Back to Article

  • Lerianis - Thursday, February 20, 2014 - link

    3 different processor port designs? Intel, what the hell were you thinking here? You come out with ONE port design or at the most two that are mutually exclusive, one for servers and one for regular desktops/laptops.

    I am seeing a lot of people with blown processors in the future due to this.
  • GL1zdA - Thursday, February 20, 2014 - link

    And how do you expect one socket for processors with different features? The EX uses 4 QPI links, the EP 2 QPI links, the Haswell-EP will use DDR4 - how many contacts should your "one socket to rull them all" have? Ten thousand? Reply
  • fokka - Thursday, February 20, 2014 - link

    you are right, but simply naming all lga 2011 still is a stupid idea. Reply
  • dylan522p - Thursday, February 20, 2014 - link

    Why? Anyone who cares knows. Reply
  • repoman27 - Thursday, February 20, 2014 - link

    And if the we're talking about a 2011-land Land Grid Array package, then LGA2011 is how we'd refer to it. There are plenty of semiconductor manufacturers that produce similar packages that are not pin compatible or even related, e.g. QFN64 or TSOP28.

    Intel refers to the sockets as "Socket R", "Socket R1" and "Socket R3".
  • repoman27 - Thursday, February 20, 2014 - link

    And the packages and corresponding sockets (or ILMs in Intel parlance) are keyed differently, so you'd likely break something before you could cause electrical damage to the processor by inserting it into an inappropriate system. Reply
  • HisDivineOrder - Thursday, February 20, 2014 - link

    Someone who cares won't know. It's inevitable. There'll be that one guy complaining on a forum somewhere about how he was completely surprised by this.

    Inevitably, his name will be Larry. Or Bob.
  • LordOfTheBoired - Friday, February 21, 2014 - link

    Why name them at all, then? If the name is useless, and they've created a situation where IT IS, just don't give the socket a name.

    I'd rather see them say nothing at all about the socket than than give a slew of different and largely unrelated connections a shared name. This is actually a WORSE situation than if the motherboard and processor specs said nothing about compatibility, because now someone who is unsure can check a bullet list, see that he has an LGA2011 processor and an LGA2011 motherboard, and expect them to work. With NO ready compatibility information, he will be FORCED to either do the necessary research or buy AMD out of sheer frustration. And this is dumb enough that "AMD has an agent inside Intel sabotaging the company" is the MOST rational explanation, which says a LOT about how insane this mess is.
  • Kevin G - Thursday, February 20, 2014 - link

    There can be multi-purpose pins that are configured upon boot. For example, pins for 8 lanes of PCI-e could be configured to be an extra QPI link (note that Ivy Bridge-EX only has 32 PCI-e lanes instead of the 40 on EP version). Ditto for the memory bus pins: reuse them for DDR3, DDR4 and the serial memory buffers. The pin count would increase but not drastically so.

    This is especially more pronounced with socket 1156/1155/1150. The main difference between them was display out and clock generation. Intel could have easily a few extra pins for display IO growth, ignored the external clock input on later CPU's and kept the consumer line on the same socket for the past 5 years.
  • ZeDestructor - Thursday, February 20, 2014 - link

    Utterly pointless though, since you still have to change motherboards because you need a new chipset.

    On the other hand, within LGA1155, you can use both Sandy Bridge and Ivy Bridge with no issues (bessides possible UEFI updates, and being limited to PCIe 2.0 on x6x compared to PCIe 3.0 on x7x boards)

    I for one do not lament moving away from the LGA775 years of choosing which chipset and which motherboard implementation of said chipset supports the CPU I have.

    So far Intel has kept the sockets inline with chipset and power requirements since LGA1366. It is true that I can't replace my motherboard with a newer board if my current one dies, but tbh, it's made my life easier: Gen X with chipset series Y will work across the entire generation, and by the time a motherboard dies, I want a new CPU as well.....
  • Kevin G - Thursday, February 20, 2014 - link

    Oh? I could see why a Lynnfield chip wouldn't work with a Z87 chipset: no integrated graphics and it requires a clock input that the Z87 doesn't provide. However, what prevents the opposite? Say a Haswell board on a P55 chipset? Things like the external clock generator could be simply ignored on Haswell. PCI-e speeds would be likely be limited to 2.0 speeds. Otherwise, I'm not seeing much of a real limitation that couldn't have been addressed with a bit of foresight for a one-socket-to-rule-them-all scenario.

    Things are even more similar between socket 2011 variants. The main differences here don't deal with the chipset but rather socket level features (memory type, PCI-e lanes and QPI links).
  • TheinsanegamerN - Thursday, February 20, 2014 - link

    The only thing that comes to mind is that haswell moved some voltage-control circuits onto the CPU itself. that would probably play havoc with the external voltage control on older boards. Reply
  • piroroadkill - Thursday, February 20, 2014 - link

    Intel is obsessed with different sockets, but AMD tries to keep them the same.

