Interesting. Intel's 22-core Nova Lake-S with 108MB of big LLC is a single-tile design, so all the cores get the extra cache without the chiplet-parking complexity you see on AMD's multi-CCD X3D chips. No need for Game Mode, core parking, or worrying about the scheduler sending games to the wrong die. That could actually be simpler to use than AMD's 12- and 16-core X3D parts.

RejZoRThey are not. Maybe they are on a level of 10th gen cores. Which is not bad but far from newer generations P cores.

Arctic Wolf e-core IPC perhaps better than zen6, should be significantly better than Lioncove and zen5, with its IPC closer to Coyotecvoe rather than zen6
In other words, it is better than the cpu uarchs used by everyone here right now.

N/A8 p cores or forget it. Fine.

18 out of 22 cores, each core is much better than the best core you are using right now

TheinsanegamerNThe LPE cores are identical to normal E cores

No, they're not. The LPEcores run at slower clock speeds and lower voltages than the normal Ecores, thus the LP(low power) distinction.

[crs]so the big question is: Will this be a good upgrade from my 9800X3D?

Probably not. It'll be faster, but not a whole platform swap's worth of faster

efikkanweak front-end. latency sensitive

Skymont has low bandwidth but decent latency. BP performs well but is weaker than the p core. Decode are comparable to the p core. if Arctic Wolf is upgraded to 12-wide decode, which is the most advanced.

[crs]so the big question is: Will this be a good upgrade from my 9800X3D?

You should,have a direct path to zen 6 on that board so, why would you switch

lexluthermiesterNo, they're not. The LPEcores run at slower clock speeds and lower voltages than the normal Ecores, thus the LP(low power) distinction.

The design is the same, but it uses SLC instead of ring for energy saving purposes, so it has a significant impact on performance.

kondaminYou should,have a direct path to zen 6 on that board so, why would you switch

Because U5bllc might be much cheaper than R7x3d and have more cores.
lp-e，npu，igpu，etc

Sol_BadguyAnd the award for the most moronic naming scheme goes to Intel. Anybody surprised?

The P-core is stronger than the E-core, yes?
The wolf is stronger than the coyote, yes?
And this is an Arctic wolf, therefore bigger than one from a region closer to the Equator.
en.wikipedia.org/wiki/Bergmann%27s_rule

Then how does this naming scheme make sense? Huh Lip, how?
Spotted hyena P-core or GTFO.

Oh yeah and the 6P approach, wow I can almost smell how future proof that is (crying in next gen console games). Mhm that "10800X3D" won't know what hit it. 🙃:rolleyes:
Get your act together you goddamn amateurs.
I'm getting that gut feeling that you're going to fuck up this launch like you did with Arrow Lake. All hype and then no bite.

6p+12e will provide 18 high-quality threads for the OS, while zen6 can only provide 12, with the rest being low-quality smt, thereby affecting performance. For ue6, intel 24core is the most advantageous, while zen6 12c24t and novalake 18c18t should perform closely in scheduling. p core is not the critical factor, and Arctic Wolf e core performance should reach over 90% of p core.

zusMakes sense, although I don't really know what's the difference between Arrow Lake and Arrow Lake Plus. I know that Intel Binary Optimization Tool only works on Arrow Lake Plus, and that they're tuned closer to the 200S Boost profile out of the box, but that's about it.

The main difference lies in the n3b node yield, dtco, which allows intel to offer a cheaper 285k.

tsurugiThe design is the same, but it uses SLC instead of ring for energy saving purposes, so it has a significant impact on performance.

Um, ok. They're still a bit slower and not identical, that's what I'm saying. That said, this CPU will still be a winner for the right money.

Let3rI’m still so annoyed that I bought into LGA 1851 and then they only gave it one generation.
😭 Wish I had gone with AM5.

Its extremely rare for Intel to give any socket a second CPU. Look at the 1700 Socket, 4 generations and 3 refreshes of the same CPU lol

[crs]so the big question is: Will this be a good upgrade from my 9800X3D?

Beside the CPU the new LGA 1954 platform will be much better than the aging and neglected AM5 platform, even the current Arrow lake platform is. AMD users are not getting a new platform any time soon.

