News Intel Processor N-series CPU specifications leaked — same old architecture with paltry clock speed improvements

[Admin]

Intel is preparing to refresh its Alder Lake-N series of processors, but it appears that the company will reutilize the same architecture with minor bumps in frequency.

Intel Processor N-series CPU specifications leaked — same old architecture with paltry clock speed improvements : Read more

 

[Jame5]

It has been a while since Intel abandoned its long-adored and nostalgic Celeron and Pentium brandings

Pentium being adored/nostalgic maybe. Celeron has always been synonymous with junk bargain bin low performance CPUs.

 

[thestryker]

Pentium being adored/nostalgic maybe. Celeron has always been synonymous with junk bargain bin low performance CPUs.

The generation who grew up with the Celeron debut would beg to differ as the 300A was unprecedented at the time. Intel made sure to never repeat that mistake again though.

I believe the Netburst Celerons were also OC frequency record holders for a long time (not that this made them good).

 

[thestryker]

It's looking like Intel might not update the low power CPUs until it comes time to increase Atom performance. Given that they never shifted to the increased cache from RPL Gracemont and the minimal performance uplift from Crestmont it seems logical the next ones would be based on Skymont/Darkmont. There's no chance Intel would spend the money required to produce them on TSMC N3 which would likely mean waiting for Intel 3 capacity. My bet is that we see a new architecture for these lower performance/power parts no earlier than late 2025, but most likely 2026 sometime.

 

[Notton]

Honestly, the specs that were leaked are inconsequential.
The important ones were memory bandwidth, iGPU configuration, and idle power usage.

I'm guessing it still uses the same 1-ch DDR4 + 1-ch DDR5/LPDDR5 configuration from before
And I'd be surprised if the iGPU got any bump from 24EU, but it'd be nice for confirmation.

N100 struggles when rendering 4K video or running 3 monitors.

 

[cyrusfox]

Honestly, the specs that were leaked are inconsequential.
The important ones were memory bandwidth, iGPU configuration, and idle power usage.

I'm guessing it still uses the same 1-ch DDR4 + 1-ch DDR5/LPDDR5 configuration from before
And I'd be surprised if the iGPU got any bump from 24EU, but it'd be nice for confirmation.

N100 struggles when rendering 4K video or running 3 monitors.

This is exactly the issue, the n line of Intel CPUs are too cut down on memory bandwidth and power throttled to perform as well as they are capable of. Also the pricing of the higher sku is ridiculous to even consider, the n97/n100 are ubiquitous on the low end and for a media consumption/internet machine, more than capable and very efficient.

All I see here is a minor clock bump refresh, unless pricing of the higher end skus come down to reasonable levels, and/or tdp is tunable, N150 will be the new budget king to fight against the prior offerings, but the performance will still be equivalent to quad core skylake of yore, sufficient but stuttering at times.

These chips could fly if their wings were not clipped by starving the memory bandwidth and the self throttling the iGPU is hamstrung by the tight TDP limits imposed. Even the higher EU iGPU models perform similarly on the N line, the TDP is the key setting that determines how well the iGPU can perform, Only when it has enough power budget do the higher EU models pull ahead. I have seen many reviews of the n95 with its 16EU besting the N200 with its 32EU (15W TDP vs 6W TDP). On the 6W TDP models, CPU is capable enough and performs well but iGPU will not be able to perform as well as it could if given only a few watts more. These aren't ending up in tablets, so the 6W TDP is ridiculous, 15W is where these all shine, with diminishing returns once you go past 25/30W, I believe Intel wants to keep them in the budget section and not have them cannibalize its low end mobile offerings, so I believe this is intentional handicapping.

 

[abufrejoval]

Atoms don't make money. Not for Intel and not for the OEMs.

There have been stories just how much OEMs hated Intel shoving them down their throats in "package deals" they couldn't refuse...

...back then, when Intel dominated the PC market and was bent very hard on banning ARM from invading "their" territory from the low end.

And things aren't getting any better because cores scale down with die shrinks, nearly everything else in an Atom doesn't: they need a full set of connectivity and that won' shrink.

So Intel could do chips with lots of Atom cores, roughly four E-cores for every P. But those would poison the entry server market, which is already pretty much disappearing anyway via the clouds. And gamers would only buy them for their NAS.

16 or 32 core Atoms for the price of an entry level notebook chip, would be kind of attractive for "edge" use cases such as home-infra µ-servers. But investing engineering resources into such a product makes ever less sense for Intel, because nobody wants to pay a lot of money for that.

Snow Ridge Atom-P SoCs are made by Intel for niches like that, but very pricey and very niche with slow refreshes if any.

If you're looking for something economical there, you're probably better off getting an older generation Zen APU, mobile and soldered to some ITX or NUC board or even in a socket with cheap mainboard: there is tons of really cheap deals from China, where these systems evidently flourish also on the local market. Some of these include lots of SATA and Ethernet ports so you can combine them for NAS, firewall and VM server, while they still use little power and are way more economical than what Intel offers new.

