News Intel's lackluster Arrow Lake appears to have a refresh inbound — Arrow Lake Refresh appears in reference document

Toggle sidebar Toggle sidebar
Will we get a better gaming experience the second time around?
Click to expand...
Nothing interesting will happen on the CPU side. Even the long dead rumor of 8+32 cores wouldn't help gaming performance. I guess 1-4% gains from clocks would be helpful for comparisons.

On the GPU/NPU side, they could improve it, but probably won't in a substantial way.
 
  • Like
Reactions: jp7189 and bit_user
I sincerely hope this starts a new attitude at Intel that maybe their sockets and platforms might be useful for longer than just the two cycle.

Being not in the strongest market position right now, having their sockets go over more generations would be a show of good-will and improved customer relations.

There are absolutely no downsides for OEM builders with this, especially since they are often times on autopilot with their business purchases anyways. But this helps those of us who do custom builds and are the mouthpieces on the streets generating buzz with word of mouth praise.
 
If this had been like Raptor Lake was to Alder Lake, it could be a real hit. There are some great cores, inside of Arrow Lake, that really want to shine. Sadly, I think we're looking at like a Gen 14 Raptor Lake vs. Gen 13 situation, here.

No matter what, I'm planning on waiting until Nova Lake. If I buy anything Intel, this year, it'll be a B770 dGPU.
 
  • Like
Reactions: closs.sebastien, bolweval, TheSecondPower and 2 others
>>Nothing interesting will happen on the CPU side. Even the long dead rumor of 8+32 cores wouldn't help gaming performance.

>I sincerely hope this starts a new attitude at Intel that maybe their sockets and platforms might be useful for longer than just the two cycle.

>Being not in the strongest market position right now, having their sockets go over more generations would be a show of good-will and improved customer relations.


Intel doesn't give a diddly about socketed CPUs for desktop. Its turnaround doesn't depend on the tiny minority of DIY gamers. That said, it's a refresh, so of course it'll use the same socket.

It's fine for wanting things to help further your hobby, which presumably is PC gaming. But understand where you are in the overall picture, which is one minnow in a pretty big ocean. Intel has much bigger fish to fry than to worry about what some gamers on some forum think.

Ditto on the GPU side. Nvidia cares about staying on top (of AI), AMD cares about chasing after Nvidia, and Intel cares about just staying afloat. Gaming is an afterthought--and even less than that for Intel.

Given the AI wave, one would think that having NPUs to satisfy MS' "40 TOPS" requirement would be an add, but apparently not. Neither the mobile parts intro'ed at CES'25 (200U/H/HX) have sufficient NPUs, nor do the upcoming desktop refresh. It'll be up to Panther Lake in late '25 to improve NPU. This is a good indicator of MS' pace of AI inclusion for Windows, which looks like the big push will be in '26 with Win12.

https://www.pcmag.com/news/ces-2025-intels-2025-arrow-lake-core-ultra-chips-for-laptops-prioritize
 
  • Like
Reactions: closs.sebastien
baboma said:
It's fine for wanting things to help further your hobby, which presumably is PC gaming. But understand where you are in the overall picture, which is one minnow in a pretty big ocean. Intel has much bigger fish to fry than to worry about what some gamers on some forum think.
Click to expand...
I mentioned gaming because it's in the Tom's Hardware article subhead: "Will we get a better gaming experience the second time around?" The answer is no, unless +1-4% counts.

There were rumors months ago of an 8+32 core Arrow Lake Refresh, which would have improved multi-threaded performance, but further rumors asserted it was cancelled. Things get tested behind the scenes that won't make it to market, like Meteor Lake on desktop.

Arrow Lake has moved to a chiplet design, which in theory could allow Intel to update the NPU independently from the CPU cores, for example. However, anything involving differently sized chiplets is a lot more work than bumping up clocks by 100-200 MHz and calling it a day.

I'll give Intel this: the Arrow Lake-S iGPU is substantially faster than the Radeon 610M found in Ryzen 7000/9000 desktop CPUs. Because of the improved iGPU and the efficiency improvements, the Arrow Lake office PCs will be nice purchases when they hit the used/refurbished market.
 
  • Like
Reactions: bit_user
>A gigantic chunk of office PCs are socketed...

Yes. My response was in context of DIY'ers wanting socket longevity.

