PCIE Bifurcation

ChronoDog said:
Upward risers do exist, just not bifurcated ones.
The goal is bifurcating a PCIe slot while keeping a low-profile expansion card in the same place without using multiple risers.
Click to expand...

News to me, that's awesome, how do you screw them in then?

C_Payne said:
I have actually just recently made a low profile to full height riser and also made a long time ago a bifurcated one for an M.2 slot. (search this thread)
Spacing is very tight though and I don't have the tools to work with small twinax cables. (yet) Maybe I can figure something out with an angled slot...
Click to expand...

C_Payne you are the best!
 
Check those adaptor brackets. IMAG3073.jpg
 
C_Payne said:
I have actually just recently made a low profile to full height riser and also made a long time ago a bifurcated one for an M.2 slot.
Click to expand...

I have to say, your risers seem amazing, you really do things nobody else seems to do :)
And if you mean this one, it's what gave me the idea above in the first place:

5eY4pfv.jpg


How difficult would it be to make an x8x4x4 version of it, with 2 mechanical x16 slots and an M.2 slot?
Because that would be amazing for NCase M1 owners, it'll allow having a GPU, an(other) M.2 drive and any low-profile expansion card all on a single PCIEx16 slot, assuming the board supports it. If it only supports x8x8, make it so that it'll function as just two PCIe slots, x8 and x4.
 
How difficult would it be to make an x8x4x4 version of it, with 2 mechanical x16 slots and an M.2 slot?
Because that would be amazing for NCase M1 owners, it'll allow having a GPU, an(other) M.2 drive and any low-profile expansion card all on a single PCIEx16 slot, assuming the board supports it. If it only supports x8x8, make it so that it'll function as just two PCIe slots, x8 and x4.
Click to expand...

I don't know about your motherboard, but I've only seen motherboards that support 8x/8x and very rarely 4x/4x/4x/4x (x299 and Gigabyte's x570) If you know of a motherboard that runs 8x/4x/4x I would love to know because the set up you are dreaming about sounds awesome and I'm super interested.
 
TheArkratos said:
I don't know about your motherboard, but I've only seen motherboards that support 8x/8x and very rarely 4x/4x/4x/4x (x299 and Gigabyte's x570) If you know of a motherboard that runs 8x/4x/4x I would love to know because the set up you are dreaming about sounds awesome and I'm super interested.
Click to expand...
The board I have has these two options:

IIukYVd.jpg


RxLmblq.jpg


...which can be both enabled at the same time, like so:

f6RQSeD.jpg


Whether this means the board will support mixed riser configurations or strictly x8x8 / x4x4x4x4, I have no idea - I don't have an x4x4x4x4 riser to test - but when testing with an x8x8 riser it seems to just work as x8x8 either way.
Wouldn't it be fascinating if this meant it could support "double bifurcation" (for lack of a better term), though?
 
Last edited:
Intel mainstream per se supports x8x4x4.
Intel Extreme x8x4x4, x4x4x8 and x4x4x4x4

The boards that support x8x4x4 I know are:

Asrock
Z370 itx
Z390 itx
X99E itx
X299E itx

AMD Ryzen 3000:
Gigabyte X570 itx EDIT: not sure here maybe only x4x4x4x4 and x8x8

This is what I have in my head and of course depends on bios version and is incomplete, other mainboards may have support as well.
 
Last edited:
C_Payne, is such a riser (half-height x8x4x4 with M.2 and two x16 slots) something you could possibly design and manufacture? Would it, in your opinion, be a worthy investment of your time?
The potential market, as I've mentioned, would include many NCase M1 owners - perhaps a collaboration with its' creators (potentially selling it as an accessory on sfflab?) could help with sales.

I'm indeed driven by my own personal interest in such a product, but I reckon it already is (or about to be) a fairly wide-spread usage scenario, with so many new ITX boards supporting such bifurcation out-of-the-box, and quite a few NCase owners.

Edit: If not, I'd gladly buy at least one for myself. I realize it probably won't be cheap.
 
Last edited:
I will think about a solution, however I have no easy+cheap version that I can think of currently, reason for that is the tight spacing, and my lack of expertise in working/manufacturing with cables. Will keep you posted.
 
I loved the original idea of half-height x8x8 riser and I'd be interested in it soon as well (I don't have the MB yet). Instead of working with cables would it be possible to do both with actual PCI-e connectors? That way I could use the top port for HBA (low profile) and run the flex riser to GPU in future.

