• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Intel Skips NPU Upgrade for Arrow Lake Refresh, AMD Cancels Medusa Halo in Latest Rumors

The RX 8060S in the top Strix Halo AMD Ryzen AI Max+ 395 is already as fast as RTX 5060.
It was rumoured that the newer RX 9060S in Medusa Halo AMD Ryzen AI Max+ 495 would hit RTX 5070 Ti performance levels.


View attachment 408337


Does anyone of you even know that and bought Strix Halo ?
Why are you posting performance figures from a 15 year old synthetic benchmark and not the gaming numbers?

I mean who wants to play Doom at 60fps, right?

1752879188343.png


Oh, and have you heard of dark mode?
 
The same amount of people who would want to play the Nvidia sponsored Doom on an iGPU.
So none? That was my point, but thanks for reinforcing it for me.
 
What about Nvidia? Qualcomm? What about RISC-V breathing behind everybody's neck?

What about Xiaomi's ARM based cpu? What if tomorrow the sanctions will be lifted and finally this chip will enter the world wide competition like everybody else?

P.S:
Nvidia here sends a clear signal about it's intentions for the future.
Nvidia is preparing to dump ARM with it's royalty fees to switch on RISC-V instead.

So a lot of What-If's and 0 actual CPU makers aside from AMD and Intel right now. Anyone with at least comparable desktop level CPU is a decade if not more away in best circumstances, when you basically start from zero.

I really really do not understand how people can fanboy so hard about corpos.
 
Exactly!!!
& this process is absolutely not reversible.

The show must go on, but only among the remaining players.
Unfortunately in this game - Who is dead can't be revived anymore, sorry bro, but it's like that.

I think you might have missed the forest for the trees in that post

Intel's demise made by Intel it self.
I don't think anybody is denying that intel's problems are of its own making.... But the lack of intel will be a monopoly for the diy market for amd cpus, which is what lead intel astray in the first place, you are cheering for a repeat in the process.
 
Last edited:
So Strix Halo was just an one-off without a successor? Bummer
We don't know this. It's an online gossip. Strix Halo is a testing vehicle for new Infinity Fabric, an experimental line-up for some technologies that will be adopted widely in Zen6.

The only way to get more traction for any future Halo APUs with powerful graphics is to make it socketed for desktop platforms too, on AM6, so that anyone can buy it as a normal CPU with powerful graphics. There could be a fork of AM6 motherboards with four memory channels to cater for such segment, but this is conditional on rethinking the value and purpose of HEDT segment.

The solution with soldered LPDDR memory and without an option to upgrade components is NEVER going to be mainstream. Buyers don't want yet another Apple-like, locked-in ecosystem. It's too expensive and niche in mobility devices, so if it does not become available in DIY, it's not going to work in a long run. Mini-PCs will also need to evolve into socketed systems, beyond current BGA lock-in.
 
How large is the group buying systems with dedicated gpu's vs the group that gets systems with an igpu?
as stated in my previous post, I believe npu's for local processing will be a thing needed in the coming years
techies like us don't mind spending some money on new gear every now and then. we are a minority.

That's really not relevant and not where I was going.

I was pointing out that these NPUs are not powerful and are not providing any new capabilities or 'killer' app functionality.

If they were, we'd know about them on our much more powerful desktop systems.

Also hinting at the fact that, even if there is a killer use over the horizon we don't know about, the NPUs out now are going to be turds running it. Same as always with 1st / 2nd gen new hardware. Intel already quadrupled performance on the NPU between Meteor Lake and Lunar Lake.

So what can we expect coming up, double current performance maybe? Sounds right.

Even then they will be weaksauce compared to a 3 year old laptop dGPU, so again I question what is the killer app that needs the compute offered by an NPU.
 
rDNA4 was not optimized for mobile performance at all. Strix Halo has certainly generated some interest, given how hard these devices are to find.
The only thing that kept me from them was the insane cost. The only real laptop so far is the one HP G1a whatever, and we're talking of spending $4k+ to get a 128GB model.


After this latest driver update, I’m wondering if AMD is kicking themselves for not at least trying to make a mobile RDNA4 dGPU. But then again it’s also possible/likely that going forward their focus in the mobile GPU space is now going to be in the form of Halo APUs. Their mobile dGPUs never did do well so in the end I’m not surprised they didn’t bother.

