TheLostSwedeThe question is if there will be any DRAM module maker that will be willing to be first and/or if Intel will push it, as AMD is still terrible at pushing new standards and the Taiwanese board makers pretty much don't care about what AMD wants when it comes to new tech.

It's more complex. Taiwanese board makers have finally looked into a mirror, a bit. Suddenly, for the first time ever, we have more models of motherboards on AMD platform than for Intel. Even they can see market share and sentiment changes. As regards to new standards, it's not surprising that Intel has been traditionally pushing more initiatives for adoption. This is due to their dominant position over the years. However, despite this advantage, it was AMD that initially pushed:
- PCIe 4.0 adoption first onto all motherboards in desktop, workstation and server space
- PCIe 5.0 adoption into server boards
- integrate USB10 into chipset and CPU
- offer NVMe drive IO on the CPU, including Gen5 NVMe drive on the CPU
- offer 4 displays on iGPU via desktop APU
- offer NPU via desktop APU
- support DP40 and HDMI 2.1 FRL on iGPU
- important adoption of UALink open ecosystem in server space
Obviously, Intel has a longer list, buy hey, consumers benefit from both.

john_AM5 needed time because AMD delayed the 7000 X3D chips for about 6 months.

Yep. That was a mistake. They did not get their packaging plant working on time, to stitch V-cache dies with the rest sooner. I hope they don't make the same mistake again. Up to 3 months difference in vanilla and X3D SKUs is tolerable, but not any longer.

That H1 2026 estimate is based on current Steam Survey. My guess is that it would be a bit lower. ~40%, but more is possible if Arrow Lake refresh does not impress buyers.

Tek-CheckFor the upcoming LGA1954 to be a successful platform, Intel will have to play AMD's playbook and offer way more value, including far better platform longevity and ease of upgrade rather than changing sockets every 2-3 years. It's absolutely fundamental that they change their past approach to frequently dumped sockets.

1954 will support 3-4 generations. I refuse to accept Intel decision makers would ignore the lessons of the last decade. Nova will kick ass. But leave it to Intel to have the worst timing possible, again. It's surrealistic. When Nova and Zen 6 get here, they will be fighting over share of a rapidly contracting market. Z6 at least has the in socket upgrade option, which will help a little, if it is a good jump over Z5. Won't matter much in the big picture. NAND and DRAM pricing and availability, combined with wider economic factors, are going to spoil the party.

Tek-CheckIt's more complex. Taiwanese board makers have finally looked into a mirror, a bit. Suddenly, for the first time ever, we have more models of motherboards on AMD platform than for Intel. Even they can see market share and sentiment changes. As regards to new standards, it's not surprising that Intel has been traditionally pushing more initiatives for adoption. This is due to their dominant position over the years. However, despite this advantage, it was AMD that initially pushed:
- PCIe 4.0 adoption first onto all motherboards in desktop, workstation and server space
- PCIe 5.0 adoption into server boards
- integrate USB10 into chipset and CPU
- offer NVMe drive IO on the CPU, including Gen5 NVMe drive on the CPU
- offer 4 displays on iGPU via desktop APU
- offer NPU via desktop APU
- support DP40 and HDMI 2.1 FRL on iGPU
- important adoption of UALink open ecosystem in server space
Obviously, Intel has a longer list, buy hey, consumers benefit from both.

I think you misunderstand how a lot of this works when it comes to the motherboard makers.
Once upon a time, these companies were living on the cutting edge, took chances and were innovative.
Then the upper management got old and comfortable, their businesses had outlasted the competition, so there was no longer any need to do more than they had to, as the money was still coming in and they had their share of the market.
The only "innovation" that was going on, was when there was a new motherboard chipset or some generational shift in technology, more often than not, pushed by Intel and also encouraged by Intel through their MDF program.

What you're listing are just evolutional steps of already existing standards (except USB10?), which didn't scare these companies, as AMD never pushed for a different graphics card interface or some non standard type connector or a new type of memory. On top of that, Intel followed shortly behind on many of these things, making it a safe bet for the board makers and they could beta test some things on the AMD platform and if anything went wrong, they could always blame it on AMD.

If you look at actual motherboard specs, things like HDMI 2.1 was blocked by the board makers and often still are, especially on lower-end boards that get to make do with HDMI 1.4. AMD has always and still is struggling to get the board makers to implement new features, even more so if it's not something that isn't on Intel's road map.

