DMI 4.0 will it be a bottleneck?

[stand]

Hi all,

Alder Lake's W680 platform comes with the next generation of the DMI interface - 4.0. Moreover it is now an x8 connection instead of an x4. This makes it 4x faster than the previous DMI 3.0 x4 interface.

According to this article, as well as the Wiki entry on DMI, the throughput of the DMI 4.0 x8 interface is pretty much on a par with an PCIe4.0 x8 connection - ~16 GB/s.

This got me wondering - since it now has the throughput is it still a bad idea to connect a bunch of HDDs/SSDs through it instead of directly to the CPU?

For example the Asrock Rack W680D4U-2L2T motherboard has 4 SATAs and 3 OcuLink connections to the PCH. Theoretically even if I connect 12 SSDs (splitting each OcuLink to 4x SATA 6Gpbs connections) the bandwidth I would need is 12 * 6 = 72 Gbps. On the other hand the throughput of the DMI interface is ~16GB/s or ~128Gbps. This leaves around ~56Gbps for other peripherals which is more than enough.

Am I missing something here or does the new DMI interface finally resolve the bottleneck issues we used to have?

P.S. I know Alder Lake is bleeding edge and still not fully supported by TrueNAS. But the Linux kernel already has support for it so it is just a matter of time before SCALE's kernel gets updated and start supporting it too!

 

[Arwen]

There may be bottle necks inside the PCH chip its self. (Like a PCH chip with only a x4 connection, at DMI 4.0 speeds.) But, DMI 4.0 does sound like a good improvement.

One thing to remember, is that a ZFS vDev made up of 6 SSDs, (12 is a bit too many for a single vDev), ZFS would be reading or writing from this 6 SSD vDev. Not the whole 12 SSDs at once.

 

[stand]

Yup it does sound awesome!

I am hoping it really delivers on its promise. I have an awesome Alder Lake build in mind.

> ZFS vDev made up of 6 SSDs, (12 is a bit too many for a single vDev),
True, I said 12 because I wanted to emphasise how many SSDs you can connect without reaching the speeds it can offer.