Intel Core Ultra 200 "Arrow Lake-S" Lineup and Clock Speeds Revealed

[btarunr]

Intel is preparing at least twelve Core Ultra 200-series "Arrow Lake-S" desktop processor SKUs for the consumer segment, with more variants possible for the commercial desktop segment in the future. Q4 2024 could see the company debut its first SKUs targeting the PC enthusiast and gamer crowd with as many as five unlocked K or KF series SKUs. These, and finer details such as clock speeds, were revealed in a massive info dump by Jaykihn, a reliable source with Intel leaks. Intel is expected to debut the series later this year with the Core Ultra 9 285K, the Core Ultra 7 265K and 265KF; and the Core Ultra 5 245K and 245KF. The company is skipping a KF SKU for its top Core Ultra 9 part.

As has been consistent for several past generations of Intel processors, the top Core Ultra 9 (formerly Core i9) tier gets Thermal Velocity Boost, Turbo Boost Max 3.0, and classic Turbo Boost 2.0. The 285K maxes out the "Arrow Lake-S" B0 silicon, enabling all 8 "Lion Cove" P-cores, and all 16 "Skymont" E-cores. It comes with a P-core base frequency of 3.70 GHz, and an impressive 3.20 GHz E-core base frequency. The maximum P-core boost frequency achievable for up to two cores is 5.70 GHz, and 3-6 as well as 7-8 cores boost up to 5.40 GHz, making it the all-P-core boost frequency for this chip. The four E-core clusters are assured an all-E-core boost frequency of 4.60 GHz. The iGPU has 64 execution units, and ticks at up to 2.00 GHz.



The Core Ultra 7 265K/KF are supposed to succeed the Core i7-14700K/KF, and lose out on the TVB algorithm. Intel is giving these chips an 8P+12E core configuration. These come with a P-core base frequency of 3.90 GHz, and E-core base frequency of 3.30 GHz. The P-cores boost up to 5.50 GHz for 1-2 cores, and has 5.20 GHz as its all-P-core boost frequency, while the E-cores boost up to 4.60 GHz, same as the 285K/KF.

The Core Ultra 5 245K/KF are successors of the Core i5-14600K/KF, with a 6P+8E core configuration. This time around, Intel isn't recycling older silicon for the lower tiers of the Core Ultra 5 series, so you're assured increased IPC across the lineup. The P-cores of the 245K/KF come with a base frequency of 4.20 GHz, and the E-cores 3.60 GHz, which is the highest in the series. There's no Turbo Boost Max 3.0, and the classic Turbo Boost algorithm boosts up to 2 P-cores to 5.20 GHz, while its all-P-core boost frequency is 5.00 GHz. The E-cores boost up to 4.60 GHz.

There are several non-K/KF SKUs featured in the table, which Intel will likely launch in Q1-2025. These lack CPU overclocking features, and come with generally lower clock speeds than their K/KF siblings, besides lower power values. One SKU that caught our eye is the Core Ultra 5 225/225F. This chip appears to succeed the Core i5-14400/F, and comes with a 6P+4E configuration. The P-cores boost up to 4.90 GHz (up to 4.70 GHz all-P-core), while the E-cores go up to 4.40 GHz. We like how the Core Ultra 5 series isn't cluttered this time around, and you're only choosing between the 245K/KF and the 225/F.

View at TechPowerUp Main Site | Source

 

[Onasi]

I kinda hoped Intel would do away with the whole “there are three boost types” thing. Apparently not. This isn’t an issue in practice, obviously, the chip will boost to whatever it can, just feels like an over-complication for no reason.

 

[Hyderz]

when can we expect these cpu reviewed? 2nd week of september?

 

[Caring1]

The 245K looks decent.

 

[Crackong]

People would want to know ...

 

[Raiden85]

Hopefully some reviewers will put it under a heavy 24/7 work load for a few weeks and see how stable it is.

