the current "Meteor Lake" processor lineup. Not much surprising though.

This double-digit IPC gain over "Crestmont" is achieved through an improved branch prediction unit, a broader 9-wide Decode unit compared to the 6-wide Decode unit of "Crestmont," and an 8-wide integer ALU, compared to 4 Integer ALU on its predecessor.

There is also a dependency optimization in the out-of-order engine, and deeper queuing across the engine, just to name a few optimizations.

Skymont E core:

9-wide Decode
8 integer ALU
Two-digit IPC


DOORGuW7SmHtxhHd.jpg


cAACdk0rSThXyUYX.jpg
 

endocine

Honorable
Aug 27, 2018
77
85
10,610
I really wish all the tech companies would normalize and use 10G ethernet. Consumer based ethernet has been stagnant for decades. Every other type of component has doubled many times over since released, except ethernet.
Yeah 2.5G really isn't that great compared to 10G, its gotta be cost still, which after a decade+ is a bit strange. By the time you put in a GPU into the consumer motherboard, there is only one slot left (if you are lucky) and that's gotta handle the 10G NIC.
 

in_the_loop

Distinguished
Dec 15, 2007
165
25
18,710
Yeah but even with more lanes, they'll still keep removing PCI-e slots off of motherboards.
What I get is that there are now only two "extra" PCI-e slots available (X4) besides the normal X16, or is it?
For me considering adding another PCI-e card (and additional PRO audio card for DAW that has a special multicard driver) to my current motherboard with three extra PCI-e slots, where one of the slots can't be used since the grapchis card extends over it.
If it is the same with an upcoming 890 chipset, it will mean that I can't use the extra PCI-e card if I buy it now.
That is, if there are only two extra PCI-e slots and one of them would be hindered by my graphics card...
 
This is not the first time an Arrow Lake-S chipset diagram, or any platform detail has been shared between distributors, partners, AIB members though. There have been discussions going on before as well, but the info wasn't made public unlike this one.

Here are some of the previously leaked slides, although they are a bit old.

These were more detailed and informative than this newly shared slide, which is light on info. It would appear this new 800-series desktop platform will support much higher memory speeds natively with up to 6400 MT/s, in a dual-channel config.

Rumor: Arrow Lake-S (Lion Cove P-Core) - 3 MB L2 Cache Per Core, so that's up to 50% larger L2 cache
  • Arrow Lake-S 8+16 (24 Cores / 32 Threads)
  • Arrow Lake-S 6+16 (22 Cores / 28 Threads)
  • Arrow Lake-S 6+8 (14 Cores / 20 Threads)


aL5NqCA.jpeg



HPY4gE3.jpeg



2mgLj0u.jpeg
 
Last edited:

DS426

Great
May 15, 2024
72
45
60
I really wish all the tech companies would normalize and use 10G ethernet. Consumer based ethernet has been stagnant for decades. Every other type of component has doubled many times over since released, except ethernet.
I fully agree. At a minimum, 2.5G should be the standard on the lowest-cost boards, with 10G being the standard on mid-range and up. Wi-Fi 6 can already have real-world throughput of over 2.5G, and Wi-Fi 6E only pushed that further. Seriously, when did wired networking becoming slower than wireless became acceptable?

I know it's not a sexi selling point unless you can specifically utilize the higher speeds today, and yes, 10G PCIe adapters have been around for awhile, but it shouldn't have to be "add-in" at this point.
 
  • Like
Reactions: usertests

mac_angel

Distinguished
Mar 12, 2008
600
111
19,160
Yeah 2.5G really isn't that great compared to 10G, its gotta be cost still, which after a decade+ is a bit strange. By the time you put in a GPU into the consumer motherboard, there is only one slot left (if you are lucky) and that's gotta handle the 10G NIC.
yea, but cost can't really be used as an excuse. 1Gb Ethernet was released in 1998, over 25 years ago. And most computers, routers, etc are still mostly released with 1Gb. Some odd high end routers may have one or two 2.5Gb or 10Gb, but that's it. If they had kept up with the rest of technology, and kept up with mass production, cost wouldn't be an issue. We now have ISP's offering above 1Gb service, but a huge majority can't utilize those speeds (other than being shared, but realistically, you'd have to have multiple people trying to saturate the Internet all at the same time for it to matter. And even home networking, file transfers from computer to computer or NAS, storage speeds have gone way up that the bottleneck is always going to be ethernet, even at 2.5Gb. Just a good SATA SSD can write faster than those speeds, let alone RAID or m.2 drives.
 
Last edited:
  • Like
Reactions: usertests

in_the_loop

Distinguished
Dec 15, 2007
165
25
18,710
Here are some of the previously leaked slides, although they are a bit old.

