Intel Reports Findings on Missing Arrow Lake Performance, Targets Jan 2025 for 0x114 Microcode

[btarunr]

Intel today, in a letter to the press, presented an overview of the lower than expected performance of its latest Core Ultra "Arrow Lake-S" desktop processors, which the company released this October, to less than enthusiastic reviews. The top Core Ultra 9 285K ended up barely faster than the previous Core i9-14900K in gaming performance, and moderately faster in applications. AMD's November release of the Ryzen 7 9800X3D extended the green team's lead over Intel, prompting Intel to get to work on identifying why "Arrow Lake-S" isn't as fast as the company hoped, and if there was something the company could do about it.

Last month, Intel announced that it had identified probable reasons why "Arrow Lake-S" underperforms and that it is working on a combination of fixes that include OS-level updates from Microsoft, some motherboard-vendor level fixes to their UEFI setup programs, and a microcode update for its processors. Intel's microcode updates have been in the news throughout 2024, as one such update proved crucial in fixing the faulty voltage regulation behavior of its Core "Raptor Lake" processors, which was causing irreversible damage to the chips. Today, the company released documentation that highlights all issues the company identified by Intel, their technical root cause, and what's been done about it. Long story short—most of these issues are already addressed, but the Microcode update should be in circulation by January 2025, when the company expands its Core Ultra 200-series desktop processors with 65 W models.

Update Dec 19th: We tested the new OS patches, and unfortunately we have to report that this not a general fix for Arrow Lake, but just a fix for performance problems specific to 24H2. More in our article.

Update Dec 20th: Last night, ASUS released a non-Beta BIOS with microcode 0x114, we tested it and the findings are included in the Dec 19 article linked above.


The first set of performance issues are caused due to the way the OS scheduler works with the various kinds of CPU cores on "Arrow Lake-S," which was causing higher delta (variation) in scores among benchmark runs with identical settings. There is also less than expected scores in 1T (single thread) settings of certain synthetic benchmarks. What's even worse, there is a 50% increase in DRAM latency during the benchmarks. Performance overall is reduced on the latest Windows 11 24H2 release, compared to the older 23H2.


Intel confirms that the first set of issues are caused due to faulty collaborative power management between the OS and the processor. UEFI CPPC establishes a set of standards by which an OS can send information about the performance demands of its various processes and services to the hardware power management unit, letting it alter performance parameters such as core scheduling/parking, voltage, clock speeds, among others. Specifically, Intel says that the processor power management (PPM) settings at the OS-level for Windows 11 24H2 are faulty. This alone accounts for a massive 6% to 30% loss in performance. Intel reports that all these issues have been solved by version 26100.2161 (KB5044384), or the November 2024 "patch Tuesday" update of Windows 11 24H2.


The second set of performance issues are caused due to Intel Application Performance Optimizer (APO) not delivering on performance uplifts along expected lines. APO is an application-specific set of processor optimizations Intel has been releasing with its processors over the last few generations. A faulty implementation of APO on the OS level meant that PPM settings on APO profiles don't apply correctly. Another issue noticed was that motherboard BIOS versions released to CPU reviewers around the time of Core Ultra 9 285K reviews (October) didn't have APO enabled by default. These together cost anywhere between 2% to 14% performance in games that have APO profiles. Intel says that these issues are solved with the November "patch Tuesday" update to Windows 11 24H2, just like the first set of problems.

Another rather bizarre set of issues affected games using EAC (Easy Anti-Cheat), triggering a BSOD at launch, when running on Windows 11 24H2. This was due to a bug with the kernel-mode driver (KMD) of EAC when paired with Windows 11 24H2, which was allegedly exacerbated with virtualization-based security (VBS) being disabled by the user (Microsoft enables this by default for 24H2). Intel says that it's been working with Epic Games to distribute an updated EAC KMD to game developers.


The fourth set of performance issues are caused due to combination of incorrect default settings in motherboard UEFI setup programs. Some of these are glaring omissions, such as PCI Resizable BAR being disabled by default, APO being disabled, wrong Compute Ring Frequency, a setting that affects intercore latency, and integrated memory controller Gear mode, which affects the ratio between the IMC frequency and DRAM frequency, enabling Gear 4 even in lower frequencies that can work perfectly stable with Gear 2. These could affect performance by anywhere between 2% to 14%. Intel says that it's been pursuing motherboard vendors to release BIOS updates that fix these defaults, and these should already be in circulation.

