News Intel Panther Lake and Wildcat Lake CPU specs break cover — leak suggests up to 16 CPU cores and 180 total AI TOPS

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Wild Cat sounds very interesting to me. I currently use a home-made NAS on the basis of a Raspi 4, it's been running for 4 years now. I would like to use some Docker containers in the future, which currently isn't going very well. So I was looking at the N100, since boards with that chip are pretty cheap where I live, at 96€. Couple that with 16 - 32 GB RAM and you got a nice, inexpensive base for a NAS.

Wild Cat would be a lot more expensive at the start is my guess, the N100 certainly was, but would make for a good future upgrade, and maybe even lower N100 prices further. Though I would get the N100 rather sooner than later...
I'm not sure what all is available options wise where you are but there are some very good N305 SBCs available. That would give you a little more flexibility with regards to what all you can easily run in parallel. If you needed more single threaded performance or more GPU capability then you might be able to find something inexpensive with an 8505.

Nothing like this was available yet when I replaced my server box, but if it had I might have sacrificed overall performance for simplicity and low power: https://www.aliexpress.com/item/3256807101632310.html
 
  • Like
Reactions: KyaraM
Wild Cat sounds very interesting to me. I currently use a home-made NAS on the basis of a Raspi 4, it's been running for 4 years now. I would like to use some Docker containers in the future, which currently isn't going very well. So I was looking at the N100, since boards with that chip are pretty cheap where I live, at 96€. Couple that with 16 - 32 GB RAM and you got a nice, inexpensive base for a NAS.
My advice would be to look at ODROID-H4 Plus. It's based on the higher-performance N97, has an inexpensive ITX adapter kit, uses a regular DDR5 SODIMM (so you can use up to 48 GB, currently), and can run passively.


If you need neither SATA nor a second 2.5G Ethernet port, the regular H4 would do.

It has two features you don't always find in these products. One is a full x4 NVMe slot @ PCIe 3.0 (although, if you need to use the SATA ports on the Plus, it drops to just x2).

More importantly, their latest BIOS enables in-band ECC. This gives you data integrity, which is important for something like a NAS. It dings performance slightly, but not enough that you'll probably notice. The best part is that it works with any RAM you use, due to the way it works. It's described, here:

I have only seen the in-band ECC option exposed on 2 or 3 other Alder Lake-N boards, and they tend to be the higher-end, more industrial-focused models.

Another feature it has is "Unlimited Performance" mode, which allows it to run at full boost speeds, without a time limit. The web page shows what kind of performance benefit this provides. You'll need active cooling, if you're going to use this feature.

At low load, it's also very power-efficient. I bought a N97 board from a different manufacturer, just weeks before the ODROID-H4 came out, and mine won't idle below 7W (headless). They claim theirs just 2.0 to 2.8 W, in the same configuration, depending on the model.

Lastly, HardKernel (who make ODROID) is based in Korea, not China. I've ordered from them, directly. They have forums with a pretty good user community and have been in the business of making small computer for more than 15 years.

I'm not sure what all is available options wise where you are but there are some very good N305 SBCs available.
The ODROID-H4 Ultra is also based on the N305.

Nothing like this was available yet when I replaced my server box, but if it had I might have sacrificed overall performance for simplicity and low power: https://www.aliexpress.com/item/3256807101632310.html
Although it has dual M.2 slots, I'm sure they're not x4. In fact, I'd be surprised if they weren't just connected at x1! That 10 GbE controller is going to need probably two lanes all to itself!
 
Last edited:
Although it has dual M.2 slots, I'm sure they're not x4. In fact, I'd be surprised if they weren't just connected at x1! That 10 GbE controller is going to need probably two lanes all to itself!
They're definitely PCIe 3.0 x1 as I don't think I've seen any of these boards using anything else. I can't imagine why anyone would care about x4 on a device like this though.

I'm not 100% sure how they're doing the connectivity on the 10gb since the only detailed layouts I've seen are on the 4x 2.5gb boards. If they're giving it full bandwidth it would definitely need two lanes though.
The ODROID-H4 Ultra is also based on the N305.
Yeah that's a really good choice if you want something with better support as the connectivity loss isn't too bad.
 
  • Like
Reactions: KyaraM and bit_user
I'm not sure what all is available options wise where you are but there are some very good N305 SBCs available. That would give you a little more flexibility with regards to what all you can easily run in parallel. If you needed more single threaded performance or more GPU capability then you might be able to find something inexpensive with an 8505.

Nothing like this was available yet when I replaced my server box, but if it had I might have sacrificed overall performance for simplicity and low power: https://www.aliexpress.com/item/3256807101632310.html

My advice would be to look at ODROID-H4 Plus. It's based on the higher-performance N97, has an inexpensive ITX adapter kit, uses a regular DDR5 SODIMM (so you can use up to 48 GB, currently), and can run passively.

