Discussion Anyone else bored out of their mind due to mainstream CPU market stagnation?

[igor_kavinski]

No mainstream desktop motherboard has this, so no Strix Halo for desktop.

Unless they start soldering the mobile SoC package with RAM on the mobo...

 

[DavidC1]

JEDEC Publishes LPDDR5X Standard at up to 8533 Mbps

www.anandtech.com

View attachment 91553

Funny. That says LPDDR5/5X is single channel.

I already explained it man.

PCs have always based it on the DIMM, and that's 64-bit. By those standards LPDDR5 is 1/2 channel. You ever seen a DIMM? You see multiple chips there? That table is per chip, not per device as in a DIMM.

You'd need SIXTEEN of those to reach quad channel or 256-bits.

And each bit width requires wires, which requires routing on the motherboard. That takes up additional space and cost, especially since we've been dual channel for eons.

Generative AI? Check the output very, very carefully for any obvious mistakes, unless you want to be embarrassed in front of everyone.

This I agree. It's fine if you want to use it for things that don't require precision, like graphics. Hey, isn't the company that promotes AI also make money hand over fist by doing graphics?

3DFX is known for bringing graphics acceleration to the consumer by reducing bit width to 32-bit from IEEE 64-bit standard among other tricks, because consumer graphics didn't need that accuracy, while 3D workstations did.

I wonder if the errors generative AI makes have anything to do with the further reduced precision it's using? It's not even 32-bit. It's using Int8 now. 8-bits.

People who push AI such as Raja Koduri often claim that they don't understand how it works. Well, if you don't understand it, why are you working on it? How can you build something you don't understand?

 

[Schmide]

I don't see the malaise in the market, there are a myriad of different platforms variance in with memory, compute and bus. You have the extra level cache with the AMD 3d line. You have handhelds with enough memory bandwidth to run an APU at near low end discrete levels. Apple puts memory closer to the system to get lower power and improve prefornace. USB4 is near pcie speeds. And those pcie speeds keep becoming faster and more available. At the lower end, if you don't have a pcie, co-opt an nvme. At the high end, every motherboard has at least pcie4, multiple mvme that makes raid basically unnecessary.

Intel for all its woes produces a bunch of SBC brick computers that as much as their processors are just a bit below AMD, their network and USB are often generations above what AMD offers in the segment. If you ever look at where AMD drops the ball this is it.

I bought a rembrandt refresh Ryzen 7 7735HS laptop that AMD updated to be io rich. It's still the same zen3+ cores, but they updated all the lanes to be pcie4 with 4x nvme. Sadly the improved usb4 is not exposed on this lenovo. (I heard rumors that neferious drivers can expose this) It has a 4060 in addition to the 680m, so yeah I don't get the AI engine of the, nor the avx512 of the 7840hs, but it's a fabulously polished turd. Thinking of exposing one of the nvmes to run an external card of some sort. Why? because I can. That's whats cool with the current generation.

On the super low end, we no longer have to by a compute board to expose the pcie on the raspberry pi. The pi5s have a ribbon cable that if you can get one can expose a pice2 (maybe3 with tweeking) lane. A high speed drive on such a system is game changing.

In some ways the, core ultra processors series 1 formerly meteor lake (aka 155h) is a repackaging of intel's portfolio products into a first gen (do we still say foveros) chiplet, is a very polished... first step. It is weird that they used chiplets on the low end as AMD uses them on the high end. More uniqueness.

I could rant on, but just because the next generation isn't here doesn't detract from this generation and all it's quirks.

 

[Shmee]

A big downvote for both Sarah and Pohemi regarding subscription. Are you people nuts? Why give ideas to the idiot marketing people from the big companies lurking on these forums????

You can't down vote Sarah Kerrigan, she has been through so much! Infested twice! What an ordeal. Also she is darn fine, and a great hero. So is Nova Terra.

 

[DrMrLordX]

Intel for all its woes produces a bunch of SBC brick computers that as much as their processors are just a bit below AMD, their network and USB are often generations above what AMD offers in the segment. If you ever look at where AMD drops the ball this is it.

Does AMD even compete in that space? Sure OEMs could use their mobile hardware there but . . .

You can't down vote Sarah Kerrigan, she has been through so much! Infested twice! What an ordeal. Also she is darn fine, and a great hero. So is Nova Terra.

She has a good Widowmaker skin in OW/OW2 as well:

Too bad about that scoliosis.

 

[SteveGrabowski]

@DigDog @biostud @Torn Mind @adroc_thurston @DAPUNISHER @mikk @SteveGrabowski @VirtualLarry @uzzi38 @mxnerd

You guys wanna chime in with your wisdom?

