[Laforet]

Intel actually intended for LGA1151 to remain unchanged for Coffee Lake but found out late in the testing process that many existing motherboards did not have enough power delivery capability to support the planned 6 and 8 core parts. Hence the decision to lock them out in software only. They are probably aware of the bad optics but decided that it’s better than trying to deal with the RMAs later.

It’s very similar to what had happened in 2006 when the 65nm Core 2 series were released in the same LGA775 package used by 90nm Pentium 4s, however the former mandated a specific VRM standard that not all comtemporary motherboards supported. Later 45nm parts pretty much required a new motherboard despite having the same socket again due to power supply issues.

AMD went the other route when they first introduced their 12 and 16 core parts to the AM4 socket. A lot of older motherboards were clearly struggling to cope with the power draw but AMD got to keep their implicit promise of all-round compatibility. Later on AMD tried to silently drop support for older motherboards when the Ryzen 5000 series were introduced but had to back down after some backlash. Unlike the blue brand they could not afford to offend the fanboys.

P.S. Despite the usual complaints, most previous Intel socket changes actually had valid technical reasons for them:

- LGA1155: Major change to integrated GPU, also fixed the weird pin assignment of LGA1156 which made board layout a major pain.

- LGA1150: Introduction of on-die voltage regulation (FIVR)

- LGA1151: Initial support for DDR4 and separate clock domains

This leaves the LGA1200 as the only example where there really isn’t any justification for its existence.

[Rinzler89]

Thank you for providing valuable insight. I wish these kinds of comments would end up at the top instead of the usual low quality "hurr-$COMPANY evil, it's all because greedy obsolescence-durr", from people who have no idea how CPUs and motherboards work together and the compatibility challenges that come when spining new CPUs designs with big difference that aren't visible to the layman who just counts the number of cores and thinks there can't possibly be more under the hood changes beyond their $DAYJOB comprehension.

Here's a video from gamer's Nexus on AMD's HW testing lab, just to understand the depth and breadth of how much HW and compatibility testing goes into a new CPU, and that's only what they can talk about in public. https://www.youtube.com/watch?v=7H4eg2jOvVw

[qball]

> LGA1155: Major change to integrated GPU, also fixed the weird pin assignment of LGA1156 which made board layout a major pain.

Of course, the P67 chipset was trivially electrically compatible with LGA1156 CPUs; Asrock's P67 Transformer motherboard proved that conclusively.

That said, the main problem with 1155 was their locking down the clock dividers, so the BCLK overclocking you could do with 1156 platforms was completely removed (even though every chip in the Sandy Bridge lineup could do 4.4GHz without any problem). This was the beginning of the "we're intentionally limiting our processor performance due to zero competition" days.

> LGA1150: Introduction of on-die voltage regulation (FIVR)

Which they would proceed to remove from the die in later generations, if I recall correctly. (And yes, Haswell was a generation with ~0% IPC uplift so no big loss there, but still.)

[Laforet]

> P67 chipset was trivially electrically compatible with LGA1156 CPUs

Well it’s possible to shoehorn in support for the determined but iGPU support is definitely out of reach and I am not sure what segment of the market is that targeted to. Seems like an excuse for AsRock to get rid of their excess stock. The socket change was actually very well received by everybody in the industry.

> Haswell was a generation with ~0% IPC uplift so no big loss there

You are right that FIVR did not last long in that particular iteration. However Haswell does have a 10% to 30% IPC advantage over the previous gen depending on the test[1].

Haswell also added AVX2 instructions which means that it will still run the latest games whereas anything older is up to the whims of the developer (and sometimes denuvo, sadly)

https://www.anandtech.com/show/9483/intel-skylake-review-670...

[Dylan16807]

Why does a socket that can support DDR3 or DDR4 need to be different from a socket that only supports DDR3?

And with the current socket being 1700, they're going to change it again for the next generation to 1851, and with a quick look I don't see any feature changes that are motivating the change. (Upgrading 4 of the PCIe lanes to match the speed of the other 16 definitely does not count as something that motivates a socket change.)

So by my reckoning, half their desktop socket changes in the last decade have been unnecessary.

[qball]

Because DDR4 is electrically different and memory controllers are all on-die.

Intel could get away with doing that pre-Nehalem because the memory was connected via the northbridge and not directly (which is what AMD was doing at the time; their CPUs outperformed Intel's partially due to that), so the CPU could be memory-agnostic.

AMD would later need to switch to a new socket to run DDR3 RAM, but that socket was physically compatible with AM2 (AM3 CPUs would have both DDR2 and DDR3 memory controllers and switch depending on which memory they were paired with; AM3+ CPUs would do away with that though).

There were some benefits to doing that; the last time Intel realized them was in 2001 when RD-RAM turned out to be a dead-end. Socket 423 processors would ultimately prove compatible with RDRAM, SDRAM, and DDR SDRAM.

[Dylan16807]

> Because DDR4 is electrically different and memory controllers are all on-die.

Them being on-die is exactly why you don't need a socket change to take full advantage of DDR4, since they directly showed a socket can support both at once. Unless you're particularly worried about people trying to buy a new motherboard for their existing CPU, but who does that? You can tell them no without blocking CPU upgrades for everyone else.

> pre-Nehalem

> AM3 CPUs would have both DDR2 and DDR3 memory controllers

LGA1151 supported both DDR3 and DDR4.

[xxs]

>many existing motherboards did not have enough power delivery capability to support the planned 6 and 8 core part

Coffee lake's 8700k had 6 core not 8, the refresh lake-s did feature 8 cores. The release TDP of 8700k was the same as 6700k - 95W, of course it consumed more at peak, and when overclocked.

The support would still depend on the motherboard manufacturers adding the CPU support (along w/ the microcode) to their BIOS. Many high-end boards, e.g asrock z170 extreme6/7, would have supported 8700k.

The situation is no different today that many (most) boards do not top support top end processors anyways, due to poor VRMs, even when the socket is the same. (or if they support the CPUs are effectively power throttled)