• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Arrow Lake leaks + post-launch discussion

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
I find any CPU above 200W a mistake for a home PC. If popular coolers can't handle them, then something isn't right. Ryzens generally stop before 200W (170-180W for 16 cores without OC). X3D is pure magic with that gaming performance at ~120W. The 265K has a maximum TDP of 250W, but in motherboard reviews, I could barely see it passing 190W, and in games, it's ~150-170W max. I have no idea how Intel calculated it, given its high TDP in general specs. If you unlock limits, then it can pass 250W, but it's possible to keep it at lower wattage and still overclocked to the max (which is ~5.6GHz P/~5GHz E cores). 285K, to me, is a mistake at the current store price and barely better performance than the 265K. It's still hard to get.
I see a lot of complaints from people who didn't even test one Core Ultra 200 CPU. As much as I hated 13/14th gen, then the new one is pretty good. I wouldn't decide to put it into my daily PC if it was bad (at least today I feel it's good ;) ). Another thing is that I had a Z890 ITX mobo after review, and no one wants to buy them anyway. I would post something in classifieds as I have 2-3 more mobos that I don't need, but shipping out of the EU costs too much to be worth it (literally ~$100 shipping only).
 
I find any CPU above 200W a mistake for a home PC. If popular coolers can't handle them, then something isn't right. Ryzens generally stop before 200W (170-180W for 16 cores without OC). X3D is pure magic with that gaming performance at ~120W. The 265K has a maximum TDP of 250W, but in motherboard reviews, I could barely see it passing 190W, and in games, it's ~150-170W max. I have no idea how Intel calculated it, given its high TDP in general specs. If you unlock limits, then it can pass 250W, but it's possible to keep it at lower wattage and still overclocked to the max (which is ~5.6GHz P/~5GHz E cores).
There's multiple things going on above so can we be clearer what exactly are we looking at? Do Intel even list it as TDP any more? I think that is replaced by Processor Base Power (PBP = PL1?) but otherwise is the same definition. Either way, it is not power consumption! Max Turbo Power (MTP = PL2?) is the highest sustained value you expect to see. Still, these are power limits. Want to put in a "200W" cooler, you can, and it'll limit. While enthusiasts might cry at that thought, that's how pretty much mass market systems run, with a lower sustained power limit. Cooler is minimally PBP, not MTP. Historically on Intel CPUs, highest power is seen when running workloads similar to Prime95 or Linpack. This doesn't apply to AMD in my experience and other workloads may be worse case.

Also while AMD has great efficiency, you don't usually see massive power readings from their CPUs since they actually enforced a "stock" power limit, which Intel didn't start doing until the Raptor Lake thing. Even now, Intel remains more generous than AMD. Look at the 65W to 105W allowance AMD made for Zen 5.
 
Whatever you call it, CU 7 265k's max power is 250W at default (PL2), but it never goes above 200W at auto/stock/default Intel profile (as I remember, ED said that the 285K runs at up to ~220W, but the PL2 is the same 250W). Last gen i7/i9s had 253W (PL2), and were going up to ~253W or slightly more at the default Intel profile. AMD still calls it TDP and runs a bit lower or higher, depending on the CPU model.
A summary as simple as a lower wattage is good, and Core Ultra runs at a lower than declared maximum power.
 
I just switched 113 to 114 microcode, and there is no difference in Cinebench 2024. The same limits, temps, core utilization ... everything I see in hwinfo64 is the same on both versions. Below is one of the screenshots as the second one would look the same. It's more about showing temps and wattage. The cooler is a mentioned custom loop with a single slim 240mm rad. The i7 14700K would hit 95C and throttle in the same test, which is a huge improvement over the last gen for the same positioned product.


cu265k.jpg
 
Look forward to any results for more potentially latency sensitive things. Which I'm not aware of any modern ones other than gaming? Or maybe I just don't do latency benchmarks!
 
Try Linpack Xtreme v1.15 that is the only way I can get big power numbers out of Ryzen. Higher than P95..

I run the 10GB load, option 5..
 
I'm not sure if it's a bug or something else, but hwinfo64 reads that E-cores are power-limited. However, they don't throttle and are not frequency-limited. The same is when power limits are unlocked. At first I thought it's regarding all cores, but when the list is expanded, it shows only E-cores.

