It's going to be quite a while before "workstation" type laptops stop carrying an RJ45. That said... I can't remember the last time I was inconvenienced by the speed of gigabit ethernet on my work laptop. And pretty soon the wireless link on the laptop is going to be faster than gigabit anyway.
> And pretty soon the wireless link on the laptop is going to be faster than gigabit anyway.
Perhaps but right now we are much much much much much slower. I have never seen a wifi working at the advertised speed, even between two devices in the same room.
@thetuna: "I can't remember the last time I was inconvenienced by the speed of gigabit ethernet on my work laptop."
Agreed. For most use cases, the end user link isn't what really needs to be 10G. At this point, I'd like to have 10G for uplinks between switches and 10G to my file server. Gigabit isn't really a problem with a single person accessing it, but multiple people accessing separate volumes on the file server can be bottlenecked by a single gigabit link to the file server.
@thetuna: "And pretty soon the wireless link on the laptop is going to be faster than gigabit anyway."
I wouldn't count on it any time soon. Even if you take the so called 2000Mbps routers at face value, you still need four antenna's to pull it off. Laptops don't generally have that and carrying around a dongle and a set of four antenna's seems even more burdensome than carrying an ethernet cable.
Of course, that's not the only issue. These speeds assume unidirectional traffic. As soon as traffic needs to go the other way, the transmitter has to transition to receiver (and vice versa). This kills more than half of your theoretical bandwidth for a balanced load. These speeds also assume the use of four adjacent 20MHz channels to make their so called 80MHz channel. As soon as another devices starts transmitting in any of the designated channels, your speed tanks as it tries to retransmit and reorder data. Having other devices connected is a problem as well as even small amounts of traffic can cause backoffs. There are technologies that could allow multiple devices to transmit on different channels simultaneously without interfering with each other, but again, the loss of these channels results in the immediate loss of the bandwidth associated with those channels. Of course, you loose even more if you make the devices share the channels aggregated as you encounter more backoffs and transmit to receive transitions. So I caution you not to get your hopes up.
In most cases full duplex doesn't actually kill more than half the bandwidth. Most router test out at slightly higher full duplex speeds than their half duplex.
So a wifi router might hit something like 486Mbps in actual unidirectional speeds, but hit something like 500Mbps for bidirectional.
Since 80/160MHz is only used in 5GHz, there is rarely traffic that isn't your network on there due to 5GHz low penetration.
No, you don't hit real listed speeds, but in a supportive EMI environment close to the router, you can hit a reasonable fraction. My older Intel 7260ac with newer 802.11ac routers can easily net 76% payload through put on 2.4GHz, which is right at the theoretical limit (that missing 24% are acknowledgement and control packets, beaconing and forward error correction overhead).
On 5GHz, it isn't nearly as good, but it can still hit about 500Mbps with a tail wind, that is 57.6% of the signaling rate. I've seen wifi bridges between routers hit about 600Mbps for 2:2 802.11ac routers. That is 69% and probably also closer to where an even newer wifi client chipset can manage.
600Mbps doesn't "threaten" GbE, but it is still pretty fast and it was not obtained same room, that was through a wall and a bookcase. I'd imagine most newer 802.11ac 3:3 chipsets can hit a realistic 850-900Mbps or even somewhat higher. 4:4 802.11ac with 80MHz can absolutely saturate 1Gbps wired links (I've seen it in testing between AC2600 routers, I just don't have two to test with, just an AC1750 and an AC2600 and I've frankly never bothered to try out wireless bridges between them, just an AC1200 and AC1750 bridge, so 2:2 only).
We are inching up on 160MHz 802.11ac chipsets and 802.11ax isn't too far in the future, and that will likely see at least a real world increase of 50% in throughput with good EMI environment, maybe better.
Sooooo that said, we actually are probably just 2-4 years away from a gigabit Ethernet link slowing down network transfers with the latest and greatest wireless gear in good EMI environments (and I don't mean testing labs).
Still waiting on 2.5 and 5GbE gear. Been languishing in standards hell for the last couple of years with the promise of "coming really soon, we promise".
Schedule to debut 2019 and Huwai has obtained 10Gbps speeds in labs with pre-draft equipment. You'll need an even more ideal environment than you do now to see speeds like that, because of the very high modulation rate, but I would still expect to see real world 3-4Gbps speeds same room and 2+ next room over.
Never mind 160, are there even any 80MHz channel products on the market?