These were more detailed and informative than this newly shared slide, which is light on info. It would appear this new 800-series desktop platform will support much higher memory speeds natively with up to 6400 MT/s, in a dual-channel config.

Rumor: Arrow Lake-S (Lion Cove P-Core) - 3 MB L2 Cache Per Core, so that's up to 50% larger L2 cache
  • Arrow Lake-S 8+16 (24 Cores / 32 Threads)
  • Arrow Lake-S 6+16 (22 Cores / 28 Threads)
  • Arrow Lake-S 6+8 (14 Cores / 20 Threads)


aL5NqCA.jpeg



HPY4gE3.jpeg



2mgLj0u.jpeg

Two big surprises here:

1. Hyperthreading still seems to be there.
2. The i7 now gets only 6 P-cores instead of 8 but 16 e-cores instead of 12.

So if hyperthreading is still there, then the totally new architecture will be "hyperinteresting" (no pun intended) with these gains in IPC.
But what about the i7? Isn't it unbalanced with 6 p-cores and the same 16 e-cores ass the i9?
Don't know how to judge it. Against the i5 it gets double the e-cores.
Against the i9 it loses only two p-cores.
 
Last edited:
Hyperthreading still seems to be there.

The i7 now gets only 6 P-cores instead of 8 but 16 e-cores instead of 12.

Actually, the HT stuff is outdated, and since that was old info, I have now omitted the core configurations from my previous post. My bad.

Although, not officially confirmed by Intel yet, Arrow Lake processors won't support HT.

VD0h1MB.png



For some unknown reason, these Gfx drm patch links from Intel are not working now. I took and saved the above screenshot before these went down, almost 2 months ago.

https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10976/bat-arls-2/boot0.txt

https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10968/bat-arls-3/boot0.txt
 
Last edited:
The pleasant surprise here is that ARL is going to have Thunderbolt built in. I'm surprised that Intel. kept the dedicated PCIe 4.0 lanes from the CPU while adding the 4 PCIe 5.0. Hopefully motherboard manufacturers will take advantage of this rather than just letting the lanes go unused.

Looks like the chipset is probably the same as Z690/Z790/W680/W700 which is unfortunate as I was hoping for PCIe 5.0 DMI.
 

in_the_loop

Distinguished
Dec 15, 2007
165
25
18,710
Actually, the HT stuff is outdated, and since that was old info, I have now omitted the core configurations from my previous post. My bad.

Although, not officially confirmed by Intel yet, Arrow Lake processors won't support HT.

VD0h1MB.png



For some unknown reason, these Gfx drm patch links from Intel are not working now. I took and saved the above screenshot before these went down, almost 2 months ago.

https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10976/bat-arls-2/boot0.txt

https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10968/bat-arls-3/boot0.txt
Okay. No HT.
And judging by the number of total cores, it will be the same as with the 14th series:
20 cores should be 8 P+12 E? (and not the strange 6 P+ 16E. Could still be 6P and 14E, if that strange 6P from your previous post had some truth in it, huh? If the strange 14E would be allowed?)
24 cores should be 8P+16E?

HT outdated? Do you mean that the info about it was outdated or that using actual HT is outdated?
HT often can give you an extra 30-32% if multithreading is used efficiently.
With that gone, the new cores must be A LOT faster. Perhaps the E-cores are way over 30% faster compared to last gen. But they can't be that fast (more than, say 40% faster).
Feels like this upcoming gen is going to be a repeat of the 11th gen dud...
 

usertests

Distinguished
Mar 8, 2013
624
579
19,760
I hope that the NPUs/Copilot in desktops catch on with businesses so that previous stock (Alder Lake, etc.) gets dumped onto the refurbished market faster. IIRC the Arrow Lake NPU will be well under 40 TOPS though. You know what, dump Arrow Lake too!

I really wish all the tech companies would normalize and use 10G ethernet. Consumer based ethernet has been stagnant for decades. Every other type of component has doubled many times over since released, except ethernet.
I'll take 2.5G or 5G as the minimum if there are legitimate cost savings (starting with those working on Cat 5e cables). 10G would be great, but we might as well fully transition to the literal stopgap standard first.

If it's only a matter of pennies per device, then yes, move straight to 10G please.

Two big surprises here:

1. Hyperthreading still seems to be there.
2. The i7 now gets only 6 P-cores instead of 8 but 16 e-cores instead of 12.

So if hyperthreading is still there, then the totally new architecture will be "hyperinteresting" (no pun intended) with these gains in IPC.
But what about the i7? Isn't it unbalanced with 6 p-cores and the same 16 e-cores ass the i9?
Don't know how to judge it. Against the i5 it gets double the e-cores.
Against the i9 it loses only two p-cores.
There was talk about a 6+16 configuration in January. Seems weird, but how bad is it really? 6 fast P-cores is probably fine for gaming and 22 total cores is fine for multi-threaded. The picture might become more clear when we learn about the tiles in Arrow Lake. Decisions related to yields that made sense in the past may not apply to Intel desktop CPUs anymore now that they're not using monolithic dies.