BoggledBeagleBeside the CPU the new LGA 1954 platform will be much better than the aging and neglected AM5 platform, even the current Arrow lake platform is. AMD users are not getting a new platform any time soon.

For practical purposes this only matters if you saturate your NVMe bandwith, which is a very rare thing.

This is the cpu i am probably going to upgrade to. I will wait for reviews between Ryzen 9800x3d successor vs this chip, whoever wins, that's what I am buying

i don't intend to upgrade again for a solid 5-10 years, so i don't care if this is one gen mobo. either way, I will be keeping my old mobo/7800x3d as backup rig.

RejZoRThey are not. Maybe they are on a level of 10th gen cores. Which is not bad but far from newer generations P cores.

Intell really should rebrand e cores because commenters on tech sites should not be this misinformed about them.

Gracemont e cores were compared to 10th gen p cores almost FIVE years ago. Skymont e cores have 30% higher IPC and 20% faster clocks speeds than Gracemont, which means they are comparable to 12th gen p cores. There is even a generous reddit user who went through the trouble of proving this.

overclocking/comments/1p2vaif/sharing_a_foolish_test_i_did_all_cores_vs_ecores
17% performance drop for e-cores-only means that only 12 Skymont e cores are faster than an entire 12700k.

tsurugiSkymont has low bandwidth but decent latency. BP performs well but is weaker than the p core. Decode are comparable to the p core. if Arctic Wolf is upgraded to 12-wide decode, which is the most advanced.

Which latency is being discussed here, L3/memory access or decoder (or another part of the pipeline)?

Also there were some rumours long ago about L2 sharing in P cores (4MB per two cores)... has this been confirmed?

tsurugiArctic Wolf e core performance should reach over 90% of p core

Quite believable in ST testing. MT is more unpredictable, especially if P cores will be clustered too.

Intel is really getting itself ready for the unified core quite soon, and that one will be a descendant of the more modern E core, not the P core, whose architecture has been exploited in full by now.

kinikuInteresting. Intel's 22-core Nova Lake-S with 108MB of big LLC is a single-tile design, so all the cores get the extra cache without the chiplet-parking complexity you see on AMD's multi-CCD X3D chips. No need for Game Mode, core parking, or worrying about the scheduler sending games to the wrong die. That could actually be simpler to use than AMD's 12- and 16-core X3D parts.

What's more important is whether Nova Lake's 6/8 P-core parts will have close to the clock speeds of the top parts. In recent generations their "5" and "7" tiers have been too constrained, preventing them from being amazing sweetspot buys. It's kind of an artifical segmentation. And while some would argue overclocking is an option, you wouldn't get safe calibrated voltage curves then, and you don't know the quality of the silicon. I would much rather they broght back "golden samples" again, but tuned more for sustained clocks.

tsurugiArctic Wolf e-core IPC perhaps better than zen6, should be significantly better than Lioncove and zen5, with its IPC closer to Coyotecvoe rather than zen6
In other words, it is better than the cpu uarchs used by everyone here right now.

It's not apples to apples, as those synthetic benchmarks don't represent real world use cases, and this is a much bigger problems for E-cores who's performance is very closely tied to what the other cores are doing. It's simply foolish and misleading to use a benchmark of the core in isolation to extrapolate real world performance of cores with such constraints.

Approximations for "IPC" these days mostly consist of various batch loads, which aren't latency sensitive or logically complex, which means two CPUs with similar scores may have very different performance characteristics in real world usage. What's made Intel's CPUs for generations more responsive is a very effective front-end, even back with Coffee Lake/Comet Lake vs. Zen 3; Intel's were more snappy while AMD's were better for larger batch loads, but in most synthetics the latter looked much better. And the more CPU makers are focusing on looking good on paper, the less useful single metrics are for gauging performance.

Geofrancisthey are finally doing another 5775c, it only took them 11 years.

Yeah its going to be interesting. Intel always invests a lot more into l2 cache than amd. I'm guessing they're going to do the same thing here.