 

[AusWolf]

I kinda hoped Intel would do away with the whole “there are three boost types” thing. Apparently not. This isn’t an issue in practice, obviously, the chip will boost to whatever it can, just feels like an over-complication for no reason.

I agree. It's hard to follow what the chip can do this way. Maybe that's the intention here?

 

[Vincero]

I'm gonna guess they didn't release TDP numbers alongside these clock speed listings...

If you aren't interested in attempting to over lock/kill the silicon quicker, the 265 looks like a solid offering - would be interesting to how much the non-K part differs in power ratings Vs the K variant.

 

[R0H1T]

Hopefully some reviewers will put it under a heavy 24/7 work load for a few weeks and see how stable it is.

You probably need it under a heavy load for months! Any way I doubt their regular posse HP/Dell/Lenovo(?) would mind screwing the end users any less than what Intel's done with the last "two" gens.

 

[Daven]

So basically the top clock is 5.4 GHz and lower since everything uses more than two cores nowadays.

 

[Lycanwolfen]

If I win the Lotto and want a real intel CPU for desktop I'm buying a Xeon. I'm so sick of P cores and E cores. If I want something pure now have to buy a xeon.

Intel® Xeon® w9-3495X Processor (105M Cache, 1.90 GHz) - Product Specifications | Intel

Intel® Xeon® w9-3495X Processor (105M Cache, 1.90 GHz) quick reference with specifications, features, and technologies.

www.intel.com

4 grand but still a real CPU.

But so I'm poor AMD is the only company that shows they are for gamers now with pure 16 core cpu's

 

[MaMoo]

If I win the Lotto and want a real intel CPU for desktop I'm buying a Xeon. I'm so sick of P cores and E cores. If I want something pure now have to buy a xeon.

Intel® Xeon® w9-3495X Processor (105M Cache, 1.90 GHz) - Product Specifications | Intel

Intel® Xeon® w9-3495X Processor (105M Cache, 1.90 GHz) quick reference with specifications, features, and technologies.

www.intel.com

4 grand but still a real CPU.

But so I'm poor AMD is the only company that shows they are for gamers now with pure 16 core cpu's

I have the w2495x. I also have a 12700k. Let me tell you, my 12700 k with only p-cores enabled working at 5.1Ghz all cores managed to do a Linpack bench at 350 Gflops. My Xeon at stock does 530 Gflops. Using e-cores slows down the bench because Linpack uses symmetric work distribution in threads (my version for internal performance control). I tested this using actual production workflows and replicated the benchmark results (Xeon takes 5.5 days to run 1 iteration and 12700k takes 7.5 days).

 

[docnorth]

when can we expect these cpu reviewed? 2nd week of september?

Supposedly around 10 October.

 

[Onasi]

@Lycanwolfen
That’s unfathomably silly. The Xeon will be absolute pants in any normal day-to-day desktop tasks compared to the P/E-core processors. Not to mention that pretty much no desktop task will be able to leverage 56 cores. The whole “E-cores aren’t real cores” discourse is ridiculous. They are an absolute valid way of increasing MT performance and handling background tasks.

 

[Carlyle2020hs]

... but will it run a minecraft server without degredation for 100 days?

That´s the kind of test that would give me peace-of-mind.

 

[Hyderz]

Supposedly around 10 October.

I thought September sigh…. The wait continues

 

[Vincero]

If I win the Lotto and want a real intel CPU for desktop I'm buying a Xeon. I'm so sick of P cores and E cores.

Semi-agree - Intel are pushing the E-cores to make up for lack of good process for having more P cores. Some of these Arrow Lake chips will have 16E and only 8P... at which point the E cores are no longer being leveraged for 'efficiency' and low power tasks, but essentially just to make up the thread count higher regardless of if it brings down the overall multi-thread performance of certain tasks.
I'd rather have seen a 10P 8E variant instead but without hyper-threading it would be a lower performing part in terms of multi-thread performance.