Power delivery and additional options are very different from those of last-generation Intels. For example, CPU Input Voltage (as described in ASRock BIOS), when too low, will cause CPU throttle. When too high, it may cause overheating when, at the same time, power limits are unlocked. There are also new voltages and new naming, like VDD2, which is a memory controller voltage, and it's not even described this way.

I expect many users will have problems with all the new options and required voltages. On the other hand, for a CPU like the 265K, 5.4-5.5GHz on all P-cores is easy with almost everything at auto and on average quality chips. 5.6GHz is already problematic, and can count it will work on above-average CPUs. 5.7GHz will make only top ~5% CPUs. Cooling doesn't help much (assuming we are still on ambient temps and not sub 0).
 

Info drop on perf. Reading it myself now.

INTEL-ARROW-LAKE-UPDATE.jpg

Edit: the tldr is Intel screwed up what they sent to reviewers. A Windows package was scheduled to be available for retail, not helpful if you're reviewing before. They also say settings in pre-release bios were not well controlled so various settings may have been less than optimal. It seems that if you're running Win11 24H2 and up to date on patches you'll be good. The other part is bios with microcode 0x114 for now. In "first half Jan 2025" there will be a further bios update with the CSME version above with additional improvements.

I guess now we wait and see how ARL performs vs AMD.
 
Last edited:
I don't think they solved some of the listed issues. For example, the first one still happens on Win 11 26100.2605, microcode 0x114, and the latest firmware kit, and they said it was fixed some updates ago.
I have no stability issues, and the performance has been pretty good since I got the CPU on the day of the premiere. However, I have seen no performance improvements since then, and I switched 5 microcodes and various other things.
 
For example, the first one still happens on Win 11 26100.2605, microcode 0x114, and the latest firmware kit, and they said it was fixed some updates ago.
What do you mean by "first one"? The whole top line of the table? If I'm reading it right, a retail buyer running updated 24H2 with a bios released at or after launch should not be affected. Some of those are arguably a bit subjective too.

The list seem to primarily address the unexpected performance seen by pre-release reviewers. When reviewers get around to retesting, the change will be the most interesting. Especially if gaming isn't as bad as launch reviews make out, that could help Intel out going forwards. Reviewers might have testing fatigue by the time the Jan release comes out, and I wonder if some will wait for that rather than test what we have today.
 
Last edited:
What do you mean by "first one"? The whole top line of the table? If I'm reading it right, a retail buyer running updated 24H2 with a bios released at or after launch should not be affected. Some of those are arguably a bit subjective too.

The list seem to primarily address the unexpected performance seen by pre-release reviewers. When reviewers get around to retesting, the change will be the most interesting. Especially if gaming isn't as bad as launch reviews make out, that could help Intel out going forwards. Reviewers might have testing fatigue by the time the Jan release comes out, and I wonder if some will wait for that rather than test what we have today.

As I said, it still happens on my test rigs with all the requirements they mentioned. If I enable the power saver power plan, I get ~30% of the expected performance in tests like Cinebench. It still works fine using the performance power plan. However, it worked fine even on earlier Windows versions and older microcodes. I didn't notice it at first, as tests require the performance plan to compare it right, and I kept it this way.
You have yesterday's Cinebench 2024 result (post #145), which I made with the latest BIOS, microcode, ME, and Windows Updates. It's the same result (performance-wise) as with the BIOS, microcode, ME, and Windows updates from early November (any Z890 mobo review on the front page, except for the Z890I Nova, and Z890 OCF will be published soon).