I'm loving how cheap good gaming cpus are today and that the $160 i5-12400F I bought in May 2022 has already aged better as a gaming cpu than the $185 i5-4460 I was thinking of getting back in 2014 when the cpu market was so stagnant. Thank god I spent $65 more for a Xeon E3-1231v3 for the extra threads to better match the core counts of the last gen console gaming cpus. But also seems great that going cheap on cpu hasn't been limiting me at all so far this console gen and might not period since parallelization is much less critical this gen now that the consoles use a cpu with respectable IPC and clocks unlike the Jaguar cores of last gen.

 

[Schmide]

Does AMD even compete in that space? Sure OEMs could use their mobile hardware there but . . .

If you watch tiny mini micro at serve the home, there are AMD offerings. If you want it to double as at TV or such computer AMD is often better. Intel just always had the right 2.5-10gbe network with usb3-4 ports, and wifi. edit: here

 

[VirtualLarry]

Got a bread machine for Xmas, never owned one before, so I'm having a good time with it.

Those can be very fun too.

 

[VirtualLarry]

idk, maybe you need a 16 core because you do photo editing, and that's £500 .. dude that is NOTHING.

VFM is really high right now in high-end consumer / pro-sumer, aka higher-end Ryzen.

Part of that is that AMD is no longer (well, c'mon, get a move on) currently supporting ThRipper. If they were, you would be paying a lot more per-core for a true pro-sumer platform.

 

[DrMrLordX]

If you watch tiny mini micro at serve the home, there are AMD offerings. If you want it to double as at TV or such computer AMD is often better. Intel just always had the right 2.5-10gbe network with usb3-4 ports, and wifi. edit: here

Oh man, A6 and A10 CPUs? Hurk! Ugh bleh. Mendocino would be so much better in one of those.

 

[VirtualLarry]

https://www.newegg.com/p/2SW-003Z-00005?Item=9SIBFJJJ6G8409&source=f&cm_sp=web_notification-_-retention-_-Pers_RetenEmailWebNotify_NVTC_Exclusive_Care-_-PDP-_-9SIBFJJJ6G8409-_-SKAGZAGZJFZGS1704602485689&utm_campaign=web_notification-_-retention-_-Pers_RetenEmailWebNotify_NVTC_Exclusive_Care-_-PDP-_-9SIBFJJJ6G8409-_-SKAGZAGZJFZGS1704602485689&utm_medium=push&utm_source=web%20push&rec_value=-_-309-_--_-9SIBFJJJ6G8409-_--_-9SIBFJJJ6G8409_1___-_-Pers_WP_PriceDrop-_--_-&cr_rid=web_notification

 

[RnR_au]

Hopefully this will be a boredom killah...

 

[moinmoin]

I'm so glad that I was so busy in those years that it didn't matter to me one bit what was going on in the CPU world.

Get busy again then? You essentially complain about having the memory of a fruit fly and post an exploding filler thread while it's summer only down under!

 

[uzzi38]

@DigDog @biostud @Torn Mind @adroc_thurston @DAPUNISHER @mikk @SteveGrabowski @VirtualLarry @uzzi38 @mxnerd

You guys wanna chime in with your wisdom?

Tbh the only thing boring me a little is that we're in this little lull where nothing interesting is launching.

Meteor Lake is fine but feels like a very transitionary product. Zen 4 is old news now and now that I've got a 7840U on hand, I'm not even very interested in STX either now. It's clear how bad the memory bottleneck really is when we operate nowhere close to peak boosts at 15W yet we already have most of the performance.

It's really Zen 5 desktop (in particular V-Cache, because that might be when I upgrade my desktop), Strix Halo (not really regular Strix) and Lunar Lake that have me super excited.

 

[yuri69]

Well, yea. There are no interesting leak channels these days. All we have are those 2 mindless YT channels, a few hit-miss Twitter accounts. That's it. The rest is a horrible sewer-level junk.

Roadmaps, internal info, rumored plans, cancelled products, etc. These are quite scare nowadays. Instead we get baseless speculations painted as leaked info.

The idea of AMD's with the V-Cache and TR releases filling the "release gap" sounds cool in theory but nope. It brings pain, since you simply have to wait for the release of the product which is more or less known... Everybody knows the Zen 5 + V-Cache will be better for gaming than the vanilla Zen 5, right?

Anyways, Zen 5 will most probably be held back by the infrastructure. So Zen 6 it is... Which makes Intel kinda interesting. Arrow Lake is rumored to suffer from low clocks, but it opens a nice way to push multiple refreshes/revisions. That's interesting.