However, while it is true that Strix Halo is hard to find, is that due to sheer sales of the parts, or thanks to a lack of interest in it? I’m betting it’s the latter to be honest; there hasn’t been much buzz about it since it launched to the public.


My hope was we'd eventually get one that traded the two 8c CCDs with a single 12c CCD with a x3d cache and the 40 CU GPU.

One sure can hope. But that would make an already expensive part even more expensive imo.
 
Too bad, I currently have 265K and would have considered an upgrade with an improved NPU. I should be happy that I'll get an extra year of use out of my 265K before upgrading to a new platform in 2026/2027. After doing a bit of overclocking I'm very happy with my 265K setup.
What exactly is the point of the refresh? More Ecores?
Higher clocks and perhaps it will end up getting better reviews if reviewers opt to use 200S Boost (which boosts fabric clocks substantially from launch reviews). Should be something around a 10-15% gaming improvement from launch reviews if 200S Boost (and faster RAM) is used.
 
Intel is finally dying & I'm super happy about this.
After all!
10 years of absolute stagnation with a surplus bonus of cultural degradation is what Intel deserves to get today = Death.
Rest in.... hell Intel, I will not even miss you.
View attachment 408270

For Intel everything is gone.
Are you even aware that in 2028 the ram slots will slowly start to disappear? (In summary: the OTS memories are basically here.)
So what Intel is preparing for 2028?
Intel is still working on it's own Spintronics/Valleytronics?
What's the roadmap for the future if the only future which Intel gives is only about layoffs????

It's over, Rest In Sufferance Shintel View attachment 408272

Are you seriously thinking like that or just playing the dumb ass?
Do you really think that AMD can afford to be relaxed with all these new players around?
Seriously???

Hahahaha ... well, I'm no one to judge you for this.

What about Nvidia? Qualcomm? What about RISC-V breathing behind everybody's neck?

What about Xiaomi's ARM based cpu? What if tomorrow the sanctions will be lifted and finally this chip will enter the world wide competition like everybody else?

P.S:
Nvidia here sends a clear signal about it's intentions for the future.
Nvidia is preparing to dump ARM with it's royalty fees to switch on RISC-V instead.

Intel's demise made by Intel it self.
I will never pay a single cent from my pocket to save a degenerative corporation like Intel with AMD included in the package.
My true & genuine wish is only for Nvidia's N1x being a nice soc for it's price so I can give a very big Hasta la vista to Shintel with AMD altogether.

Last thing about all this mess is: the people should never be supposed to pay from their pocket the price for the mistakes of corporations!

Bye & good luck for everybody!
This fanboy ragebait doesn't even understand about competitions and marketing, yet that person would be the one of getting backlashed. Talking trash and ran away from this thread wasn't good for communities. Several new accounts and bots in other communities are roasting Intel without any reasons.
 
Thank goodness desktop chips don't meet Copilot+ certifications and Microsoft won't use discrete GPUs instead.
 
NPU on desktop is not really useful.

What intel needs is to fix their design of Arrow Lake, in particular the very high memory latency.

by the way:
"Intel's 20A process." ????

does that even exist?
N3B no?
 
  • Like
Reactions: N/A
by the way:
"Intel's 20A process." ????

I had false hope for a monolithic 6+4 core 20A. The next big thing is a unified Titan core which is an E core with better IPC. End of the P core.

continued production on Intel's 20A process

There was no production to begin with and unfortunately 20A process was scrapped.
 
Intel is finally dying & I'm super happy about this.
After all!
10 years of absolute stagnation with a surplus bonus of cultural degradation is what Intel deserves to get today = Death.
Rest in.... hell Intel, I will not even miss you.
View attachment 408270
Intel is needed. Intel's fabs being able to produce hundred of millions of X86 CPUs is what keeps X86 still on top. If Intel doesn't fix it's fabs and relies on TSMC, or if Intel is gone and AMD remains the only one producing X86 CPUs, that will lead to the end of X86 and it's replacement by ARM or/and Risc-V. That will also mean trouble for AMD, because as long as X86 is the most used architecture in servers/desktops/laptops, AMD enjoys huge success, meaning it also makes billions like Intel. If X86 starts losing ground to other architectures, not just Intel but AMD also, will lose a huge advantage against companies like Qualcomm, Nvidia, Broadcomm, Mediatek, Huawei etc. While that would look good, because competition for ARM in servers/desktops/laptops will drive these companies to start building much more powerful SOCs, it will mean a decade of a transision from X86 to ARM. I don't know about you, but in my age, I love stabillity. Not stagnation, but stabillity.