But if we have all these cores, pcores for their single threaded performance, ecores for multithreaded die area efficiency, low power cores for background tasks or something?.... multiple large caches.... Well, its like there's no compromises here... Well except maybe lack of hyperthreading, if you consider that an advantage. So, I hope they are affordable for normal humans. High prices I expect, I just hope its not ram all over again.

Maybe thats why they are preparing some of the midrange chips first. Don't want people to see the price of the flagship yet.... maybe even the midrange will be hard to swallow.

So... the question is, do I buy the 270k PLUS while I still can?

ir_cowIts extremely rare for Intel to give any socket a second CPU. Look at the 1700 Socket, 4 generations and 3 refreshes ...

Alder Lake
Raptor Lake
Raptor Lake refresh
Bartlet Lake
Raptor Lake Next

How did I manage to miss some?

ir_cow... of the same CPU

12100
13100
14100

They even look like the same piece of silicon with updated info on package.

BoggledBeagleBeside the CPU the new LGA 1954 platform will be much better than the aging and neglected AM5 platform, even the current Arrow lake platform is. AMD users are not getting a new platform any time soon.

Do they need it?

✅ ... CPU socket
✅ ... PCIe Gen 5 x16 slot (used or not)
✅ ... PCIe Gen 5 x4 slot
Covers 95% of use cases.
In the right size you may put it into shoe box and call it a day.

PrettyKitten800Intell really should rebrand e cores because commenters on tech sites should not be this misinformed about them.

Gracemont e cores were compared to 10th gen p cores almost FIVE years ago. Skymont e cores have 30% higher IPC and 20% faster clocks speeds than Gracemont, which means they are comparable to 12th gen p cores. There is even a generous reddit user who went through the trouble of proving this.

overclocking/comments/1p2vaif/sharing_a_foolish_test_i_did_all_cores_vs_ecores
17% performance drop for e-cores-only means that only 12 Skymont e cores are faster than an entire 12700k.

Yeah, I did a similar test, 8 ecores on 12900k (so 3.7 ghz) were ~as fast as my 10900k in games. :eek:

RejZoRThey are not. Maybe they are on a level of 10th gen cores. Which is not bad but far from newer generations P cores.

Clock for clock the Skymont E-cores in Arrow-Lake matched the P-Cores in Raptor Lake in almost all tasks, except, ones where the E-Cores lacked instructions.
The Darkmont E-Cores in Nova Lake have instruction parity with the P-Cores and on top of that are decently improved over the previous Skymont E-Cores in both PPC and higher clock speeds.

I'm not terribly surprised that the E cores could have good IPC. After all, Raptor Lake P-cores had strong IPC and had the potential to be very efficient, but Intel pushed the design well past the performance/efficiency curve intersect. That was largely because they couldn't get the node shrinks working and had to push an old node (which also took them a while to get working) too hard. Arrow Lake was a half-step in the right direction, but they had to backtrack a bit on the power consumption to stop burning up CPUs. If Intel finally has the fab issues resolved, then I would expect Nova Lake to address this issue while also gaining a bit in IPC. It ultimately comes down to cache and clocks, and Intel has the opportunity to improve both with Nova Lake. AMD has already played the cache card, and it looks like that advantage is about to go away. The problem for Intel now is timing--it's a very expensive time to upgrade.

Darmok N Jalad... AMD has already played the cache card, and it looks like that advantage is about to go away. The problem for Intel now is timing--it's a very expensive time to upgrade.

I am not sure that many people realise that AMD had just this one fundamental advantage - they were able to stick an additional chip on top or below a CPU - the CPU itself was nothing special or in any significant way better than what Intel had and has.

Expensive to upgrade? I think that the RAM sticks and SSDs are reusable.

I believe that Intel may want to regain more market share and better image in consumer PCs and they may go on in an agressive great value for money strategy defined by 250K and 270K. So at least some of these new CPUs could be a good value for money and the only remaining piece of the puzzle is the cost of the motherbords.

zusIt's a pretty small difference still. I don't even know if we can attribute it to P-cores, or if it's due to the slight increases in clock and L3 cache.

