I haven’t been directly involved in networking for a while, so it was funny to tune back in and see that now not only is 10Gbps apparently no big deal, they even put it on the management interface!
I believe the 30W figure was power usage before SFPs or DACs. Looks like 400G sfps are about 10W, so you'd double it when fully loaded. I believe passive twinax doesn't use anywhere near that though.
Essentially, yes. The DGX Sparks, at 128 gigs for ~$4700 are one of the cheapest ways to get enough high-ish speed memory cobbled together to run one of the more capable open weight models to run at home/for a small biz. This switch lets you combine the memory of 2-4 of them.
They are QSFP56-DD which can be split using breakout cables. So it's 4 x 400GbE ports, which could in theory be split into as many as 32 x 50GbE ports. (Although whether that then ends up being an economical thing to do is a different matter.)
Ahh yeah, I've seen people using those. I was thinking you'd ideally want to connect them full speed, but it looks like they're 200gbps ports anyway, and I guess even if they were 400gpbs, you don't really need that much for cross-GPU traffic for MoE models doing tensor parallel.
I'm not defending the use of money; I think it's questionable. It's still a home lab by my definition: it's computer stuff that costs less than a sports car in your home.
I purchased my SSD based on price and I have a AM4 APU so I'm limited to pci-e 3, but a quick check with crystaldiskmark says I'm hitting about 3.6 GB/s on sequential reads in seq1m q8t1 ... That would saturate 25gbps (assuming you have the right access patterns). seq1m q1t1 does 2.3 GB/s on my system, close enough to 25 gbps.
By my math, pci-e 5.0 should do 4x the throughput, so if you had 4 pci-e ssds you wanted to serve over the network and you had the pci-e lanes, you could saturate 400 gbps.
I'm pretty happy with my little baby network with a handful of 10G and 2.5G ports, and I'm cheap so if I made a big upgrade it'd probably be something used that has more 10g ports plus a 40G I could DAC to the adjacent home server ... but my home servers are barely powerful enough to push 10g anyway. (Something like a DCS-7050TX-48 goes for ~ $100 and would be a big upgrade in capacity I don't use over the S3300-28X I currently run)
DGX Sparks have 200Gbps interlinks for clustering. Apparently switches are the only practical way to scale such clusters beyond 2 nodes even though each Spark has 2 such ports.
Well, the problem is that you want each node to be able to directly connect to each other, so you run into a mesh situation. https://www.youtube.com/watch?v=QJqKqxQR36Y shows 8 running at 100gbps and clustered.
I understand that part, what I don't quite get is why we can't divide the layers such that each node only needs to talk to at most 2 nodes. I assume I'm misunderstanding how work is distributed between them, but haven't looked into the details yet.
I think this is the distinction between pipeline parallelism (first N layers on machine 1, next N layers on machine 2, etc.) vs tensor parallelism where all machines contribute to the same matrix multiplication operation within the same layer. Pipeline parallelism doesn't need exotic high speed networking, but also doesn't deliver the kind of speed increase that tensor parallelism provides.
(tl;dw - guy connects first four DGX and then another 4 "DGX-alikes". Uses 400gbps breakout cables (you can split interfaces, TIL) so he has 2 ports on the first switch, and then gets this one, I believe. You can buy cables that are 1x400 to 4x100gig, etc.
I mean at lot of what a homelab is experimenting. For example I have an absurd amount of communication 56GbE Mellanox switches but never really saturate it.
I'd really like them to take one or two of those 400G ports and split them into internal port splitters to generate a 32x10/25g switch with 400g uplinks for $2k. They do a 8x50G one (crs812_ddq)
Ultimately this is a Marvell 98dx7335 with a mikrotik management interface on the top. Most switches you tend to encounter use Broadcom chipsets rather than Marvell.
Setting aside what features routeros exposes, This chip seems to be somewhere between a Trident 3 and Trident 4 in capability (hardware support for things like evpn/vxlan, q-in-q, mpls, macsec etc)
Last mikrotik switch I tried only supported PTP Transparent clock, this chip does do boundary clock, but I don't think it's implemented through the linux layer let alone exposed via the RouterOS layer.
I think the chip is mainly aimed at service providers, which fits in well with mikrotik's main customer base (ISPs)
I haven’t been directly involved in networking for a while, so it was funny to tune back in and see that now not only is 10Gbps apparently no big deal, they even put it on the management interface!
Noting actually says those ports have to be for management.
Most people probably would use them for something else.
They're not connected to the ASIC so anything on those ports is taking the slow path.
> The exciting part of this is that there are four 400GbE QSFP56-DD ports for a total of 1.6Tbps of network bandwidth.
That I can't break down to 10gbit or 1gbit without another switch...
> The list price is $1295, but these often sell at a street price of $1100 or so, making this under $0.70/Gbps
It doesn't work that way...
I like mikrotik, but who is this switch really made for?
Small-mid scale company who want to setup local LLM i guess ?
> While $1100 is not a low number
It really is when you're talking 400G.
OK there's very few ports on this, but in general if you're looking at 400g interfaces. $1100 won't even get you a genuine Arista 400G passive DAC.
On their video Patrick points out that this is now lower than $1 per Gbps, cheaper than any of the cheapest AliExpress switches.
Power usage is also very reasonable! 400GE switches and routers pull a ridiculous amount of power.
I believe the 30W figure was power usage before SFPs or DACs. Looks like 400G sfps are about 10W, so you'd double it when fully loaded. I believe passive twinax doesn't use anywhere near that though.
https://www.fs.com/eu-en/products/82464.html
Claims "Near-Zero Power", but I have zero experience with >100g
Is there an actual use for this in a homelab? Isn't 25gbps enough that no SSD will saturate it?
Please be careful with units. There are high-end consumer SSDs that will deliver data at 14+GB/s, which is much higher than 25Gb/s.
The second sentence of the article:
> We have a number of these now and have been running our RDMA network backend for the 8x NVIDIA GB10 cluster for several months now, using one.
I have a friend who uses something similar for his 4 NVIDIA GDX Spark cluster.
I see, so only if you are running an AI cluster. And what is it transferring between the nodes? Partial computations?
Essentially, yes. The DGX Sparks, at 128 gigs for ~$4700 are one of the cheapest ways to get enough high-ish speed memory cobbled together to run one of the more capable open weight models to run at home/for a small biz. This switch lets you combine the memory of 2-4 of them.
4-8 of them. 2 Sparks can operate without a switch.
Well, this one only has 4 ports, would you daisy chain them somehow to get to 8? Or bridge three of these switches?
They are QSFP56-DD which can be split using breakout cables. So it's 4 x 400GbE ports, which could in theory be split into as many as 32 x 50GbE ports. (Although whether that then ends up being an economical thing to do is a different matter.)
Ahh yeah, I've seen people using those. I was thinking you'd ideally want to connect them full speed, but it looks like they're 200gbps ports anyway, and I guess even if they were 400gpbs, you don't really need that much for cross-GPU traffic for MoE models doing tensor parallel.
Oh I'm sorry, just let me casually spend 30k...
I'm not defending the use of money; I think it's questionable. It's still a home lab by my definition: it's computer stuff that costs less than a sports car in your home.
I purchased my SSD based on price and I have a AM4 APU so I'm limited to pci-e 3, but a quick check with crystaldiskmark says I'm hitting about 3.6 GB/s on sequential reads in seq1m q8t1 ... That would saturate 25gbps (assuming you have the right access patterns). seq1m q1t1 does 2.3 GB/s on my system, close enough to 25 gbps.
By my math, pci-e 5.0 should do 4x the throughput, so if you had 4 pci-e ssds you wanted to serve over the network and you had the pci-e lanes, you could saturate 400 gbps.
I'm pretty happy with my little baby network with a handful of 10G and 2.5G ports, and I'm cheap so if I made a big upgrade it'd probably be something used that has more 10g ports plus a 40G I could DAC to the adjacent home server ... but my home servers are barely powerful enough to push 10g anyway. (Something like a DCS-7050TX-48 goes for ~ $100 and would be a big upgrade in capacity I don't use over the S3300-28X I currently run)
DGX Sparks have 200Gbps interlinks for clustering. Apparently switches are the only practical way to scale such clusters beyond 2 nodes even though each Spark has 2 such ports.
Well, the problem is that you want each node to be able to directly connect to each other, so you run into a mesh situation. https://www.youtube.com/watch?v=QJqKqxQR36Y shows 8 running at 100gbps and clustered.
I understand that part, what I don't quite get is why we can't divide the layers such that each node only needs to talk to at most 2 nodes. I assume I'm misunderstanding how work is distributed between them, but haven't looked into the details yet.
I think this is the distinction between pipeline parallelism (first N layers on machine 1, next N layers on machine 2, etc.) vs tensor parallelism where all machines contribute to the same matrix multiplication operation within the same layer. Pipeline parallelism doesn't need exotic high speed networking, but also doesn't deliver the kind of speed increase that tensor parallelism provides.
Yes, AI... this video does a good job walking through: https://www.youtube.com/watch?v=QJqKqxQR36Y
(tl;dw - guy connects first four DGX and then another 4 "DGX-alikes". Uses 400gbps breakout cables (you can split interfaces, TIL) so he has 2 ports on the first switch, and then gets this one, I believe. You can buy cables that are 1x400 to 4x100gig, etc.
I mean at lot of what a homelab is experimenting. For example I have an absurd amount of communication 56GbE Mellanox switches but never really saturate it.
Yup, I have a pair of CRS504, so I can learn MLAG and my three node cluster doesn't really tickle the sides of that.
I'd really like them to take one or two of those 400G ports and split them into internal port splitters to generate a 32x10/25g switch with 400g uplinks for $2k. They do a 8x50G one (crs812_ddq)
Ultimately this is a Marvell 98dx7335 with a mikrotik management interface on the top. Most switches you tend to encounter use Broadcom chipsets rather than Marvell.
Setting aside what features routeros exposes, This chip seems to be somewhere between a Trident 3 and Trident 4 in capability (hardware support for things like evpn/vxlan, q-in-q, mpls, macsec etc)
Last mikrotik switch I tried only supported PTP Transparent clock, this chip does do boundary clock, but I don't think it's implemented through the linux layer let alone exposed via the RouterOS layer.
I think the chip is mainly aimed at service providers, which fits in well with mikrotik's main customer base (ISPs)
Nic needs pcie5 x16 slot, so on am5, take out your gpu