What do you think?

Still, more than I would spend as a home user.

Given that 1G is on PCIe on the current CCR1072, I imagine CCR1072-16G-2S-4S+ is doable and serves additional important use cases with a single model.

Although it might be better for Mikrotik to develop a simpler and thus cheaper CCR1072-12G-4S-4S+ than anything else first.

Your choice?

taking in count the architecture

http://www.tilera.com/products/?ezchip=585&spage=618

even a 32 gigabit ethernet ports are possible

in theory for each 10g port you can obtain 4 1g ports

too many possibilities...

I am a home user and I can see 3x 10G ports in the near future, with 10x 1G about right. 16x 1G would be too many for me. A business might also want at least 2x SFP instead of wasting 2x SFP+.

I would like to have 1x 10G free, that's why the seed specification is 4S+, and having 2x 1G more should be enough.

I don't think I would use the 4S, but those would fit someone else.

Maybe the more popular model would be CCR1072-14G-2S-4S+. I don't know if you can find a 2x7 connector block, but I imagine 2x4 + 2x3 would be a possibility. I did not mention it before because of the particular number of GBaseT ports.

Of course, for me, this is mostly theoretical at this point, as I would not pay 3K$ for home and is overkill, when I would be better served by a cheaper "10G" WAN (because ISPs are already offering more than 1Gbps for home users) and 10G or 10G+1G switch chips with "10G" links to the CPU.

I am not interested in anything less than a 10G NAS, and I don't care at all about NBASE-T.

Vote!

I could allocate space for such a 1U router in my rack, but a fan is out of the question too, not just a money/overkill thing.

For home users, you're much more likely to need a switch with more 10G ports than a router. Unless you're running an enterprise network out of your house, the CCR1072 is just way overkill (and the configuration you're suggesting doesn't even use the hardware to its fullest extent). Layer 2 and layer 3 costs are extremely different. I would imagine that MT could develop a 24G-4S+ switch for marginally more money than their current CRS226, and considering that most people just need more layer 2 ports, would be a much more sensible way to go.

I personally think that CCR1072 only need 10G interfaces for data, you can always add switch , configure a trunk and some Vlans, and have any port configuration you need. CCR1072 are only there for processing power not to fit all possible port combinations.

If port combinations did not matter, we would only have CCR1036-4S+

For home users, you're much more likely to need a switch with more 10G ports than a router. Unless you're running an enterprise network out of your house, the CCR1072 is just way overkill (and the configuration you're suggesting doesn't even use the hardware to its fullest extent). Layer 2 and layer 3 costs are extremely different. I would imagine that MT could develop a 24G-4S+ switch for marginally more money than their current CRS226, and considering that most people just need more layer 2 ports, would be a much more sensible way to go.

It would not be such a big overkill once it is running Suricata after a 16nm shrink, I think.

I know EzChip will not just shrink, but switch to ARM.

There are smaller versions than the 100 core on the pipeline, but I don't know if one capable of 12G-4S+.

I think what is needed is a shrunk 3x7 TILE-Gx21.

I think what is needed is a shrunk 3x7 TILE-Gx21.

Which would allow to reuse a lot of the engineering from CCR1036-12G-4S

I see the core rows are not always in correspondence with the port blocks. Maybe it needs some more, so perhaps 3x8 or 4x7.

Keeping in mind that target is fanless.

2Gbps symmetric WAN performance (512-byte, a few dozen rules, no fasttrack) will be enough for a while.

But it should be capable of switching at 10Gbps.

Hmm... I think if I am going to be switching one 10Gbps stream while at the same time saturating that WAN, I would need at least 32 cores.

Unless a lot of the cores are idle while switching on CCR perhaps.

No, I think it is using only the last core column for switching, which is what one would expect.

Maybe 3x7 is indeed it, then.

Keeping in mind that target is fanless.

2Gbps symmetric WAN performance (512-byte, a few dozen rules, no fasttrack) will be enough for a while.

But it should be capable of switching at 10Gbps.

fanless with tile gx 72 core is very difficult because power consumption is around 100watt

even the ccr1009 fanless comes with 200mhz cpu reduction to maintain power dissipation under control, power monitor indicates 15-20watt on ccr1009.

ccr1072 = ccr1009 passive X 5 power consumption (aprox)

a good record because it has eigth times more cores, is more efficient

That's why I said SHRUNK.

Maybe 21x 1GHz cores would not be too bad for fanless in a 28nm LP process, and a more expensive node could be avoided.

TILE-Gx72 is already 28nm.

So maybe a Gx21 instead of an Mx21 is not unthinkable of.

take a look at chip manufacturer documentation

http://www.tilera.com/products/?ezchip=585&spage=614

I know those pages. I don't know what you expect me to look at.

Mx is thin in details and they don't even mention the smaller models they are preparing.

From what I've seen elsewhere, they will also be 28nm.

For businesses, I imagine what Mikrotik could engineer now with maximum similarity to existing designs would be CCR1072-12G-5S+ .

And if a CCR1072-12G-5S+ is produced, a 3x8 Gx24 might be a better option than a Gx21 if a CCR1024-12G-5S+ could be made fanless.

And a regular CCR1024 could just have a different chip than the corresponding CCR1072, reusing the complete design.

It's a shame TILE are only BGA, which prevents just upgrading the chip.

Good time for contracting some wafers.

Yes this is would be awesome!

I know those pages. I don't know what you expect me to look at.

Mx is thin in details and they don't even mention the smaller models they are preparing.

From what I've seen elsewhere, they will also be 28nm.

I think Mx will ship in 2017, that's why I was wondering if a quick 'son of Gx72' stripping would be possible.