AMD doesn't have much of an MDF program, AMD doesn't really have the engineering resources Intel has had in the past, although I don't know what it's like today, but Intel have had local FAE's that come over and train the motherboard makers, reference boards and really detailed documentation, whereas AMD hasn't invested in these type of things. There's a reason why the Initial AM4 motherboards weren't so good, as AMD had apparently sent over a few trays of chipsets, some rough instructions and asked the board makers to make some boards. These are things that aren't general knowledge for obvious reasons, but that I've been told by people in the industry, due to both having worked as a tech journalist for a long time, but also having lived in Taiwan for a good chunk of my life.

Should things be like this? No, but these companies have old, slow leadership that wants to make safe products that they know there where the ROI is. Taking chances isn't something they do easily any more and this is most likely also a reason why we won't see CAMM2 boards, unless someone like Kingston does it together with Intel.

Just look at how dumb the motherboard market is that the four "big" motherboard makers (Asus, ASRock, Gigabyte and MSI) can't sit down and agree on some common standards. As such, we now have four different back side connector layouts, we have four different ways of attaching Wi-Fi antennas and there could be common connector standards for a lot of other things, but not. They really need a kick in the arse for this.

RinSenpaiIf you buy spare parts like that, you're already at the very least in the minority that does DIY PCs. Doesn't matter where you buy your spare parts after that. Getting 50% of sales from a market that is like 5% at best is nothing compared to the entire PC market. At least the brand gets loyalty/fidelity, but we're already too deep for the average consumer to care about that, and neither do they care about this feeling as well.

If DIY'ers dont care about core counts then regular consumers care much less.
The sales numbers prove that. While it would be nice to get 12c on a single CCD with Zen 6 i also dont feel like it's strictly necessary for most people.
Much less the 52c Intel plans and that's going to cost and arm and a leg. (just looks at Panther Lake prices).
Not that i think Zen 6 will be cheap either because even if it's economically cheaper to make due to much smaller chiplets than Intel, AMD will likely still price it high because margins...

john_It did. It helped them go against AMD with the same or more numbers of cores.

Something a tiny minority of people care about even in the DIY space.
Most people consume content. They dont create the content.
At least not the kind of content that would meaningfully benefit from high core counts.

john_And because Zen3 was configured for efficiency, when Intel was pushing 250W-300W power consumption, it was winning.

I fail to see how pushing 2-3x the power can be considered winning. A pyrrhic win at best.

john_Because consumers are looking at DDR4 platforms the last few months, because of DDR5 RAM prices, that's why 8 core AM4 CPUs are top options. 12 or 16 core AM4 CPU models are probably considered expensive and unnecessary for those looking at AM4 options.

8c/16t etc topped the charts last year too. Way before DDR5 prices went to hell. It's not a recent development.

john_The first one who realized that 8 P cores are probably enough, was Intel. That's why they chose to go with a maximum of 8 P cores.

They chose to go with 8 P-Cores because of power limitations. 13th and 14gn were already power hogs even with 8p.
I shiver to think what the results would have been with 16p. And even so with 8p they still pushed these too far.
Tho to be fair the degradation issue was not because of core counts or even core power consumption.

TheLostSwedewe have four different ways of attaching Wi-Fi antennas and there could be common connector standards for a lot of other things, but not. They really need a kick in the arse for this.

Which are these?

I've seen the old standard screw in the cables variant (annoying) and the new quick connect variant everyone seems to be moving to. What are the other two?

TomorrowWhich are these?

I've seen the old standard screw in the cables variant (annoying) and the new quick connect variant everyone seems to be moving to. What are the other two?

There are three different quick connect ones, that are incompatible between brands.
MSI at least retained the option of using screw in antennas, which Asus and Gigabyte didn't bother with.

DAPUNISHER1954 will support 3-4 generations. I refuse to accept Intel decision makers would ignore the lessons of the last decade. Nova will kick ass. But leave it to Intel to have the worst timing possible, again. It's surrealistic. When Nova and Zen 6 get here, they will be fighting over share of a rapidly contracting market. Z6 at least has the in socket upgrade option, which will help a little, if it is a good jump over Z5. Won't matter much in the big picture. NAND and DRAM pricing and availability, combined with wider economic factors, are going to spoil the party.