Personally, I'm hoping 80/160MHz channels are never implemented. There's only enough spectrum for 5/2 channels that large (with another 160 possible if we give up on a highspeed intercar communication network - wifi handshakes are too slow for opposing traffic on the highway). Enabling them would make the 5GHz band as clogged up with channel conflicts as the 2.4 GHz band is today.
Keep in mind how low penetrating 5GHz is. Sure, in an apartment situation, especially one with paper thing walls it would be a big issue...but in most typical apartment construction, let alone single family homes 5GHz conflict isn't an issue. In my rancher, I am getting <-90dB on the other side off the house from my basement router and vice versa with my upstairs access point. 2.4GHz on the other hand it is a low, but still viable -72dB or so. My house is not that big.
or 80MHz channel products? Errm...all 802.11ac is at least 80MHz. I fit is running 40MHz, then it is running at reduced rate (200Mbps I think it 40MHz 256QAM?)
> And pretty soon the wireless link on the laptop is going to be faster than gigabit anyway.
No, very likely this will never happen, at least not for realistic data rates in real world usage scenarios. While 802.11ad potentially allows for >1Gbs sustained data rate, the use of 60GHz effectively means you will need to have line-of-sight to the communication partner and better do not have people crossing the LoS. Getting >1Gbs on 802.11ac is only possible on 5GHz with 160Mhz wide channels (not to be shared with any neighbour) and a crazy number of antennas (4 on each side should do) and of course LoS; in that case it just might be possible to get a real and sustained >1Gbs out of the Wifi link. It's important to realise that advertised speed or "capacity" is nowhere near the real speed...
You are smoking some good stuff if you think wireless is every going to surpass wired. I've yet to see a Gigabit wifi that can even come close to a sustained transfer on wired 1G. The only time they can even hit the theoretical rate is at 15 inches seperation in a faraday cage. In real world conditions AC can barely exceed 100mbit wired and it frequently can't sustain it for longer than a few seconds. Even then it takes 3-6 antennas, multi channel and a low noise floor. Things that will never be possible in a crowded urban environment.
Maybe it will exceed it some day but honestly physics is against it ever happening. Wired will always be a step ahead. 10G copper will widely deploy when prices hit a threshold price just like 1G did. If you don't think 10G is going to deploy you aren't old enough to remember the transition from 10M to 100M to 1G. Prices just have to come down. The beauty of 10G over copper is that if your wire length is less than 30M you can probably already do it on Cat5e without a cable upgrade. Either way when switch prices come down to $100 for an 8 port you will see wide deployment of 10G copper.
> The beauty of 10G over copper is that if your wire length is less than 30M you can probably already do it on Cat5e without a cable upgrade.
No, that's almost impossible. A properly installed Cat.6 cabling might /just/ work... Cat.5 is only certified for 100MHz operation, Cat.6 for 250MHz (and hence might work for shorter distances), Cat.6a for 500MHz. Proper installation is a hard requirement, so either a point-to-point connection with a single ready-made cable or a properly done fixed installation is a hard requirement; if you don't have a measurement protocol for your installation chances are you do not apply for 10GBase-T.
Cat5e may not be certified for 10G - however for short distances it will work. (Just like an incorrect type of coax would work for a short thin Ethernet connection back in the early days of networking.) According to Intel's specs for their current 10Gbase-T NICs: - For cable distances up to about 55m Cat6 is acceptable (YMMV) - For cable distances up to about 20m Cat5e is acceptable (YMMV) - For cable distances up to about 3m, plain old Cat5 is acceptable (YMMV) (Info copied from https://hardforum.com/threads/cat-5e-vs-cat-6.1600... )
And with SSDs getting cheaper, they will make their way to NAS pretty quickly and then we actually are going to use 10GbE regularly.
I am not bullish on 100GbE though. That would be more data than a CPU can pump in and out and those aren't really going any faster. I don't see how we could use that. Not absurd to assume 10GbE is the last deployment at least for end points (I am not talking about inter router links).
Every so often I need to move a 60GiB file. Waiting 1min for a 10GB/s server is jarring to my work flow, but waiting 10 minutes will interrupt my work flow.
There is *literally* nothing productive that you can do for 10 minutes every once in a while? That sounds more like a time-management issue than a technology issue.
Once chipsets reach 22nm, we should start to see 10Gb on laptops. Intel went from 32nm to 28nm on their NICs and went from 10Gb to 40Gb, reduced the price 30%, and reduced power consumption 50%. That was only a 12% reduction in size.
