EVGA

How come there's no full x16 lanes for multiple GPU?

Author
futuremark2
Superclocked Member
  • Total Posts : 204
  • Reward points : 0
  • Joined: 2006/11/09 14:01:34
  • Status: offline
  • Ribbons : 6
2014/07/31 23:04:02 (permalink)
before there was x16 x16 then like x16 x16 x16 x16 maybe with 2.0

but with Z97 theres only X16 then the rest are 8 4 2?

why is there a limitation with the lanes purposely?

the best motherboards cannot have x16 more than once because intels specifications say only x16 once

.
#1

9 Replies Related Threads

    EVGATech_JaesonW
    EVGA Tech Support
    • Total Posts : 910
    • Reward points : 0
    • Joined: 2013/02/06 10:41:56
    • Status: offline
    • Ribbons : 5
    Re: How come there's no full x16 lanes for multiple GPU? 2014/07/31 23:14:33 (permalink)
    You really don't need 16x even with 4 cards. 4 way SLI at PCIe 3.0 8x is more than enough bandwidth for modern cards without bottlenecking. Add in the cost of PLX chips for extra lanes, and you're raising the cost of the board for no reason.

    EVGA Customer Service Manager
    Have a moment? Please rate our service 
    #2
    futuremark2
    Superclocked Member
    • Total Posts : 204
    • Reward points : 0
    • Joined: 2006/11/09 14:01:34
    • Status: offline
    • Ribbons : 6
    Re: How come there's no full x16 lanes for multiple GPU? 2014/08/02 14:10:40 (permalink)
    PLX? but if intel specs out no more than pcie 3.0 16x x 1 it's not possible for boards manufacturers to twist that and make more? any specs that have more is fake? or actual performance of more?

    .
    #3
    daddyd302
    CLASSIFIED Member
    • Total Posts : 2135
    • Reward points : 0
    • Joined: 2006/09/15 17:10:20
    • Location: West Allis, WI
    • Status: offline
    • Ribbons : 5
    Re: How come there's no full x16 lanes for multiple GPU? 2014/08/03 07:43:19 (permalink)
    futuremark2
    PLX? but if intel specs out no more than pcie 3.0 16x x 1 it's not possible for boards manufacturers to twist that and make more? any specs that have more is fake? or actual performance of more?




    I wanted to be able to run 2 SLI and have both running at 16X instead of 8X. I was looking at this but the price was too high. 
    http://www.newegg.com/Product/Product.aspx?Item=N82E16813188160&cm_re=evga_z97-_-13-188-160-_-Product
     
     
    I was also interested in the FTW model but it didn't have PLX. I instead got this because it was one of the cheapest motherboards to have PLX on it.
    http://www.newegg.com/Pro...x?Item=N82E16813128702

    **Intel 13900K**Gigabyte Z790 Aorus Master**Gigabyte 4090 Aorus Master**G.Skill Trident Z5 RGB 32GB DDR-6800**Corsair H150i Elite LCD**2 Inland M2 SSD 2TB**2 Samsung 870 QVO 8TB**MSI MEG Ai1000P PCIE5**Li Lian Evo**ASUS PG42UQ**Corsair K100 **Logitech G502X**Windows 11 Pro** Evo**ASUS PG42UQ**Corsair K100 **Logitech G502X**Windows 11 Pro**
    #4
    MSim
    Omnipotent Enthusiast
    • Total Posts : 14685
    • Reward points : 0
    • Joined: 2005/05/22 23:13:30
    • Location: Earth
    • Status: offline
    • Ribbons : 38
    Re: How come there's no full x16 lanes for multiple GPU? 2014/08/03 09:58:01 (permalink)
    I thought the Classified version was suppose to support 16x 16x.
     
     
    #5
    lehpron
    Regular Guy
    • Total Posts : 16254
    • Reward points : 0
    • Joined: 2006/05/18 15:22:06
    • Status: offline
    • Ribbons : 191
    Re: How come there's no full x16 lanes for multiple GPU? 2014/08/03 18:27:07 (permalink)
    Chips like nF200 and PLX are "bridge chipsets", they don't invent more lanes, they effectively tell your graphics card what to run at, as normally a card will default to the rate that it is connected to, and sometimes runs slower as a result.  I think for SLI, slots need to be x8, that could be an nVidia licencing requirement; for the client socket LGA115x, 16 in the CPU means needing a bridge chipset for more than 2-way SLI.  LGA2011 has 40 lanes to spread out, originally build for high-end servers.  There is an LGA1356 socket that provides 24 lanes, but Intel uses that as a mid-level server platform, consumers don't get one.
     
    But, this performance is dependent on the application, so this is why you may hear the idea you don't need more for certain applications.  Games, for instance, don't use as many PCIe lanes under most circumstances, while purely computational work using the GPU as a co-processor does and carries the weight of pushing the boundary for new standards.  From a gaming standpoint, we don't need the upcoming PCIe 4.0 spec yet, as mentioned PCIe 3.0 x8 is great which equals PCIe 2.0 x16-- but since modern controllers are integrated with the CPU platform, its the older generation CPU that will eventually hold back future graphics cards.  
     
