The whole stability mess would have been mitigated had Intel actually tried to compete with AMD's X3D chips by not brute forcing clock speeds just to eke out that tiny little performance advantage (so it can look good on presentations)
IMHO the LGA1700 platform is a weird place; while the 12th gen chips are legitimately good, the 13th gen chips feels like a 12th gen chip with mostly minor improvements, and the 14th gen is just a pure waste of sand. It really speaks volume about Intel's platform 'longevity' especially when you consider Arrow Lake will come on a new platform/socket

I really wish both AMD/Intel didn't lock themselves in an all out performance war and they should start designing desktop chips will efficiency in mind - those 5.5-6+ GHz clock speeds are sort of unsustainable and reminds of the Netburst era where little emphasis is put on efficiency. Especially when you consider most x86 chips are clocked way beyond the efficiency point which leads to excessive power consumption. Yes I know you can power tune those CPUs or even use Eco mode but then the hivemind only wanted longer performance bars at any cost so they are more than happy to oblige

This post has been edited by chocobo7779: May 11 2024, 12:50 PM

Probably not noticeable outside of very heavy, multithreaded workloads (in games it will be fine)

Yup, same goes to AMD as well with their '95C Tjmax' temperatures

On the other hand, the whole power limit mess has been already happening for about a decade or so, according to Hardware Unboxed:


Yes, you can tune Intel chips to very efficient levels, but the same goes to AMD as well (it's even better when you consider you can just use 105W Eco mode on the BIOS for the Ryzen 9 chips and you'll have very little performance loss outside of synthetic benchmarks with a significant uplift in efficiency)
That being said however it just kind of proves that modern x86 CPUs are massively tuned for performance with very little regard in efficiency

This post has been edited by chocobo7779: May 11 2024, 08:02 AM

From a ISA point of view, both x86 and ARM are quite performant and efficient (there is a common myth that x86 is 'inefficient' because of the legacy bloat), and ISA doesn't really mean much where's there's no native apps to run. Back in the 1990s there are plenty of RISC ISAs that outperformed x86 significantly like DEC Alpha and yet x86 prevailed due to the large library of software written. I know binary translation/dynamic recompilation will help but this will introduce a performance overhead
Modern CPUs these days are so complex from the architectural standpoint that the ISA doesn't mean much anymore

Ultimately, implementation matters (there's also the economic/business side that needs to consider), not ISA - there's also a YouTube channel that talks about semiconductor design and manufacturing that goes deep dive in CPU architectures:
https://www.youtube.com/@HighYield

https://chipsandcheese.com/2021/07/13/arm-o...-doesnt-matter/

This post has been edited by chocobo7779: May 11 2024, 08:05 AM

Probably for the better, but this just makes the 14th gen K CPUs look like a rebranded 13th gen and even more of a waste of sand

No PCIe 4? Not that much of a problem for most GPUs unless you are talking something about RX6400/6500XT with their nerfed PCIe x4 interface. Even the mighty 4090 only loses about 2% performance on PCIe 3.0 x16:


On the SSD side of things there's not much difference between PCIe 3.0 and 4.0 either for gaming

USB 3.2? That's more or less belong to the 'nice to have' territory rather than must haves, and even that can be done with a PCIe card if you really need one

? Depending on where you live you can get a B650 board that will happily run a 7950X at full power, here's one for USD109:
https://www.microcenter.com/product/664700/...atx-motherboard

The problem with most of these boards are the fact that they simply fail at pricing them right - I mean why bother buying those boards where the HDV will happily outperform them at the same price, if not lower? Perhaps it may work better for those who only uses 65W chips or even the 7800X3D, but that kind of defeats the point of AM5, so that's why HUB more or less advised users to stay away from them
AM4 is still very viable and there's really nothing wrong with that especially with their X3D chips that will still happy perform quite well compared with modern Intel equivalents

Software side is tricky, but let's not forget that Intel/Nvidia has their fair share of software/driver issues

I mean, the M2 Ultra is a very expensive chip in terms of cost and transistor budget - it had what, 100 billion transistors which is much closer to something like AMD's MI300A than most consumer hardware (Nvidia's AD103 only has around 76 billion transistors). Flawless software integration isn't hard to do with Apple as they have much more R&D budget to spend with a highly vertically integrated ecosystem (this is the nice part of a walled garden)
Sure AMD/Intel can design those chips but why bother selling them at the consumer market where they can make vastly more selling them as a HPC/AI chip as the consumer market is fundamentally a low margin business

Again, ISA doesn't matter, see above post, and is not like standard ARM cores are a gold standard in performance/efficiency

This post has been edited by chocobo7779: May 11 2024, 08:48 AM

Unless Snapdragon X actually eats up market then no - the whole thing about ARM/RISC-V will take over x86 almost reminds on the age old sentence [insert year] will be the year of the Linux desktop

This post has been edited by chocobo7779: May 11 2024, 08:56 AM

To be fair to AMD/Intel, it's really not hard to match Apple in terms of peak performance/efficiency in mobile chips, but what Apple dominates is the performance at low/mid power range which matters a lot more in real world as it's very unlikely for most modern workloads to utilize peak performance for any long period of time, due to power/heat constraints