Agreed core scaling in games is, to a large degree, a black box.
Looking at AMD CPUs comparing 9600X to 9700X or further to a 9950X is the wrong way because frequency isn't constant for the same core/thread count.
tpucdn.com/review/amd-ryzen-5-9600x/images/boost-clock-analysis.png
tpucdn.com/review/amd-ryzen-7-9700x/images/boost-clock-analysis.png
tpucdn.com/review/amd-ryzen-9-9950x/images/boost-clock-analysis.png
Beyond that there is SMT which could skew the result, when it comes to thread count even the 9600X has enough threads for games, so if the game engine would properly take advantage of these threads you can't really get obvious scaling (due to only the cores) from 9600X to 9950X because the 9600X has enough virtual threads to compensate for the lack of physical cores. So those would have to be exhausted before a hard cut off in the performance is observed.

Thus the right approach would be to use a homogeneous CPU and disable cores. That way the cache and frequency are constant (well ignoring the boosting algorithm anyway).
Right now the 270K/285K could be used for Intel with P-only config and use 4, 6 and 8 cores.
For AMD either 9700X or 9800X3D with SMT off and using 4, 6 and 8 cores.
Only in this manner we can reliably identify the games that can genuinely use many cores.
The Zen6 "10800X3D" will be an even better candidate due to its 12 core CCD, it will have more than enough homogeneous physical cores to cover outliers.

zusMakes sense, although I don't really know what's the difference between Arrow Lake and Arrow Lake Plus. I know that Intel Binary Optimization Tool only works on Arrow Lake Plus, and that they're tuned closer to the 200S Boost profile out of the box, but that's about it.

In the review the older Arrow Lake SKUs are using 200S Boost enabled so there's no disadvantage there.
It's something alright, I don't know/think if it's the BOT or something else but it's working because from what I've seen comparing the 285K vs 270K at the same core clocks (P and E), same NGU, D2D, Ring whatever yields worse results in CB multi. In single thread the 285K still has an advantage due to the Thermal Velocity Boost algorithm.

zusSurely Intel will make a CPU with 8 P-cores too, I would be very disappointed if they didn't.

I'm counting on it.

iameatingjamSo... the question is, do I buy the 270k PLUS while I still can?

Will you be satisfied by its performance for your use case?
If yes do it, if not then get ready for single full tile with bLLC NL at $600.
If so then the only chance for a significant discount is if it's beaten by the Zen6 "10800X3D" in games.
Also keep in mind that NL might need higher speed RAM (compared to the 270K) to achieve its full potential. This is a minor drawback but since you'd be willing to wait then every aspect counts.

efikkanWhat's more important is whether Nova Lake's 6/8 P-core parts will have close to the clock speeds of the top parts. In recent generations their "5" and "7" tiers have been too constrained, preventing them from being amazing sweetspot buys. It's kind of an artifical segmentation.

AMD is doing the same.
I agree it's a major inconvenient.
250 W MTP with the current clocks is stupid for the 265K, 200 max should have been the target.
The 9700X is power constrained at 65/80 W it should've had 80/110 W something like that with slightly higher clocks.
I doubt Intel would change its strategy although I would bet that is the winning approach for next gen.
Suppose AMD is stupid and puts higher clocks on the "10950X3D" compared to the "10800X3D".
tpucdn.com/review/amd-ryzen-7-9800x3d/images/boost-clock-analysis.png
tpucdn.com/review/amd-ryzen-7-9850x3d/images/boost-clock-analysis.png
tpucdn.com/review/amd-ryzen-9-9950x3d/images/boost-clock-analysis.png
That means if Intel does the opposite, the single full tile with bLLC can win in games vs the "10800X3D", whereas the NL dual tile flagship will lose in games against the "10950X3D" but still win in MT, which is well the point of that CPU. I mean buying that primarily for gaming is definitely a "choice".

Darmok N JaladIf Intel finally has the fab issues resolved, then I would expect Nova Lake to address this issue while also gaining a bit in IPC.

The compute tile is made by TSMC just like with Arrow Lake.

WirkoWhich latency is being discussed here, L3/memory access or decoder (or another part of the pipeline)?

Also there were some rumours long ago about L2 sharing in P cores (4MB per two cores)... has this been confirmed?


