Gigabyte M.2 PCIe Riser Cards Take Storage Performance to New Level

[Guest]

Gigabyte reveals its new CMT4034 and CMT4032 M.2 PCI-Express riser cards.

Gigabyte M.2 PCIe Riser Cards Take Storage Performance to New Level : Read more

 

[hotaru251]

hmmm I have an older desktop that has no m.2 slots on motherboard but i dont use any of the pci slots...how would I be able to tell if they were 3.0 or not? (its about 5 yrs old or so now)

 

[takeshi7]

Do these require PCIe bifurcation support on the motherboard, or does the card handle it on its own?

 

[Non-Euclidean]

Look up the specs of the MB or PC.

 

[Brian_R170]

Does this mean we'll soon be seeing another Optane-RAID review with four of the 905p M.2 SSDs in RAID 0?

 

[popatim]

@Brian, why stop at 4 when 2 of these cards can net you an 8-way raid 0. LoL

 

[bit_user]

[quotemsg=21116325,0,1800928]hmmm I have an older desktop that has no m.2 slots on motherboard but i dont use any of the pci slots...how would I be able to tell if they were 3.0 or not? (its about 5 yrs old or so now)[/quotemsg]
If the CPU is a desktop Sandybridge (or older), then it's only PCIe 2.0. If it's Ivy Bridge or newer, then they might be PCIe 3.0.

As @Non-Euclidean said, just look around at the motherboard until you see what looks like the model number. Do a search (or visit the manufacturer's website) and you'll probably find the owners' manual, which will spell out the different slots' specs and configuration options.

 

[SkOrPn]

No mention of the need for Bifurcation requirement? Without PCIe port bifurcation enabled in the BIOS, either hidden or via switch, the Motherboard and OS will have no clue what to do with an adapter such as that. PCIe was designed for one device per slot regardless how many lanes it has. PCIe port bifurcation was developed by Intel to get around this limitation. Cmon Liu, you should know this stuff. How does this adapter work on a x8 or x16 slot without bifurcation support? Not very many consumer boards will have that, if any at all.

EDIT: OK, Gigabyte clearly states these riser cards are only meant for their Purley Server systems, which obviously have Bifurcation built in.

 

[compprob237]

[quotemsg=21116991,0,310904]@Brian, why stop at 4 when 2 of these cards can net you an 8-way raid 0. LoL[/quotemsg]
Well, only the Threadrippers really have enough PCI-E lanes to do that. You could squeak by on the i9 Skylake-X chips but the GPU would be limited to x8 and that assumes all but 4 lanes are used exclusively by these. Meanwhile, the Threadripper would only use up half of its lanes so you could have dual x16 (theoretical, but not realistic) GPUs in SLI on top of the 8-way raid.

 

[Diji1]

What happens if you do not have PCI-E lanes available? Does it not work at all or work but with degraded performance if something else is sharing or what?

 

[emeraldsmines1990]

[quotemsg=21118053,0,960078]What happens if you do not have PCI-E lanes available? Does it not work at all or work but with degraded performance if something else is sharing or what?[/quotemsg]

If you dont have lanes available then the empty slot you have will be disabled. and cant be used.

Some motherboards have PLX Switches to share lanes ... but are very expensive ones.

 

[bit_user]

[quotemsg=21118053,0,960078]What happens if you do not have PCI-E lanes available? Does it not work at all or work but with degraded performance if something else is sharing or what?[/quotemsg]
Some devices will still work in slots that have fewer lanes enabled (such as using graphics cards in x16 slots that only have x4 lanes active). However, these cards seem to route x4 lanes to each M.2 drive.

So, it's a good question whether you could still use a single drive in a PCIe slot with only x4 lanes active, for instance.

 

[supremelaw]

In general, the computer industry really needs to standardize a set of features that are compatible with this relatively new "4x4" option for PCIe x16 expansion slots. One of the main and most important reasons is the proven fact that Intel's DMI 3.0 link has the exact same upstream bandwidth as a single M.2 NVMe SSD. There is an obvious engineering elegance that obtains with 4 x NVMe M.2 SSDs @ x4 PCIe 3.0 lanes = x16 edge connector. And, the availability of idle CPU cores has rendered obsolete the need for dedicated Input-Out processors on AICs (add-in cards). Another compelling reason is the 16 GHz clock rate now approved for the PCIe 4.0 standard. Even if it ends up being a "premium" feature, there is a need for high-end desktop computers to support modern RAID modes with this "4x4" functionality, in the same manner that native SATA ports ended up supporting modern RAID modes. This is a natural, and inevitable, evolution to the NVMe standard which was invented in part to eliminate the overhead inherent in the SATA protocol (read "to maximize speed").

 

[emeraldsmines1990]

[quotemsg=21120468,0,43340]In general, the computer industry really needs to standardize a set of features that are compatible with this relatively new "4x4" option for PCIe x16 expansion slots. One of the main and most important reasons is the proven fact that Intel's DMI 3.0 link has the exact same upstream bandwidth as a single M.2 NVMe SSD. There is an obvious engineering elegance that obtains with 4 x NVMe M.2 SSDs @ x4 PCIe 3.0 lanes = x16 edge connector. And, the availability of idle CPU cores has rendered obsolete the need for dedicated Input-Out processors on AICs (add-in cards). Another compelling reason is the 16 GHz clock rate now approved for the PCIe 4.0 standard. Even if it ends up being a "premium" feature, there is a need for high-end desktop computers to support modern RAID modes with this "4x4" functionality, in the same manner that native SATA ports ended up supporting modern RAID modes. This is a natural, and inevitable, evolution to the NVMe standard which was invented in part to eliminate the overhead inherent in the SATA protocol (read "to maximize speed").[/quotemsg]

The industry already has this . it is called PLX Switch Chips. and they exist in many Motherboards/products.Cards

 

[supremelaw]

Assuming bifurcation is available, one possible configuration is to install a 4x4 AIC into the first PCIe 3.0 slot, and choose a video card that only needs x8 PCIe lanes e.g. AMD Radeon Pro WX 2100 i.e. x16 edge connector but only x8 electrical. If you can find a photo of that video card, you can see how the edge connector is only populated with half of the normal conductors on its physical x16 edge connector.

 

[bit_user]

[quotemsg=21120476,0,2752725][quotemsg=21120468,0,43340]In general, the computer industry really needs to standardize a set of features that are compatible with this relatively new "4x4" option for PCIe x16 expansion slots. One of the main and most important reasons is the proven fact that Intel's DMI 3.0 link has the exact same upstream bandwidth as a single M.2 NVMe SSD. There is an obvious engineering elegance that obtains with 4 x NVMe M.2 SSDs @ x4 PCIe 3.0 lanes = x16 edge connector. And, the availability of idle CPU cores has rendered obsolete the need for dedicated Input-Out processors on AICs (add-in cards). Another compelling reason is the 16 GHz clock rate now approved for the PCIe 4.0 standard. Even if it ends up being a "premium" feature, there is a need for high-end desktop computers to support modern RAID modes with this "4x4" functionality, in the same manner that native SATA ports ended up supporting modern RAID modes. This is a natural, and inevitable, evolution to the NVMe standard which was invented in part to eliminate the overhead inherent in the SATA protocol (read "to maximize speed").[/quotemsg]

The industry already has this . it is called PLX Switch Chips. and they exist in many Motherboards/products.Cards[/quotemsg]
Aren't those just "dumb" switches, though? @supremelaw mentioned RAID...

 

[emeraldsmines1990]

[quotemsg=21120497,0,328798][quotemsg=21120476,0,2752725][quotemsg=21120468,0,43340]In general, the computer industry really needs to standardize a set of features that are compatible with this relatively new "4x4" option for PCIe x16 expansion slots. One of the main and most important reasons is the proven fact that Intel's DMI 3.0 link has the exact same upstream bandwidth as a single M.2 NVMe SSD. There is an obvious engineering elegance that obtains with 4 x NVMe M.2 SSDs @ x4 PCIe 3.0 lanes = x16 edge connector. And, the availability of idle CPU cores has rendered obsolete the need for dedicated Input-Out processors on AICs (add-in cards). Another compelling reason is the 16 GHz clock rate now approved for the PCIe 4.0 standard. Even if it ends up being a "premium" feature, there is a need for high-end desktop computers to support modern RAID modes with this "4x4" functionality, in the same manner that native SATA ports ended up supporting modern RAID modes. This is a natural, and inevitable, evolution to the NVMe standard which was invented in part to eliminate the overhead inherent in the SATA protocol (read "to maximize speed").[/quotemsg]

The industry already has this . it is called PLX Switch Chips. and they exist in many Motherboards/products.Cards[/quotemsg]
Aren't those just "dumb" switches, though? @supremelaw mentioned RAID...
[/quotemsg]


PLX chips will make the system see more lanes (double or more than double) and there are many PLX chips available.

you can see them all here

https://www.broadcom.com/products/pcie-switches-bridges/pcie-switches/

download the PDF of each model and read.

and they arfe not dummy they act like ACTIVE Lanes with ports. Actually some one made an M2 Card with PLX chip that supported RAID and was reviewed here a while ago. That card does not even need X299 chipset to boot in Raid. it works on any Standard X16 Slot.

 

[supremelaw]

See Highpoint's SSD7101, SSD7110 and SSD7120 add-in cards. AFAIK, the model SSD7110 is the only one that is bootable, according to the last time I read the relevant Highpoint specifications. Also, a Forum User at servethehome.com tried booting from the SSD7101 with the driver for the SSD7110, and it worked for a short while. But, then Highpoint reportedly changed their driver downloads for those 2 AICs, and that change prevented the SSD7101 from being bootable. There is a review of the SSD7120 here at Tom's Hardware. Contact Highpoint for the very latest status of the "boot-ability" of those 3 AICs.