Leaning to Arrow Lake?! (need 2nd opinion)

Hardware Hub Build a PC
1

Hi all, new here, hope I’m in the right place. :slight_smile: I’m currently using a very old Intel system and after following tech advancements for [too many] years, I figured I’m overdue for a new system. I want to get back into software development, experimenting with heavier CPU workloads, and play some more demanding games, for once!

It appears that in that time, AMD has left Intel in the dust when it comes to CPUs. I have therefore considered building a 9700X or 9900X based system. However, after looking into the platform/motherboard differences between the two, I wonder if Intel has more advantages overall, when not considering just gaming? (If I was building this mainly for gaming, I would just get the 9800X3D!) This is where I’d appreciate getting a 2nd opinion from the experts here. :smile:

Here’s my reasoning in more detail, in comparing 265K ($560) vs 9900X ($570 on sale):
Performance:
Applications: Tie. Gaming: 9900X +3.3%. (Ref: 3dcenter - Launch Analysis Intel Arrow Lake, Page 2)
Power consumption:
Applications: 265K @ 128W vs 9900X @ 135W. Gaming: 265K @ 80W vs 9900X @ 105W. (Ref: 3dcenter - Launch Analysis Intel Arrow Lake, Page 3)
So, based on comparable performance and efficiency, the 265K is actually a decent CPU? (Yet reviews and online discourse in most places seems to be overwhelmingly negative?)

Then there are I/O advantages:
In summary, Arrow Lake Z890 appears to have a massive advantage.

Details

Arrow Lake gives you 2 TB4 ports. X870 gives you a USB4 port based on a Realtek chipset (likely a lower quality solution.)
For example, I’m considering the following two MSI Pro boards, as I love their style and both have 5 GbE built in:

PRO Z890-P:
PCIe slots
3x PCI-E x16 slot
1x PCI-E x1 slot
PCI_E1 Gen PCIe 5.0 supports up to x16 (From CPU)
PCI_E2 Gen PCIe 4.0 supports up to x1 (From Chipset)
PCI_E3 Gen PCIe 4.0 supports up to x4 (From Chipset)
PCI_E4 Gen PCIe 4.0 supports up to x4 (From Chipset)

NVMe slots
4x M.2
M.2_1 Source (From CPU) supports up to PCIe 5.0 x4 , supports 2280/2260 devices
M.2_2 Source (From CPU) supports up to PCIe 4.0 x4 , supports 22110/2280/2260 devices
M.2_3 Source (From Chipset) supports up to PCIe 4.0 x4 , supports 2280/2260/2242 devices
M.2_4 Source (From Chipset) supports up to PCIe 4.0 x4 / SATA mode, supports 2280/2260/2242 devices

PRO X870-P:
4x PCI-E x16 slot
PCI_E1 Gen PCIe 5.0 supports up to x16 (From CPU)
PCI_E2 Gen PCIe 3.0 supports up to x1 (From Chipset) :frowning:
PCI_E3 Gen PCIe 3.0 supports up to x1 (From Chipset) :frowning:
PCI_E4 Gen PCIe 4.0 supports up to x4 (From Chipset) (x2 if using SSD in M2_3) :frowning:

M.2 slots:
3x M.2
M.2_1 Source (From CPU) supports up to PCIe 5.0 x4 , supports 2280/2260 devices
M.2_2 Source (From Chipset) supports up to PCIe 4.0 x4 , supports 22110/2280 devices
M.2_3 Source (From Chipset) supports up to PCIe 4.0 x2 , supports 2280/2260 devices :frowning:

And I do have a couple high end Gen 4 SSDs I’d like to put to good use… maybe mirror them with ZFS or Linux RAID? :smiley:
I understand X870E boards have better I/O, but these are in a different price category and get expensive fast.

Arrow Lake is commonly referred to as a disaster, but from what I can tell, it seems… fine? Pretty good even, on balance - unless you’re only interested in gaming. And Intel promised to improve gaming with upcoming firmware/software/OS patches - if they can deliver on that, then even better.

Am I missing something critical here?

Oh and one more question, regarding memory choice for a potential 265K build:
I’m not into memory overclocking, but if a XMP kit works 100% stable out of the box, I’m willing to pay a bit of a premium to get that. But not sure what to look for, what the sweet spot is as far as cost and stability. Phoronix did some benchmarks and found that 2 x 32GB DDR5-6400 Corsair CMK64GX5M2B6400C32 ($280) worked well, even faster than another DDR5-8000 kit (go figure?!) so I’m considering that. Is that reasonable, or should I be looking at a different kit?
(Ref: Phoronix - Intel Core Ultra 9 285K Linux Memory DDR5 Performance Testing)

2

I feel that after 13& 14 th gen CPUs flying to close to the sun. it was handled poorly untill it became a class action. ( in India people still buy 13gen intel without updating the bios because normiee’s don’t know about the issues and many System Integrators only care about commissions here.)

I have personally dealt with a few relatives who are struggling to get RMA or lost work due to crashes.

unless some cloud provider/ thirdparty releases a long term reliability report of these CPUs running stable I don’t think people are lining up get one when AMDs X3D stuff is topping charts and getting all that hype and still boasting good enough power efficiency.

