Friday, August 16th 2024
Top Intel Core Ultra 9 "Arrow Lake-S" Part Boost Clocks Surface
Intel Core Ultra "Arrow Lake-S" desktop processors are expected to debut later this year, and introduce the new Lion Cove P-cores, along with Skymont E-cores to the desktop platform. Engineering samples and qualification samples with specs close to retail chips seem to already be in the hands of PC OEMs and motherboard vendors, given the volume of leaks over the past few days. Jaykihn0, one of the more influential sources of these leaks, revealed a few interesting details of the maximum boost frequencies of these chips.
The QS of a top Core Ultra 9 "Arrow Lake-S" SKU, probably the flagship model that succeeds the current Core i9-14900K, is described as having a maximum P-core boost frequency of 5.70 GHz, and an all-P-core boost frequency of 5.40 GHz. The maximum E-core boost frequency, which is also the all-E-core boost frequency, is said to be 4.60 GHz. Let's unpack this. "Arrow Lake" uses the same mix of "Lion Cove" P-cores and "Skymont" E-cores as "Lunar Lake," albeit arranged along a ringbus, and sharing an L3 cache, unlike on "Lunar Lake," where the P-cores have their own exclusive L3 cache, and the E-cores are arranged in a low-power island, with the fabric of the SoC tile connecting the two.
We know from the "Lunar Lake" deep-dive from Intel, that the company claims a 14% IPC gain for "Lion Cove" over the previous generation "Redwood Cove" P-core found in "Meteor Lake." Given that "Redwood Cove" cores have been tested in the real world to offer roughly similar IPC to the "Raptor Cove" P-cores powering "Raptor Lake," if Intel's IPC claims for "Lion Cove" hold, then at 5.70 GHz, the P-cores of "Arrow Lake-S" should be 14% faster than "Raptor Cove." It's worth noting here that "Lion Cove" cores lack Hyper-Threading, but "Arrow Lake-S" has 8 of these, and as our recent "Zen 5 without SMT" article has shown, games largely aren't affected with the lack of SMT/HTT if the core count is as high as 8.
The cache sub-system of "Arrow Lake-S" is another interesting factor that could influence its gaming performance. Each "Lion Cove" P-core on the Core Ultra 9 "Arrow Lake-S" is expected to have 3 MB of dedicated L2 cache, and the 8 P-cores share 36 MB of L3 cache along with the four "Skymont" E-core clusters. Thread Director tends to avoid scheduling game workloads on the E-cores, unless there are specific optimizations within the game that use them (eg: for processing game physics, audio DSPs, network stack, etc).
Intel has promised a massive IPC leap for the "Skymont" E-cores over the current "Gracemont," with the company claiming an IPC resembling that of the "Raptor Lake" P-core. Of course there are some riders—"Skymont" cores don't boost nearly as high as "Raptor Cove" P-cores do, even in this top Core Ultra 9 SKU, the maximum E-core boost frequency is a moderate 4.60 GHz. Also, the SPECrate2017 benchmark Intel uses in its IPC calculations isn't memory intensive; "Skymont" cores are clustered into groups of four cores, and made to share a 4 MB L2 cache on "Arrow Lake-S."
All in all, with these frequencies, the top Core Ultra 9 "Arrow Lake-S" part seems to be gunning for the gaming performance leadership crown from AMD, which has held the bragging rights of selling the fastest gaming processor for 16 months now (since the April 2023 launch of the Ryzen 7 7800X3D).
Sources:
Jaykihn0 (Twitter), VideoCardz
The QS of a top Core Ultra 9 "Arrow Lake-S" SKU, probably the flagship model that succeeds the current Core i9-14900K, is described as having a maximum P-core boost frequency of 5.70 GHz, and an all-P-core boost frequency of 5.40 GHz. The maximum E-core boost frequency, which is also the all-E-core boost frequency, is said to be 4.60 GHz. Let's unpack this. "Arrow Lake" uses the same mix of "Lion Cove" P-cores and "Skymont" E-cores as "Lunar Lake," albeit arranged along a ringbus, and sharing an L3 cache, unlike on "Lunar Lake," where the P-cores have their own exclusive L3 cache, and the E-cores are arranged in a low-power island, with the fabric of the SoC tile connecting the two.
