PCIe Gen 4 bifurcation risers and NVMe SSDs

mirrormax

Active Member
Apr 10, 2020
141
62
28
thats brilliantly simple as long as motherboard supports bifurication on the x16 slot to 4x4x4x4 you should be good
 

lunadesign

Active Member
Aug 7, 2013
220
26
28
Yeah I like the simple aspect but I recall reading somewhere that the cable length can really impact things with PCIe Gen 4, and thus the need for things like retimer. But I'm finding a hard time finding that source now.
 

im.thatoneguy

New Member
Oct 28, 2020
26
4
3
I've been searching for weeks for something exactly like that!

I'm curious too about the need for retimers. I've been keeping an eye on Micro Sata since they had a Gen3 card:
PCIe Gen3 16-Lane to SlimLineSAS (SFF-8654) 8i Adapter - Newegg.com

They have a redriver for Gen4 which seems to suggest that a redriver at the very least is necessary.

Gigabyte has a product but they aren't apparently releasing it outside of a full system purchase. But it looks like it's passive.

There is an active retimer card:
Gen4 Retimer Cards - Serial Cables

But $500 and requires a fan.
 
Last edited:

NateS

Active Member
Apr 19, 2021
137
76
28
Sacramento, CA, US
Yeah I believe whether you need a redriver or not depends on how long of a cable you're planning to plug into it, but I don't know an exact length figure. Probably how far the slot is from the CPU or a motherboard retimer will also make a difference.

I think standard server chassis with PCIe slots all the way at the back and drive bays up front will pretty much always need retimers. In a desktop tower case, you might be able to get away without it, but I haven't tested it myself. And of course motherboards that natively support NVMe will generally have retimers on board if they'd be required for a standard case form factor.
 
  • Like
Reactions: im.thatoneguy

NateS

Active Member
Apr 19, 2021
137
76
28
Sacramento, CA, US
This website claims.
PCIe 4.0 and 5.0 require the use of retimers for compliance.
Choosing the Right Redriver or Retimer Device
Retimer as opposed to redriver, assuming you need one. It's not saying you always need a retimer; rather if you're driving long cables or traces, you need a more complicated, PCIe protocol aware retimer, rather than a simple protocol-agnostic redriver as was common with PCIe 3.0.

That website also gives some helpful info for figuring out whether you're likely to need one or not. The end-to-end connection needs to have no more than 36dB of attenuation, and with PCIe 4.0, you're losing about 2.3dB per inch of trace on a standard FR4 PCB. Connectors add a loss of about 1.5dB. So, for a standard AIC, we have 1 connector for the card slot, and let's say two inches to route the traces from the edge of the card to the ASIC, then that means the slot can be a maximum of 13 inches from the CPU. OTOH, if we want to use a NVMe HBA, then we have three or four connectors in the chain (MB to HBA, HBA to cable, cable to drive or backplane, and backplane to drive if a backplane is used), which eats up 6dB of our budget right off the bat. If we assume cables have a similar dB loss per inch as a PCB (which is probably not a great assumption, but I can't find a good number for cables), then we have 13 inches available for CPU->slot, slot->cable connector, and cable->(backplane)->drive. Anything beyond that would need a retimer, which would reset our budget to the full 36dB at the point in the chain where it's inserted.

For PCIe 5, all these numbers are much worse. I expect when that's out we'll start seeing optical PCIe links become more common.
 

lunadesign

Active Member
Aug 7, 2013
220
26
28
Now we just need Icy Dock to release a 5.25" bay hot swap enclosure for gen 4 as well.
I've been interacting with Icy Dock for a long time about Gen 4 models. It seems these models are coming very soon with some actually advertised on their site now (i.e, MBB720-MK-B V2). However, last I heard, the Gen 4 version of MB699VP-B has been delayed again until later this year.
 
  • Like
Reactions: im.thatoneguy

lunadesign

Active Member
Aug 7, 2013
220
26
28
Retimer as opposed to redriver, assuming you need one. It's not saying you always need a retimer; rather if you're driving long cables or traces, you need a more complicated, PCIe protocol aware retimer, rather than a simple protocol-agnostic redriver as was common with PCIe 3.0.

That website also gives some helpful info for figuring out whether you're likely to need one or not. The end-to-end connection needs to have no more than 36dB of attenuation, and with PCIe 4.0, you're losing about 2.3dB per inch of trace on a standard FR4 PCB. Connectors add a loss of about 1.5dB. So, for a standard AIC, we have 1 connector for the card slot, and let's say two inches to route the traces from the edge of the card to the ASIC, then that means the slot can be a maximum of 13 inches from the CPU. OTOH, if we want to use a NVMe HBA, then we have three or four connectors in the chain (MB to HBA, HBA to cable, cable to drive or backplane, and backplane to drive if a backplane is used), which eats up 6dB of our budget right off the bat. If we assume cables have a similar dB loss per inch as a PCB (which is probably not a great assumption, but I can't find a good number for cables), then we have 13 inches available for CPU->slot, slot->cable connector, and cable->(backplane)->drive. Anything beyond that would need a retimer, which would reset our budget to the full 36dB at the point in the chain where it's inserted.
This is really helpful. I suppose this is one of those "won't know until you try it" type things. But how does one test a real-world setup? Is it as simple as getting one of these "dumb" risers and trying it out and see if the drive is detected and provides the expected performance? Is there any danger to damaging the drive if the distance is too long?
 

lunadesign

Active Member
Aug 7, 2013
220
26
28
Looks like there are cables available to go between the slimsas 8i and the 4x Oculink on the IcyDock.

模型 (serialcables.com)
Maybe.