1 Like
3

I think these CPU work great for that. For the stuff I use it for, like light gaming but mostly programming and compilation, it is actually really good.

Currently running a 285K Arrow Lake now and it does indeed seem… fine.
Just dont dare enable the iGPU on Windows, when the driver downloads and installs silently in the background it will hard reset without a bluescreen message :pensive:

Luckily I dont use Windows very often.

Running two kits of Corsair Dominator Titanium DDR5-6400 ( CMP32GX5M2B6400C32W ), The XMP Profile 1 works just fine, however the second profile doesn’t work.

I wouldn’t buy a product based on future promises, I bought the 285K because I thought it was a intriguing product and not because I thought it was good. I also could have bought anything new and it would be faster than my 9900K.

You could adopt a wait and see mentality to see if some future microcode updates fixes the performance issues, then you might be able to get them at a lower price point too.

That happens when your product has performance regressions relative to your previous gen. Though it might have been unavoidable since it is one of their first tiled designs for the consumer market.

Since it is a tiled design, tile to tile communication will cause latency and that hurts gaming performance the most. AMD has had a few years to refine this, but it’s clear Intel hasn’t.

If you dont really care about min-maxing every little bit of performance you get for your money, I think you are gonna be happy regardless.

2 Likes
4

Most modern CPUs are “fine” but if every tech review was “yeah the CPU is fine, you’ll barely notice any difference outside of max frame rate gaming at 1080p low details” they wouldn’t get any clicks and therefore no ad revenue. Take most review sites with a pinch of salt, they’re mostly hyper-gaming focused.

Maybe wait for @wendell to do an arrow lake review if he hasn’t already? likely to be more productivity focused I would wager.

Far more important imho for the average person, especially for development is platform stability. And in that respect I’d be careful to look into Arrow lake with regards to:

  • kernel stability/support
  • platform/chipset features/support
  • how the big.LITTLE style heterogenous core layout works in your specific applications

AMD has uniform same-size cores with the same instruction set across them and at this point is a slightly more mature platform than arrow-lake. I’m not sure whether that’s still a concern with non-uniform cores under linux at this point; windows is still a shit show.

I’d consider those things to be more important/worth looking into more than benchmark results.

2c.

Pretty much that.

I hear some arrow lake boards support thunderbolt 5 - which could be useful for you if you’re planning to deal with external SSDs. 120 Gigabit vs. 40 on previous versions.

Either system will be a massive upgrade from what you’re probably running right now.

1 Like
5

On paper, the 265K(F) makes a ton of sense. It’s cheaper than the 9700X while providing pretty much the same performance, and Z890 is definitely cheaper than X870 boards.

Two minor drawbacks do exist, if the past is any indicator the LGA1851 platform will most probably not have a decent upgrade path (and if it does, it will possibly be an Arrow lake refresh), while AM5 will have at least Zen 6, possibly Zen 7.

The second drawback is that it draws more power than the 9700X.

Neither has to be important, but it is there. For value builds I can definitely see the appeal, especially with the 265KF. :slight_smile:

6

This is not confirmed at all. Possibly zen 6, mayyybe but unlikely zen 7 IMO.

While rendering or so yes. If you calculate idle or light work AMD power consumption is through the roof. My AM5 system takes 150w to play a YouTube video, 120w idle.

2 Likes
7

According to the latest information AM5 will be supported with new CPUs until, at the very least, 2027. This all but guarantee Zen 6 support, though as you say, not confirmed yet.

AM4 received new CPUs this year (GT line) so…

8

Yes, that’s the thing. AM4 seems like it’s still supported, but no new architectures since a couple years.

It will all hinge on DDR6. Historically AMD has not brought architectures to multiple sockets, nor supported multiple RAM generations on one architecture. Zen 6 is probably still DDR5, zen 7 very likely DDR6 and on a new socket.

9

Yeah, probably. Bit early though as Arrow’s still in beta and it remains to be seen what Intel’s able to deliver for fixes. Not that Zen 5 wasn’t also a beta launch but, since it’s been longer, it’s more clear what AMD’s offering.

3dcenter’s metareview may understate Arrow’s power consumption a bit due to some reviewers reporting only EPS power on Asus boards where the CPU partly pulls 24-pin. If anything, it’s IMO a little disappointing Intel didn’t get more out of finally ditching Intel7 and leapfrogging Zen 5 (N4P) with TSMC N3B. Arrow’s idle power’s ~10 W lower than Zen 3, 4, and 5, though, possibly due to being on Foveros rather than an organic substrate.

Intel’s also worsened their financial position by losing their N3B discount, which I suspect inclines Intel to look at Arrow as more of a stopgap gen than it already was. So far as I know the company’s avoiding saying anything about LGA1851’s lifespan and there was already speculation it’d be a single gen socket.