Zen 6 will probably be at the tail end of AM5's lifecycle, so that will probably only be an advantage for Zen 4 owners. Also, with Socket 1954, the gap in IO will be even greater, so at least on paper that platform will be much more appealing for the "prosumer" crowd. Unlike with AM4, AM5 haven't offered high featured motherboards significantly cheaper than Intel counterparts, if anything, they have been more expensive.

As for the market contracting, those claims are grossly exaggerated. While the selection and pricing of memory and SSDs are bad, they are absolutely more obtainable (if you adjust your expectations) than during the 2020-2021 problems, and there are bundles as well as great deals for CPUs and motherboards. I expect the sales to be just fine, even though the selection of some parts will be challenging.

TheLostSwedeThey really need a kick in the arse for this.

Ah, yes. I agree with a wider narrative. I thought it was about specific features from standards.

DAPUNISHER1954 will support 3-4 generations. I refuse to accept Intel decision makers would ignore the lessons of the last decade. Nova will kick ass. But leave it to Intel to have the worst timing possible, again. It's surrealistic. When Nova and Zen 6 get here, they will be fighting over share of a rapidly contracting market. Z6 at least has the in socket upgrade option, which will help a little, if it is a good jump over Z5. Won't matter much in the big picture. NAND and DRAM pricing and availability, combined with wider economic factors, are going to spoil the party.

Just like how Intel was late with Sapphire Rapids-SP/WS and Granite Rapids-WS... though, at least Granite Rapids-WS is not super late in terms of relevancy. Hope that socket longevity target from Intel transfers to the workstation side so that LGA 4710-2 is supported for Diamond Rapids-WS and Coral Rapids-WS.

dgianstefaniArrow Lake E cores are equivalent IPC to Raptor Lake P cores,

To @W1zzard Unfortunately, we're questioning the @dgianstefani expertise: he's seriously pointing out the similarities in performance between Intel EcoRes and PCores, which is completely untrue for every parameter in any benchmark. These are two completely different architectures for completely different purposes, with diametrically different latencies. This data can't possibly be unknown to the author, and he keeps spouting this nonsense. We request that the employee take appropriate action.
To @dgianstefani. Dear, stop spewing this Intel marketing vomit and show us how you run real-time applications (ohh, may be games?) on your favorite ecores cores with the same performance as pcores.

ContraTo @W1zzard Unfortunately, we're questioning the @dgianstefani expertise: he's seriously pointing out the similarities in performance between Intel EcoRes and PCores, which is completely untrue for every parameter in any benchmark. These are two completely different architectures for completely different purposes, with diametrically different latencies. This data can't possibly be unknown to the author, and he keeps spouting this nonsense. We request that the employee take appropriate action.
To @dgianstefani. Dear, stop spewing this Intel marketing vomit and show us how you run real-time applications (ohh, may be games?) on your favorite ecores cores with the same performance as pcores.

Skymont E cores are indeed at the level of Raptor Cove P cores in IPC/performance clock normalized, they're clocked lower and have some latency differences, but you can test this yourself if you simply set affinity to E cores on an Arrow Lake CPU and do some benchmarking in games/programs etc.

If you disagree, you're welcome to supply some evidence to the contrary.

chipsandcheese.com/p/skymont-intels-e-cores-reach-for-the-sky
blog.hjc.im/spec-cpu-2017
chipsandcheese.com/p/skymont-in-gaming-workloads

dgianstefaniSkymont E cores are indeed at the level of Raptor Cove P cores in IPC/performance clock normalized, they're clocked lower and have some latency differences, but you can test this yourself if you simply set affinity to E cores on an Arrow Lake CPU and do some benchmarking in games/programs etc.

If you disagree, you're welcome to supply some evidence to the contrary.

chipsandcheese.com/p/skymont-intels-e-cores-reach-for-the-sky
blog.hjc.im/spec-cpu-2017
chipsandcheese.com/p/skymont-in-gaming-workloads

www.techpowerup.com/review/intel-core-i9-12900k-e-cores-only-performance/

Friend, it's all about L2cache and memory latencies loss. For ecore, it's a 2x penalty compared to pcore. No amount of IPC will help them.

Contrawww.techpowerup.com/review/intel-core-i9-12900k-e-cores-only-performance/

Friend, it's all about cache loss. For ecore, it's a 2x penalty compared to pcore. No amount of IPC will help them.

That's Gracemont, not Skymont.

ContraFriend, it's all about cache latencies loss. For ecore, it's a 2x penalty compared to pcore. No amount of IPC will help them.

