Intel considering abolishing CPU sockets for consumer market?

[Grall]

I recently speculated in some other thread that CPU sockets may not survive the end of the decade; now it may look like they possibly might not even survive the middle of this decade.

According to Xbit Labs, Intel might drop socket support in favor of all-BGA packages (IE, soldered straight onto the motherboard) for its haswell-successor broadwell. I can only imagine that Intel is getting antsy in the pants over the continuing (and worsening) slump in PC sales, and is preparing to face a post-PC market like Apple has been going on about for the last couple years now.

In very small form-factor systems, there's neither room, nor need for a CPU socket, and with broadwell targetting a multi-chip module setup with I/O hub integrated alongside the CPU - probably together with voltage regulator hardware as well - and the need for a socket might not be very big, one might speculate.

This sure is an interesting - and potentially frightening - development. The end of PC enthusiast systems could be near - you wouldn't want to spend several hundred dollars on a gamer-grade mobo with an equally expensive CPU soldered in on it, and then risk losing it all to something simple like say, a bad capacitor on the mobo for example.

 

[Npl]

Whats a pc enthusiast? One that spends alot of money, or one that knows a it about PCs and could just replace the capacitor.
To me its all good, one kind can throw even more money away, and the other can fix their thing just like the old times.

But seriously, sockets lost their purpose the time AMD gone Intel and changes them faster than most people upgrade CPUs. They may still be around for a while, but ultimately they are just a cost and technical limitation (signal quality).

 

[Grall]

...And you would of course be able to identify the problem as caused by a bad cap, and also know which cap it is?

(No, it's not always obvious from the outside when a cap goes bad.)

 

[pjbliverpool]

I don't think ive ever replaced a cpu without replacing the mobo at the same time so its no biggie to me, probably just makes things easier.

 

[pcchen]

Personally the only problem I can see with it is that motherboards are more likely to break compared to CPU. However, even this is not a "problem" concerning normal consumers, because if your motherboard gone bust, you just send it back to the manufacturer for repair or replacement. Whether the CPU is on the board or not doesn't seem to matter a lot.

As for "upgrading" the CPU, I haven't been able to do that for years, because Intel were constantly changing their sockets (not just pin count, but also VRM and so).

 

[kache]

Intel is seriously going too far, and any hope we had of AMD doing anything is dieing as well...
The future looks bleak...

 

[Sulik]

Since they've already required a new socket for pretty much every CPU in the last 5-10 years, I don't see it as a big impact for most DIY customers.

 

[Malo]

I don't think ive ever replaced a cpu without replacing the mobo at the same time so its no biggie to me, probably just makes things easier.

This. I don't remember the last time I upgraded the CPU only.

 

[I.S.T.]

Since they've already required a new socket for pretty much every CPU in the last 5-10 years, I don't see it as a big impact for most DIY customers.

Prescott/Cedar Mill/Conroe/Penryn would like a word with you.

Of course those ran into chipset issues sometimes, but that's not the same thing.

 

[nutball]

I don't think ive ever replaced a cpu without replacing the mobo at the same time so its no biggie to me, probably just makes things easier.

^ This.



Surely the real major implication of this is that the number of CPU variants will have to dramatically reduce? Are motherboard manufacturers seriously going to produce a dozen different variants of each motherboard, each with a slightly different flavour of CPU soldered on?

 

[Lightman]

^ This.



Surely the real major implication of this is that the number of CPU variants will have to dramatically reduce? Are motherboard manufacturers seriously going to produce a dozen different variants of each motherboard, each with a slightly different flavour of CPU soldered on?

I think this problem will be dealt with at Dell, HP, Lenovo or any other OEM.
Self-builds would have to move from current market model with tons of various models to more simplified one.

On one hand I will be sad when this happens as a long time enthusiast who changes components on regular basis, and yes, I still change CPU's from time to time. For example my Gigabyte MA-790FX-DQ6 board serverd me for 2.5 years and hosted Athlon X2 6000+, Phenom I 9600 and Phenom II 940. My current S1155 Z68 MSI board hosts i5 2500K but when choosing board I already planned to upgrade to Ivy at some point.

On another hand BGA as mentioned will bring better electrical signaling, which means lower power, faster interface speeds, possibly lover cost for which I'm happy.

Besides, I think that Intel is committed to enthusiast market and will maintain special lines of products for it. Might be expensive and with 2011 pins, but it will!

 

[nutball]

Besides, I think that Intel is committed to enthusiast market and will maintain special lines of products for it. Might be expensive and with 2011 pins, but it will!

Well I don't know about this. It would be interesting to know how significant the enthusiast market is to Intel in dollar terms.

It's one thing to support a niche market with products which are a slight tweak of what you're making by the million anyway (eg. unlocked overclockable CPUs). It's a whole other proposition to serve a niche market with products that require a whole other way of engineering. Regardless of the price premium of the individual items.

Obviously the enthusiast market has the potential for the halo effect, which has dollar value in a real sense that may be hard to quantify. But with Intel's competitor imploding spectacularly they may feel in the future that they don't have to care about this so much.