Quite believable in ST testing. MT is more unpredictable, especially if P cores will be clustered too.

Intel is really getting itself ready for the unified core quite soon, and that one will be a descendant of the more modern E core, not the P core, whose architecture has been exploited in full by now.

cache from L1 to ram

Jaykihn mentioned it, credibility is quite high.

I think p core was more dragged down by lioncove's crappy issues and slow R&D progress. Lioncove directly ruined a generation of DC, replaced by redwoodcove x, DMR is still very late and smt was canceled, while griffincove's performance might be very good. But the e core team can stably bring quite decent IPC every 2 years, and the area is much smaller. What more can the next gen of p core do, maybe another lioncove, delays, cancellations, the risk is too high.

efikkanIt's not apples to apples, as those synthetic benchmarks don't represent real world use cases, and this is a much bigger problems for E-cores who's performance is very closely tied to what the other cores are doing. It's simply foolish and misleading to use a benchmark of the core in isolation to extrapolate real world performance of cores with such constraints.

Approximations for "IPC" these days mostly consist of various batch loads, which aren't latency sensitive or logically complex, which means two CPUs with similar scores may have very different performance characteristics in real world usage. What's made Intel's CPUs for generations more responsive is a very effective front-end, even back with Coffee Lake/Comet Lake vs. Zen 3; Intel's were more snappy while AMD's were better for larger batch loads, but in most synthetics the latter looked much better. And the more CPU makers are focusing on looking good on paper, the less useful single metrics are for gauging performance.

I agree that using SPEC to measure cpu uarch is outdated, but I don't know a better way to measure it. The e core uarch is also a general-purpose cpu works for any task.

in SPEC2017 fp 519, the performance of the e core is clearly inferior to the p core, else is fine.
I was wrong about the latency; skymont has a higher latency than p core when it is below 3000kb. even if this is the case, it is nothing to be happy about that the p core, which occupies more area, performs similarly in SPEC IPC to the e core, which occupies less area.
the p core needs to be more well-rounded. But where the e core underperforms, it is a trade-off made for area/efficiency, I can understand it. since the e core is part of a hybrid design, the real world scenario might be that such tasks are allocated to the p core.

Perhaps it is an advantage in hybrid designs and low-power devices. Intel might improve unified p core L1/L2 to low latency and high bandwidth, but retain this part in unified e core.

BoggledBeagleI am not sure that many people realise that AMD had just this one fundamental advantage - they were able to stick an additional chip on top or below a CPU - the CPU itself was nothing special or in any significant way better than what Intel had and has.

Expensive to upgrade? I think that the RAM sticks and SSDs are reusable.

I believe that Intel may want to regain more market share and better image in consumer PCs and they may go on in an agressive great value for money strategy defined by 250K and 270K. So at least some of these new CPUs could be a good value for money and the only remaining piece of the puzzle is the cost of the motherbords.

Sure, you can reuse some stuff, but if you're not being impulsive, chances are your current DDR5 setup is a stone's throw away from anything new that comes out under most use cases. Not to say there isn't a place for an upgrade if all those cores are actually useful to you, but if one had an AM5 X3D CPU already, then I'm not sure anything next gen coming from either vendor is going to be worth the upgrade costs. The IPC gains won't be monumental--they haven't been for many years now. You'd have to be coming from ADL or AM4 to make it worth the cost, unless you just like having the new stuff with money to burn. This will be a new socket on top of that, and performance-grade motherboards are not cheap.

And I highly doubt these 2 particular CPUs will be cheap in the slightest with that big cache pool. They will be halo tier, meaning as expensive as the current X3D chips, probably more so if they do indeed outperform them. Wouldn't be surprised to see the $999 tier make a comeback since the 9950X3D2 is sitting at $900.

WirkoAlso there were some rumours long ago about L2 sharing in P cores (4MB per two cores)... has this been confirmed?

I guess you're talking about the L2 sharing of 2 mini cores 2c-2T in Hammer Lake, as opposed to the big 1C-2T cores. Best of both worlds, same die area. Same thread performance, no lag or stutter. The good news is that it may arrive for S1954 after the DDR5 apocalypse.