On the hyperthreading issue, Intel's Lunar Lake presentation made it clear that they could configure the P-cores to have hyperthreading, but have removed the physical hardware for it in Lunar Lake. It seems like it could be found in Arrow Lake desktop, contradicting many previous rumors, but if what @Metal Messiah. says is true, I guess hyperthreading will be limited to P-core Xeons instead?

Whatever. The most important aspect of Arrow Lake for most folks will be the IPC/single-threaded gain.
 
10Gbe does have a decent amount of disadvantages and I don't think the power consumption one can be fixed. If we could convince everyone SFP was the way to go this would largely negate that. Laptops aren't going to be able to go 10Gbe without some sort of really big change so we're stuck with controllers that tend to cost almost as much as a chipset.

2.5Gbe for the most part covers home usage and from what I understand the 5Gbe controllers use about the same amount of power. I don't know about the transmission power costs though as I haven't seen reviews yet.
 
  • Like
Reactions: usertests
HT often can give you an extra 30-32% if multithreading is used efficiently.
With that gone, the new cores must be A LOT faster. Perhaps the E-cores are way over 30% faster compared to last gen. But they can't be that fast (more than, say 40% faster).
Feels like this upcoming gen is going to be a repeat of the 11th gen dud...
30% on what the p-cores can do, which are a small percentage of the full performance ever since the e-cores are more than the p-cores.

But that's the part that the e-cores can cover if there are more of them or they are just more performant than before. Like if we have 8+16 then the e-cores only have to be 15% faster to cover 30% loss from the p-cores because we have twice the amount of e-cores.

What you are really losing is the ability to run one additional light thread per p-core at the full speed of that p-core. That of course is without knowing how rentable units is going to work out for this kind of thing.
 

in_the_loop

Distinguished
Dec 15, 2007
165
25
18,710
On the hyperthreading issue, Intel's Lunar Lake presentation made it clear that they could configure the P-cores to have hyperthreading, but have removed the physical hardware for it in Lunar Lake. It seems like it could be found in Arrow Lake desktop, contradicting many previous rumors, but if what @Metal Messiah. says is true, I guess hyperthreading will be limited to P-core Xeons instead?

Whatever. The most important aspect of Arrow Lake for most folks will be the IPC/single-threaded gain.

Not for me, at least. I want both good single and multi-threaded performance gains.
My use case is DAW, music creation/production, ie real time rendering with low latency soft synths and effects. Both aspects (single/multi) are important for me in that regard. I also do other things that are non gaming related that eats a lot of CPU power....

(I always find it amusing with people buying the absolute top CPU:s only for gaming. In almost all cases the GPU is what matters, if you don't game at very low res with low settings and a top GPU at the same time in some competitive games with several hundred FPS.)
 

in_the_loop

Distinguished
Dec 15, 2007
165
25
18,710
30% on what the p-cores can do, which are a small percentage of the full performance ever since the e-cores are more than the p-cores.

But that's the part that the e-cores can cover if there are more of them or they are just more performant than before. Like if we have 8+16 then the e-cores only have to be 15% faster to cover 30% loss from the p-cores because we have twice the amount of e-cores.

What you are really losing is the ability to run one additional light thread per p-core at the full speed of that p-core. That of course is without knowing how rentable units is going to work out for this kind of thing.
So the CPU that will suffer the most if hyperthreading is gone is the i5 with only 8 additional e-cores to make up for the 6 lost HT on the P-cores.
They have to be a lot faster to cover for the lost P-core HT that also runs at a higher frequency.
So if hyperthreading is gone, it makes a lot more sense to go for the top tier CPU:s with lots of the newer e-cores...
 

KnightShadey

Reputable
Sep 16, 2020
135
79
4,670
Some odd high end routers may have one or two 2.5Gb or 10Gb, but that's it.
That was the case for 2.5Gb until about 2 years ago (although still the case for 10Gb), although that likely accelerates faster 2.5->10 than the long wait from 1->2.5. The thing was commercial networks needed 1Gb long before home internet (especially in the US for the most part).

Last year D-Link added cheap desktop switches with 2x 2.5Gb+ 5x 1Gb for about $60 (bought one for the parents' Fibre/Tv setup [which has separate 2.5Gb and a 10Gb port on the modem]), and 5x2.5 and 8x2.5Gb for about $100 +/-, even their gaming switch with 1x 10Gb and 5 x 2.5Gb is about $200. So Those are cheaper than most WiFi6E modems, which are now starting to come standard with 2.5Gb ports.