Let's say that Intel going down doesn't affect X86 because for the X,Y reason. AMD having no competition in the X86 market means that finally they can turn their focus on the GPU front. That will be bad for CPUs because no competition means, no reason to keep prices down, no real reason to keep pushing performance up. At least not until another architecture starts becoming a threat. That will be bad for us and the CPU market. On the other hand that could be good for GPUs and could in a matter of years balance the GPU market again, go back from a monopoly to a duopoly. Intel is gone remember? No Intel GPUs, so duopoly. We might start getting better GPUs, but I doubt we will start getting cheaper GPUs. Intel is gone, the company that has a motivation to offer cheap GPUs is gone. AMD will never price it's GPUs too low, because it also wants to sell APUs. Pricing GPUs too low will also mean lowering APU prices and I doubt they will/can do this. In fact they will probably price everything they make at the maximum price they beileve that the market will be willing to swallow, because you need money to compete with a 1 2 3 4? 5? 6? trillion company, like Nvidia.


What needs to happen is this.
AMD to start eating from Nvidia's AI huge cake. AMD making 5 billions per quarter, not per year but per quarter from AI, will be good for them and it might end up good for us too. More R&D in GPUs, might help them improve not just Instinct cards, but also Radeons. Or at least it could help them remain competitive with Nvidia. They could also have a stronger hand when negotiating with TSMC pricing and number of waffers, meaning they might get more waffers at lower prices that could help them offer higher quantities of products at lower prices.
Intel to fix it's fabs, start getting orders from others, start making their CPUs and GPUs again in their fabs. That could lead Intel into being able to flood the market with cheap GPUs and CPUs. If they could flood the market with GPUs that they build, that could be huge news for us. Imagine B570 and B580 selling at their MSRPs. If Intel was producing those chips, they might be doing it now. Buying waffers from TSMC probably makes B570 and B580 low profit products, meaning lower quantity in market, much higher prices compared to MSRP.
 
The RX 8060S in the top Strix Halo AMD Ryzen AI Max+ 395 is already as fast as RTX 5060.
It was rumoured that the newer RX 9060S in Medusa Halo AMD Ryzen AI Max+ 495 would hit RTX 5070 Ti performance levels.
No, except if read theoretical speed of pixels and texels. But memory bandwidth is ~50% from this in RTX 5060. In real world average FPS of RTX 5060 is around 20% higher than Radeon 8060S and depends of game and resolution difference may increase quite more.
I have not in my mind idea how will make bandwidth equal to RTX 5070 Ti which is ~900 GB/s with LPDDR5X, will needs not 4 channels like in Strix Halo, but 16 channel LPDDR5X + infinity cache in Medusa Halo with twice or more capacity than in Strix Halo.
 
What the actual hell is going on in this thread. Intel is just re-aligning their product stack in order to save resources for a potential future jump to equalize them in the market. It’s a perfectly sane business decision and it absolutely doesn’t mean they are “dead”. For pete’s sake, they have ALREADY been behind with a shit product (P4) in living memory of a lot of us and they survived just fine. And yes, as mentioned, even if you are for some reason a deranged madman you still should absolutely not want Intel dead because you just swap around the companies in that regard - AMD will become Intel and behave similarly. Or, worse for you, they become the x86 equivalent of NVidia. Fun.
 
Intel is needed............................................. and so on
So in summary:
Intel till 2030 remains completely noncompetitive, meanwhile Apple silicon is getting only better, Nvidia's N1x will get to be better and so on...

So my dear techpowerup enthusiast!
I appreciate your effort, but unfortunately the current situation is only because of Intel's סדום & also because of that infamous decade of their corporate degradation seasoned by the milking of their own customers.
If Intel's destiny is to disappear completely from the face of the earth, then let it be, otherwise just never mind, because nobody here can make miracles.
 