What I've been learning is that while SFF-8654 (SlimSAS) and SFF-8611/8612 (OCuLink) are connector standards, different manufacturers wire them up differently so there's no guarantee things will play with each other. That is, unless you buy the whole chain (adapter, cable, backplane) from the same manufacturer.

Apparently there's a newer SFF-9402 standard that dictates how those connector standards (and perhaps some others) should be electrically wired to ensure compatibility (and in some cases, cable reverse-ability).

It looks like the Gen 4 Icy Dock bays are going to use SFF-8612 OCuLink connectors that expect the SFF-9402 compliant wiring. The challenge is that I'm not seeing a lot of adapter and cable makers claiming SFF-9402 compliance. Maybe it's too new?

If any of this is wrong, please feel free to correct me!
 

im.thatoneguy

New Member
Oct 28, 2020
26
4
3
This is really helpful. I suppose this is one of those "won't know until you try it" type things. But how does one test a real-world setup? Is it as simple as getting one of these "dumb" risers and trying it out and see if the drive is detected and provides the expected performance? Is there any danger to damaging the drive if the distance is too long?
Damage isn't going to be an issue, just like a bad ethernet cable, you'll see reduced performance. Worst case scenario potentially random disconnects.

Retimer as opposed to redriver, assuming you need one. It's not saying you always need a retimer; rather if you're driving long cables or traces, you need a more complicated, PCIe protocol aware retimer, rather than a simple protocol-agnostic redriver as was common with PCIe 3.0.

That website also gives some helpful info for figuring out whether you're likely to need one or not. The end-to-end connection needs to have no more than 36dB of attenuation, and with PCIe 4.0, you're losing about 2.3dB per inch of trace on a standard FR4 PCB. Connectors add a loss of about 1.5dB. So, for a standard AIC, we have 1 connector for the card slot, and let's say two inches to route the traces from the edge of the card to the ASIC, then that means the slot can be a maximum of 13 inches from the CPU....

For PCIe 5, all these numbers are much worse. I expect when that's out we'll start seeing optical PCIe links become more common.
Actually it looks like for PCIe 5 things are actually better. the 36db is for PCIe 5. PCIe 4 is 28db according to this SIG graphic.

1622165312137.png
 

NateS

Active Member
Apr 19, 2021
137
76
28
Sacramento, CA, US
Damage isn't going to be an issue, just like a bad ethernet cable, you'll see reduced performance. Worst case scenario potentially random disconnects.

Actually it looks like for PCIe 5 things are actually better. the 36db is for PCIe 5. PCIe 4 is 28db according to this SIG graphic.
Agreed that trying it shouldn't cause any problems -- most likely it'll just link up at Gen3 speeds if the length for Gen4 is exceeded.

Interesting that Gen4 has less link budget than Gen5, but the part I meant was worse is the losses along the way -- Gen5 is losing 4.0dB/inch of FR4 trace instead of 2.3.
 
Last edited:

lunadesign

Active Member
Aug 7, 2013
220
26
28
Agreed that trying it shouldn't cause any problems -- most likely it'll just link up at Gen3 speeds if the length for Gen4 is exceeded.

Interesting that Gen4 has less link budget than Gen5, but the part I meant was worse is the losses along the way -- Gen4 is losing 4.0dB/inch of FR4 trace instead of 2.3.
Good to know. And what if the SlimSAS / OCuLink cables are wired incompatibly? IE, if the port uses "legacy" electrical wiring but the cable uses SFF-9402 compliant wiring. Any risk of damaging the SSD?
 

j.battermann

Member
Aug 22, 2016
50
7
8
41
Has anyone tried hooking up NVMe SSDs to something like this? Or is a retimer required with PCIe Gen 4?


View attachment 18638
FYI - just like many other of the IOI products, this card is re-sold by Delock: Delock Products 89030 Delock PCI Express x16 Card to 4 x internal SFF-8654 4i NVMe - Bifurcation . You can however contact ioi directly and they will help you out to get the product directly from them. This one or any other of their more exotic cables (as I did).
 

Stephan

Active Member
Apr 21, 2017
272
145
43
Germany
Retimer as opposed to redriver, assuming you need one. It's not saying you always need a retimer; rather if you're driving long cables or traces, you need a more complicated, PCIe protocol aware retimer, rather than a simple protocol-agnostic redriver as was common with PCIe 3.0.
So with a passive m.2 to U.2 Adapter + cable for PCIe 3.0, how long can the cable itself in theory be? Bought this Delock Products 62721 Delock Adapter M.2 Key M > SFF-8643 NVMe a while back (version without L1 power management) together with a 50cm cable and could never figure out if this is already beyond the bus' link budget. Board is a Supermicro X11SAT. I also saw a 75cm cable on offer but I went for 50cm to have less attenuation. Device is an Intel P3700 U.2 2.5".
 

NateS

Active Member
Apr 19, 2021
137
76
28
Sacramento, CA, US
So with a passive m.2 to U.2 Adapter + cable for PCIe 3.0, how long can the cable itself in theory be? Bought this Delock Products 62721 Delock Adapter M.2 Key M > SFF-8643 NVMe a while back (version without L1 power management) together with a 50cm cable and could never figure out if this is already beyond the bus' link budget. Board is a Supermicro X11SAT. I also saw a 75cm cable on offer but I went for 50cm to have less attenuation. Device is an Intel P3700 U.2 2.5".
It depends on the cable, and like I mentioned above, I haven't been able to find a good figure on the attenuation per unit length for PCIe cables. But Gen3 is a lot more forgiving, and I know at least Intel has made a 95cm PCIe3 cable, so if the cable is decent, and you're not also trying to drive long traces at both ends, 75cm should theoretically be fine, and 50cm even better.
 
  • Like
Reactions: Stephan