Intel said that 22nm we should see 10Gb NICs in low end laptops, and 14nm we should see 40Gb. Major players are still too busy cranking out $500 CPUs, not $1 integrated NICs. Once 10nm makes it, we should see the 22nm manufacturing free up enough for low end parts to be made.
"That was only a 12% reduction in size." Isn't it more like a 23% reduction, since it is an area (32nm in one direction and the other, vs 28nm in one direction and the other) and not a line? :)
When pricing is at the point where manufacturers think they can sell laptops, or desktops, with this technology, and, for laptops, when the power requirements drop enough, we will begin to see it.
I remember back when Apple was the first to adopt 100Mbe for their entire lineup, and then later 1Gbe. That helped to kickstart Intel into supporting it on everything as well, back when they were still making about half of OEM motherboards.
WiGig comes in the newer premium Intel wireless chip-sets, and can hit about 7G over short-range. You may see a dock that support 10G given that, or a dedicated bridge.
I'm just curious, what is it that makes 10GbE so expensive and power hungry? Is something radically different being done versus 1 GbE? At a cost of >$100 per port, doesn't fiber optics start to make more sense? I can certainly see one barrier to entry with fiber being the requirement that you re-cable your business/home.
1Gb had the same issue until modern fabrication processes could be used. I remember 1Gb NICs with massive heatsinks and fans. Now they're integrated into a chipset and only consume 1mmx1mm, and passively cooled.
@3DoubleD: "I'm just curious, what is it that makes 10GbE so expensive and power hungry?"
The power consumption of transistors can be divided into two groups: Switching power, and leakage power. Leakage power is power loss when the transistor is holding a state. It is generally a much smaller quantity than switching power, but has become a much larger quantity as transistor sizes have shrunk. FinFETs can do a lot to combat this rising leakage. Switching power is the power spend to "switch" the transistor from one state to the other. This is generally the dominant power usage. Assuming the exact same operations, switching the transistors 10 times as often should result in larger part of an order of magnitude more power draw. Of course, that's all before you even get off of the silicon. Practicalities in copper quality, construction tolerances, electromagnetic interferences, etc. force the NICs to use more complex algorithms to reach these speeds, causing even more transistor switches.
The good news is that, like processors, NICs benefit from process size reduction. This lowers the switching threshold of the transistors requiring less supply voltage to operate. Lower supply voltages reduce power for both switching and leakage.
> I'm just curious, what is it that makes 10GbE so expensive and power hungry? Is something radically different being done versus 1 GbE?
Yes, the amount of signal processing required to fish a stable link out of the "noise" coming out of the cable is just insane. After all existing cable (qualified for 500MHz operating frequency) was supposed to be able to carry the signal so they had to crank up the signal encoding quite a bit to make that possible.
> At a cost of >$100 per port, doesn't fiber optics start to make more sense?
Yes, it does but who cares?
> I can certainly see one barrier to entry with fiber being the requirement that you re-cable your business/home.
For 10GBase-T you'll at least need properly(!) installed Cat.6 (short distance only) or Cat.6a cabling anyway and who does already happen to have that anyway?
Spec is 45 meters for 5e, 55 meters for 6 and 100 meters for 6a.
The issue of course is that in a high cross talk or outside EMI environment, that 5e and 6 will fall apart at much shorter distances than 6a will. So if you have a bundle of cables, 5e might only realistically support 10GbE at 10-15 meters, but under ideal conditions can carry it to 45 meters.
So your average power user at home can probably use old 5e or 6 cables in their walls with no problems, but in a data center, it would fall apart.
2.5 and 5GbE are supposed to support Cat5e to 100 meters, which is one of the reasons it is being pushed towards a ratified standard, along with the lower port cost and lower power dissipation. There isn't any real need in consumer market for 10GbE ANY year soon, other than we need something faster than 1GbE in the next couple of years and in the SMB market, most don't really need 10GbE. However, having 2.5GbE or even 5GbE and suddenly you have a product to address next generation wireless networking backhauls for consumers as well as SMB's that need >1GbE, but don't need 10GbE. All in a smaller, less expensive package.
> No, actually Cat5e supports 10GbE. > Spec is 45 meters for 5e, 55 meters for 6 and 100 meters for 6a.
I unfortunately don't have access to IEEE documents but all other sources of information disagree strongly with what you're claiming, let's take e.g. Wikipedia for example (https://en.wikipedia.org/wiki/10_Gigabit_Ethernet)...