I think that's the real problem - you no longer have hyper-threading on the P cores but get lots of a bit slower E cores to make up for it.
Will be really curious to see how that plays out with thread heavy tests vs LGA 1700 offerings.

... but will it run a minecraft server without degredation for 100 days?

That´s the kind of test that would give me peace-of-mind.

I'd have a bit more peace of mind if one would expect a Xeon variant of these products / platform, but I don't think there will be. That leaves out a bit of extra validation testing that Intel/OEMs would be doing for their respective HEDT/WS/Mini-server product lines.

 

[Raiden85]

They really need to come up with a new name for those E cores because they are far from efficient, even more so when they are starting to dwarf the P core amount. Honestly surprised Intel haven't come up their own version of the chiplet layout yet because surely this would make it far easier for them to manufacture instead of one monolithic chip.

 

[Daven]

I thought September sigh…. The wait continues

Yeah waiting for Zen 5 3D chips is a bummer.

 

[TumbleGeorge]

Hmm, rumours for iGPU around 1.9Tflops at 2GHz? Be able to light gaming.

 

[P4-630]

2nd week of september?

Mobile Lunar Lake in September.

 

[Vincero]

They really need to come up with a new name for those E cores because they are far from efficient, even more so when they are starting to dwarf the P core amount. Honestly surprised Intel haven't come up their own version of the chiplet layout yet because surely this would make it far easier for them to manufacture instead of one monolithic chip.

Foveros packaging is pretty much their version of the chiplet layout - they can mix/match tiles how they want so long as they fab the necessary interposer part also.
Not sure Intel want to necessarily move to a similar/same seperate chiplets on CPU substrate - they have done it before and still do it when they need to, e.g. i3/5/7 1st gen LGA 1156, right through to tiger lake chips.

 

[dgianstefani]

Foveros packaging is pretty much their version of the chiplet layout - they can mix/match tiles how they want so long as they fab the necessary interposer part also.
Not sure Intel want to necessarily move to a similar/same seperate chiplets on CPU substrate - they have done it before and still do it when they need to, e.g. i3/5/7 1st gen LGA 1156, right through to tiger lake chips.

It's actually more advanced than the chiplet strategy AMD currently uses, should theoretically be closer to monolithic performance than the Zen approach, with much better power management, we'll see. AMD is planning to move to a modern interposer solution with Zen 6, apparently.

 

[Vincero]

@dgianstefani my point was more Intel have been there and done it already with chiplets, arguably (ok very different substrate / packaging method on these old ceramic chips) even going all the way back to the Pentium Pro...

Their more modern reintroduction was the 1st gen Core i3/5/7 - I exclude the Pentium D (P4 Dual core) as it was a bit of a panic move and not really a proper chiplet design with both cores just tied to the same FSB connections. I'll be honest I wasn't sure why everyone back in the day when Zen2 was released were like "Chiplets, OMFG!!! AMD are amazing... Wow "
Not saying it wasn't a good engineering approach on their part, just "yeah, ok, makes sense".

Just that Intel's response to AMD's chiplet approach has been to move to foveros instead of pursue a chiplet approach for the desktop, although they did with the laptop (as mentioned tiger lake for example but repeating what they did with the 1st gen Core where you had a CPU chiplet and GPU chiplet).

 

[Minus Infinity]

I thought September sigh…. The wait continues

Lunar lake is September 3 IIRC.

It's actually more advanced than the chiplet strategy AMD currently uses, should theoretically be closer to monolithic performance than the Zen approach, with much better power management, we'll see. AMD is planning to move to a modern interposer solution with Zen 6, apparently.

Agreed, whatever one thinks of Intel this is far more advanced than AMD's process and with BSPD coming to 20A node gap further widens. I wonder what AMD has as an answer going forward. TSMC won't have BSPD until N2 at earliest, or GaaFET IIRC. Zen 6 is on N3P not N2 (well outside Epyc).