Maybe they fixed it compared to the pre-release, but most reviewers already got the somehow fixed version. So only "leakers" who work with manufacturers could really show any problems based on old versions that shouldn't even be shown in public. All motherboards that I tested arrived with a BIOS after some improvements, and some were shipped ~2 weeks before the premiere. That BIOS was prepared ~3 weeks before I got the motherboards. There were already 2-3 newer official versions the day I started tests (I had to wait for the CPU, even though mobos were already waiting for about 2 weeks).
Comparing it to the pre-release versions is stupid, as no respectable website would present results on a two-month-old BIOS on the premiere date unless it wanted to show how bad the new Intel is. Also, in stores, there were no motherboards with old BIOS that could include some of the listed issues (Gear modes, ReBAR, and more things that couldn't work). Since I got all ASRock, Gigabyte and even Biostar with newer microcode (EarthDog got his mobos with the same BIOS versions), other reviewers couldn't get their samples with very old BIOS. So we are talking not about official press samples but early pre-release samples that could show problems with early BIOS/mc/Windows updates.
I'm not saying I'm 100% right here. I'm saying how I see it based on what I tested and my contacts with various vendors. I also have info from other brands, not only motherboard manufacturers, like how and when some RAM manufacturers received their mobo samples for internal CUDIMM validation.
 
Maybe they fixed it compared to the pre-release, but most reviewers already got the somehow fixed version. So only "leakers" who work with manufacturers could really show any problems based on old versions that shouldn't even be shown in public.
I'm trying to make sense of this. Intel said the power module was not pushed until retail date. That means all seeded reviewers before launch would have been missing that, since they got early access. Also mobo bios on release date would almost certainly be out of date too. To have it ready to ship to customers it would have been manufactured weeks in advance. Any updates since then would have to be applied. It isn't clear what was the first "reasonable" bios, and was that made available to reviewers by the mobo manufacturers in time? From Intel's version, that might not have been the case.

I'm not suggesting any site retest using old versions. The ones that had early access still have their original testing data. That can easily be compared to testing today, with latest bios and other updates. That's what is interesting to me. How much did Intel shoot themselves in the foot by messing up the launch reviews? Is it enough to improve the perception for gaming?
 
I'm trying to make sense of this. Intel said the power module was not pushed until retail date. That means all seeded reviewers before launch would have been missing that, since they got early access. Also mobo bios on release date would almost certainly be out of date too. To have it ready to ship to customers it would have been manufactured weeks in advance. Any updates since then would have to be applied. It isn't clear what was the first "reasonable" bios, and was that made available to reviewers by the mobo manufacturers in time? From Intel's version, that might not have been the case.

I'm not suggesting any site retest using old versions. The ones that had early access still have their original testing data. That can easily be compared to testing today, with latest bios and other updates. That's what is interesting to me. How much did Intel shoot themselves in the foot by messing up the launch reviews? Is it enough to improve the perception for gaming?

Intel couldn't release the full BIOS/ME package at the retail date, as I had reviewed samples before the release date. As I remember, they had a 110/112 microcode and a complete ME package. That BIOS version was on the ASRock website before I got my samples (nearly all ASRock mobos had 2.08.AS BIOS).
ASRock delivered motherboards with the latest available BIOS (the whole first wave, Taichi, Taichi Lite, Riptide, Steel Legend, ...). Only the Z890I Nova and Z890 OCF were shipped some days later, and when I got them, I updated the BIOS to a newer one with the microcode 113. ASRock flashed all review samples just before the shipping - early October. ASRock also lists the ME/firmware update (so the whole power management package) for the beginning of October (Z890I Nova has it on the 4th of October). As far as I remember, the release date was the 24th of October. The first BIOS release is dated the 23rd of September, but there is a large gap between the 1.31 version and the next official one - 2.06, so the 2.06 or 2.08 could get the whole working package (first week of October). This I'm not sure. I can only confirm that review samples were shipped with the 2.08 version in early October and that the whole ME/firmware was working fine.

The Gigabyte Z890 Master was shipped even a few days before ASRock, and it also had the whole working BIOS, ME, and other things, but it had an older microcode - 110. This is a screenshot from my review. Tests started about October 31st because I had to wait for a CPU, but the mobo was waiting for two weeks.

Gigabyte_Z890_Master_BIOS31.jpg


All the info we get from Intel and various online sources is misleading. It looks like Intel's management and marketing wanted to push the premiere and made a lot of mess with non-release-ready hardware and software support. Hardware without software is useless, and their cooperation with Microsoft failed, too. Now, all are trying to cover their failed premiere.