    Intel needs to stop screwing things up. Fuck, add a pin that does nothing and call it 2012 if you want, but they should never have several "2011" which aren't even the same...
  • willis936 - Thursday, February 20, 2014 - link

    They did that with 1155. Reply
  • nekoken - Thursday, February 20, 2014 - link

    AMD used to be pretty great for socket compatibility but the aborted FM1 socket kind of ruined that. Reply
  • Flunk - Thursday, February 20, 2014 - link

    There was no way around that, You can't get the integrated GPU functionality into an AM3 form-factor. Reply
  • Death666Angel - Thursday, February 20, 2014 - link

    Huh? FM1 APUs already had iGPUs (Llano), but AMD decided not to support that anymore with the launch of Trinity. Has nothing to do with the AMx-line and iGPU functionality not being available. Reply
  • TiGr1982 - Thursday, February 20, 2014 - link

    Not only FM1 - FM2+ CPU do not fit/work at all in FM2 M/B.
    That was NOT the case for AM2+/AM2 in the past - these frequently worked together.
    So AMD isn't better in this regard these days.
  • ZeDestructor - Thursday, February 20, 2014 - link

    You think this is bad.... Look at the laptop CPUs... Socket 479 from the 110nm Pentium M/Pentium 4 (backwards compatible with S478 chips as I recall too) all the way to the quad-core 45nm mobile Penryn cores, with nothbridges spanning from AGP8x/PCI/IDE/DDR all the way to PCIe/SATA/DDR3...

    On the other hand, we do have rPGA-988 and rPGA-988B for SB/IVB, so they could have done a lot better naming-wise...
  • mapesdhs - Thursday, February 20, 2014 - link

    The trouble with AMD though is that even when a chip physically fits
    a socket ok, often the mbd vendor does not release a BIOS update to
    support the chip in question. This happened a lot with the Phenom II;
    there are numerous AM2 boards which don't support them due to lack
    of BIOS updates.

  • FieryUP - Thursday, February 20, 2014 - link

    AFAIK Haswell-E/EP socket is designated (by Intel) as LGA2011-3, and SNB-E/EP/IVB-E/EP socket is LGA2011-0. If IVB-EX would have LGA2011-2, then it would just be quite logical, although I'm not sure what happened to LGA2011-1 then :) Reply
  • repoman27 - Thursday, February 20, 2014 - link

    Yep, you got it. The original SNB-E/IVB-E socket was referred to by Intel as "LGA2011-0" or "Socket R" and the HSW-E variant as "LGA2011-3" or "Socket R3". More details here:

    Somewhat oddly, the Intel parlance for the Xeon E7 v2 socket seems to be "LGA 2011" or "Socket R1".
  • Berzemus - Thursday, February 20, 2014 - link

    Just check the supported CPU list at your motherboard's vendor's website. The required BIOS version is a bonus.

    As for AMD: FX-9000 series 220W TDP made them incompatible with (virtually ?) all previous AM3+ motherboards.
  • TiGr1982 - Thursday, February 20, 2014 - link

    In case of FX-9000 it's not an electrical/logical compatibility issue, but a TDP issue (for M/B - specific high requirements for CPU socket power delivery).

    There is nothing totally new here with FX-9000 - this kind of issue existed several years ago for Phenom I X4 / II X4 140W models - these particular models were also not supported by a bunch of cheaper/simpler M/Bs.
  • TiGr1982 - Thursday, February 20, 2014 - link

    Seems to be a non-issue at all, because:
    1) professional IT staff is dealing with Xeons, so by default, being professionals, this staff knows what is for what.
    2) enthusiasts are dealing with desktop LGA2011 CPU's. These guys, being enthusiasts/prosumers, also happen to read the web and to know what is for what in advance - before actually buying the stuff.

    So, it's just an interesting engineering curiosity, but not a problem at all.
  • maecenas - Thursday, February 20, 2014 - link

    Not a problem until someone messes up and ruins chip and/or board. You have to admit this is a ridiculous system. At the very least they could call them 20011a, 2011b, etc. Reply
  • TiGr1982 - Thursday, February 20, 2014 - link

    Yes, I admit :)
    For the consumer/end users desktop segment, there still should be some designation differing current LGA2011 (Sandy Bridge-E/Ivy Bridge-E) vs future LGA2011 (Haswell-E).