I'll take 2.5G or 5G as the minimum if there are legitimate cost savings (starting with those working on Cat 5e cables). 10G would be great, but we might as well fully transition to the literal stopgap standard first.

If it's only a matter of pennies per device, then yes, move straight to 10G please.

For short runs (under 60ft) Cat5e is fine for 5Gb, and will even do 'OK' with 10Gb at half that, but usually we are talking about last few dozen feet for these routers/switches. For 2.5Gb+, if you're feeding a centralized modem throughout an old house network, it'll struggle past 2.5 beyond 100ft, so you might need to prioritize ports.

Right now, I think you have to work hard to find just Cat5e anymore, almost everything is by default Cat6 for the cheapest and Cat7 in general with Cat8 being the one where you notice the price jump. Are they true to spec, meh, kinda like HDMI/USB/TB/DP compliance, usually good enough, sometimes not, but most will get it done.

As to the price difference, I'd say for most network hardware vendors, 5Gb isn't worth the focus of a spec stop, most devices I see are 2.5Gb then 10Gb in their offerings, and the split is usually 1/2 vs many as for consumer vs commercial.

Where 5Gb is still going to be the mid-way point IMO, will likely be laptop integrated connections, and even that likely goes to 10 Gb in a coupla years. The limiting issue is the form factor support and dislike of external dongles. (edit: although realistically MOST laptops have completely done away with an ethernet port, with just the higher end & gaming/workstation laptops getting 2.5Gbs+ , some of those have started showing up with 5Gbs last fall).
Most Desktop Mobos will likely go straight from 2.5Gb to 10Gb on the high end and stay 2.5Gb in the middle for near term, just because it's so easy to get as an add-in card as a desktop it won't matter..... except for gamers whose spare slots are taken up by GPU blowers, and those mfrs won't want to look like they are lacking when people compare vendors, especially not for Enthusiast/Gaming MoBos that demand a premium price, it's an easy spec bump that might show utility in many ways for that group vs much more limited use in the low/mid-range IMO.

Not everything needs it though, I still have majority of my smart gear and security on quality T10/100 POE switches, because I'd rather pay for reliability over speed if the devices don't need it, just 2 computers, TV, AppleTV and an Extender are on GB+ (Why 4K/120 [even 8K] TVs still have only 10/100 ports ANNOYS me [again pennies per TV], had to put USB dongle on my Sony ! 🤬)

I just wish new network gear was as widely durable as old stuff, now it is a little more hit/miss even with 'known' brands IMO.
 
Last edited:
  • Like
Reactions: usertests
So if hyperthreading is gone, it makes a lot more sense to go for the top tier CPU:s with lots of the newer e-cores...
That purely depends on how much performance you need.
If an i3 covers you then even the i5 is overspending.

Also again we have no idea how rentable units is going to work in practice.

Also we still have no official word on if HTT is really gone on ALL models, they might have some with and some without.
 
  • Like
Reactions: KyaraM

abufrejoval

Reputable
Jun 19, 2020
451
306
5,060
On the 10Gbit topic...

When 10GBase-T was first defined, 10 Watts or more per port were needed to make it work, because pretty much like telephone modems they needed to exploit very complex "analog" modulations to push more data through a copper medium that simply wouldn't support higher frequencies.

10 Watts won't work with a notebook, and I remember the first time I powered on a 48-port 10GBase-T switch in the office... like a jet engine taking off from the desk!

I've always had trouble getting power consumption figures for the full range of NBase-T, with Green Ethernet and other power consumption figures painted in. I guess mostly, because I believe they are dynamically negotiated and thus depend on the end points and the cable between.

Very likely, they will be more exponential than linear between 1/2.5/5/10 Gbit/s supported by NBase-T.

But as a vendor you have little choice but to cater for the extremes to qualify for the standard, even if your average consumption might these days be more around 3 Watts for 10Gbit/s.

Today notebook simply seem to have closed the book on Ethernet, because it's a near endless bother and potentially an energy drain bigger than their SoC.

Perhaps it would have been better if vendors had sold 1/2/5/10 Max-Watt NICs and then say that speed would deepend on your infra... but, yeah that would have been very confusing to most consumers, almost like those WiFi-x numbers.

Quite frankly, 10Gbit/s would have been great to have 10 years ago. Today with NVMe storage rarely slower than 50Gbit/s, that's still a bottleneck even on a laptop.

So to do this right, vendors need to make TB/USB networking work properly. It's near Inifiniband point-to-point, but we need TB/USB switches rather than hubs including power delivery for docked operation.

Too bad there is nobody out there who'd be able to and motivated to push such a thing: a gain here is too often a loss in some other division.
 
  • Like
Reactions: KnightShadey