Zx90 lane structures and pricing are indeed nice and, yeah, AMD kind of owned themselves by changing X870 to single Promontory 21 from X670’s dual. Nearly all current gen mobos, whether AMD or Intel, compromise the second CPU NVMe’s thermals by placing it under a dGPU. DMI 4x8 should get around that for IO intensive tasks plus, in addition to the uplink narrowing to 4x4, there’s a ~1 GB/s NVMe penalty for going through Promontory 21 that doesn’t happen on Z chipsets. However, I don’t know x8 uplink tests on the Zs.

For now it seems most likely my next build’ll be Zen 5 anyways, though. We’ve enough AVX-512 enabled workloads Arrow’s not appearing particularly competitive. I share some of @Tanmay’s concerns over how well Intel might have ported their power and voltage management from 20A to N3B+N5P+N6 and, as @thro noted, Windows 11 is a shitshow.

The ASM4242 is ASMedia, not Realtek. I’m not aware of comparative data but ASMedia’s other execution’s pretty solid, so a tie with or better than Intel’s plausible. An X870(E) constraint is the ASM4242 takes four CPU PCIe lanes, which I don’t see as that big of a deal generally, but may be significant to some builds.

For NICs, Realtek’s been at parity or better than Intel for a while. But some people still assign a premium to Intel regardless.

Wherever you’re looking for prices, probably good to look somewhere else. I paid US$ 450 for a 9900X in August and it’s been US$ 383 (before tax) for a bit. The 265K is ~$390.

Normal for gear 1 versus gear 2, same for Zen 5. But the marketing works because people assume bigger number better.

Intel seems unwilling to confirm or deny whether their their 6400 support encompasses 1DPC 2R. Absent something official I’d assume it doesn’t, that CMK64GX5M2B6400C32 is thus an overclock, and that the XMP’s therefore not guaranteed stable. Likely it would be stable and, if it isn’t, the DDR’s quite unlikely to be the limiting factor.

You Probably Won’t Buy Intel’s New Ultra 9 285k CPU

Wendell hasn’t done the 265K or 245K, perhaps because the analysis would be fairly similar.

2 Likes
10

I can’t speak too much to the 265KF as my 265K has been something of a lemon, but my experience with the 285K and 245K has been very good. Intel definitely doesn’t hold any sort of performance lead in gaming, but the productivity performance is great depending on the application.

I’d double check if the application you are interested in gets confused by the big.little architecture (as thro pointed out) but also if it is AVX-reliant. On Windows, I haven’t had too many issues w/ the cores as long as high performance power profiles were configured, but I don’t do too much on linux. Historically, Linux support has been better than Windows, though. For AVX, AMD hold a big advantage.

In terms of efficiency, I think what confuses people is that both AMD and Intle made claims as to gains this generation, but did so in different areas. Intel is not too much more efficient at heavy workloads, as the TDP remained constant and overall perf gains were low. AMD gained a lot of efficiency at those same workloads, as they dropped tdp while maintaining perf. At medium/light workloads, AMD maintained average power draw, essentially, while increasing perf slightly, while Intel reduced average power draw substantially while maintaining perf.

As to memory, I have had no issues with a similar dual-rank 6400 kit, but based on the last Intel spec I saw, 6400 with CUDIMMs is only “officially” supported on boards with 2 slots, so 6400 is basically always going to be an “OC”. I expect most mobos will ignore this and, frankly, think most users should as well.

2 Likes
11

Re-badging the same product doesn’t really count though.

1 Like
12

Yeah, ITX boards or Gigabyte Tachyon. Plus the slowest CUDIMMs I can find are 8000 with most being 8400+. And they’re all 1R.

13

Thanks everyone for sharing your views. :slight_smile: It gives me some things to think about.

Have you tried the new firmware? Supposedly some iGPU/dGPU conflict issue is fixed.

Oh, def not! I see it as more of a bonus if it happens.

I think it’s mostly fine on linux. Though there are occasional oddities like in the Phoronix browser test of the 285K where it scores last by a huge margin but the CPU is also drawing almost no power. Seems like the benchmark was running on 1 e-core. From Wendell’s linux review and Phoronix’s in general, however, it sounds like overall the linux experience is OK.

Fair, though for me… I will likely use this thing for the next 5+ years again, so probably would want to build a whole new system by that point anyway, like I am doing now.

Good to know. I’m a bit iffy because it’s not on the board’s QVL. Do you think picking something from the board’s QVL is important at 6400 CL32? Or is that a safe bet for Arrow Lake? Or too soon to tell (as not many people have been buying/testing/sharing feedback on these?)

14

My suspicion is that 6400 should be fine regardless of QVL, but it is too soon for me to tell (granted, I haven’t looked into it a lot).

I’ll also note, since it has come up, that I have yet to experience any serious issues with the iGPU itself/drivers in Windows. I don’t doubt other reviewers’ issues, but it certainly seems like it depended on how you installed drivers, and an issue that’s mostly been addressed by now. Not a great launch experience though—Intel dropped the ball pretty bad overall.

15

First, welcome! If you are building for gaming only, I did this on the cheap last year. 7600x 6core 12 thread, 7800xt 16gig, all on a MSI B650m Mortar wifi board, case and all was under $1500 and she will play most AAA titles in ultra settings.