    But for those that either think they need more lanes or actually do, there are both boards with bridge chipsets.

    For Intel processors, 0.122 x TDP = Continuous Amps at 12v [source].  

    Introduction to Thermoelectric Cooling
    #6
    futuremark2
    Superclocked Member
    • Total Posts : 204
    • Reward points : 0
    • Joined: 2006/11/09 14:01:34
    • Status: offline
    • Ribbons : 6
    Re: How come there's no full x16 lanes for multiple GPU? 2014/08/04 17:37:29 (permalink)
    I see there are boards that support more than 1 16X lane but it bypasses the CPU controller one all together creating lag

    that's why other manufacturer said when using only 1 GPU it turns off the PLX that's outside the CPU one

    .
    #7
    futuremark2
    Superclocked Member
    • Total Posts : 204
    • Reward points : 0
    • Joined: 2006/11/09 14:01:34
    • Status: offline
    • Ribbons : 6
    Re: How come there's no full x16 lanes for multiple GPU? 2014/08/04 17:55:57 (permalink)
    oh I see

    the cheap 1/3 the price of the actual 40 lane PCIE 3.0 4960x vs 16 PCIE 3.0 4770K 16

    you can only get full PCIE 3.0 lanes if you spend $1k+ on their intel CPU

    I was thinking how some Titan Z's are in 4x doing world record runs w/o full 16x lanes then I realize it's cause they on a bigger more expensive CPU
    post edited by futuremark2 - 2014/08/04 17:59:40

    .
    #8
    lehpron
    Regular Guy
    • Total Posts : 16254
    • Reward points : 0
    • Joined: 2006/05/18 15:22:06
    • Status: offline
    • Ribbons : 191
    Re: How come there's no full x16 lanes for multiple GPU? 2014/08/06 11:23:24 (permalink)
    futuremark2
    you can only get full PCIE 3.0 lanes if you spend $1k+ on their intel CPU
    No, any CPU in that socket meant for X79 will give you 40 lanes including the $300 i7-4820K quad-core option.  But how those lanes are split up depends on the board manufacturer, some will give you a pair of x16 and an x8, or even five x8's; it has nothing to do with which LGA2011 CPU you get for the board.
     
     
    Having said that, apparently and according to rumors so far, the upcoming X99 platform will be different.  The cheapest 6-core option will only have 28-lanes enabled, and only with the $580 6-core and the $1000 8-core option will 40-lanes be available.  My guess is that Intel is trying to figure out who really needs what buy what they are willing to spend towards.  At that point, with X99 boards, you can't judge the board's slot configuration as independent of the CPU.

    futuremark2I was thinking how some Titan Z's are in 4x doing world record runs w/o full 16x lanes then I realize it's cause they on a bigger more expensive CPU
    Not necessarily.  PCI Express standards are really for those customers that use it fully, i.e. computation using the GPU as a co-processor for scientific calculations, not games or benchmarks.  The amount of space in each standard is so wide that there are several generations worth of scaling for gamers.   For instance, while there is enough room in the older PCIe 2.0 spec to hold a single Titan Black, you'd have to have a newer generation processor to properly feed the graphics card and not something from when PCIe 2.0 appeared, i.e. Core 2 Quad Q9000 series.  Granted a 3.2GHz QX9770 = 2.66GHz Core i7-920 in performance, you can't just keep the older CPUs at stock speed considering the 4GHz Core i7-4790K = Core i7-920 at a hypothetical 5.5GHz-- impossible overclock without refrigeration.
     
     
    This is why is doesn't matter what PCIe specification or lane configuration you have, by the time it matters, your CPU will be so old that no overclock will help feed those graphics cards-- unless you don't play CPU-bound games, GPU-bound games don't care about CPU type or frequency or or age, only graphics card and display resolution.

    For Intel processors, 0.122 x TDP = Continuous Amps at 12v [source].  

    Introduction to Thermoelectric Cooling
    #9
    feniks
    CLASSIFIED ULTRA Member
    • Total Posts : 7983
    • Reward points : 0
    • Joined: 2008/08/02 22:07:39
    • Location: NYC
    • Status: offline
    • Ribbons : 38
    Re: How come there's no full x16 lanes for multiple GPU? 2014/09/03 10:07:01 (permalink)
    bottom line is that you dont' need pcie 3.0 x16 for each card, because none of gaming video cards is able to saturate pcie 3.0 8x (same as pcie 2.0 x16)...
     
    EDIT:
    besides, with mainstream (not X?? platform) chips you get only 16 native pcie lanes, always. It can get multiplexed with PLX if that chip is present on motherboard (it jacks the price up tho so don't look for it in cheap models), but fact is that it's not really needed as most cards can't benefit from extra synthetic lanes, sometimes it even gives worse scores in benchmarks, because multiplexing brings a slight lag into equation (depending on MB circuitry and engineering), in most cases it's better to go with native 8x/8x than multiplexed 16x/8x or 16x/16x.
    post edited by feniks - 2014/09/03 11:39:20

    #10
    Jump to:
  • Back to Mobile