It doesn't matter when they released what. What matters is that they claim that something was fixed, but it's not.
My ITX PC has run for three days without issues with the latest BIOS with microcode 114. However, I can't see any difference from the earlier two releases. I can also compare my early results of ASRock Z890 Taichi Lite or Gigabyte Z890 Master, and all of them are +/- 1% in CPU tests and rendering like Cinebench. I still have no time to compare the 114 microcode in 3D.
 
It doesn't matter when they released what. What matters is that they claim that something was fixed, but it's not.
Another case of we're looking at things differently. From my personal perspective, what I want to know is how does ARL perform when it is working as expected. If the "fix" didn't, then others will also see it and Intel can have another go at it, but I'm not so interested in that.

As far the bios, maybe it isn't the microcode version that is at fault, but the bios default settings around it. I'd expect this update and/or the January one to trigger rounds of retesting. See where we're at then.
 
I mean, I don't care what they do; they should provide a working solution, not claim that they did something when there is no difference. I'm not complaining about the current performance, but I'm curious how it will work when they actually correct all issues.

There is a weird behavior while changing power settings. For example when I leave CPU input voltage at auto, then I see it goes up to 1.5V+, when in real, it could stay at ~1.35V. At the same time, when the input voltage is higher, then VCore goes up. This is something I haven't seen before. When I manually set both at 1.35V, then VCore is ~1.3V, and input is ~1.35V. When I set VCore at ~1.3V and input at ~1.5V, then VCore readings are about 1.45V. There are other weird things, like many settings depending on others.

I guess we'll have to wait a few weeks to find out if some things are Intel's design or just problems. Memory latency is weirdly low and I hope they fix it somehow. AIDA64 shows 15-20ns worse results at about the same settings, compared to Intel's 13/14th gen.

Edit:
Two days ago, ASRock released 2.23.AS01 beta BIOS with 114 microcode for the Z890I Nova. Today, they have already released 2.23.AS02. Maybe they improved something.
 
Last edited:

TechPowerUp tested several scenarios:
  • 23H2 / Oct 24: This is a configuration that roughly matches what we originally reviewed Arrow Lake with (except for a slightly newer BIOS that came out in late October)
  • 23H2 / Dec 24: This is the previous config, but with Windows Update fully updates as of yesterday.
  • 23H2 / Dec 24 / Newest BIOS: Next, we updated the BIOS on the Z890 Hero to 1101, which was release on November 29 and uses microcode 0x113. All settings in BIOS are identical.
  • 24H2 / Fresh Install: Now we installed Windows 11 24H2, a fresh installation (not an upgrade from 23H2). The motherboard BIOS version remains at the newest 1101. We didn't allow Windows Update to install all the updates and started benching instead.
  • 24H2 / Dec Patches: Now we ran Windows Update several times with reboots, to ensure we have the latest updates, up to yesterday, which includes the KB mentioned by Intel.
Skimming the results, some tests didn't show much difference between any of them. Some show un-updated 24H2 to fall notably behind. The "new" bios with microcode 0x113 didn't seem to do much of anything. Isn't 0x114 available now, just without the other update expected in Jan?

Sounds like another swing and miss from Intel.
 
Isn't 0x114 available now, just without the other update expected in Jan?

114 is available for about a week in BIOS published on product websites (beta or not, somehow official releases). ASRock already released 2-3 beta versions with 114 microcode. Maybe today I will check the second BIOS for the Z890I Nova, but I don't expect it to run better.
I expect a whole BIOS package with a new microcode in January, as it comes with all updated components. ME is not provided separately, even though you can update it separately. There is an option to flash the whole BIOS with ME or only ME ... but it's in the BIOS package.

The TPU article pretty much confirms what I said. There is no improvement. I wonder why performance gains up to 7% were mentioned in some news about two weeks ago when there is +/- 0. Either way, we will wait until January, and see what upcoming patches bring.
 
Sorry guys I'm looking for enlightenment before purchasing the 245k.
Basically I read some news from last week and Intel China announces a microcode and various updates that greatly increase the performance of the new Arrow Lakes, today a guy on Twitter wrote that even with the new microcode the performance gets worse, there's a lot confusion, can anyone tell me some truth about this? thanks in advance
 
The "final" fix isn't out yet, and the interim fix is quite the mixed bag although I don't recall any reports of it getting worse than before. Look at testing once that is available and decide then.
 
Back