I'm gonna grab a Ryzen 3600 to unlock gen 4 to make a little efficient (still hobbled just not as badly) dual 5060Ti LLM rig. The fact that I'm going to downgrade the 5600G to a 3600 to gain a massive LLM inference tensor parallel performance unlock is absolute cinema.
Multi-NVMe (m.2, u.2) adapters that do not require bifurcation
- Thread starter andrewbedia
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
i guess it will be cheaper to swap the motherboard with one with bifurcation supportI know they are no longer made however is there any chance of finding PLX 8747 or 8750 PCIe switch card that splits PCIe x16 into two x8 slots, not a quad M.2 version.
or this one ? need MCIO -> PCIe slot adapters.I know they are no longer made however is there any chance of finding PLX 8747 or 8750 PCIe switch card that splits PCIe x16 into two x8 slots, not a quad M.2 version.
Last edited:
@RolloZ170 did you mean to paste a link?
I've been looking into this as well. It seems like nowadays finding cheap x8/x8 boards locally is no longer a thing. Too many people also going around building local AI rigs...
For now my assumption is this thing probably gives the most affordable approach: https://www.aliexpress.us/item/3256...ewayAdapt=glo2usa4itemAdapt#nav-specification
Need to add another x16 -> 2xSFF8654 8i card, but that is $20-30.
I do not know if it is pcie 4 capable though, which really matters.
Overall a gen 3 PLX solution could make sense, but to really make sense it has to take 3.0 x16 and serve 4x 3.0 x8 at least. Otherwise you should really just bifurcate. Also, 3.0 x8 is 8GB/s and somewhat pitiful. It may be good enough for tensor parallel inference with 4 mid range GPUs though, I need to evaluate this...
I hope there is a PLX that targets x8 outputs. because it will unlock things like scaling GPUs on x99 better. I think x99 does not do flexible bifurcation so it has these fat x16 slots. I think there are two, so it would be very clean if we can turn 2x x16 into 4x x8.
I've been looking into this as well. It seems like nowadays finding cheap x8/x8 boards locally is no longer a thing. Too many people also going around building local AI rigs...
For now my assumption is this thing probably gives the most affordable approach: https://www.aliexpress.us/item/3256...ewayAdapt=glo2usa4itemAdapt#nav-specification
Need to add another x16 -> 2xSFF8654 8i card, but that is $20-30.
I do not know if it is pcie 4 capable though, which really matters.
Overall a gen 3 PLX solution could make sense, but to really make sense it has to take 3.0 x16 and serve 4x 3.0 x8 at least. Otherwise you should really just bifurcate. Also, 3.0 x8 is 8GB/s and somewhat pitiful. It may be good enough for tensor parallel inference with 4 mid range GPUs though, I need to evaluate this...
I hope there is a PLX that targets x8 outputs. because it will unlock things like scaling GPUs on x99 better. I think x99 does not do flexible bifurcation so it has these fat x16 slots. I think there are two, so it would be very clean if we can turn 2x x16 into 4x x8.
ooops. yes. have update my postdid you mean to paste a link?
Did you ever end up buying a PEX88096? I like many others are looking for solutions to the lack of pcie lanes haha. The PEX88096 seems like a pretty decent buy that you could use for a long time@RolloZ170 did you mean to paste a link?
I've been looking into this as well. It seems like nowadays finding cheap x8/x8 boards locally is no longer a thing. Too many people also going around building local AI rigs...
For now my assumption is this thing probably gives the most affordable approach: https://www.aliexpress.us/item/3256...ewayAdapt=glo2usa4itemAdapt#nav-specification
Need to add another x16 -> 2xSFF8654 8i card, but that is $20-30.
I do not know if it is pcie 4 capable though, which really matters.
Overall a gen 3 PLX solution could make sense, but to really make sense it has to take 3.0 x16 and serve 4x 3.0 x8 at least. Otherwise you should really just bifurcate. Also, 3.0 x8 is 8GB/s and somewhat pitiful. It may be good enough for tensor parallel inference with 4 mid range GPUs though, I need to evaluate this...
I hope there is a PLX that targets x8 outputs. because it will unlock things like scaling GPUs on x99 better. I think x99 does not do flexible bifurcation so it has these fat x16 slots. I think there are two, so it would be very clean if we can turn 2x x16 into 4x x8.
While everyones needs/solutions are different, for Dell 3640 mainboard transplanted into a different case, I ended up converting the 2 M.2 M-key slots on motherboard to 2 PCIe x4 and then using a Glotrends PA20 in the only motherboard PCIe x4 slot to connect NVME SSD + 2.5GbE NIC.