Which isn't even a DIY consensus. I'm a DIY'er and I replace my (desktop) PCs only after 5+ years. Socket longevity is irrelevant.

>(Yeah I know there are a lot of miniPCs and handhelds and stuff like that, still)

Speaking of mini-PCs, specifically NUC-sized ones, that was on my "future PC" list. But with the 30% tariff going strong, I'll just have to wait out Trump. He's older than I am, so chances are good I can outlast him.


>I mentioned gaming because it's in the Tom's Hardware article subhead: "Will we get a better gaming experience the second time around?" The answer is no, unless +1-4% counts.

Re gaming CPUs, I'm of the school of thought that you buy a "good enough" CPU to not bottleneck the GPU, which only equates to "best gaming CPU" when you already have a 4090/5090, which I (or most) will never have. So "best gaming CPU" to me, and for most, is just more hype to sell wares. I don't pay any attention to it, just as I don't pay attention to RGB doodads or see-through tower cases.

>There were rumors months ago of an 8+32 core Arrow Lake Refresh, which would have improved multi-threaded performance, but further rumors asserted it was cancelled.

Which would make sense. Avg consumers don't need more desktop CPU power. Even for gaming, more cores aren't a help.

>I'll give Intel this: the Arrow Lake-S iGPU is substantially faster than the Radeon 610M found in Ryzen 7000/9000 desktop CPUs.

Yeah, better iGPU was the one thing that interested me about ARL. My gaming needs are light enough that ARL in a NUC (sans dGPU) may have sufficed. But my upgrade cycle will be after that, so I'm looking at Nova Lake & successors for new features. AI is looking to be the main feature add going forward. IMO, any gaming improvement will be incidental, at least for Intel, and at least for the next 3-5 years.
 
  • Like
Reactions: usertests
These most likely are going to exist as an OEM centered refresh. I'd be surprised if anything fundamentally changed. There's no room for meaningful clock scaling on N3B with the design so I'm not even sure a refresh would be as good an improvement as RPL to RPL-R was.

The only possibilities I see for additional performance are a redesigned SoC tile and/or using an Intel node for the Compute tile. Of those two the only one I think might be a remote possibility is SoC tile because it may be possible to use the same one as PTL will if the rumors regarding the memory controller being moved to the Compute tile are wrong.

I would love for my cynicism to be wrong, but I feel like if something was really coming there would be indications either official or leaks.
 
thestryker said:
There's no room for meaningful clock scaling on N3B with the design so I'm not even sure a refresh would be as good an improvement as RPL to RPL-R was.

The only possibilities I see for additional performance are a redesigned SoC tile and/or using an Intel node for the Compute tile.
Click to expand...
I think Intel can almost always find another 100-200 MHz if they look for it. 285K (P-core) is boosting to 5.7 GHz, 265K to 5.5 GHz, 245K to 5.2 GHz. At least the lower parts can inch up.

NPU could be boosted from 13 to 16 TOPS to match Hawk Point desktop APUs, as if that's a fight that matters.

I'll throw out this wild card: They could put a 6P+12E 250/255K in the gap between the 6P+8E 245K and 8P+12E 265K. From a yield perspective that product could make more sense than a 245K.

But if they did redesign tiles, that would be much more interesting than any of this.
 
At best, you'll see 50TOPS NPU, a 100MHz clock boost, all limited to K series IMO. They can't redesign the chip for a refresh. You'll have to wait for Nova Lake. Honestly, I'd prefer they did Panther Lake S which supposedly puts the memory controller back on the cpu tile and improves the ringbus.
 
  • Like
Reactions: bit_user
usertests said:
I think Intel can almost always find another 100-200 MHz if they look for it. 285K (P-core) is boosting to 5.7 GHz, 265K to 5.5 GHz, 245K to 5.2 GHz. At least the lower parts can inch up.
Click to expand...
You've nailed the problem. I'm sure they could improve the 265K/245K bins, but there's just not enough room on the 285K. Skatterbencher was able to get their 285K to 5.8Ghz and this is the start of their conclusion:
The Core Ultra 9 285K is Intel’s top of the line Arrow Lake processor and it certainly feels they squeezed the chip for all it has. It’s one of the most difficult chips I’ve ever had the pleasure of tuning.

Especially overclocking the P-cores is like trying to squeeze blood out of a stone. It doesn’t seem to scale with voltage or have any headroom with the default curve. All that’s possible is getting the non-favored cores to run close to 5.7 GHz and lift the all-core frequency as high as possible.
Click to expand...
 