If you need neither SATA nor a second 2.5G Ethernet port, the regular H4 would do.

It has two features you don't always find in these products. One is a full x4 NVMe slot @ PCIe 3.0 (although, if you need to use the SATA ports on the Plus, it drops to just x2).

More importantly, their latest BIOS enables in-band ECC. This gives you data integrity, which is important for something like a NAS. It dings performance slightly, but not enough that you'll probably notice. The best part is that it works with any RAM you use, due to the way it works. It's described, here:

I have only seen the in-band ECC option exposed on 2 or 3 other Alder Lake-N boards, and they tend to be the higher-end, more industrial-focused models.

Another feature it has is "Unlimited Performance" mode, which allows it to run at full boost speeds, without a time limit. The web page shows what kind of performance benefit this provides. You'll need active cooling, if you're going to use this feature.

At low load, it's also very power-efficient. I bought a N97 board from a different manufacturer, just weeks before the ODROID-H4 came out, and mine won't idle below 7W (headless). They claim theirs just 2.0 to 2.8 W, in the same configuration, depending on the model.

Lastly, HardKernel (who make ODROID) is based in Korea, not China. I've ordered from them, directly. They have forums with a pretty good user community and have been in the business of making small computer for more than 15 years.


The ODROID-H4 Ultra is also based on the N305.


Although it has dual M.2 slots, I'm sure they're not x4. In fact, I'd be surprised if they weren't just connected at x1! That 10 GbE controller is going to need probably two lanes all to itself!
Thanks, you two! Will definitely look at those options. The ECC-mode on the ODROID-board does sound very interesting indeed.
 
  • Like
Reactions: bit_user
Thanks, you two! Will definitely look at those options. The ECC-mode on the ODROID-board does sound very interesting indeed.
It seems they haven't updated their product pages to mention it, but here's the forum post where they stated that support for it was included in BIOS version 1.02 for the H4 series:

It's also mentioned in the description of the 1.02 BIOS release:
 
I've got a question… maybe someone can answer or correct me. Wildcat Lake will allegedly come with Darkmont E-cores as LPE, probably with 4 cores, in addition to 2 (or 1?) P-cores. In the news it is often stated that Wildcat Lake could be the successor to Intel's N-series CPUs. Maybe. Maybe not. The latter are often used for DIY low-power home servers, which mostly works well, but just barely because Gracemont offers only 9 PCIe 3.0 lanes, two of which are for a maximum of 4 SATA ports (see diagram below). For 4 more SATA ports, you'd need a chipset like the ASM1164, which is PCIe 3.0 x2, leaving you with 5 remaining PCIe 3.0 lanes, e.g. 2 for one full-speed 10GbE port, 2 other for two x1 M.2 NVMe slots, and the last one for a PCIe 3.0 x1 slot. Now, Wildcat Lake is supposed to come with PCIe 4.0, but only with 6 lanes. From a pure perspective of bandwidth, that would be better, equivalent to 12 PCIe 3.0 lanes. But afaik there are no PCIe 4.0 SATA controllers, e.g. PCIe 4.0 x1 to 4 x SATA. So I don't see how Wildcat Lake will be a good thing for the DIY low-power NAS community. It could actually be a detriment. On the other hand, maybe Darkmont will also have a couple of PCIe 3.0 lanes in addition to the 6 PCIe 4.0 lanes. Anyone who knows more or can comment/correct?

intel-soc-adl-n.png
 
Last edited:
  • Like
Reactions: bit_user
Wildcat Lake is supposed to come with PCIe 4.0, but only with 6 lanes. From a pure perspective of bandwidth, that would be better, equivalent to 12 PCIe 3.0 lanes. But afaik there are no PCIe 4.0 SATA controllers, e.g. PCIe 4.0 x1 to 4 x SATA. So I don't see how Wildcat Lake will be a good thing for the DIY low-power NAS community.
Yeah, I agree. Unless you can pair it with a chipset that embeds a PCIe switch, that sounds too low.

On the other hand, maybe Darkmont will also have a couple of PCIe 3.0 lanes in addition to the 6 PCIe 4.0 lanes.
I assume the lane-count you mention is for the entire package. So, the only way you could turn that into more lanes would be by adding a switch outside of the CPU, except that adds cost and power and is usually not done on laptops.
 