 

[Otto Dafe]

As long as no one is talking about soldering on the stock HSF I say more power to them. Seriously though I would be shocked if this affected (sp?) 1/1000 of a % of consumers in their lifetimes.

Edit: upon thinking further the only real use case I can imagine for sockets anymore would be small resellers who do their own refurb. If any of them still exist.

 

[CouldntResist]

Prescott/Cedar Mill/Conroe/Penryn would like a word with you.

Of course those ran into chipset issues sometimes, but that's not the same thing.

With FSB varying from 533 to 1333 MHz, interchangeability was actually very limited. Only 945/975 family of chipsets supported both Prescott/Cedar Mill and Conroe/Penryn. And even that only for small, overlapping range of FSB.

 

[I.S.T.]

Most core 2s on the retail market used 800 or 1066 FSB, IIRC.

Edit: Also, I did mention chipset issues in my post.

 

[HMBR]

on my previous 3 motherboards, I've used 4 different CPUs on the same LGA 775 MB, 2 different on the AM2, and I'm currently using a 1155 MB, (with a sandy bridge i3), which I may well upgrade to some other 1155 CPU later...

so for me it's not a good thing, it also means less options, nowadays you can pair a cheap MB with a faster CPU, this could well end...

3 years ago a $50 MB could support a high end CPU, and even be used for overclocking,

some AM2 boards from 2006-7 can run even a Phenom II X6 from 2010,
I remember also some 440BX boards from 98, with the correct adapters still running new Tualatin CPUs, with 200%+ higher clock many years later...

 

[Davros]

Didnt intel have a plan to sell locked cpu's and then charge for unlocking them, this could tie in with that
just have a single cpu (or 2) that is capable of being clocked at several different speeds. that way board makers buy 1 cpu and unlock it to the performance level the customer wants

 

[fellix]

I recently speculated in some other thread that CPU sockets may not survive the end of the decade; now it may look like they possibly might not even survive the middle of this decade.

Xbit's article apparently got a bit lost in translation. Here's a brief from the original source (PC Watch):

-- Intel will not provide new products for Desktop and non-BGA laptop segments in Broadwell era
-- Instead, they will provide higher clocked Haswell for those segments in 2014
-- Broadwell is "more than tick", and it will include some technologies that were previously planned for Skylake
-- This is because Intel needs to be more competitive in the tablet market, and this may mean the end of Tick-Tock strategy
-- It mentions nothing about Skylake and later or if they will be LGA or not for the desktop

 

[3dilettante]

The increasing integration at the die and socket level is leaving little for board manufacturers to differentiate themselves. Bringing the southbridge on die is leaving them barely anything but the slots, connectors, and traces.

If the upcoming consoles have 4-8 GB of RAM, this points to one less thing that the bulk desktop market will need to care about in the future. With at least some rumor of the consoles using 2.5D integration at the outset or eventually, we'll have mass-produced packages with about as much RAM as the average desktop user is going to require prior to replacing a whole machine.
That means there will be little going off-package for the desktop except PCIe, networking, and power/ground.

I'm not sure how a board manufacturer is going to have a place here, or whether we'll have more than a few speed/feature bins as opposed to the number wheel of features/clocks we have now. There's going to be a higher emphasis on turnkey features or fuse-blown chips that have very high yields in production, which may imply more conservative binning for the mainstream market.

Expandable RAM may be the province of workstation and high-end servers only. This may be a premium feature, and I'm puzzling over if this means we'll have high-end chips with both integrated package/interposer DRAM and external DRAM connections for a slower, massive pool of memory. The latency and bandwidth possibilities for the interposer seem hard to pass up in a lot of cases. The downside is an expensive chip that puts enthusiasts solely in the territory of boards not aided by large volumes or the subsidy of the general market. The PC would track with the embedded market.

The difficulty of having CPU dies with chimera memory controllers could be resolved more readily in the future if 3D stacking takes off. If DRAM stacks under the CPU become possible, the DRAM itself could use a form of hybrid memory cube technology where the logic layer of the memory can house the off-package connections and the in-stack DRAM.
That seems far-off, though.

EDIT: Another possibility is some kind of pass-through HMC, which wouldn't require stacking with the CPU die.

 

[fellix]

If Intel, at some point in the future, truly ditches socketed processors for the consumer market and moves to BGA-only bulk shipments, that doesn't mean there shouldn't be some proprietary interfacing standard to replace it. Big companies, like Asus and Gigabyte could come with some form-factor of, let's say, a slot-in CPU board much like MXM, fit for both mobile and desktop/WS SKUs. I always liked the promising idea of the original Slot 1 by Intel, but the thing never caught with the economy of scale in the 90s. Intel envisioned a grand plan for that form-factor, like integrating DSP/Media co-processors & etc., but to no avail. Now with the advancement of the lithography and integration one could have a powerful CPU+IGP and multi-gigabyte of fast memory (high-density DDR4) soldered on a single small PCB. The main board would still handle most of the periphery and maybe direct integration of SSD storage in some form.