The cache sub-system of "Arrow Lake-S" is another interesting factor that could influence its gaming performance. Each "Lion Cove" P-core on the Core Ultra 9 "Arrow Lake-S" is expected to have 3 MB of dedicated L2 cache, and the 8 P-cores share 36 MB of L3 cache along with the four "Skymont" E-core clusters. Thread Director tends to avoid scheduling game workloads on the E-cores, unless there are specific optimizations within the game that use them (eg: for processing game physics, audio DSPs, network stack, etc).
Intel has promised a massive IPC leap for the "Skymont" E-cores over the current "Gracemont," with the company claiming an IPC resembling that of the "Raptor Lake" P-core. Of course there are some riders—"Skymont" cores don't boost nearly as high as "Raptor Cove" P-cores do, even in this top Core Ultra 9 SKU, the maximum E-core boost frequency is a moderate 4.60 GHz. Also, the SPECrate2017 benchmark Intel uses in its IPC calculations isn't memory intensive; "Skymont" cores are clustered into groups of four cores, and made to share a 4 MB L2 cache on "Arrow Lake-S."
All in all, with these frequencies, the top Core Ultra 9 "Arrow Lake-S" part seems to be gunning for the gaming performance leadership crown from AMD, which has held the bragging rights of selling the fastest gaming processor for 16 months now (since the April 2023 launch of the Ryzen 7 7800X3D).
46 Comments on Top Intel Core Ultra 9 "Arrow Lake-S" Part Boost Clocks Surface
The 14900K has turbo clocks up to 6GHz, so there is indeed a 300 MHz degradation, just not 500 MHz as it would be from 14900KS
The big one is the E-cores. The theoretical IPC gain looks to be very high, but with smaller cache it's not likely to be fully realized in all use cases, possibly not at all in situations where the e-core idles waiting on cache. Still, I can't imagine it not having a significant impact on highly threaded scenarios.
The other unknown is the impact of moving from monolithic to tiles. That is going to exact some kind of performance penalty.
Overall I have some expectation of a +10% overall performance increase, much lower than that with single/light thread apps and higher than that with more threaded apps.
This is pretty well borne out by some of the early benchmark leaks :
www.igorslab.de/en/performance-jump-arrow-lake-9-285k-up-to-18-faster-than-14900k/
I don't know what you are talking about for smaller cache in the E-cores, they still have the same amount of Cache which actually has some improvements, which is the improved shared L2 bandwidth and L1 to L1 transfers, where before it would need to do so through L3.
A big issue was that there was simply not enough bandwidth when the 4 E-Cores were being used with only 64B/cycle, but now with 128B/cycle that is going to be a big improvement.
The E-cores themselves were 4.5 GHz on 14900KS and 4.4 GHz on 14900K, so there is a 100~200MHz improvement there.
Those aren't going to be a making a big impact on gaming though.
It remains to be seen if they achieve a good balance there with the E-cores, IPC increase or not.And, it'll run through cache quicker.Maybe not, but Tom's is reporting that the Ultra 7 265KF (I see this as the 14600K replacement) outpaced the 14900K in single thread by 4%, and the 14700K by 7% while tying the 14700K in multi-thread.
There should be an Ultra 9 275 and Ultra 9 285K above the 265K in the Arrow Lake lineup.
I'm guessing they do not but with how every launch seems to be rushed these days better to just let it play out.
Intel really can't afford another blunder with how their earning reports are going and how much staff they're having to cut so you'd think they be damn sure it doesn't prior to launch. Maybe it'll be delayed from October we will see.
It didn't harm AMD when Jim left after designing Zen but with Royal core's Beast lake cancelled what will intel replace it with are we going to see stagnation after Panther lakes cougar core's?
I'm interested in this platform as a possible upgrade later this year.
For me since Zen 5 wound up being a nothingburger, its main competition is the 14700K which is a quick easy $350 plug in upgrade. I don't think +5% single and +10% multi will be worth a $200 motherboard upgrade and all the hassle that goes with that (drivers, maybe need to re-install Windows and so on), unless there's something else there.
Doesn't matter if it should be an i5 like the 9700X should be an R5 7600X replacement in my book but price dictates what it actually is.
Same for you you'll have to look at it vs a 14700k and decide if a platform upgrade is justified based on it performance.