Lastly, we reach the heart of the matter, the 0x114 Microcode Update. This update is said to include several "performance enhancements," and the company's motherboard partners are working to integrate it with their future BIOS updates. Intel says that the Microcode should cause a "single digit percentage performance enhancement" across games. This is a geomean obtained over 35 games, which means the 285K should end up faster than the Core i9-14900K, and inch closer to the Ryzen 7 9800X3D in gaming performance.

Intel says that the 0x114 Microcode Update is being targeted for January 2025.

View at TechPowerUp Main Site

 

[Timbaloo]

Maybe, maybe, their core architecture is getting a little bit too complex to manage properly?

Maybe, maybe, go back to something more manageable that does not have to mature at the customer?

We already have plenty of games/software that are unplayable/unstable at release, let's not start this with hardware also...

 

[Darmok N Jalad]

Sounds like Arrow Lake S should have launched in January then. That way MS, board partners, and Intel themself could have had their products ready for game time.

This was the number one reason I’d be hesitant to buy into this platform. It’s a big departure from the previous generation, and the current strategy in this market is to release an okay product and let consumers find the issues. There’s also a high likelihood that this is the only generation on the current socket, save a refresh.

 

[Daven]

I'm definitely seeing a lapse in product refinement before release across multiple industries. Under extreme competition, hardware companies are releasing newer, untested tech with each generation. This is putting a strain on both the OS and the application. Microsoft can't seem to keep up and is always unoptimized for the latest architectures. Unless software developers make their code aware of new architecture enhancements, there is no performance increase and sometimes degradation.

I don't see any of this getting better any time soon especially given how raw core clocks and counts are stagnating. These were easy ways to generically improve performance.

 

[Onasi]

So, if their findings are to be believed the most significant performance degradation was OS side and should already be fixed, right? Would be curious to see a performance review comparing older launch 24H2 results with the current build. Something tells me there won’t actually be a significant performance uplift, let alone one of “up to 30%”. Maybe someone already did the test and I just missed it.

 

[ir_cow]

I'm fully updated with x13 code and i'm getting the same FPS in CS2 with AMD 7950X and 285K. But I also have all the power plans / C-State disabled on both. This performance issue seems to be mostly related to how turbo / c-states work on the new platform.

It still lags behind the 13/14th Gen because of higher latency.

 

[john_]

Well, at least they are moving in the right direction, especially looking at also how they try to market the new graphics cards.

 

[ir_cow]

This is what I'm seeing.

 

[Onasi]

@ir_cow
Didn’t Intel tell you? Part of the issue is fixed, totally. Up to 30%. Just wait till January and it will all be solved, gonna overtake the 14900K fo sho this time.

 

[W1zzard]

So, if their findings are to be believed the most significant performance degradation was OS side and should already be fixed, right? Would be curious to see a performance review comparing older launch 24H2 results with the current build. Something tells me there won’t actually be a significant performance uplift, let alone one of “up to 30%”. Maybe someone already did the test and I just missed it.

Working on this right now.

23H2 older install vs 23H2 newest patches vs 24H2 fresh install vs 24H2 newest patches, also new BIOS (not that magical new microcode yet though, this comes in Jan)

 

[vgm]

Ryzen 7 9800X3D extended the green team's lead over Intel

@btarunr Shouldn't it be Red Team instead of Green/Nvidia?

 

[Onasi]

@W1zzard
Thanks, absolute giga-chad move. Just don’t tire yourself out and don’t forget to enjoy the holidays.

 

[phanbuey]

Nvidia announcing am ARM CPU and GPU platform for high end desktop coming September 2025... Intel and AMD are set up to get absolutely destroyed here with the stagnation.

 

[Onyx Turbine]

im seeing likely another issue contributing to these results, not mentioned, which stems from the chip design, you can guess it, when you compare chip characteristics you get the direction

 

[Onasi]

@phanbuey
I am sure. Went absolutely swimmingly for Qualcomm, didn’t it? A first gen ARM CPU relying on a translation layer for most software is about as likely to destroy current x86 players as Zimbabwe launching a Mars expedition.

 

[phanbuey]

@phanbuey
I am sure. Went absolutely swimmingly for Qualcomm, didn’t it? A first gen ARM CPU relying on a translation layer for most software is about as likely to destroy current x86 players as Zimbabwe launching a Mars expedition.

Nvidia != Qualcomm. Nvidia is much closer to apple than it is to qualcomm.

Also if they design the platform specifically for their combo (which they will), then 90% of the GPU market will run better on it than the competition. This is a way bigger deal than qualcomm.

 

[MachineLearning]

@btarunr Shouldn't it be Red Team instead of Green/Nvidia?

This is the classic AMD logo, red or green team both are accurate. Green team is the super old school term, like ATI and before

I guess if / when NVIDIA enters the CPU market we'll all have to be more specific, though.