EDIT: Should've read more carefully, I think that is exactly what C_Payne suggested by angled slot?
 
Sorry for keeping you hanging.

Unfortunately my idea using the original cable doesn't fit. However using a right angled flex riser in conjunction with my regular single slot rev. 9 riser does fit. See pics.
I am going to reduce the rev9 riser size by another 2mm and then it will be perfect, but it still works with the current one as well. I have 3D printed brackets as well, so if you want one I can ship it right away.

I also have stuff for the Ncase coming as well.

IMAG3234.jpg IMAG3235.jpg
 
Great to hear!
What did you take to screw it tight? (the white thing?)

What about maximal power consumption?
Right now I would put a single Slot GTX 1050 https://geizhals.de/inno3d-geforce-...e-slot-n10502-1sdv-e5cm-a1651587.html?hloc=de and a Sonnet Tempo PCIe bracket https://geizhals.de/sonnet-tempo-ssd-tsata6-ssd-e2-a949163.html?hloc=de

Are there any max. TDP for the slots?

Edit: Which length of the Riser did you take? That one: https://www.amazon.de/LINKUP-Extreme-Shielded-Twin-axial-Extension/dp/B07B9FKPYS/ref=sr_1_10?__mk_de_DE=ÅMÅŽÕÑ&crid=17EXTJSHLVSPD&keywords=linkup+riser&qid=1571917815&s=computers&sprefix=linkup,computers,150&sr=1-10 ?
 
Last edited:
As an Amazon Associate, HardForum may earn from qualifying purchases.
Jup I 3D printed the white parts.
Can print black ones as well.

I used the 30cm cable you mentioned. But you can use the cheaper one as well. Doesn't have to be the gen4 one.

Both cards together shall not exceed 3.3A of 3.3V and 5.5A of 12V.
Otherwise there is power pads on the back to solder wires to, for example a SATA power pigtail.

The new revision I plan will have a SATA power connector.
 
Ok so I would take the new one revision with SATA Power Connector. Long time ago since I soldererd by my own.
Also the whtie parts printed in black.
How much would that be including Sending fee (I'm Germany based)?
 
I am waiting for the connectors, so it will take a while. Price will be €70 Including the 3D printed parts.
 
C_Payne you are a wizard! I haven't bought a riser from you yet because I'm trying to figure out which I want and what project to work on, but do you have a way I can donate to you? I love your work and want to support where I can, I just don't have the perfect project quite yet....
 
Big thx to C_Payne the Pcie x8x8 works very well. I did a Firestrike Benchmark and copied some data to the Sonnet SSD. Both worked. This brings many new ideas and work to my DAN case.

Keep up the good work!




IMG_20191030_192529.jpg IMG_20191030_191718.jpg
 
I just came here after Googling about PCIe Googling - i see this single thread in a single forum seems to be the go-to place! I also found C_Payne's products on ebay.

I have an MSI B450i Gaming Plus AC with a Ryzen 2700X. As that doesn't have an iGPU, I was looking to add a second basic one, that I could use while passing through the main one to a Windows VM (Linux host).

I tried the setting, but now linux just see 4 x 4x rather than 2 x 8x.

Code: 
 grep "GPP Bridge" basic.log  
00:01.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) PCIe GPP Bridge
00:01.3 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) PCIe GPP Bridge
00:03.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) PCIe GPP Bridge
00:07.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) Internal PCIe GPP Bridge 0 to Bus B
00:08.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) Internal PCIe GPP Bridge 0 to Bus B
 
 grep "GPP Bridge" 2x8.log
00:01.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) PCIe GPP Bridge
00:01.3 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) PCIe GPP Bridge
00:03.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) PCIe GPP Bridge
00:03.2 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) PCIe GPP Bridge
00:03.3 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) PCIe GPP Bridge
00:03.4 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) PCIe GPP Bridge
00:07.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) Internal PCIe GPP Bridge 0 to Bus B
00:08.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Family 17h (Models 00h-0fh) Internal PCIe GPP Bridge 0 to Bus B

w2x8j8cegqv31.jpg

I wonder what the limitation is here? Mobo, CPU, or OS?
 
Last edited:
This looks like a BIOS issue to me.

Have you tried setting it to x4x4x4x4.... Maybe you will end up with two x8 ports. :)

Otherwise raise a ticket with your mainboard manufacturer, maybe they will fix it.
 