    There probably will be one (like 2011-3 or something for the new desktop LGA2011),
    when Haswell-E for desktop will be released (rumours telling in Q3 this year).
  • repoman27 - Thursday, February 20, 2014 - link

    They do have names (see the above comment by FieryUp). Granted, they're not necessarily good names, but it is Intel we're talking about here. They are also notched differently so the chips cannot be inserted into an inappropriate socket/system without the use of considerable force. Reply
  • TiGr1982 - Thursday, February 20, 2014 - link

    Of, course they do have internal names inside Intel (say, 1156 is H1, 1155 is H2, and 1150 is H3). The question is how the new desktop LGA 2011 (Socket R3) will be named consumerwise - what will be written on the CPU box at retail :)
    LGA2011-3, probablly?
  • Assimilator87 - Thursday, February 20, 2014 - link

    Well hopefully, someone spending $6500 on a processor will check board compatibility beforehand. Reply
  • ZeDestructor - Thursday, February 20, 2014 - link

    If he/she isn't doing so (or buying prebuilt boxes), he/she deserves a good reprimand. Reply
  • JlHADJOE - Friday, February 21, 2014 - link

    This. The vast majority of Ivy-EX systems will come in pre-built servers, and anyone who finds himself inserting an Ivy-EX chip manually will probably know what he's doing. Reply
  • ShieTar - Friday, February 21, 2014 - link

    The amount of money available in the US has increase by 300% over the last two decades, and by about 600% in the UK. You bet there are plenty of investment bankers out there buying >1000$ CPUs for themselves or their kids without much understanding of what they're doing. Reply
  • psyq321 - Monday, March 10, 2014 - link

    One cannot mess the chip and/or board because the sockets are keyed differently.

    Unless he/she is an idiot, in which case even the different pin count and/or socket name would probably not help, either.
  • smithrd3512 - Thursday, February 20, 2014 - link

    hmm might need a big book to keep up with Intel's sockets changes Reply
  • Aslan7 - Thursday, February 20, 2014 - link

    Sockets should have different name 2011a, 2011b, 2011c, or they should be compatiable with configurable pins. Why?

    I bought a Laptop just as dual core mobile chips were coming out. This put price pressure on single core laptops. The Dell Inspiron 9200 was a 17 inch laptop in a business suit, that was actually the same exact laptop as the Dell XPS for gaming. It had a GeForce go6800 with 256mb dedicated graphics ram in it that came within 95% of the performance of a stock Geforce 6800 desktop card. At the time The Pentium-M had a good long run, and Intel's process technology had greatly improved (Newer processors needed less voltage for the same work and had more cache). After a quick check to make sure it worked I pulled the 1.6GHZ with 512KB cache processor out and swapped in a $90 from eBay 1.7GHZ Pentium-M 400MHZ bus 2MB cache processor. I dropped some thin copper wire between some holes in the socket for the processor to tell the system that this was a processor with a 533MHZ system bus, thus the processor ran at 2.26GHZ. It needed a little extra voltage so I dropped another little copper wire across some pins that permanently increased the voltage. I then put the heatsink back on with AS-5 (Artic Silver 5) amazing stuff I though it was BS, but yes you could see temps 10C cooler with AS-5 over stock. (AS-5 breakin was a bitch though. I didn't see the lowest temps until 120 hours in and about a dozen power off cooling cycles) I did the same with the graphics card. Better dissipation = less power, and lower fan speed, meaning less power again.

    I started the laptop back up. It worked the first time. I installed Inspiron fan control I think it was. I created a temperature dependent fan map quieting the computer. Inspiron fan control or whatever program it was also controlled clockspeeds and voltages. I set it to run with realtime priority, and tested each speed stepping until I found the lowest voltage it would operate at, then set the clockspeed to load dependent so if a program could use more clockspeed it got it. I created a gaming profile and an office profile, and settings changed based on if it was on mains electricity or battery.

    For the Graphics card the dell provided drivers were 18 months out of date this was a "business laptop after all and Nvidia's drivers wouldn't work on laptops. There were sites that existed to mod the inf file for the driver telling it that it worked with your card, and to enable all the hidden features including overclocking. I found stable overclocks and gained almost 10% in fps. Again I set up an office and a gaming profile.