Sure, for Gracemont that's true but we're talking about Skymont, where I said the IPC is equivalent to Raptor Cove, and it is. Obviously Lion Cove P cores are still faster, that's not the point I'm making. Gracemont (Raptor Lake E cores) were about Skylake level IPC. Skymont (Arrow Lake E cores) are about Raptor Cove level IPC. That's true, I'm not sure what you're disagreeing with here.

Contrashow us how you run real-time applications (ohh, may be games?) on your favorite ecores cores with the same performance as pcores.

dgianstefaniThat's Gracemont, not Skymont.


Sure, for Gracemont that's true but we're talking about Skymont, where I said the IPC is equivalent to Raptor Cove, and it is. Obviously Lion Cove P cores are still faster, that's not the point I'm making. Gracemont (Raptor Lake E cores) were about Skylake level IPC. Skymont (Arrow Lake E cores) are about Raptor Cove level IPC. That's true, I'm not sure what you're disagreeing with here.

Performance = Instruction per clock * clock speed

Clock speed of E-cores are much less than normal cores, and that's it.

qcmadnessIPC = Instruction per clock * clock speed

Clock speed of E-cores are much less than normal cores, and that's it.

IPC is indeed instruction per clock.
Performance is IPC * clock.
Skymont E Core IPC is essentially the same as Raptor Cove P core, but of course they run at 4.x GHz instead of 5.x GHz, and as has been pointed out there's some latency and cache differences, but it's not false to say the IPC is essentially the same. A massive achievement considering the die area of the cores being compared.

Going to bed now, but I'm satisfied with the sources I've provided to defend that statement.

dgianstefaniThat's Gracemont, not Skymont.

As long as they use one bus stops for 4 cores with a scheduler/router, nothing will change.

dgianstefaniIPC is indeed instruction per clock.
Performance is IPC * clock.
Skymont E Core IPC is essentially the same as Raptor Cove P core, but of course they run at 4.x GHz instead of 5.x GHz, and as has been pointed out there's some latency and cache differences, but it's not false to say the IPC is essentially the same. A massive achievement considering the die area of the cores being compared.

I edited the post.

Yes, you can say that the IPC is similar, but first clock speed is not the same.

Secondly, and more importantly, is if the next e-core supports 512-bit FPU instructions with 512-bit FPUs, then the advantages of "small" size will diminish.
Or if it executes with 256-bit FPUs, then it still lags behind P-cores and/or Zen 5/6.

john_90% of consumers, if not more, do not spent time reading tech stuff online. Their approach on choosing PC hardware is usually very simple.
The CPU with the more cores is faster.
The graphics card with the more VRAM is faster.
And that's it.

It's how I will choose for example a refrigerator.
"This is bigger, it's non-frost and at a better price, while carrying a known brand's name. I should choose that".
Someone reading tech staff about refrigerators will probably have a reason, or two, or even ten, to explain to me why I should avoid that model.

General users yes not users on this site or PCMR.

And those same general users will buy a 52 or 48 core Nova lake CPU and 2/3 of those cores will be idle majority of the time.

efikkanZen 6 will probably be at the tail end of AM5's lifecycle, so that will probably only be an advantage for Zen 4 owners. Also, with Socket 1954, the gap in IO will be even greater, so at least on paper that platform will be much more appealing for the "prosumer" crowd. Unlike with AM4, AM5 haven't offered high featured motherboards significantly cheaper than Intel counterparts, if anything, they have been more expensive.

As for the market contracting, those claims are grossly exaggerated. While the selection and pricing of memory and SSDs are bad, they are absolutely more obtainable (if you adjust your expectations) than during the 2020-2021 problems, and there are bundles as well as great deals for CPUs and motherboards. I expect the sales to be just fine, even though the selection of some parts will be challenging.

That life cycle will be alittle longer if Zen 7 gets released for AM5 and based on memory pricing and when DDR6 is going to come out I think its highly likely.

dgianstefaniIPC is indeed instruction per clock.
Performance is IPC * clock.
Skymont E Core IPC is essentially the same as Raptor Cove P core, but of course they run at 4.x GHz instead of 5.x GHz, and as has been pointed out there's some latency and cache differences, but it's not false to say the IPC is essentially the same. A massive achievement considering the die area of the cores being compared.

Going to bed now, but I'm satisfied with the sources I've provided to defend that statement.

20% is not a small difference.