This is sometimes why synthetic benchmarks like Cinebench/Geekbench can be misleading on power constrained devices as those only measure peak performance, but not sustained performance which is a lot more important (most x86 CPUs tend to focus a lot on peak performance, due to the fact on desktops they can be run at an indefinite period as long as the power/heat budget allows) idea

I really wish Intel didn't can their Y series chips that early though, as those chips could really become Apple M series competitor if they kept on iterating on it

There's also the very large, captive markets that require x86, like government/education/corporate/manufacturing sectors that often uses specialized, in-house software and are not COTS, and cannot be ported to ARM easily, even if those software had their source codes available

This post has been edited by chocobo7779: May 11 2024, 03:04 PM

It will be quite interesting to see where Strix Point and Lunar Lake headed, pretty excited to see what x86 can offer in the next few years
That being said though, the X Elite chips are promising, but compatibility, pricing and availability will make or break this chip

This post has been edited by chocobo7779: May 11 2024, 01:07 PM

...more like core design, really
Apple cores are specifically targeting high IPCs, so they can clock them lower relative to x86 incumbents to achieve high single threaded performance with excellent efficiency. This is why Apple cores tend to be quite large and wide compared to your garden variety x86 chips. One of the downsides is that they are often 'inefficient' in terms of SoC/die size and transistor budget, so it is very expensive to make, and this partly explains why Apple charges very high prices for RAM/storage upgrades for their hardware, presumably that they can subsidize the cost of manufacturing those SoCs

Of course x86 cores can be 'fatter and wider' to increase performance without increasing clock speeds to stratospheric heights, but that's not going to be cheap to make, and note that AMD/Intel will need to cater to vastly different markets and use cases ranging from cheap laptops to multimillion dollar supercomputers as opposed to Apple where their targeted audience and software ecosystems are much more focused and locked down

Those are speculations of course, so correct me if I'm wrong


Simply put, the enthusiasts wanted longer performance bars on benchmarks so there is an all out performance war with very little regard to efficiency

The big problem with modern CPUs right now isn't really the CPU core itself, but rather the memory subsystem, as the improvements on CPUs are vastly outclassing the improvements on memory speeds/bandwidth which leads to memory bottlenecks, and moving data between the CPU and memory isn't exactly power efficient either. This is why you have things like memory on package (like Apple chips/upcoming Lunar Lake CPUs) and die stacking like what AMD does to their X3D lineup of CPUs


That's why you have things like chiplets, backside power delivery such as PowerVia (could be revolutionary) and new chip packaging/interconnect methods to make further die shrinks possible

This post has been edited by chocobo7779: May 11 2024, 01:58 PM

Lunar Lake is mobile only though

There's also Arrow Lake with PowerVia backside power delivery

Usually it's often multithreading that often causes lots of security loopholes, that's why Intel is rumored to remove them in Arrow Lake chips, and replacing them with rentable units (if the rumors are correct)
That being said though it's very unlikely for your regular machine to get hacked using CPU vulnerabilities unless you do a lot of *ahem* stuff, or you fail to do the bare basics of computer security

Alder Lake would like a word with you though, but yeah x86 is going to be a lot more exciting in the next few years

This post has been edited by chocobo7779: May 11 2024, 01:50 PM

You almost forgot the Ryzen 3000 boost clock bug which AMD was able to fix in the ABBA version of their AGESA

Mind you, the Snapdragon X series is about 1 year late, and it seems like Qualcomm is sort of sandbagging right now (heck even the original intent for Qualcomm to acquire Nuvia was to compete in servers, not laptops)

That being said though I'm not sure how x86 incumbents can compete with it though (perhaps wider core designs and decoding/execution units, but that's not exactly cheap to implement without potentially cannibalizing much more lucrative markets)

But yeah, that's the nice part of having practically unlimited R&D and transistor budget to play with

This post has been edited by chocobo7779: May 11 2024, 03:13 PM

You can also has inefficient ARM cores either if you clocked them to the moon, note that power scales linearly with frequency, and a factor of 2 with voltage
This is why high clock speeds can be a bad thing if the process node or the architecture are not built for it (even Intel admits it on their slide deck, see their flat-ish curves on the power scaling chart):

https://download.intel.com/newsroom/2022/cl...nalyst-deck.pdf

But then for some reason, despite the huge advances in x86 efficiency we still have ridiculously inefficient chips because both Intel/AMD practically clocked them up to near unsustainable levels (I mean, why does 5GHz+ ULV mobile chips and 6GHz+ desktop chips exist?)

This post has been edited by chocobo7779: May 11 2024, 04:10 PM

? Find one game that actually makes use of high speed I/O then - even games like Rift Apart (which claims to utilize disk I/O heavily) will do just fine on a SATA SSD, and considering the very long dev times for AAA titles I think your machine will be long out of date before you need a faster SSD to run the game
People here are vastly overestimating the I/O requirements for disks in modern titles

This post has been edited by chocobo7779: May 11 2024, 04:14 PM

Anyway let's get back to the topic, we're getting a bit derailed there - the original topic is Intel's 14th gen stability issues, not a discussion between ISA and power efficiency

Yeah, but that'll require wider, larger core designs, which are not area efficient and will need larger dies unless you want AMD/Intel to cannibalize their far more profitable server/HPC business that is