"10GBASE-T, or IEEE 802.3an-2006, is a standard released in 2006 to provide 10 Gbit/s connections over unshielded or shielded twisted pair cables, over distances up to 100 metres (330 ft).[35] Category 6a is required to reach the full distance of 100 metres (330 ft) and category 6 may reach a distance of 55 metres (180 ft) depending on the quality of installation, determined only after re-testing to 500 MHz. "
You just validated two of his 3 numbers. Cat 5e isn't listed as compatible but in a low noise environment it's capable of 10G at shorter distances. The number I remember is 30m not 45m. The difference between Cat 5e and Cat 6 is just the number of twists per inch. I doubt anyones tested it but even Cat 5 might work at 10Gbe over a foot or two. The better the cable the further it goes. If you go up to shielded Cat7 you can get (IIRC) 200m at 10G.
Technically if a cable can't do 100M it doesn't meet the specification so even Cat6 isn't "approved" for 10G because it can only do 55m.
As far as I am aware Cat.5e is not even specified for use with 10GBase-T and I don't think the onus is on me to prove that it is.
> Technically if a cable can't do 100M it doesn't meet the specification so even Cat6 isn't "approved" for 10G because it can only do 55m.
Well, I'm not sure at all whether 10GBase-T is actually specified to work on Cat.6 (at a lower maximum length) at all but what I do know is that is it specified to work with Cat.6a. And to ensure this is is exactly why performance specifications by TIA and ISO exist which need to be met (and proven with a test protocol). If an installation (meaning *all* of it) does not meet the requirements for Cat.6a you're not allowed to call it a Cat.6a cabling, in that case 10GBase-T might work, or might not work but who cares... my whole point is: Despite contrary claims 10GBase-T has very stringent requirements -- which I know for a fact many installations will not fulfil -- and thus in many cases it will not be the drop-in replacement many are hoping for. That's a big difference to the introduction of 1000Base-T where people simply needed to swap NICs and switches and everything was go...
NB: A real Cat.7 installation is undesirable to have in many cases because you MAY NOT use the common RJ45 sockets and connectors anymore and thus it can be a very expensive experience with little gain.
Has anyone here got real-world experience of a 10GBase-T PCI card that isn't an Intel X540T1/2? I'd like to find something cheaper for our workstations.
There are piles of older models that'll suffice if you don't mind 20W+ TDP and little screamer fans. The X540 series is among the few that's cooled passively, though I also have an Intel AT2 card here from 2010 that's the same.
What distances? There are a number of GBIC cards out there. If you just need 10GbE cheap and low power, 10GbE Twinax GBIC modules, plus 10 meter cables is about $130 for two modules and the cable, and you can get GBIC 10Gbps PCI-e cards for like $30-40 each for the modules to go in.
Perfect for rack to rack or small distance work areas. Twinax 10GbE also only consumes about 1-1.5w a port, instead of about 4-5w a port of the newest and best 10GBASE-T stuff.
> Perfect for rack to rack or small distance work areas. Twinax 10GbE also only consumes about 1-1.5w a port, instead of about 4-5w a port of the newest and best 10GBASE-T stuff.
I've seen X520's with SFP+ (no optics included) go for $230. No fan, but it probably requires case fans blowing over it. Then buy some cheap SFP+ twinax cables and you're good to go.
Those prices aren't really great at the high end. I've just looked up some Cisco quotes from our sales guy, and after discounts it's less than $7000 for a Catalyst 3850 4x10 SFP, with 36 multi-gig POE ports, including redundant power supplies. And it's modular so you can swap things or stack them.
Granted we buy in bulk so we're getting a good deal.
Call me when you release a hardened/industrial model with +12 to +48VDC inputs, all SFP+ 10Gbps ports and a pair of 40Gbps uplinks that supports Ethernet rings (G.8032).
Very small niche. It is probably useful for a highly dense office floor with high bandwidth requirements for media workstation exchanging video files to one another.
At $1600 this 8-port switch is the same price as a Mellanox QDR Infiniband switch; if you're building a cluster you would be daft not to use Infiniband. $100 more for a QDR HCA versus the cheapest Intel 10GbaseT network card, but four times the speed!
Just looking at those fan cut-outs on the left hand side makes we want to cover my ears!
