PCI-E Bifurcation
- Thread starter HeroXLazer
- Start date
Works great from the builds I've seen use it. Boloisbolo's build is my favourite:
https://hardforum.com/threads/project-chrome-leopard-10l-cfx-nanos.1902242/
https://hardforum.com/threads/project-chrome-leopard-10l-cfx-nanos.1902242/
Got it to work with a Z170 motherboard and Amerirack riser :
https://hardforum.com/threads/pcie-bifurcation.1870298/page-11#post-1042729050
https://hardforum.com/threads/pcie-bifurcation.1870298/page-11#post-1042729050
Three factors to the PCIe Bifurcation problem:
- Motherboard support.
Of released boards, there is either the ASRock Z170 Fatal1ty ITX (Bifurcation support out of the box), ASRock X99 ITX (Bifurcation support via special firmware on request), and ASRock Z87E-ITX (firmware not publicly available). IIRC, a Z270 board from Gigabyte was shown at CES with Bifurcation support listed in the specs.
- Riser support
The riser is required to contain a Clock Buffer chip. There are risers available that use a clock buffer chip, risers that use a PLX chip, and risers that contain neither. Only clock-buffered risers are suitable for use with PCIe Bifurcation.
- Software support
This is the remaining holdout, and little information is available due to so few people experimenting with DIY PCIe Bifurcation. Successes and failures have been reported both with AMD and Nvidia cards, on both Windows and Linux. If you are not using Crossfire or SLI, then compatibility seems pretty high, but if you are intending to use SLI/Crossfire then you will probably run into issues.
- Motherboard support.
Of released boards, there is either the ASRock Z170 Fatal1ty ITX (Bifurcation support out of the box), ASRock X99 ITX (Bifurcation support via special firmware on request), and ASRock Z87E-ITX (firmware not publicly available). IIRC, a Z270 board from Gigabyte was shown at CES with Bifurcation support listed in the specs.
- Riser support
The riser is required to contain a Clock Buffer chip. There are risers available that use a clock buffer chip, risers that use a PLX chip, and risers that contain neither. Only clock-buffered risers are suitable for use with PCIe Bifurcation.
- Software support
This is the remaining holdout, and little information is available due to so few people experimenting with DIY PCIe Bifurcation. Successes and failures have been reported both with AMD and Nvidia cards, on both Windows and Linux. If you are not using Crossfire or SLI, then compatibility seems pretty high, but if you are intending to use SLI/Crossfire then you will probably run into issues.
And this is the only Bifurcation riser that's been verified to work right? http://www.ameri-rack.com/ARC2-PELY423-C7_m.html
On the [H] thread, I listed successful study cases with alternative risers from Supermicro, which are available on eBay.
MaximumBurrito :
ASRock Fatal1ty Z170 Gaming-ITX/ac with bifurcation BIOS support enabled
RSC-R2UT-2E8R
+ Lian-Li risers
Breathless
ASRock X99E-ITX/ac with 3.63 bios
RSC-R2UG-A2E16-A
+ 3M Risers
I was able to order the riser straight from Ameri-Rack but apparently I got lucky as they do not always answer to private individuals !
BirdofPrey said:
@QinX explains all this on the first page of the [H] thread
A PLX chip could be though of a little like a network switch for PCIe. It operates on every PCIe lane, making two discrete downstream devices appear to be a single device to the host, so the host does not necessarily need to support PCIe bifurcation.
The clock buffer chip's purpose is to ensure clock signals from the host are correctly passed to the two bifurcated ports (host is only designed to drive one client). It does nothing to the PCIe lanes themselves, so the host needs explicit support for PCIe bifurcation.
That was is such a fun thread on [H]. Watching Chemist_Slime become the first to get PCIe bifurcation working, and then seeing him get EKWB to knock-out a custom dual Nano waterblock.
What a treat. That thread also had some custom silicon. User C_Payne made a custom PCIe riser card with a "clock buffer" chip to support his build. Just wow.
And more recently, still in that same thread - user Maximum_Burrito stuffed two Nanos into a DonDan A4, hot damn!
There's probably more. It's just too much too read through.
@EdZ, you were in that [H] thread too, yes? Can you clue me in? I came away from that thread understanding that -provided your chipset and motherboard had PCIe Bifurcation support - any passive splitter would work. So, that's not true, right? You need chipset/mobo support AND a splitter with a "clock buffer" chip? Any examples? Best one to get?
What a treat. That thread also had some custom silicon. User C_Payne made a custom PCIe riser card with a "clock buffer" chip to support his build. Just wow.
And more recently, still in that same thread - user Maximum_Burrito stuffed two Nanos into a DonDan A4, hot damn!
There's probably more. It's just too much too read through.
@EdZ, you were in that [H] thread too, yes? Can you clue me in? I came away from that thread understanding that -provided your chipset and motherboard had PCIe Bifurcation support - any passive splitter would work. So, that's not true, right? You need chipset/mobo support AND a splitter with a "clock buffer" chip? Any examples? Best one to get?
Last edited:
You did not ask me but let me share my own example please. based on the same forum and the knowledge collected here, I made this work.
So the motherboard supports it, has the option in the BIOS and a passive Ameri-Rack riser splitter helped me to run two x8 PCIe network cards.
EDIT: my splitter is not passive in the meaning of no chips there. As EdZ pointed out it also has a clock buffer chip.
Last edited:
You need:
- Motherboard with PCIe bifurcation support exposed in BIOS/UEFI
- Splitter with clock-buffer
- Cards that work with that splitter (e.g. if using an unpowered splitter and two bus-powered 75W cards, that's not going to go well)
- Drivers for those cards that will work with bifurcation.
In particular, GPU SLI/Crossfire under Windows is rather hit-and-miss, mostly miss so far on the Nvidia side. If you're not running two GPUs (or if you're running them independently, e.g. as a Physx card or for GPU compute) things are more likely to work.
- Motherboard with PCIe bifurcation support exposed in BIOS/UEFI
- Splitter with clock-buffer
- Cards that work with that splitter (e.g. if using an unpowered splitter and two bus-powered 75W cards, that's not going to go well)
- Drivers for those cards that will work with bifurcation.
In particular, GPU SLI/Crossfire under Windows is rather hit-and-miss, mostly miss so far on the Nvidia side. If you're not running two GPUs (or if you're running them independently, e.g. as a Physx card or for GPU compute) things are more likely to work.
A PLX chip? In theory, the only problem with it is that PLX chips are rather expensive. In practice, having an extra device sitting in the middle of a high-speed bus can cause all sorts of odd issues. People sometimes have issues with multiple GPUs on motherboards where some PCIe slots are driven from PLX chips, and that is even after the whole assembly was designed that way from the start and has gone through validation with the motherboard manufacturer! Sticking a PLX chip in where it is not expected is in no way guaranteed to work.
It's mentioned in post #237 that the ASRock Z170M-ITX/ac is also capable of PCIE Bifurcation.
-Edit-
According to @iFreilicht in post #349 the Gigabyte GA-Z170N-WIFI revision 2.0 is capable also.
-Edit-
According to @iFreilicht in post #349 the Gigabyte GA-Z170N-WIFI revision 2.0 is capable also.
Last edited:
Similar threads
Motherboard
X570 Bifurcation