    That laptop roared. I bought it for $1200, did $800 in upgrades and ended up with a computer that was worth $3500 retail. That's why I want my sockets compatible.
  • Aslan7 - Thursday, February 20, 2014 - link

    Dell Inspiron 9200 17" stock vs modded
    Totally stable end result.
    cost $1200 stock + $800 in upgrades = $2000 total
    1.6GHZ Pentium-m 533FSB 512KB cache to 2.26GHZ 2MB cache 533FSB (1.7GHZ 400FSB stock)
    512MB memory, upgraded to 2048MB
    Stock mpci wifi upgraded to atheros Wireless G (better signal processing and sensitivity, higher transmit power)
    Stock wifi antenna replaced with high gain internal antenna
    40GB 5200RPM slow HD upgraded to fast 160GB PMR HD that matched the best 7200rpm drives in the benchmarks
    38 watt hour batter swapped for 80 watt hour
    80 watt power supply swapped for 150
    Dual layer DVD burner
    17" 1440x900 screen for native resolution gaming

    I could go all out gaming for 2 hours on battery with the screen at 50% brightness.
    That's why I like socket compatibility and standard parts.
  • Aslan7 - Thursday, February 20, 2014 - link

    Another reason I like socket compatibility, though even that isn't a prerequisite for a mod,

    I was given a Gigabyte P35-DQ6 motherboard with a 2.4GHZ Core 2 Quad, and because I wanted something faster I got a a Xeon LGA 771 processor to LGA775 shim and swapped it for a $85 3.33GHZ Harpertown Xeon. The LGA 771 Xeons have the save internal CPU ID's as the LGA 775 Core 2 Quad series and thus will work with the mod in the Gigabyte motherboard.

    We should be free to do what we want with out hardware.
  • Aslan7 - Thursday, February 20, 2014 - link

    Last comment, the machine has been running World Community Grid for two months now without a restart. Reply
  • bigdog1984 - Thursday, February 20, 2014 - link

    Ok I read everything in the above posts and while I find you story of copper wire and thermal paste a fun read on the risk/reward of modding a laptop...I can't help but wonder what in the world this has to do with what we're talking about here. One generation of CPU is understandable but we're talking about vastly different processors with multi-thousand dollar price tags...the people buying these chips are putting them in server rooms and I workstations and more than likely don't care about socket compatibility since they are probably buying whole systems. I like compatibly for something like a generation or two of i7s but when it's 8 socket xenons I just want my server to work when I buy it. Let the vendors worry about compatibility Reply
  • chizow - Thursday, February 20, 2014 - link

    This is part of the reason I moved off Intel's "Enthusiast" Platform. Far too many delays, incompatibilities, old/lagging tech, and a full year or more off their mainstream performance platform's microarchitecture.

    Bought two Haswell Z87 builds with 4770Ks and couldn't be happier. The added benefit is that if/when I decide to upgrade either of these builds, they will still make fine HTPCs or thin clients. Unlike the two X58 builds I upgraded from which are completely undesirable in that respect due to their much higher thermals.
  • JlHADJOE - Friday, February 21, 2014 - link

    Ivy Bridge-EX is not part of the "Enthusiast" platform. Right now there's only one socket 2011 anyone not on a big corp's IT team is likely to have to deal with, and that's Sandy/Ivy-E. Reply
  • chizow - Friday, February 21, 2014 - link

    Right, this one isn't the Enthusiast platform, but the original 2011 platform is old/aging and was already feature-deficient even at launch (no native USB 3.0, only 2 SATA 6G, no official PCIe 3.0 GPU support). LGA 2011-3 which will support Haswell-E is only slated for later this year, but again, Intel has pushed these dates back numerous times that I simply got tired of waiting for the E platform to catch up to the mainstream performance platform. Reply
  • iwod - Friday, February 21, 2014 - link

    Since Most CPU aren't even upgradeable anymore, Why dont they just sell the CPU with the MB. Reply
  • chizow - Friday, February 21, 2014 - link

    They have started to do this on the mainstream performance platforms (non-K) with soldered-on BGA chips. Reply
  • g00ey - Wednesday, March 19, 2014 - link

    Because firstly, there is a value to be able to configure a system with different CPUs and motherboard types. Even within one generation there are different types of CPU where one CPU is focused on low-budget or low power consumption, another is focused on high per-thread performance and another is focused on high throughput/workload.

    Secondly, motherboards do break, even within warranty, so it is a considerable value to be able to replace the motherboard without having to replace the CPU. Sometimes the motherboard may come from a bad batch or model revision which would prompt replacing the motherboard with a different model, a model that may not be available with the same type of built-in CPU as the current faulty motherboard.

    Thirdly, having a CPU attached to the motherboard would mean that each unit would bind more capital due to the extra cost of the CPU, both for the reseller, the distributor and also for the motherboard manufacturer. That is capital that could be more efficiently allocated to stock other products that better cater for end-users' needs than an utterly expensive built-in CPU on a motherboard.
  • C.Henrique - Tuesday, September 02, 2014 - link

    Now intel announce the X99 chipset for LGA2011-3, I think someone from intel read this post. And specify that LGA2011-3 is not compatible with LGA2011-1 and 2 (names R1, R2 and R3, if not change RX in future) PC´s market is decreasing year after year, is cheap to keep the same socket, change the version and the user needs change the motherboard. Sales!! Reply

Log in

Don't have an account? Sign up now