Is there any particular reason why those switches have MIPSBE CPU as opposed to dual core ARM in CRS317 and CRS309? I mean well CRS317 and CRS309 may not be best routing performers but 3gbps, with 9k jumbo probably much closer to actual 10G is not too shabby for simple inter-vlan routing or uplink routing.

While CRS312 and CRS326-24S+2Q+ basically have CPUs that allow them to... run RouterOS instead of just SwOS I guess? Comparing to switching capacity of those devices 800mhz MIPSBE will do what - 1gbps routing? Okay for CRS326-24S-2Q+ I guess it doesn't make sense to put any reasonable CPU since anything to match its switching capacity would probably cost more than whole this switch but CRS312 doesn't sound THAT powerful tbh. It actually has lower capacity than CRS317. So idk it could make use of some dual core ARM probably.

I'm mostly bringing this topic up since for a while it looked like you guys are slowly heading towards L3-ish switches but now it went back all the way to CRS2xx series with ultra-poop CPUs incapable of any decent L3 processing. Not that I have problem with router-on-the-stick configurations for inter VLAN routing but you're still somewhat limited to uplink speed (which is not a problem for CRS326 with 2x 40G uplinks but for CRS312 it's not so cool). It's kind of like gigabit CRS326-2S+ where lack of routing is non-issue since you have 20G uplinks and 24G access ports so it's totally fine and for new CRS326 it's 80G uplinks for 240G access ports so quite fine I guess but then again you don't have routers that could easily handle 80G of simple inter VLAN routing (except CCR1072 maybe but it's probably overkill for simple inter VLAN routing) while 20G with jumbo should be easily doable on any 2S+ CCR.

I'm basically confused by L3 inter VLAN routing options because your gear doesn't seem to have capacity to bring inter VLAN routing at any decent performance to network using CRS326-2Q+. In matter of fact I don't see any option AT ALL to provide inter VLAN routing to CRS326-2Q+ because even other vendors assume that at this speed inter VLAN routing is performed on L3 switch and any routers with QSFP+ interface will be some hardcore carrier grade behemoths handling such traffic with all bells and whistles. Well there's always just option to get generic server with 40G QSFP+ network card and do such simple jumbo frames routing on budget Linux box but well... it's kinda ghetto tbh.

Does anyone else see some other reasonable way to provide inter VLAN routing to CRS326-2Q+?

These aren't marketed (or priced) as full L3 switches. Yes you can route ports to CPU and run some L3 functions, but it is not a fully featured / full wire rate L3 switch, so if thats what you want this product for then this product is not for you.

You'd have to send in your recommendations to Mikrotik for a more full-featured / wirespeed L3 switch. With enough demand they would make such a product.

These aren't marketed (or priced) as full L3 switches. Yes you can route ports to CPU and run some L3 functions, but it is not a fully featured / full wire rate L3 switch, so if thats what you want this product for then this product is not for you.

You'd have to send in your recommendations to Mikrotik for a more full-featured / wirespeed L3 switch. With enough demand they would make such a product.

I know it's not full L3 switch and it's fine I guess. Nevertheless when breaking past 10-20G uplink barrier it's starting to be difficult to find budget routing device that could handle 40-80G inter VLAN routing. I'm not really complaining about device itself but I rather wonder how to actually use it reasonably. When we're talking about full networking solution it's getting tricky to work with pure L2 high speed switch such as CRS326-2Q+.

It's also worth to mention that inter VLAN routing is significantly less demanding than full featured routing since it's:
- typically stateless
- typically doesn't use firewall
- at this speed you can assume many of use cases will be internal communication backbone so 9k jumbo or 4k at least is norm

I'm using CCR1009 and it pulls off 10G full duplex for inter VLAN routing without sweat - even without jumbo and with single TCP stream (so almost single core). I'm sure if CCR1032 had QSFP+ interface or two it could easily pull off simple inter VLAN routing at 40G+ speeds. So the problem here is not really with power but rather with interfaces capacity of currently produced devices.

I wish CCR routers had more uplink interfaces or at least higher capacity interfaces. Connecting CRS326-2Q+ to CCR1072 using 2x QSFP+ -> 4x SFP+ breakout cables and 8x LACP sounds like bad joke xD

I'm sure QSFP+ enabled routers (CCR2xxx) range will be in the pipeline, these switches are basically the introduction to them. A 1072 equivalent with 2x QSFP and 6+ SFP+ ports will be magical for core routing.

I'm sure QSFP+ enabled routers (CCR2xxx) range will be in the pipeline, these switches are basically the introduction to them. A 1072 equivalent with 2x QSFP and 6+ SFP+ ports will be magical for core routing.

Now you got me hyped! xD

And if you want something REALLY fast just built your own x86 machine stuffed full of PCIe network cards and install RouterOS on it. One the best, and most under-rated features of Mikrotik is they RouterOS separately so you can install it on your own hardware.

Anyone tried if a QSFP+ breakout DAC can be used on CRS326 to join 4 Mikrotik SFP+ ports to a QSFP+ host, as 3rd (or 4th) QSFP+ connection, on top of the builtin 2 QSFP+ ports?

Anyone tried if a QSFP+ breakout DAC can be used on CRS326 to join 4 Mikrotik SFP+ ports to a QSFP+ host, as 3rd (or 4th) QSFP+ connection, on top of the builtin 2 QSFP+ ports?

Breakout support is host property. In a sense that device with QSFP+ port must be able to 'see' QSFP+ as 4 separate SFP+ ports. It has nothing to do with side using 4x 10G links. breakout support is rare for pci-e cards in servers but I believe it is sometimes supported. That said breakout always turns 1 QSFP+ into 4 SFP+. It's not reverse operation that will magically turn 4x 10G into one QSFP+. So to answer your question directly - no I believe it's not possible to turn 4x SFP+ ports on CRS into 1 QSFP+. BUT you can still use LAG like LACP to create 40G link between QSFP+ server and MikroTik with 4x SFP+ though you need to be aware of LACP limitations regarding single TCP tunnel speed (max 10G per single connection, rarely bottleneck unless it's storage server utilized by few really strong clients though it's still only partially true since most of storage protocols support multipathing which makes this issue irrelevant and enables to utilize full 40G even to 1 client).

But I'd love to be proven wrong