 

[SarahKerrigan]

Thread was about Intel eDRAM which used a separate 22nm die and more modern DRAM made it pretty much pointless. Using on die eDRAM requires special process flows to enable the deep trench capacitors which increases the cost. Intel doesn't have any advanced nodes that enable eDRAM and really no one is willing to continue to implement it now that other technologies have come around and are proving an overall better solution. I also don't think eDRAM would be able to scale at all due to the required capacitors but I haven't followed it's development for the past few years. IBM's mainframe products are incredibly niche and for very specific markets that don't translate well to a broader customer base. I guess it made sense for them, but even they had to move off of eDRAM eventually as well or they'd still be stuck on GF 12nm FinFETs. I guess there still could be a possible niche for a stacked eDRAM but that would have it's own hurdles to overcome and I haven't seen any serious efforts in that direction.

I was never saying that eDRAM tech as it exists today (so mostly on obsolete lithography) would be competitive - but that's not due to eDRAM being an undesirable tech, but due to the IBM/GlobalFoundries fallout after 14HP. eDRAM was a major element of IBM's roadmap prior to the cancellation of the GF 7 node, and having to switch to SRAM caches was a big part of why Power10 has 30 cores instead of the roadmapped 48.

eDRAM allowed big, dense caches with a tiny latency penalty without resorting to advanced packaging. That's worth something.

 

[SteveGrabowski]

Hopefully this will be a boredom killah...

AI hentai in Stable Diffusion is the way of the future!

 

[GaiaHunter]

As someone that just finished replacing my last skylake for a zen4, I'm fine with the CPU state... Now those GPUs...

 

[coercitiv]

Yo, psssst. You, do you think 2024 CPUs are boring? Come here, come closer: I have 2014 CPUs to show you. Don't be afraid, I don't bite... you can pick ANY quadcore you want!

 

[adamge]

Products that have made much progress and we have arrived in Shangri-La:
- SSD
- consumer CPU speed

Products that stagnated and bring pain:
- consumer CPU max RAM capacity (128GB)
- large, high refresh monitor (i.e. 32" 4k 120hz)
- Windows (size on disk, size in RAM, size of updates, length of time to perform updates, vulnerability to hackers)

 

[aigomorla]

Im sad we went from "overclocking so easy your grandma can do it"
To "overclocking so dumb'd down, you really dont need to do it."

 

[DAPUNISHER]

Im sad we went from "overclocking so easy your grandma can do it"
To "overclocking so dumb'd down, you really dont need to do it."

Matter of perspective me thinks. Competition has gotten so fierce that neither vendor can leave anything off the table. I consider that a good thing. I get lulz from the debates evolving from - "You have to overclock and deal with the extra power, heat, and $$ cooling on your CPU to keep up!" To - "You should not run OOB settings because of the power, heat, and $$ cooling it needs to keep up!"

It's still worthwhile with some CPUs, e.g. cheap Ryzens like the 5500. Can do it on cheap boards too. It's why I disagree with reviewers picking the 12100 for the ultra budget king. PCIe 4.0 is much less impactful than having the extra cores and threads at the $100 and under price point. The 5500 having 400-500MHz extra on tap is hard to argue against too. I'm testing 4600MHz with 4000MT/s synced at the moment. It should merk the 12100 setup this way.

 

[Markfw]

Im sad we went from "overclocking so easy your grandma can do it"
To "overclocking so dumb'd down, you really dont need to do it."

especially when the title of this forum is "CPUs and Overclocking"

 

[aigomorla]

especially when the title of this forum is "CPUs and Overclocking"

Well we went from tuners... to enthusiest.... to gamers (because some wierd reason gamers need to overclock, and all the overclocking boards are for gamers)... to Turbo Core/EXPO - Turbo Boost/XMP.

I really don't see the need to overclock anymore as dynamic boosting is probably better for most users... meaning very fast boosting of a partition core count, instead of a full constant core clock across the entire CPU.

Meaning id rather have cores 1-4 clocked at 4.8ghz and 5-8 clocked at 4.2ghz vs having all cores 1-8 clocked at 4.4ghz.

 

[Markfw]

Well we went from tuners... to enthusiest.... to gamers (because some wierd reason gamers need to overclock, and all the overclocking boards are for gamers)... to Turbo Core/EXPO - Turbo Boost/XMP.

I really don't see the need to overclock anymore as dynamic boosting is probably better for most users... meaning very fast boosting of a partition core count, instead of a full constant core clock across the entire CPU.

Meaning id rather have cores 1-4 clocked at 4.8ghz and 5-8 clocked at 4.2ghz vs having all cores 1-8 clocked at 4.4ghz.

I agree except for EXPO/XMP. Its created/sanctioned by AMD/Intel, just not the official memory speed. Maybe thats reserved for OEM ??