 

[thestryker]

Also the pricing of the higher sku is ridiculous to even consider

The publicly available pricing and what is on Ark is not accurate. I don't know what the benefit of this obfuscation is other than maybe giving OEMs cover for higher prices. I got a passively cooled router box with an N305 and 6x 2.5gb NICs for around $275 shortly after the CPU became available. This was about $80 more than the same thing with N100 at the time.

These chips could fly if their wings were not clipped by starving the memory bandwidth and the self throttling the iGPU is hamstrung by the tight TDP limits imposed.

From my experience the TDP limit is the big problem so long as DDR5 is being used. The CPU boost clocks are so high that they'll easily use up all of the available power and prevent the IGP from boosting. The N97 is definitely the best 4 core model even with the lower CPU boost clock, assuming you can cool it, due to the 12W TDP and higher IGP boost (it has 24 EUs). The N95 IGP boost clock is the same as the N97 which is what lets it compete with the N200 despite half the EUs. This is also what makes the N300 a bad joke compared to the N305 (N305 is definitely the best ADL-N and I assume the N355 will supplant this based on clocks).

 

[usertests]

There's no chance Intel would spend the money required to produce them on TSMC N3 which would likely mean waiting for Intel 3 capacity. My bet is that we see a new architecture for these lower performance/power parts no earlier than late 2025, but most likely 2026 sometime.

You nailed it. They are a product that can exist as long as they use a cheap Intel node. Crestmont is of no consequence, Skymont is a huge leap.

If we saw Atom die off, and lower mobile SKUs (Meteor Lake-U equivalent) migrate down in price to take its place, that could be good too. Alder/Raptor Lake mobile is being refreshed yet again. An Intel Core 250U gives you 2+8 cores, dual-channel memory, and 96 execution units instead of 32. I don't think these dies are even much larger than the Alder Lake-N die.

 

[watzupken]

I feel this is a product that don’t sell well and so don’t get attention or resources to improve it. The key benefits are its low power requirement and AV1 support, but I don’t find the price competitive for the performance it offers. Even cheaper AMD SOC like the 5600u will run circles around it when it comes to CPU and GPU performance.

 

[thestryker]

If we saw Atom die off, and lower mobile SKUs (Meteor Lake-U equivalent) migrate down in price to take its place, that could be good too.

I think it'd actually be more likely that the 2P/8E silicon died off and was replaced with Skymont or better E-core only parts. I don't really think either is particularly likely at this point, but Intel has not used hybrid for any of their enterprise SKUs yet and Atoms very much are enterprise focused.

An Intel Core 250U gives you 2+8 cores, dual-channel memory, and 96 execution units instead of 32. I don't think these dies are even much larger than the Alder Lake-N die.

They actually are significantly larger even though they don't share a die with the larger mobile CPUs as they also have over double the PCIe lanes and Thunderbolt integrated too. The ADL-N CPUs are all based on the 8 E-core 32 EU die, but even then I'd be very surprised if they broke half the size.

 

[Kondamin]

Can’t they just use what ever capacity they have on intel 3 to make some e core chips

 

[usertests]

They actually are significantly larger even though they don't share a die with the larger mobile CPUs as they also have over double the PCIe lanes and Thunderbolt integrated too. The ADL-N CPUs are all based on the 8 E-core 32 EU die, but even then I'd be very surprised if they broke half the size.

The only discussion I've ever seen of Alder Lake-N die size is this estimate:
https://locuza.substack.com/p/info-snack-alder-lake-m-raptor-lake

121.72mm^2 for Alder Lake-N, 161.48mm^2 for Alder Lake-M (2+8+96).

If that's accurate, +32% is significant but not a massive difference for what you get. And there are more interesting ways to cut down a 2+8 chip. Feel free to supply a better die size if you have one.

Good point on enterprise demanding non-hybrid. Maybe E-cores will go away within a few years instead of a hybrid Atom being made.

Can’t they just use what ever capacity they have on intel 3 to make some e core chips

Is there any profit in doing so? Alder Lake-N is practically given away for free (although no price is too low for a miser).

 

[thestryker]

The only discussion I've ever seen of Alder Lake-N die size is this estimate:
https://locuza.substack.com/p/info-snack-alder-lake-m-raptor-lake

121.72mm^2 for Alder Lake-N, 161.48mm^2 for Alder Lake-M (2+8+96).

If that's accurate, +32% is significant but not a massive difference for what you get. And there are more interesting ways to cut down a 2+8 chip. Feel free to supply a better die size if you have one.

Let's just say that those measurements are accurate (nobody has measured the N0 die that I could find despite them being bare) that means ~34% more die per wafer (assuming equal yield) on a low margin part. Anything released in this segment has to compete on price first so using a larger die doesn't really make sense. Customers might want the higher performance (and the Pentium 8505 is basically a steal right now in the minipc boxes), but it doesn't make sense for Intel as a company.

 

[Kondamin]

The only discussion I've ever seen of Alder Lake-N die size is this estimate:
https://locuza.substack.com/p/info-snack-alder-lake-m-raptor-lake

121.72mm^2 for Alder Lake-N, 161.48mm^2 for Alder Lake-M (2+8+96).