C_Payne said:
This looks like a BIOS issue to me.

Have you tried setting it to x4x4x4x4.... Maybe you will end up with two x8 ports. :)

Otherwise raise a ticket with your mainboard manufacturer, maybe they will fix it.
Click to expand...

I tried all the options in there - it's either 1x16 or 4x4 - no 8x8, even when I select it.

Guessing they just reused the BIOS when they added Ryzen 2 support across their products, and forgot to remove that from the menu. It wasn't there in the original BIOS, and it's not formally mentioned in the change list at https://www.msi.com/Motherboard/support/B450I-GAMING-PLUS-AC#down-bios

I read they had to move to a simpler BIOS menu system, to make space in the ROM.

Iguess I'll have to change mobo if I really want to do this.
 
As far as I know b_50 chipsets don't support splitting the 16x PCIe grouping on the motherboard. That's one of the big reasons to go with the x_70 chipsets. Even on full sized ATX boards from what I understand, that's why multi gpu setups basically require x_70.

Edit: I stand corrected
 
Last edited:
I seem to recall form earlier reviews that this has been proven wrong.
And that makes sense to me as the CPU lanes are nowhere near the chipset. Sounds like a marketing stunt to me. (just like disabling gen4 on x470)
 
'
C_Payne said:
Try contacting them... It may be a very easy fix for them. Also have you tried different bios versions?
Click to expand...

I've opened a technical ticket with them - will see if they get back.

In the meantime looking at https://www.msi.com/Motherboard/support/B450I-GAMING-PLUS-AC#down-bios it's not obvious which to chose, other than the latest.

Perhaps the one where they mentioned "Change BIOS UI to Click BIOS(GSE-Lite) mode" - as that might be where they went back to a common code base for the ROM, and maybe re-removed the capability without taking it out of the menu.
 
Response from MSI:

Regarding your concern, we are sorry, did you install two or more M.2 devices on the PCI_E1 slot? What do you mean by OS sees 4x4 lanes? Please take a picture for it.

Thanks for your cooperation in advance!
Click to expand...

:facepalm:

Tech support is so often not technical at all.
 
Just had another thought: maybe AMD has locked x8x8 in their AGESA at one point to disable dual gpu usage, but left x4x4x4x4 working for quad m.2 solutions. The mainboard vendor hasn't removed the settings and x8x8 falls back to x4x4x4x4.
Maybe find out which agesa has been used successfully for x8x8 and check the older versions for your board.
 
Thanks for your help.

So I started on the latest: 7A40vAAComboPI1.0.0.3abba (not sure why they name it ComboPI rather than AGESA)

Then I downgraded to: 7A40vA7 ComboPI1.0.0.3ab

Then I tried downgrading further to: 7A40vA2 AGESA Code 1.0.0.6 - but it gave me a message about not allowing to downgrade

Finally, I tried bumping up to latest, and then directly down to the last one, and that still gave me the message about not allowing downgrades (Not sure why it allows AA -> A7 then...)

I think probably the menu option is just a mistake - I'll probably go ahead and swap out for the ASRock - Fatal1ty B450 Gaming-ITX, as I don't like the latest boards anyway - expensive, need active cooling, and no features I really need.
 
Just be aware that it might not work with the latest asrock bios either... If my proposition about AMD locking it is true.

You could try flashing directly in Linux with flashrom for the downgrade.

What about an x470?
 
I'm fairly confident in the ASRock one, as it's explicitly mentioned as a feature in their specs: https://www.asrock.com/MB/AMD/Fatal1ty B450 Gaming-ITXac/index.asp#Specification

1 x PCI Express 3.0 x16 Slot (PCIE1: x16 mode) (Supports PCIe riser cards to extend one x16 slot to two x8 slots if you install AMD Ryzen series CPUs)
Click to expand...

There's one more thing I'm going to try - will see if I can patch the MSI BIOS to enable it, as I think it's just a switch rather than code.
 
The next thing actually - once I've got a motherboard that can do it - I need one of C_Payne's splitters that will physically fit the Louqe Ghost S1.

Here's a picture of one from the net (not mine) - you can see there's no clearance at the top. The socket for the GPU is screwed in to the central spine.

I was hoping to be able to get a small cheap GPU to go behind a mini RTX card, to run the Linux host while I passthrough the main card to a Windows VM.

HNShGR9.jpg
 
Against my better judgement (as it's probably a time sink), I've been looking in the BIOS with UEFI tools. As it happens the options are indeed there.

The section is called:

Code: 
0x82B8          AMD PBS (0x2 from string package 0x0)

And there's an option in there for 2x16:

Code: 
0x82E2          Suppress If {0A 82}
0x82E4                  QuestionId: 0x2D equals value in list (0x1) {14 88 2D 00 01 00 01 00}
0x82EC                          QuestionId: 0x2 equals value in list (0x1, 0x3, 0x4) {14 0C 02 00 03 00 01 00 03 00 04 00}
0x82F8                          Not {17 02}
0x82FA                          Or {16 02}
0x82FC                  End {29 02}
0x82FE                  One Of: PCIe x16/2x8 Switch, VarStoreInfo (VarOffset/VarName): 0xD, VarStore: 0x1, QuestionId: 0x9, Size: 1, Min: 0x0, Max 0x1, Step: 0x0 {05 91 19 00 1A 00 09 00 01 00 0D 00 10 10 00 01 00}
0x830F                          One Of Option: 2X8, Value (8 bit): 0x0 {09 07 1B 00 00 00 00}
0x8316                          One Of Option: 1X16, Value (8 bit): 0x1 (default) {09 07 1C 00 10 00 01}
0x831D                  End One Of {29 02}
0x831F          End If {29 02}

The trouble is I'm not sure how to determine what value 'false' would be, as presumably '0A 82' is true here.

I had a look at the same section from the ASRock BIOS, and found almost exactly the same - so it seems to be some standard AMD display features. I don't know what 'PBS' stands for.
 
I did take a quick peek as well:

This is in a different section and what is visible from the GUI: (setup dxe)

Code: 
Suppress If {0A 82}
0x4C255                 QuestionId: 0x2F3 equals value in list (0x0, 0x2, 0x3, 0x4, 0x5) {14 10 F3 02 05 00 00 00 02 00 03 00 04 00 05 00}
0x4C265                 Suppress If {0A 82}
0x4C267                     QuestionId: 0x345 equals value in list (0x1) {14 08 45 03 01 00 01 00}
0x4C26F                     One Of: PCI_E1 Lanes Configuration, VarStoreInfo (VarOffset/VarName): 0x5DC, VarStore: 0x1, QuestionId: 0x82, Size: 1, Min: 0x0, Max 0x2, Step: 0x0 {05 91 EF 07 F1 07 82 00 01 00 DC 05 10 10 00 02 00}
0x4C280                         One Of Option: Auto, Value (8 bit): 0x0 {09 07 F3 07 00 00 00}
0x4C287                         One Of Option: x8+x8, Value (8 bit): 0x2 {09 07 F5 07 00 00 02}
0x4C28E                         One Of Option: x4+x4+x4+x4, Value (8 bit): 0x1 {09 07 F4 07 00 00 01}
0x4C295                         Default: DefaultId: 0x0, Value (8 bit): 0x0 {5B 06 00 00 00 00}
0x4C29B                     End One Of {29 02}
0x4C29D                 End If {29 02}
0x4C29F                 Suppress If {0A 82}
0x4C2A1                     QuestionId: 0x345 equals value in list (0x0) {14 08 45 03 01 00 00 00}
0x4C2A9                     One Of: PCI_E1 Lanes Configuration, VarStoreInfo (VarOffset/VarName): 0x5DC, VarStore: 0x1, QuestionId: 0x83, Size: 1, Min: 0x0, Max 0x1, Step: 0x0 {05 91 EF 07 F1 07 83 00 01 00 DC 05 10 10 00 01 00}
0x4C2BA                         One Of Option: Auto, Value (8 bit): 0x0 {09 07 F6 07 00 00 00}
0x4C2C1                         One Of Option: x4+x4, Value (8 bit): 0x1 {09 07 F7 07 00 00 01}
0x4C2C8                         Default: DefaultId: 0x0, Value (8 bit): 0x0 {5B 06 00 00 00 00}
0x4C2CE                     End One Of {29 02}
0x4C2D0                 End If {29 02}
0x4C2D2             End If {29 02}
0x4C2D4             Suppress If {0A 82}
0x4C2D6                 QuestionId: 0x2F3 equals value in list (0x1, 0x8) {14 0A F3 02 02 00 01 00 08 00}
0x4C2E0                 One Of: PCI_E1 Lanes Configuration, VarStoreInfo (VarOffset/VarName): 0x5DC, VarStore: 0x1, QuestionId: 0x84, Size: 1, Min: 0x0, Max 0x1, Step: 0x0 {05 91 EF 07 F2 07 84 00 01 00 DC 05 10 10 00 01 00}
0x4C2F1                     One Of Option: Auto, Value (8 bit): 0x0 {09 07 F6 07 00 00 00}
0x4C2F8                     One Of Option: x4+x4, Value (8 bit): 0x1 {09 07 F7 07 00 00 01}
0x4C2FF                     Default: DefaultId: 0x0, Value (8 bit): 0x0 {5B 06 00 00 00 00}
0x4C305                 End One Of {29 02}
0x4C307             End If {29 02}


There is more switches in the PBS setup dxe:

Code: 
Suppress If {0A 82}
0x8AB9             QuestionId: 0x2D equals value 0x0 {12 86 2D 00 00 00}
0x8ABF                 QuestionId: 0x2 equals value in list (0x0, 0x2) {14 0A 02 00 02 00 00 00 02 00}
0x8AC9                 Not {17 02}
0x8ACB                 Or {16 02}
0x8ACD             End {29 02}
0x8ACF             One Of: PCIe x8 Switch (J10), VarStoreInfo (VarOffset/VarName): 0x6E, VarStore: 0x1, QuestionId: 0x2E, Size: 1, Min: 0x0, Max 0x1, Step: 0x0 {05 91 45 01 46 01 2E 00 01 00 6E 00 10 10 00 01 00}
0x8AE0                 One Of Option: 1X8, Value (8 bit): 0x0 (default) {09 07 47 01 10 00 00}
0x8AE7                 One Of Option: 2X4, Value (8 bit): 0x1 {09 07 48 01 00 00 01}
0x8AEE             End One Of {29 02}
0x8AF0         End If {29 02}
0x8AF2         Suppress If {0A 82}
0x8AF4             QuestionId: 0x2D equals value 0x0 {12 86 2D 00 00 00}
0x8AFA                 QuestionId: 0x2 equals value in list (0x1, 0x3, 0x4) {14 0C 02 00 03 00 01 00 03 00 04 00}
0x8B06                 Not {17 02}
0x8B08                 Or {16 02}
0x8B0A             End {29 02}
0x8B0C             One Of: PCIe x16 Switch (J10), VarStoreInfo (VarOffset/VarName): 0x6E, VarStore: 0x1, QuestionId: 0x2F, Size: 1, Min: 0x0, Max 0x1, Step: 0x0 {05 91 49 01 4A 01 2F 00 01 00 6E 00 10 10 00 01 00}
0x8B1D                 One Of Option: 1X16, Value (8 bit): 0x0 (default) {09 07 4B 01 10 00 00}
0x8B24                 One Of Option: 4X4, Value (8 bit): 0x1 {09 07 4C 01 00 00 01}
0x8B2B             End One Of {29 02}
0x8B2D         End If {29 02}


Unfortuanely my BIOS hacking skills go only so far. Maybe someone else can make some sense of this.

I checked the Asrock BIOS as well and it only has options in the PBS file, nothing for me to find in the setup dxe, rather asrock exposes the PBS menu.

I would insist with tech support and have them forward your request to a BIOS engineer...
Asrock is helpful, why schouldnt MSI be? If they want a picture take a screenshot of your bios menu with all 3 options and highlight the one which is not working. ;)
 
Someone here claims to have been able to enable it: https://www.win-raid.com/t5504f16-A...ion-in-MSI-B-ITX-Gaming-Plus-AC.html#msg96096

I'm not sure I understand how, so have asked for clarification. It sounds like you blank out some GUIDs, but why that works (if that indeed is what it's saying) I don't know. Intuitively I'd expect something like the opposite.

I'm with you on MSI support! I've asked, but they've not been at all helpful. I don't hold out much hope, as unlike ASRock, it's not an advertised feature of the board.
 
C_Payne - I'm assuming I'm going to get my board working, or at least get the ASRock one, so in the meatntime, can I clarify which adapter I'll need please?

I have exactly the same case as this one. I've linked the time in the video he's installing it:



Is ebay or your shop the best option?
 
You must log in or register to reply here.
Back
Top