 

[Onasi]

@phanbuey
Qualcomm is literally the biggest ARM SOC vendor in the world with decades of experience. Comparing NV to Apple in this regard is silly - unless they intend to create their own OS with all that entails (good luck) they will have to contend with all the same issues QC did. Besides, I am not seeing many desktop enthusiasts, especially gamers, flocking to M-powered Macs.

 

[Fazio98]

The 0x114 CPU microcode update is already rolling out with AsRock Z890 bios updates available yesterday.The Z890 motherboard bios updates contain the CPU microcode 0x114 and an ME update.

 

[ir_cow]

ASRock people already tired the x114 code. Nothing worth talking about. Until Intel moves the IMC back to where it was before (next CPU) or raise the D2D, NSU and ring clocks, we aren't going to get that 13/14th gen performance back.

 

[Fazio98]

ASRock people already tired the x114 code. Nothing worth talking about. Until Intel moves the IMC back to where it was before (next CPU) or raise the D2D, NSU and ring clocks, we aren't going to get that 13/14th gen performance back.

Where did you see this.The new bios only came out yesterday.There's nothing on the AsRock community.What does "nothing worth talking about" mean.
I haven't seen any updated reviews that used the new bios.

 

[_roman_]

I read that topic on the smartphone today.

Intel is a promoter for UEFI. reference. https://uefi.org/members

I did not really bother taking my time with the details. I had many intel based notebooks with bogus firwmare.

I assume Intel did not implement correctly the UEFI functionality. I assume INTEL does the full firmware for the uefi which is processor based. I assume INTEL does the circuit design for the processor.
As always Windows has to fix the broken Intel implementation. Same as always with other hardware. A correct implementation does not need a software fix.

-- The final nail in the coffin: Intel does not test their products before shipping. A ~30Percent or more performance penalty was not seen?

Blame everyone else. That makes it easy

It's the bios fault

Another issue noticed was that motherboard BIOS versions released to CPU reviewers around the time of Core Ultra 9 285K reviews (October) didn't have APO enabled by default.

It's the mainboard fault - Intel needs certain options enabled

The fourth set of performance issues are caused due to combination of incorrect default settings in motherboard UEFI setup programs. Some of these are glaring omissions, such as PCI Resizable BAR being disabled by default, APO being disabled, wrong Compute Ring Frequency, a setting that affects intercore latency, and integrated memory controller Gear mode, which affects the ratio between the IMC frequency and DRAM frequency, enabling Gear 4 even in lower frequencies that can work perfectly stable with Gear 2.

My intel wlan card seems to need the power saving feature disabled for every hardware in the uefi menu of my mainbaord. Else the hole box freezes randomly and that happens regularly.

It'S microsoft fault

Performance overall is reduced on the latest Windows 11 24H2 release, compared to the older 23H2.

Who's else fault is it?

 

[phanbuey]

Qualcomm is literally the biggest ARM SOC vendor in the world with decades of experience. Comparing NV to Apple in this regard is silly - unless they intend to create their own OS with all that entails (good luck) they will have to contend with all the same issues QC did. Besides, I am not seeing many desktop enthusiasts, especially gamers, flocking to M-powered Macs.

I wouldn't count Nvidia out -- qualcomms biggest issues were with the GPU not the CPU... nvidia has wont have those problems. As far as translation goes in addition to hardware, nvidia has an insanely strong software team, something qualcomm severely lacks as evidenced by their driver woes.

They have more cash than they know what to do with and are designing the full platform, deciding how that platform handles io and interrupts, and are optimizing it for their video cards -- if this yields gaming performance that has fewer stutters, for example, you will see a mass exodus from x86.

Even if they roll out mid range cpu performance but a monster in gaming, it will take a huge chunk out of intel and AMD.

 

[dirtyferret]

I'm fully updated with x13 code and i'm getting the same FPS in CS2 with AMD 7950X and 285K. But I also have all the power plans / C-State disabled on both. This performance issue seems to be mostly related to how turbo / c-states work on the new platform.

It still lags behind the 13/14th Gen because of higher latency.

We here at Intel have a message for you @ir_cow and all our faithful Intel fans

 

[ir_cow]

Where did you see this.The new bios only came out yesterday.There's nothing on the AsRock community.What does "nothing worth talking about" mean.
I haven't seen any updated reviews that used the new bios.

It was leaked last week.

We are at Intel have a message for you @ir_cow and all our faithful Intel fans

The CPU is great for a lot of things. Just not for games that are mainly latency dependent. But if you play at a higher resolution, max settings it doesn't matter which CPU you pick.