I've used cross wired dual 10Gbit Ethernet NICs to interconnect VMware ESX systems running on high-end PCs to play around with vMotion and VSAN and I can certainly see the benefit of a desktop 10Gbit switch in my home-office.
But only if it is no louder than the PCs themselves, which run nearly silent on idle.
It can't be all that difficult to put a nice big PC cooler on those merchant silicon chips and a few more holes into the chassis to support desktop use at least on the entry model.
I even considered a DIY when somebody did a teardown of the previous model and it seems straightforward enough. But since it's four digits even for the smallest model, hesitation prevailed so far.
At work I remember doing an inital setup of some 48port 10Gbit HP switches in my office: When I started that switch, it was like an airplane starting up on my table.
At the price they are asking, I expect them to do better than lots of noisy small fans and nothing in terms of heat conduction, when even a €250 notebook has heat-pipes and thermal management.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
46 Comments
Back to Article
cm2187 - Tuesday, May 17, 2016 - link
When are we going to see laptops with 10GbE ports? It's nice to have a 10GbE infrastructure but not very helpful if we cannot consume it...DanNeely - Tuesday, May 17, 2016 - link
Probably never unless you count USB-C dongles. The thinner is better koolaid has murdered the RJ45 port.thetuna - Tuesday, May 17, 2016 - link
It's going to be quite a while before "workstation" type laptops stop carrying an RJ45.That said... I can't remember the last time I was inconvenienced by the speed of gigabit ethernet on my work laptop.
And pretty soon the wireless link on the laptop is going to be faster than gigabit anyway.
cm2187 - Tuesday, May 17, 2016 - link
> And pretty soon the wireless link on the laptop is going to be faster than gigabit anyway.Perhaps but right now we are much much much much much slower. I have never seen a wifi working at the advertised speed, even between two devices in the same room.
BurntMyBacon - Tuesday, May 17, 2016 - link
@thetuna: "I can't remember the last time I was inconvenienced by the speed of gigabit ethernet on my work laptop."Agreed. For most use cases, the end user link isn't what really needs to be 10G. At this point, I'd like to have 10G for uplinks between switches and 10G to my file server. Gigabit isn't really a problem with a single person accessing it, but multiple people accessing separate volumes on the file server can be bottlenecked by a single gigabit link to the file server.
@thetuna: "And pretty soon the wireless link on the laptop is going to be faster than gigabit anyway."
I wouldn't count on it any time soon. Even if you take the so called 2000Mbps routers at face value, you still need four antenna's to pull it off. Laptops don't generally have that and carrying around a dongle and a set of four antenna's seems even more burdensome than carrying an ethernet cable.
Of course, that's not the only issue. These speeds assume unidirectional traffic. As soon as traffic needs to go the other way, the transmitter has to transition to receiver (and vice versa). This kills more than half of your theoretical bandwidth for a balanced load. These speeds also assume the use of four adjacent 20MHz channels to make their so called 80MHz channel. As soon as another devices starts transmitting in any of the designated channels, your speed tanks as it tries to retransmit and reorder data. Having other devices connected is a problem as well as even small amounts of traffic can cause backoffs. There are technologies that could allow multiple devices to transmit on different channels simultaneously without interfering with each other, but again, the loss of these channels results in the immediate loss of the bandwidth associated with those channels. Of course, you loose even more if you make the devices share the channels aggregated as you encounter more backoffs and transmit to receive transitions. So I caution you not to get your hopes up.
azazel1024 - Tuesday, May 17, 2016 - link
In most cases full duplex doesn't actually kill more than half the bandwidth. Most router test out at slightly higher full duplex speeds than their half duplex.So a wifi router might hit something like 486Mbps in actual unidirectional speeds, but hit something like 500Mbps for bidirectional.
Since 80/160MHz is only used in 5GHz, there is rarely traffic that isn't your network on there due to 5GHz low penetration.
No, you don't hit real listed speeds, but in a supportive EMI environment close to the router, you can hit a reasonable fraction. My older Intel 7260ac with newer 802.11ac routers can easily net 76% payload through put on 2.4GHz, which is right at the theoretical limit (that missing 24% are acknowledgement and control packets, beaconing and forward error correction overhead).
On 5GHz, it isn't nearly as good, but it can still hit about 500Mbps with a tail wind, that is 57.6% of the signaling rate. I've seen wifi bridges between routers hit about 600Mbps for 2:2 802.11ac routers. That is 69% and probably also closer to where an even newer wifi client chipset can manage.