Last edited by a moderator:
No, except if read theoretical speed of pixels and texels. But memory bandwidth is ~50% from this in RTX 5060. In real world average FPS of RTX 5060 is around 20% higher than Radeon 8060S and depends of game and resolution difference may increase quite more.
I have not in my mind idea how will make bandwidth equal to RTX 5070 Ti which is ~900 GB/s with LPDDR5X, will needs not 4 channels like in Strix Halo, but 16 channel LPDDR5X + infinity cache in Medusa Halo with twice or more capacity than in Strix Halo.
Giving those Igpu's access to their own layered cache could fix some of those bandwidth issues, I guess they would be useless for inferencing but for gaming where the workloads are predicatable
 
Intel's demise made by Intel it self.
I will never pay a single cent from my pocket to save a degenerative corporation like Intel with AMD included in the package.
My true & genuine wish is only for Nvidia's N1x being a nice soc for it's price so I can give a very big Hasta la vista to Shintel with AMD altogether.

Last thing about all this mess is: the people should never be supposed to pay from their pocket the price for the mistakes of corporations!

Bye & good luck for everybody!
You know that Nvidia's corporate practices make AMD and Intel look like angels right? Ever wonder why there are no ASUS AMD-based Strix GPUs anymore? Or why EVGA stopped making Nvidia GPUs despite being one of the popular brands on the market? Or why certain games run like shit on GPUs that are not Nvidia GPUs? Or how Nvidia completely misled it's investors over how cryptomining impacted GPU sales? What about Nvidia scraping both YouTube and Netflix to train it's AI? Then there is the manipulation of GPU reviews like Nvidia giving preferential treatment to media outlets who promised positive reviews and blacklisting outlets who gave honest reviews. And the last but not least was the Geforce Partner Program which sought to block manufacturers from making non-Nvidia GPUs using preferential allocation, price reductions and marketing funds (similar to what Intel did to AMD during the late 1990s and during the 2000s but Nvidia decided that it was a good idea even after seeing AMD winning that lawsuit).

(I won't even bother mentioning Nvidia cheating in benchmarks back in the 2000s or Nvidia's hairworks)
 
An APU is a relatively simple circuit, amazing that a company like Intel can't iterate on an already existing design. How the mighty have fallen.
 
No, except if read theoretical speed of pixels and texels. But memory bandwidth is ~50% from this in RTX 5060. In real world average FPS of RTX 5060 is around 20% higher than Radeon 8060S and depends of game and resolution difference may increase quite more.
I have not in my mind idea how will make bandwidth equal to RTX 5070 Ti which is ~900 GB/s with LPDDR5X, will needs not 4 channels like in Strix Halo, but 16 channel LPDDR5X + infinity cache in Medusa Halo with twice or more capacity than in Strix Halo.

Don't forget that Nvidia cheats, and uses lower textures resolution in games which means it artificially makes the FPS counts higher than the actual hardware capabilities of their chips.
It's enabled by default DLSS - bloated, dull colours and lower LOD, worse image quality, etc.
It has been discussed many times.

 
Or why EVGA stopped making Nvidia GPUs despite being one of the popular brands on the market?
This line right here basically scorches your whole argument. EVGA - a very reputable company in the GPU space - with the tech and knowhow to make GPUs - withdrew altogether from the gpu market instead of you know, working with AMD. So if your argument is to be believed, by definition AMD has worse business practices since EVGA didn't even consider working with them, they just completely withdrew. You are not making a reasonable argument, and you know you are not making a reasonable argument. Powercolor isn't making any nvidia GPUs, let me guess, is AMD forcing them?

Come on, let's stop this nonsensical attack.

(I won't even bother mentioning Nvidia cheating in benchmarks back in the 2000s or Nvidia's hairworks)
If you know about nvidia cheating in benchkarms back in the 2000, surely you must remember amd / ati doing the same. Right? :)

So none? That was my point, but thanks for reinforcing it for me.
Just so we clear with the "nvidia sponsored" comment, doom dark ages - just like every other nvidia sponsored game runs BETTER on AMD gpus than the average non nvidia sponsored game. Yes, nvidia is THAT freaking ethical that they make sure their sponsored games are very well optimized for amd cards. 9070xt faster than the 5070ti - as is the case with other nvidia games (cp2077, alan wake etc.)

 
So in summary:
AMD till 2020 remains completely noncompetitive, meanwhile Apple silicon is getting only better, Intel will get to be better and so on...

So my dear techpowerup enthusiast!
I appreciate your effort, but unfortunately the current situation is only because of AMD's סדום & also because of that infamous decade of their corporate degradation seasoned by the stupidity of their board(overpaying for ATi, having to sell their fabs, giving us this disaster that Bulldozer is, falling behind Nvidia in GPUs).
If AMD's destiny is to disappear completely from the face of the earth, then let it be, otherwise just never mind, because nobody here can make miracles.
I changed your post to reflect the situation we where facing 10 years ago. Your comment, 10 years ago, could be like the text above.