  • Like
Reactions: JossBrown
I assume the lane-count you mention is for the entire package. So, the only way you could turn that into more lanes would be by adding a switch outside of the CPU, except that adds cost and power and is usually not done on laptops.
As for a gen4to3 switch, yes: manufacturers of those popular low-power NAS motherboards wouldn't add such a switch either. (For networking it would be great, though, because afaik there are PCIe 4.0 x1 to 10GbE chipsets: some newer NICs use those… ACQ?) As for the lane count: my hope is that the alleged 6 PCIe 4.0 lanes would only come from the P-core(s). Why would I think that? On the N-series the whole package was just efficiency cores, just Gracemont, and all the PCIe lanes came from "behind" the OPIO ("on package input/output"). So maybe there could be some additional PCIe 3.0 lanes from the Darkmont LPE cores, though I'm not sure if that's possible… but maybe from a chipset on package, like Intel did with the U300 (see diagram). My guess is that (if true) it probably couldn't be 9 or even 10, but 4 to 6 additional PCIe 3.0 lanes would suffice for 8 SATA, WiFi and maybe an additional PCIe 3.0 x1 slot. But all of that is just me hoping, and of course I don't know, if that's the way they're going with Wildcat Lake, or if it's even technically possible.

intel-mobile-rpl-hpu.png
 
my hope is that the alleged 6 PCIe 4.0 lanes would only come from the P-core(s). Why would I think that? On the N-series the whole package was just efficiency cores, just Gracemont, and all the PCIe lanes came from "behind" the OPIO ("on package input/output").
Ah, nice diagram. I found another one, here:


I guess one thing we can hope for is that they'll make an alternate South Bridge chip for Wildcat Lake, with more PCIe connectivity.

Anyway, as you can see from these diagrams, PCIe isn't tied to the CPU cores, themselves. This annotated die shot of the desktop Alder Lake shows that more clearly:
The PCIe controller is contained in the block labelled "Uncore" (also sometimes called the system agent).

The reasons why the CPUs with P-cores so far have more PCIe lanes are for cost control and market segmentation. Keep in mind that more PCIe lanes means more pins and a larger package. So, it's not only the die size that increases, but it also has downstream costs for systems that might not even need more PCIe lanes.

all of that is just me hoping, and of course I don't know, if that's the way they're going with Wildcat Lake, or if it's even technically possible.
Yeah, I think there's nothing we can do but wait and see. Let's hope Intel doesn't turn its back on the NAS market. If they do, I predict some ARM chips will fill the void, or maybe AMD's embedded APUs.
 
  • Like
Reactions: JossBrown
Let's hope Intel doesn't turn its back on the NAS market. If they do, I predict some ARM chips will fill the void, or maybe AMD's embedded APUs.
Just a couple of days ago I found out about the Radxa ARM boards. Almost there… almost there! 👍 As a dyed-in-the-wool Mac user, I'd love some good ARM options. (I could always go the Mac Mini route, of course, but that would be a real janky NAS project. lol)

And thank you: great info!
 
  • Like
Reactions: bit_user
LPDDR5-8533 equates to theoretical bandwidth of 136.5 GB/s. Do you think it'll be enough?

For graphics, I guess maybe, if they lean into that "side cache", introduced in Lunar Lake. But, it would need to scale like AMD did with Infinity Cache.

Considering its 180 TOPS, that works out to 1318 ops per byte of memory bandwidth. I think it won't be enough for that.

For comparison, consider that a RTX 4060 Ti has 177 TOPS and 288 GB/s of memory bandwidth. That's about twice the bandwidth per
LPDDR5-8533 equates to theoretical bandwidth of 136.5 GB/s. Do you think it'll be enough?

For graphics, I guess maybe, if they lean into that "side cache", introduced in Lunar Lake. But, it would need to scale like AMD did with Infinity Cache.

Considering its 180 TOPS, that works out to 1318 ops per byte of memory bandwidth. I think it won't be enough for that.

For comparison, consider that a RTX 4060 Ti has 177 TOPS and 288 GB/s of memory bandwidth. That's about twice the bandwidth per op.
Panther lake Will have 273gb/s memory bandwidth not 136.5gb/s previous gen lunar lake igpu have 256 bus width and 8533mhz ram which equate to 273gb/s memory bandwidth, 256x8,533÷8=273 you assume that pantherlake only have 128 bus width thats your mistake IM here to give you good news that you're falsely calculate the memory bandwidth and i'd say 273gb/s is more than enough 4060 only have 272gb/s btw
 
Panther lake Will have 273gb/s memory bandwidth not 136.5gb/s
Well, Intel said it's not going to have on-package memory, so if it ends up with that much memory bandwidth, I'll certainly be surprised. Got a source on that?

previous gen lunar lake igpu have 256 bus width and 8533mhz ram which equate to 273gb/s memory bandwidth,
Are you talking about its speed in accessing the 8 MB "Side Cache"? Because Lunar Lake definitely doesn't have that much bandwidth to its LPDDR5X.

256x8,533÷8=273 you assume that pantherlake only have 128 bus width thats your mistake
Yeah, um no. Intel, themselves, say it has only "2 channels", by which I think they mean 2x 64-bit = 128. Here's the top spec model: