PCIE Bifurcation

I just got in touch with the manufacture that makes the flexible splitter that did not work for me initially. He said that the V3 should work and I just ordered one. The previous was PCIE 2.0 apparently.

http://www.ameri-rack.com/ARC2-PELY423-C7_m.html

I'll try this one without the 1.20E bios flash to see if it works out of box. But at the same time, I will try the 1.20E tonight with all 3 other riser cards.
 
QinX said:
Well, 1 month into the build I had the same idea you want to do right now, only the X99e wasn't even out so it was a dead end, I also assumed that the X99e also didn't have PCIe bifurcation support. But now it seems more people have gotten the same kind of idea :p
Click to expand...

What can I say Qinx, we're trail blazers!
 
QinX said:
Well, 1 month into the build I had the same idea you want to do right now, only the X99e wasn't even out so it was a dead end, I also assumed that the X99e also didn't have PCIe bifurcation support. But now it seems more people have gotten the same kind of idea :p
Click to expand...

Qinx, do you by any chance know what the length of the PNY 970 GTX is without the fan. I'm wondering if it's the same length as the ASUS 970 GTX min and want to confirm before I purchase 2.
 
Seems to have just got confirmation from Ameri-rack, that the other splitter I have from them will work with the new bios 1.20E but at a slower speed.

"The non-V3 used stranded ribbon and IDC connector, won’t support Gen3 high-speed data transfer."

In fact, he attached the exact same Bio that ASROCK_TSD just posted on the asrock site so I'm hoping it's confirmed tonight when I try! The v3 that I just bought is confirmed to work on his end of course but at v3 speeds. Awesome!
 
Last edited:
Spoke with LiHeat a few days ago. We discussed their ability to produce a passive Gen 3.0 bifurcated flexible riser for just this application. They are very responsive via their facebook page. Finger crossed. wishing you luck, chemist_slime.
 
Runamok81 said:
Spoke with LiHeat a few days ago. We discussed their ability to produce a passive Gen 3.0 bifurcated flexible riser for just this application. They are very responsive via their facebook page. Finger crossed. wishing you luck, chemist_slime.
Click to expand...

Who's Liheat?

nm, I just googled it.
 
Alright guys, I just tested all 3 risers with the 1.20E bios and I'm happy to report that all 3 now work! I am elated right now, I have poured countless hours into researching this and pretty much found nothing online except speculation and now it all works. Let's this day 7/31/2015 be known that we have the first working PCIE bifurcation support for a mini-ITX board and my dream build machine can be achieved! However, what's puzzling is that the PLX riser from super micro also now works although previously it did not without the bios support. So it seems that some bios support is indeed needed even if a PLX chip is used!

@Qinx, looks like you no longer need to create your own custom board unless you want a specific PCB layout that isn't achievable with a flexible riser. :)
 
Last edited:
chemist_slime said:
However, what's puzzling is that the PLX riser from super micro also now works although previously it did not without the bios support. So it seems that some bios support is indeed needed even if a PLX chip is used!
Click to expand...

You see, this was the reason for my obsessive questioning about the relation between the BIOS support and PLX chips. What is interesting though that Supermicro seems to follow some kind of standard procedure for BIOS and splitter communication.

In this regard, I still do not understand how the professional PCIe expander boxes work. They are connected by cable to a PCIe slot and supposedly work right off the bat with any mobo, it is nowhere mentioned the mobo needs BIOS upgrade.

What is amazing in any case is the ASRock support, that is absolutely exceptional, it ROCKS!. If I knew it beforehand I would buy ASRock mobo instead of Asus. Now I would like to ask you for a favor due to your established contact with ASRock in this matter. I would like to know answer to this: If somebody would like to expand the PCIe bus on ASRock mobos which have the PLX chips, like the X99 WS-E or X99 Extreme11: Do these mobos would also require the BIOS upgrade or not?

Or maybe the Supermicro splitter would work with no problems???

The question is then if the BIOS on the PLX mobos is prepared generically to work with the PLX chips or the BIOS in the mobos with the PLX chips are just matched to them.

Phuncz said:
If you just need a chute to throw money into, the server market always has a solution for you: http://www.supermicro.com/products/motherboard/Xeon/C600/X10DRX.cfm

7122_43_supermicro-x10drx-intel-c612-server-motherboard-review.jpg


11 (eleven) open-ended PCIe 3.0 x8 slots. Requires dual E5-2600 CPUs and RAM 4 modules at once though.
Click to expand...

First I need PCIe 3.0x16 and thus another mobo would be needed which could be fund but this is not the whole story. That is exactly the problem: it would be highly nonoptimal and waste of money to use a dual socket mobo for my app. Note that among others, I can overclock my Xeon E5-1680v3 well over 4GHz while the E5-26xx can not be overclocked. They are just not good for apps which rely on smaller number of threads but speed of the threads is critical. Dual socket mobos have more RAM capacity but I have 128 GB ECC RAM which is fully enough. There is also the problem with the dual socket PCI buses, they are split in half, each processor has access to one bus and for my apps it is preferable to have single PCIe bus.
 
Open-ended means they fit x16 cards but are only electrically connected for x8, like most boards in the world with multiple x16 slots.

But you can't have a bunch of real PCIe 3.0 x16 lanes without multiple CPUs when they only have 40 lanes per CPU (highest spec at the moment).
That's why on the X10DRX board all slots (10 anyway) are x8 and need two CPUs to actually work.

8 slots x 16 lanes = 128 lanes
128 lanes / 40 lanes/CPU = 3.2 CPUs

So you'd need a 4-socket board to be able to use eight 16-lane PCIe 3.0 slots.
 
chemist_slime said:
Alright guys, I just tested all 3 risers with the 1.20E bios and I'm happy to report that all 3 now work! I am elated right now, I have poured countless hours into researching this and pretty much found nothing online except speculation and now it all works. Let's this day 7/31/2015 be known that we have the first working PCIE bifurcation support for a mini-ITX board and my dream build machine can be achieved! However, what's puzzling is that the PLX riser from super micro also now works although previously it did not without the bios support. So it seems that some bios support is indeed needed even if a PLX chip is used!

@QinX, looks like you no longer need to create your own custom board unless you want a specific PCB layout that isn't achievable with a flexible riser. :)
Click to expand...

In this case it seems bios support means proper PCIe support. Because I wonder now if you'd stick a 295X2 in there with an older BIOS if it would have worked.

I'm still contemplating designin a case with a proper board that is made for it.

Because right now either you
Build your own case($$$),
Mod one(time, tools and effort)
Use a mATX case where you can move all cards over 1 slot(but then why did you choose the ITX board if you are going with a mATX case)
 
Spent all night trying to get this to work with a Hackintosh and I finally did it! However, besides the supermicro board with the PLX chip, the 970 GTX is not recognized otherwise. So it seems that the other two passive splitters will need to be PCI-E 3.0 compliant in order to work with 3.0 cards. All the splitters did work with a Geforce 8400 GT and a Geforce 610 GT though which are PCIE 2.0. Good night everyone!
 
Blue Fox said:
Or you could just buy a case already designed for this: http://www.lian-li.com/en/dt_portfolio/pc-o5/
Click to expand...

I don't think they thought about PCIE bifurcation for the mini-itx when designing this. They would have had to know that a company such as asrock was going to design such a board. :) Although that case is indeed interesting, 4 slots. makes me want to now cram in 4 single slot 970 GTXs, sigh..
 
Was just about to go to bed then saw this. Fury Nano, 153mm vs 170mm 970 GTX. Crossfire anyone and even smaller form factor? Not to mention it can also be single slotted with a water block. God I love what's going on around SFF these days!

6AUukfxl.jpg


H8bANrkl.jpg


i9vwe55l.jpg


VaFLqJGl.jpg
 
And no need for an SLI bridge, important if flex risers apart from eachother are used.
 
@chemist_slime

Incredible!! So glad to see this worked. I wish I hadn't purchased a Z97 board several months ago, considering selling my system and going with the Asrock board. I wonder if they plan to have bifurcation support going forward with their Z170 boards.

Also, I'll be curious to see how the GTX 970 is recognized in OS X going forward with the PCI-e 3.0 bifurcation without PLX chip once you get the new model. Please post your results! I imagine it will work, but just want to be sure!

As for the Fury Nano, I also think it looks awesome. I'm contemplating doing a build in the A4 with two of these... one oriented the normal way, the other sitting directly behind with the ports facing upward so they can both fit along the length. The second set of ports wouldn't be accessible, but you would still get the power of the two GPUs in the A4. Or, you could water block the two and have them sitting in their normal spot, you'd just have to find a place to put the pump, radiator, etc. This could be considered crazy since AMD is coming out with a dual Fiji card. However, my guess is that this will be attached to a huge radiator like the R9 Fury X, and thus won't fit in something like the A4.

Another thing to consider for OS X is support for the Fury Nano, which has a Fiji chip. As there are no current macs that have this, OS X will likely not support it right now. However, given Apple's current relationship with AMD, I bet you'll see support as soon as they update their Mac Pros!

Another, off topic question, for you. You're running OS X on X99, correct? Do you have power management and sleep/wake working? My understanding is that this wasn't native since there are no current macs that use this chipset.
 
CONGRATS CHEMIST_SLIME!!!!!!!!!!!!!!!!!!

If my Google-fu is correct you are the first known example of an individual using dual GPU on a consumer grade mITX size board. You sir, are trail-blazer that inspires us all! The race is on!
 
I don't know how much this is of interest here but it is doable and relatively easy to convert the GPU 970 to single slot with watercooling. In the first step one has to remove the upper DVI socket, which I did by cutting out the socket support and connecting links. It is possible then to have two single slot cards in SLI but of course tubing, radiator and fans are needed.
 
wirk said:
I don't know how much this is of interest here but it is doable and relatively easy to convert the GPU 970 to single slot with watercooling. In the first step one has to remove the upper DVI socket, which I did by cutting out the socket support and connecting links. It is possible then to have two single slot cards in SLI but of course tubing, radiator and fans are needed.
Click to expand...

There are several manufacturers that have GTX970 cards that already have single slot I/O
Palit GeForce GTX 970 4GB
Palit GeForce GTX 970 4GB JetStream
Gainward GeForce GTX 970 4GB
Gainward GeForce GTX 970 4GB Phantom
PNY GeForce GTX 970 4GB
 
^Yes, that was for the case somebody already has the GTX970 with two DVI connectors :mad:.
 
chemist_slime said:
So it seems that the other two passive splitters will need to be PCI-E 3.0 compliant in order to work with 3.0 cards. All the splitters did work with a Geforce 8400 GT and a Geforce 610 GT though which are PCIE 2.0. Good night everyone!
Click to expand...

This is a bit unusual. PCIe 3.0 compliant devices ought to negotiate to 2.x and 1.x speeds if they repeatedly fail to come up at 8Gbps.
 
Does the system boot when only 1 cards is used with the splitter?
It should unless the bifurcation only happens when 2 cards are present.

Also I'm still interested in better photo's of the back and the chips.
 
QinX said:
Does the system boot when only 1 cards is used with the splitter?
It should unless the bifurcation only happens when 2 cards are present.

Also I'm still interested in better photo's of the back and the chips.
Click to expand...

Yes, the system does boot up with only 1 card for all 3 splitters for the Geforce 610 GT and Geforce 8400 GS. However, only the PLX splitter works for the 970 GTX.

I have also attached more images from the splitters here, have fun :)

BTW: the PLX chip heatsink gets super super hot!

cWBtuvll.jpg


KWymf1Kl.jpg


vOd3N6Bl.jpg


n3IcCjVl.jpg


nnWNZbjl.jpg


jEcr6Xtl.jpg


JkL1DVHl.jpg


bixotMHl.jpg
 
Runamok81 said:
CONGRATS CHEMIST_SLIME!!!!!!!!!!!!!!!!!!

If my Google-fu is correct you are the first known example of an individual using dual GPU on a consumer grade mITX size board. You sir, are trail-blazer that inspires us all! The race is on!
Click to expand...

Thanks man! I can't wait to show everyone what I build, I'll have a build log up soon when I have all the parts.
 
jb1 said:
@chemist_slime

Incredible!! So glad to see this worked. I wish I hadn't purchased a Z97 board several months ago, considering selling my system and going with the Asrock board. I wonder if they plan to have bifurcation support going forward with their Z170 boards.

Also, I'll be curious to see how the GTX 970 is recognized in OS X going forward with the PCI-e 3.0 bifurcation without PLX chip once you get the new model. Please post your results! I imagine it will work, but just want to be sure!

As for the Fury Nano, I also think it looks awesome. I'm contemplating doing a build in the A4 with two of these... one oriented the normal way, the other sitting directly behind with the ports facing upward so they can both fit along the length. The second set of ports wouldn't be accessible, but you would still get the power of the two GPUs in the A4. Or, you could water block the two and have them sitting in their normal spot, you'd just have to find a place to put the pump, radiator, etc. This could be considered crazy since AMD is coming out with a dual Fiji card. However, my guess is that this will be attached to a huge radiator like the R9 Fury X, and thus won't fit in something like the A4.

Another thing to consider for OS X is support for the Fury Nano, which has a Fiji chip. As there are no current macs that have this, OS X will likely not support it right now. However, given Apple's current relationship with AMD, I bet you'll see support as soon as they update their Mac Pros!

Another, off topic question, for you. You're running OS X on X99, correct? Do you have power management and sleep/wake working? My understanding is that this wasn't native since there are no current macs that use this chipset.
Click to expand...

Yea, sell that Z87 and get the Asrock x99e itx! What are you evening waiting for :)

Power management doesn't work for now; when the display goes to sleep it does not wake up again.
 
Last edited:
chemist_slime said:
Yes, the system does boot up with only 1 card for all 3 splitters for the Geforce 610 GT and Geforce 8400 GS. However, only the PLX splitter works for the 970 GTX.
Click to expand...

Hmm, that makes it even more puzzling. Wonder what ASRock will have to say.

chemist_slime said:
BTW: the PLX chip heatsink gets super super hot!
Click to expand...

This is known issue, the PLX chip is always splitting and/or multiplexing PCIe lanes no matter if there are cards/jobs or not. On the motherboards with two PLX chips there is special block and radiators with heatpipes. You can see a motherboard with removed radiator showing the two PLX chips . Hot PLX chip is normally not a problem, provided radiators are sufficient. This however may get different in case of tight spaces around the radiators. For example in motherboards with multiple graphics cards which tend to be hot. From this reason it is good to think about additional fan or water cooling block for the PLX chip :). Here it is shown a special full cover waterblock for the Asus X99-E WS mobo covering processor, voltage regulators and PLX/PCI chips.
 
If the splitter was intended to be used in a server, that PLX heatsink may have been specced assuming a certain amount of forced airflow across it.
 
I've got a quick question for anyone that might be interested in a case that can support this setup.

Would you rather have 4 PCI slots and keep with the standard or 4.5 PCI slots so the 2nd GPU can breathe?
I'm looking at either 4 Slots or 2x2 slots with 10mm space between them.
 
If you're already going to non-standard PCIe slot positioning, may as well take advantage of that to space the GPUs, Though for a relatively niche concept (ITX + splitter rather than straight mATX) it might be better to focus on a more compact twin-single-slot layout, with the assumption that anyone going the more complicated route is probably going to use a custom cooling setup to squeeze down the volume rather than using stock blowers.
 
EdZ said:
If you're already going to non-standard PCIe slot positioning, may as well take advantage of that to space the GPUs, Though for a relatively niche concept (ITX + splitter rather than straight mATX) it might be better to focus on a more compact twin-single-slot layout, with the assumption that anyone going the more complicated route is probably going to use a custom cooling setup to squeeze down the volume rather than using stock blowers.
Click to expand...

You can't really get the volume down going for 2 single slot cards instead of 2 dual slot card.
You win 2 slots but you'd need at least a 240mm rad with fans and a place for the pump and reservoir.

You can still use the case with a single card if you want, but it has the option for 2 dual slot cards.

Going for 1 card could allow the use of a radiator and pump config.
 
QinX said:
You can't really get the volume down going for 2 single slot cards instead of 2 dual slot card.
You win 2 slots but you'd need at least a 240mm rad with fans and a place for the pump and reservoir.

You can still use the case with a single card if you want, but it has the option for 2 dual slot cards.

Going for 1 card could allow the use of a radiator and pump config.
Click to expand...

There may not be a volume reduction, but wouldn't the thermals and acoustics be better?
 
Runamok81 said:
There may not be a volume reduction, but wouldn't the thermals and acoustics be better?
Click to expand...

I think I'll start a topic on a case that could house this but I want it to be somewhat prepared.

My current draft has these specs:
Dimensions(HxWxD): 100mmx325mmx300mm (9.75L) with feet
- Fit 2 dual slot videocards up to 11 inches(Flexible SLI bridge needed)
- 4.5 PCI slot layout, allows secondary card to breathe easier
- CPU cooler up to 90mm in height
- Any SFX-L PSU
- unknown amount of drives, but should at least a couple
- potential support for slimline ODD
- Special PCIe risers for Left angle placement and extending. if the initial design works making different risers is not difficult and also not very expensive to do.

Regarding liquid cooling
With the current layout going with 2 single slots videocards would allow you to have 50mm of space to fit a 240mm radiator with fans, I've seen bottom mounted radiators in the M1 so that should fit. You might also be able to fit a 120mm rad above the CPU area.

However, with the M1 you'd have the fittings go up into the CPU cooler area, that isn't possible with this layout because the to fittings would go into the side of the case. If you have side fittings it would be possible. the two questions would be where to put the reservoir and pump and is 1x360mm radiator enough for up to 600W of heat?( looking at the 700W Silverstone PSU)

Liquid cooling in the case would be a challenge, but that hasn't stopped some people doing to amazing stuff in the M1.

I think going with the aircooling first, liquid cooling second is the best way way.

This case would serve the extreme niche of people using the X99e and wanting SLI/Crossfire, but that is the design goal, if it can do SLI it can also to single GPU.

Also in the ASRock forum thread confirming X99e PCIe bifurcation support they say future 100 series motherboards are also going to support it, so there is some future proofness there.
 
Last edited:
I take it would support SFX-L?

360mm is plenty for just the video cards, the 295X2 uses a 120mm.
 
Aibohphobia said:
I take it would support SFX-L?

360mm is plenty for just the video cards, the 295X2 uses a 120mm.
Click to expand...

Correct changed it.

also we that 120mm rad on the 295X2 is mostly placed somewhere with good airflow. the bottom radiator in the M1 or in this future case are both very awkwardly placed in terms of airflow, blowing straight against the card(s) that is/are being cooled by it.

Also anyone else notice that the ASRock forum thread confirming PCIe bifurcation support for X99e and future 100 series board is gone?
 
Aibohphobia said:
this thread? http://forum.asrock.com/forum_posts.asp?TID=525&title=pcie-bifurcation-support-on-x99eitx
Click to expand...

No this one
PCIe bifurcation in Z97E-ITXac - ASRock Forums

In google cache:
Cached

ASROCK_TSD said:
Greetings, this is ASRock Technical Supports Department.

The model which support PCIE bifurcation function is X99E-ITX/ac at current.
http://www.asrock.com/mb/Intel/X99E-ITXac/

If you would like to use PCIE bifurcation function, please choose X99E-ITX/ac or 100 series model in the future.


Thank you for your patience.
If any, please feel free to contact us!


Kindest Regards,
ASRock TSD
Click to expand...
 
Got the new Gen 3 splitter today and it's quite a bit different from the previous one that was Gen 2. Hopefully this one will work with the 970 GTX. :)
 
You must log in or register to reply here.
Back
Top