If in 2016 AMD has gone bankrupt and Intel was just starting having the problems with it's fabs that it is facing the last 10 years, if also Intel has bought Radeon group from AMD's creditors and had kept Raja as the head of the group for much longer, what would have been the market today? A total mess probably. Intel CPUs would be 8-10 cores max for the retail market. GPUs would probably be non competitive to Nvidia's, because for years Intel didn't believed in GPUs, it was seeing them as a threat and ARM would be in a rise, meaning the market would be divided between X86 and ARM in such a way that we would be having tough time choosing which architecture to go.

So, be careful of what you wish.

and Intel look like angels right?
Intel was never an angel. Probably they started looking like angels last summer when they had to aggressively cut expenses, meaning their influence on the tech press and OEMs got reduced.
 
We don't know this. It's an online gossip. Strix Halo is a testing vehicle for new Infinity Fabric, an experimental line-up for some technologies that will be adopted widely in Zen6.

The only way to get more traction for any future Halo APUs with powerful graphics is to make it socketed for desktop platforms too, on AM6, so that anyone can buy it as a normal CPU with powerful graphics. There could be a fork of AM6 motherboards with four memory channels to cater for such segment, but this is conditional on rethinking the value and purpose of HEDT segment.

The solution with soldered LPDDR memory and without an option to upgrade components is NEVER going to be mainstream. Buyers don't want yet another Apple-like, locked-in ecosystem. It's too expensive and niche in mobility devices, so if it does not become available in DIY, it's not going to work in a long run. Mini-PCs will also need to evolve into socketed systems, beyond current BGA lock-in.
I'm not sure about that, Mini PC's have been extremely popular, just not in the enthusiast market. I would like to see a socketed APU, however bandwidth would be a limitation unless the motherboard would have dual CAMM2 sockets or 4 DIMM slots, and then you'd end up with a mini PC as large as an ITX system.
withdrew altogether from the gpu market instead of you know, working with AMD. So if your argument is to be believed, by definition AMD has worse business practices since EVGA didn't even consider working with them, they just completely withdrew.
The context you left out is important, the CEO wanted to step down, and according to all of the talk that went on when EVGA exited the market, there was likely a non-compete clause from Nvidia for EVGA to not work with any competitors, and IMO that definitely sounds like something the leather jacket would force upon their partners. EVGA was forced out of the market because Nvidia was pushing all sorts of rules on their partners while undercutting them with FE cards, and didn't seem to care when their best business partner was forced to exit. Your argument makes no sense especially when you choose to attack AMD instead to make Nvidia look like a hero.
Powercolor isn't making any nvidia GPUs, let me guess, is AMD forcing them?
More likely is Nvidia wouldn't allow them to make any Nvidia GPU's, and the same happened with Acer and Asrock, which is why they only make AMD and Intel GPU's.
If you know about nvidia cheating in benchkarms back in the 2000, surely you must remember amd / ati doing the same. Right? :)
Except we know where Nvidia cheating on benchmarks back then got them to the near monopoly they are today. Intel also cheated on benchmarks for decades, but chose to rest on their back foot for too long.
Just so we clear with the "nvidia sponsored" comment
The point is none of these GPU companies should be influencing game developers to not optimize for competitors, just look at CP2077 for example, it didn't get FSR3.1 until recently, a 5 year old game which most people wouldn't care about replaying now except enthusiast Nvidia customers.
 
You guys ever tire of GPU corpo vs GPU corpo shitflinging? No? Okay.

More on topic, I also had a stray thought that Intel might have decided on not upgrading the NPU for a very simple reason that consumer facing system integrated AI just didn’t really take off as much as the companies thought it would. Apple seemingly slowly pulls back on advertising Apple Intelligence as a “killer feature”, MS isn’t finding much uptake on the Copilot side of things, the average customer is just sort of over the whole AI hype train, at least as it seems to me. Yes, it’s massive in the enterprise space, for sure, but that’s the realm of GPGPU and specialized ASICs. While the mundane consumer oriented AI as presented by these companies is really not that different from the Voice Assistants of yesteryear and after completely failing to attract people to things like Cortana I don’t think it is feasible to explain to an average pleb why Copilot is so much better and why they totally should upgrade their CPU (or, realistically, buy a new PC) for it.
 
Back
Top