  • Like
Reactions: usertests
bit_user said:
I think that wouldn't count as a "refresh".
Click to expand...
Generally speaking I certainly agree. At the same time what's the point of tiles (in client) if not design flexibility?

So far I think the only tile that has been used across multiple Compute architectures is the Graphics though. I would love to see a die shot of ADL mobile to see if it has a different Compute tile to that of desktop. I know every other tile on it is different to the desktop part and wouldn't be surprised if that was too since maximum configuration is 6p/8e.
 
thestryker said:
Generally speaking I certainly agree. At the same time what's the point of tiles (in client) if not design flexibility?
Click to expand...
But in moving the memory controller, you're talking about refactoring the tile architecture, not just swapping out one tile for an improved alternative. That's no longer just a refresh. If they'd done that, it would be a new Lake and prominently featured on their roadmap. Not a "refresh" they've been waffling over even doing, at all.
 
bit_user said:
But in moving the memory controller, you're talking about refactoring the tile architecture, not just swapping out one tile for an improved alternative. That's no longer just a refresh. If they'd done that, it would be a new Lake and prominently featured on their roadmap. Not a "refresh" they've been waffling over even doing, at all.
Click to expand...
I mean if the rumors about PTL moving the memory controller are false. So far the only rumors regarding the PTL memory controller say it's on the Compute tile and if that was the case then a PTL SoC tile would be worthless for ARL.

edit: it's also possible that there is no desktop appropriate SoC tile coming up period since PTL was never really planned for desktop that I'm aware of.
 
Last edited:
baboma said:
>>Nothing interesting will happen on the CPU side. Even the long dead rumor of 8+32 cores wouldn't help gaming performance.

>I sincerely hope this starts a new attitude at Intel that maybe their sockets and platforms might be useful for longer than just the two cycle.

>Being not in the strongest market position right now, having their sockets go over more generations would be a show of good-will and improved customer relations.


Intel doesn't give a diddly about socketed CPUs for desktop. Its turnaround doesn't depend on the tiny minority of DIY gamers. That said, it's a refresh, so of course it'll use the same socket.

It's fine for wanting things to help further your hobby, which presumably is PC gaming. But understand where you are in the overall picture, which is one minnow in a pretty big ocean. Intel has much bigger fish to fry than to worry about what some gamers on some forum think.

Ditto on the GPU side. Nvidia cares about staying on top (of AI), AMD cares about chasing after Nvidia, and Intel cares about just staying afloat. Gaming is an afterthought--and even less than that for Intel.

Given the AI wave, one would think that having NPUs to satisfy MS' "40 TOPS" requirement would be an add, but apparently not. Neither the mobile parts intro'ed at CES'25 (200U/H/HX) have sufficient NPUs, nor do the upcoming desktop refresh. It'll be up to Panther Lake in late '25 to improve NPU. This is a good indicator of MS' pace of AI inclusion for Windows, which looks like the big push will be in '26 with Win12.

https://www.pcmag.com/news/ces-2025-intels-2025-arrow-lake-core-ultra-chips-for-laptops-prioritize
Click to expand...

Learn to quote please. It's not at all hard and it isn't the early 90's anymore.
 
ezst036 said:
I sincerely hope this starts a new attitude at Intel that maybe their sockets and platforms might be useful for longer than just the two cycle.

Being not in the strongest market position right now, having their sockets go over more generations would be a show of good-will and improved customer relations.

There are absolutely no downsides for OEM builders with this, especially since they are often times on autopilot with their business purchases anyways. But this helps those of us who do custom builds and are the mouthpieces on the streets generating buzz with word of mouth praise.
Click to expand...
Back in the 14nm days, the CPUs didn't change much each generation, but the platform made meaningful advances. It was the platform that sold new CPUs not the other way around. I'd argue were back to that now.

To many users the number of e.g. USB 4 ports is more important than 5% fps boost in a game, so they buy a platform and whatever CPU comes with it.
 
From what I can tell efficiency cores add latency to memory and cache access at the hardware level while at the software level complicate how the operating system schedules tasks.

I suspect a performance-core-only design no efficiency cores would address the software scheduling and hardware latency problems. With 10 performance cores this might turn into a best-of-class gaming CPU.

Said another way, greater performance and efficiency might be achieved by optimising a simpler design.
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom