3 ms·

Do you have an idea how much of bandwidth a MIPS cpu like that can push? The chip is actually made for routers: it has 802.11n "MAC/BBP", 5-port 10/100 switch,

by listic 12y ago

Do you have an idea how much of bandwidth a MIPS cpu like that can push? The chip is actually made for routers: it has 802.11n "MAC/BBP", 5-port 10/100 switch, etc. http://www.anz.ru/files/mediatek/MT7620_Datasheet.pdf http://www.anz.ru/files/mediatek/MT7620_Datasheet.pdf I wonder how "Hardware NAT with IPv6 and 2 Gbps wired speed" works if the switch is only 100 Mbits... I guess if what you want out of this is really a router, you are supposed to either build the rest of the hardware yourself or wait for yourself, or wait for the "motherboard" module for this.

api 12y ago
I looked into a lot of these boards a while back and most are like this one: crippled with 100mbps Ethernet. Sad to see that yes, this one is too. The Ethernet module is far slower than what the CPU can push. Doesn't destroy it, but it's certainly a black mark. Are 1gig chipsets that much more expensive than 100mbps these days?
wtallis 12y ago
It's a chip for very low-end routers: 2.4GHz only, no gigabit switch (although apparently it can be equipped with two gigabit ports), and a CPU that would struggle to do 50Mbps WAN-LAN with firewall and QoS. It's only redeeming quality is that it apparently has a PCIe lane, but the CPU is still going to have trouble feeding whatever NIC you connect to that.
ajb 12y ago
I do, thanks.Hmm, let's see if I can give you the information without breaking a confidentiality agreement. Ah, here is a discussion on the openwrt mailing list: https://lists.openwrt.org/pipermail/openwrt-devel/2013-August/020963.html https://lists.openwrt.org/pipermail/openwrt-devel/2013-Augus... As you can see, they reported 90mb/s (they also quote 710, but I think that is the switch doing that). And that's on a mips74k. I know it's made for routers, but most likely not for doing a lot in software. As you quote, it does hardware NAT. There's a reason they put that in. There isn't a fundamental reason why the mips couldn't do the job, but running stock linux it's going to be thrashing its L1 cache, and it doesn't have an L2. Also, next time you play with one of those,check the hardware counters for how many instructions/sec it actually manages to process running flat out :-)
- listic 12y ago
  90Mb/s at least? Makes sense that they won't go for the gigabit Ethernet. Any idea how that compares to ARM with the same frequency? I'd say about the same. That's not bad, right?
  - ajb 12y ago
    I would guess about the same; I didn't have two similar ones to compare. The mips might be a bit slower though. I had the sense that the limitation was really L1 thrashing of code space, so the ARM being a bit denser might help. It would also help if the SOC had a smaller latency to memory as well - don't know what this one has. Bear in mind that it's probably a packet rate rather than a bit rate limitation. That really bit us in our application because we were trying to run many lines of voip, which means lots of small packets. - YMMV.

by listic 12y ago

api 12y ago

I looked into a lot of these boards a while back and most are like this one: crippled with 100mbps Ethernet. Sad to see that yes, this one is too. The Ethernet module is far slower than what the CPU can push. Doesn't destroy it, but it's certainly a black mark. Are 1gig chipsets that much more expensive than 100mbps these days?

wtallis 12y ago

It's a chip for very low-end routers: 2.4GHz only, no gigabit switch (although apparently it can be equipped with two gigabit ports), and a CPU that would struggle to do 50Mbps WAN-LAN with firewall and QoS. It's only redeeming quality is that it apparently has a PCIe lane, but the CPU is still going to have trouble feeding whatever NIC you connect to that.

ajb 12y ago

I do, thanks.Hmm, let's see if I can give you the information without breaking a confidentiality agreement. Ah, here is a discussion on the openwrt mailing list: https://lists.openwrt.org/pipermail/openwrt-devel/2013-August/020963.html https://lists.openwrt.org/pipermail/openwrt-devel/2013-Augus... As you can see, they reported 90mb/s (they also quote 710, but I think that is the switch doing that). And that's on a mips74k. I know it's made for routers, but most likely not for doing a lot in software. As you quote, it does hardware NAT. There's a reason they put that in. There isn't a fundamental reason why the mips couldn't do the job, but running stock linux it's going to be thrashing its L1 cache, and it doesn't have an L2. Also, next time you play with one of those,check the hardware counters for how many instructions/sec it actually manages to process running flat out :-)

listic 12y ago
90Mb/s at least? Makes sense that they won't go for the gigabit Ethernet. Any idea how that compares to ARM with the same frequency? I'd say about the same. That's not bad, right?
- ajb 12y ago
  I would guess about the same; I didn't have two similar ones to compare. The mips might be a bit slower though. I had the sense that the limitation was really L1 thrashing of code space, so the ARM being a bit denser might help. It would also help if the SOC had a smaller latency to memory as well - don't know what this one has. Bear in mind that it's probably a packet rate rather than a bit rate limitation. That really bit us in our application because we were trying to run many lines of voip, which means lots of small packets. - YMMV.