Inside a 400GbE 32-port Teralynx 7 Hyper-scale Switch 

Cheap used 100GigE switches: SOON

I kinda want to get into the high end networking market. Simply so we can have more interesting colours than just dark grey/black and blue

So hard to see guts of these anywhere else. And the tech details as well without all the marketing fluff.

Ahh I was looking for a new switch for the summer house, this one looks like it will fit the Bill perfectly!

I love your enthusiasm, really passionate about technology and you clearly convey the details about everything! Thanks dude, great work - I've subscribed!

I'm still lusting after a 10Gb network, nevermind a 400Gb one! 😂

I just upgraded my backbone at home to 10Gbit/s SFP+ ports.. (old used server hardware really makes it possible) but now this .. 400GbE... ouch.. I want.. I dont' know why and know I'll never use even a fraction of it, but i want it for home use LOL Another great review.

Glorious B-roll!

Thank you for sharing.

I'll stick to SG108E for the moment. Even if they are not quite as hyper scaled... 😃

This should be a nice incremental upgrade for my home network :P

Would have loved a bit more info on what that Xeon D was for...? And also would love to see temperature graphs of the QSFP cages as they start pushing 400Gbps...

Ah yes, when the patch cables cost more than the computer I am typing this on, you know you are pushing packets.

Edgecore has released DCS810, same 12T switching but P4 programmable with Tofino 2, would be cool if you can do a video on that.

The real question is when can I put one in for my home network 🤔

That is bad ass. Wish I could put that in my home. Ha ha🤣

Afaik 400G uses 8x 50G lanes. So if you want to break out to standard 100G (4x25G) you have to reduce the lane speed by half and then you can break out 64x100G and not 128x100G.

Can't wait to see Linus upgrade to one of these next year

Patrick make a video on functions of different network ports like BNC, IPMI, CONSOLE, Out-of-Band Management and In-Band Management

At some point, equipment will just be 99% heatsink.

I wonder if this will be fast enough to run my gigabit internet. 😅

Can those 400 gbits borts be easily be split up to 4x 100 gbit? If so I‘ll need to order a few of those 😅

Marvell has since purchased Innovium, not surprising as Marvell really wants to stay at the #2 spot behind Broadcom.

"...but realistically, we're not going to see 400 GbE adapters until we get to PCIe Gen 5.0 in 2022...", 8:17. You sure about that, Patrick? Could be sooner.

You connect this to storage servers? NAS?

now that's some I/O

What's this tool called for the Traffic generation? Is this a software or Hardware Solution?

If QSFP+ can go from 100 to 400 gb/s how fast can they do before you have to change the connector?

"cool the optics"

What is crazy is that one day this will be in the bargain bin! lol

i cant even imagine what the price of something like this is

A lower level breakdown of the use of a 400Gb switch is: *Power consumption* With fewer units you save money on the power bill. Doesnt matter if the replacement switch uses 5x the power compared to the five switches it replaces, the heat dynamics arent perfectly scaled either. That means that 5 units running at 100watts each doesnt equal one unit running at 500watts. The one unit running at 500watts will end up using less power due to a more efficient power distribution system and better cooling. You have to calculate the loss to thermal mass and heat radiation as well. *Heat management* With fewer units running you can spend less money on the cooling systems, or at least scaling down the effective cooling needed. See the note above. *ROI with and without Pc and Hm* An important feature with all of this is the potential ROI. If you spend $50.000 on one unit that replaces five or ten other units, the cost of cooling, management (which is an actual cost), licences, replacement stock, shipping, replacement time, downtime, fault checking time and loads of other concerns, will be lowered. Sure, you can theoretically deploy 50.000 switches and have one person keep track, but you cant have one person serve them all if something happens. You also have to keep in mind that these switches are for DC use primarily (or really large nodes, however i dont think these would be used for anything else than dcnode cases, dont want to sever too many users from one node due to heat, power and redundancy issues), fault checks are a real part of a NOC techs job. If you can have one person serve the same capacity as it would normally take 5 people to do, you make that change happen. Meatbags are expensive. *Physical space benefits* Due to it replacing several other units you can over a span of 5 of these probably save an entire rack full of patch panels, cable management, lower switches etc. Remember, DC's dont grow on trees, they leaf spine.... sorry for that one.... Space is always an issue. *Smaller cable runs* Fewer cables that you need to keep track off means happier techs when shit hits the fan. *fewer hops / simpler logical topography* This bundles well with the previous points. You have to remember that not all packages (or frames for layer 2 devices) go in straight north/south jumps (to and from the internet), they often go between servers and switches in the internal structure. If you look at an example where this switch could be a good use of power: Imagine you have a cloud provider. When you create a new node in your AWS system you basically rent cores and memory modules from a resource cluster. Imagine the structure youd need to deploy to keep a bussiness up and running if that bussiness use 400 cores and a few hundred terrabytes of RAM. You cant just deploy 400 cores, you need to always have the power to give the user those 400 cores woth of processing cycles. To do that, and cut down on operating costs by renting out the same core several times by pooling the cores and only really selling the cycles, you need way more cores than the user pays for. Complicated? Sure... The idea here is that you need a fast and caplable infrastructure to be able to link several massive clusers of units. If you can remove 10 100Gb 48p switches and replace those with about 4 of these switches (and divisions) you can shave off a few interconnects, simplify the topology and save on valuable latency. You deal with milions of operations per port per second, remove one nanosecond from each operation and you save one milion nanoseconds of latency per port.... Summa sumarum; Its worth it, trust us.

Great but what does it cost and is this just an engineering prototype?

@ServeTheHome Why aren't 2.5 Gbe or 5 Gbe popular yet? SSD's have been around for a decade!

Where do you see Cisco ACI being used vs systems like this which are more Legacy Networking? ACI uses usually Nexus 9K 10-40Gbps connections for Leafs and 40-100Gbps for Spines, along with the native firewalling that ACI provides.

That's one clean PCB. Is the air guide 3D-printed? It doesn't look like injection mold plastic.

Content Free as usual

Isn’t that a Spirent test switch those things are expensive af and you pay per port.

You know... Big companies don't really like buying new stuff to find out it obverheats and shuts itself down...

How much does it cost?

Why do network switches always have network ports in front and most servers always have network ports in the rear

Can you really have TOO much bandwidth ?!? I say no ;)

linus would love this

Don’t show this to Linus!

Honestly I feel you guys should rename the channel to ServeTheEnterprise.... Other than your tiny mini micro series and occasional home network worthy stuff, your videos are mostly geared at enterprise gears.... I'm not sure what led to the name ServeTheHome... What home? Don't get me wrong, I'm a subscriber and I love learning about these things, watching your videos and reading your articles. But the name doesn't make sense, that's all.

Why is this on ServeTheHome? Who is gonna put this in their HOME?