Do you think you will be happy if your pay cut is cut by 20%. Or you will not be happy with you pay rises by 20%?

Tek-Check- offer NPU via desktop APU

Wasn't that Intel?
Nvm, after some extra googling I found about the 8000G series, I had totally forgotten about those.

ContraTo @W1zzard Unfortunately, we're questioning the @dgianstefani expertise: he's seriously pointing out the similarities in performance between Intel EcoRes and PCores, which is completely untrue for every parameter in any benchmark. These are two completely different architectures for completely different purposes, with diametrically different latencies. This data can't possibly be unknown to the author, and he keeps spouting this nonsense. We request that the employee take appropriate action.
To @dgianstefani. Dear, stop spewing this Intel marketing vomit and show us how you run real-time applications (ohh, may be games?) on your favorite ecores cores with the same performance as pcores.

They clearly mentioned IPC, not absolute performance, and they are indeed right on that claim.

qcmadnessPerformance = Instruction per clock * clock speed

Clock speed of E-cores are much less than normal cores, and that's it.

Se above, they never claimed that those E-cores were overall faster than the P-cores, but just compared the IPC.

ContraAs long as they use one bus stops for 4 cores with a scheduler/router, nothing will change.

You really know nothing about this stuff.

TomorrowThe sales numbers prove that

This prove only that most of users eid not have money for more cores

Visible NoiseYou really know nothing about this stuff.

What changed in Skymont? CnC
"Gracemont’s shared L2 controller and offcore interface is probably also a complex piece of logic that draws a good chunk of power by itself. Labeled in orange above, this block is responsible for handling L2 requests coming from all cores in the cluster, and arbitrating between them. That means each Gracemont core has a less direct path to the ring bus."
www.techpowerup.com/review/intel-core-i9-12900k-e-cores-only-performance/2.html
Can we ask @W1zzard to do a completely identical test for skymont?

dgianstefani

What resolution was this tested on?
I presume 1080p, but the C&H article does not specify.
Also they tested with Arc B580. A relatively weak CPU to use for a CPU limited test.

If it's a CPU test then it's not optimal as the results look more GPU limited than CPU limited.
Plus they used the built in benchmark that is not CPU limited so that's another mistake.
They also dont state if they used APO (CP2077 being one of the APO supported games) or what specific Windows version they were using as this can greaty affect AMD's performance.

This is from 285K launch. Default configuration (all P and E enabled). Tested with 4090 in a custom CPU limited area:


I very much doubt that e cores would be so close to p cores in this scenario and nor would the 285K be this close to 9900X.
In the Techspot article their difference is not 3fps, but rather 26fps.

TumbleGeorgeThis prove only that most of users eid not have money for more cores

Top #10 are between $84 and $675.
#1 is $473.
Only 3 out of 10 are below $200.

I very much doubt it's about price.
www.amazon.com/Best-Sellers-Computer-CPU-Processors/zgbs/pc/229189

TomorrowWhat resolution was this tested on?
I presume 1080p, but the C&H article does not specify.
Also they tested with Arc B580. A relatively weak CPU to use for a CPU limited test.

If it's a CPU test then it's not optimal as the results look more GPU limited than CPU limited.
Plus they used the built in benchmark that is not CPU limited so that's another mistake.
They also dont state if they used APO (CP2077 being one of the APO supported games) or what specific Windows version they were using as this can greaty affect AMD's performance.

This is from 285K launch. Default configuration (all P and E enabled). Tested with 4090 in a custom CPU limited area:


I very much doubt that e cores would be so close to p cores in this scenario and nor would the 285K be this close to 9900X.
In the Techspot article their difference is not 3fps, but rather 26fps.

Top #10 are between $84 and $675.
#1 is $473.
Only 3 out of 10 are below $200.

I very much doubt it's about price.
www.amazon.com/Best-Sellers-Computer-CPU-Processors/zgbs/pc/229189

We can know that, in general, compared to 8p, 8p16e can improve gaming performance by 5%. For games with good MT optimization such as 2077, e-core performance scales well. Using e-core alone performs poorly in purely ST games like Starcraft 2, but in most games the performance difference from p-core is not significant. 8p is affected by the thread count and performs worse than 16e in more modern games.

ContraAs long as they use one bus stops for 4 cores with a scheduler/router, nothing will change.

Intel uses 2P per ring stop in Novalake, following the e-core method, viewing the sharing of a ring stop among multiple cores as a more advanced practice.