If that's accurate, +32% is significant but not a massive difference for what you get. And there are more interesting ways to cut down a 2+8 chip. Feel free to supply a better die size if you have one.

Good point on enterprise demanding non-hybrid. Maybe E-cores will go away within a few years instead of a hybrid Atom being made.


Is there any profit in doing so? Alder Lake-N is practically given away for free (although no price is too low for a miser).

The fabs are there, it would be a small chip so yields should be OK.
Some money > No money

 

[das_stig]

Are Intel now incapable of being innovative, copy and repeat with minor bump is all they seem to do now.
When you looked back between generations, you could see advancement, now just a case of flog our old, tired tech, while we hope for a miracle design to turn us back around, aka Prescott to Core style.

 

[Kondamin]

Are Intel now incapable of being innovative, copy and repeat with minor bump is all they seem to do now.
When you looked back between generations, you could see advancement, now just a case of flog our old, tired tech, while we hope for a miracle design to turn us back around, aka Prescott to Core style.

Euhm, no.

Intel has introduced drastically new cores for desktop Gen 11 back in 2021, brought big little to the PC in 2022 started with tile based chips for laptops in 2023 and introduce totally reworked cores again in 2024.

It's not panning out that much is true, but you can't say they aren't being innovative.
AMD put fire under their buts and they are and have been working hard .

There is a company that has done very little the last couple of years, but it's not Intel or AMD

 

[bit_user]

the minimal performance uplift from Crestmont it seems logical the next ones would be based on Skymont/Darkmont. There's no chance Intel would spend the money required to produce them on TSMC N3 which would likely mean waiting for Intel 3 capacity.

I think the efficiency benefits of Intel 3 would make for a worthwhile product iteration with Crestmont. I'm not sure Intel would bother backporting Skymont to Intel 3, however.

The real trouble could be graphics, since Intel hasn't done any sort of GPU on any intel node below Intel 7.

 

[thestryker]

I think the efficiency benefits of Intel 3 would make for a worthwhile product iteration with Crestmont.

While this is no doubt true I doubt Intel particularly cares on the Atom side. They've been pretty consistent on releasing new Atom products every couple of years which would be 2026 and I can't imagine them using Crestmont at that point.

I'm not sure Intel would bother backporting Skymont to Intel 3, however.

The real trouble could be graphics, since Intel hasn't done any sort of GPU on any intel node below Intel 7.

Given that Intel is using industry standard tools for all of their EUV nodes neither should be particularly problematic. The more pressing issue should be whether or not there was capacity available for low margin parts.

 

[Pemalite]

Pentium being adored/nostalgic maybe. Celeron has always been synonymous with junk bargain bin low performance CPUs.

*wipes away tears from the Celeron 300A* It's okay, he didn't really mean it.

 

[bit_user]

While this is no doubt true I doubt Intel particularly cares on the Atom side. They've been pretty consistent on releasing new Atom products every couple of years which would be 2026 and I can't imagine them using Crestmont at that point.

Couple of years? So, you're saying this refresh lineup resets the clock, then?

Given that Intel is using industry standard tools for all of their EUV nodes neither should be particularly problematic.

I guess, if you don't really care what frequency it hits, then you might not need to do too much to port it to a different node. If you want a uarch to run optimally on a given node, then there's still some non-trivial effort involved.

 

[thestryker]

Couple of years? So, you're saying this refresh lineup resets the clock, then?

They didn't release an 8 core Atom part until this year when they finished up the SKU lineup (who knows why it took so long) so I was basing it on that.

I guess, if you don't really care what frequency it hits, then you might not need to do too much to port it to a different node. If you want a uarch to run optimally on a given node, then there's still some non-trivial effort involved.

It's also nowhere near the trouble it would have been when they were using proprietary tools for each node. Given that Intel 3 is a long term node it makes infinitely more sense to port Skymont and reap the design benefits than to use Crestmont just to gain efficiency. The exception being if the fabs would be idle as I assume Crestmont would be much faster time to manufacture.

 

[bit_user]

They didn't release an 8 core Atom part until this year when they finished up the SKU lineup (who knows why it took so long) so I was basing it on that.

Not sure if you mean "Atom" literally or generically, but the N300 and N305 have an official launch date of Q1 2023:

https://ark.intel.com/content/www/u...ame/232598/products-formerly-alder-laken.html
​

It's also nowhere near the trouble it would have been when they were using proprietary tools for each node. Given that Intel 3 is a long term node it makes infinitely more sense to port Skymont and reap the design benefits than to use Crestmont just to gain efficiency.

Skymont is a much bigger core than Crestmont and maybe they'd have to clock it quite a bit lower for it to meet timing constraints, on Intel 3. IMO, it's not a given that a Skymont backport would be very compelling for them, economically.

 

[thestryker]

Not sure if you mean "Atom" literally or generically, but the N300 and N305 have an official launch date of Q1 2023:

Literally otherwise I'd use code name/architecture not branding.

 

[bit_user]

Literally otherwise I'd use code name/architecture not branding.

Okay, so I guess you mean Amston Lake:

https://ark.intel.com/content/www/u...ame/235968/products-formerly-amston-lake.html
​