600Mbps doesn't "threaten" GbE, but it is still pretty fast and it was not obtained same room, that was through a wall and a bookcase. I'd imagine most newer 802.11ac 3:3 chipsets can hit a realistic 850-900Mbps or even somewhat higher. 4:4 802.11ac with 80MHz can absolutely saturate 1Gbps wired links (I've seen it in testing between AC2600 routers, I just don't have two to test with, just an AC1750 and an AC2600 and I've frankly never bothered to try out wireless bridges between them, just an AC1200 and AC1750 bridge, so 2:2 only).
We are inching up on 160MHz 802.11ac chipsets and 802.11ax isn't too far in the future, and that will likely see at least a real world increase of 50% in throughput with good EMI environment, maybe better.
Sooooo that said, we actually are probably just 2-4 years away from a gigabit Ethernet link slowing down network transfers with the latest and greatest wireless gear in good EMI environments (and I don't mean testing labs).
Still waiting on 2.5 and 5GbE gear. Been languishing in standards hell for the last couple of years with the promise of "coming really soon, we promise".
802.11ax https://en.wikipedia.org/wiki/IEEE_802.11ax
Schedule to debut 2019 and Huwai has obtained 10Gbps speeds in labs with pre-draft equipment. You'll need an even more ideal environment than you do now to see speeds like that, because of the very high modulation rate, but I would still expect to see real world 3-4Gbps speeds same room and 2+ next room over.
DanNeely - Tuesday, May 17, 2016 - link
Never mind 160, are there even any 80MHz channel products on the market?Personally, I'm hoping 80/160MHz channels are never implemented. There's only enough spectrum for 5/2 channels that large (with another 160 possible if we give up on a highspeed intercar communication network - wifi handshakes are too slow for opposing traffic on the highway). Enabling them would make the 5GHz band as clogged up with channel conflicts as the 2.4 GHz band is today.
azazel1024 - Friday, May 20, 2016 - link
Keep in mind how low penetrating 5GHz is. Sure, in an apartment situation, especially one with paper thing walls it would be a big issue...but in most typical apartment construction, let alone single family homes 5GHz conflict isn't an issue. In my rancher, I am getting <-90dB on the other side off the house from my basement router and vice versa with my upstairs access point. 2.4GHz on the other hand it is a low, but still viable -72dB or so. My house is not that big.or 80MHz channel products? Errm...all 802.11ac is at least 80MHz. I fit is running 40MHz, then it is running at reduced rate (200Mbps I think it 40MHz 256QAM?)
Daniel Egger - Tuesday, May 17, 2016 - link
> And pretty soon the wireless link on the laptop is going to be faster than gigabit anyway.No, very likely this will never happen, at least not for realistic data rates in real world usage scenarios. While 802.11ad potentially allows for >1Gbs sustained data rate, the use of 60GHz effectively means you will need to have line-of-sight to the communication partner and better do not have people crossing the LoS. Getting >1Gbs on 802.11ac is only possible on 5GHz with 160Mhz wide channels (not to be shared with any neighbour) and a crazy number of antennas (4 on each side should do) and of course LoS; in that case it just might be possible to get a real and sustained >1Gbs out of the Wifi link. It's important to realise that advertised speed or "capacity" is nowhere near the real speed...
rahvin - Tuesday, May 17, 2016 - link
You are smoking some good stuff if you think wireless is every going to surpass wired. I've yet to see a Gigabit wifi that can even come close to a sustained transfer on wired 1G. The only time they can even hit the theoretical rate is at 15 inches seperation in a faraday cage. In real world conditions AC can barely exceed 100mbit wired and it frequently can't sustain it for longer than a few seconds. Even then it takes 3-6 antennas, multi channel and a low noise floor. Things that will never be possible in a crowded urban environment.Maybe it will exceed it some day but honestly physics is against it ever happening. Wired will always be a step ahead. 10G copper will widely deploy when prices hit a threshold price just like 1G did. If you don't think 10G is going to deploy you aren't old enough to remember the transition from 10M to 100M to 1G. Prices just have to come down. The beauty of 10G over copper is that if your wire length is less than 30M you can probably already do it on Cat5e without a cable upgrade. Either way when switch prices come down to $100 for an 8 port you will see wide deployment of 10G copper.
Daniel Egger - Tuesday, May 17, 2016 - link
> The beauty of 10G over copper is that if your wire length is less than 30M you can probably already do it on Cat5e without a cable upgrade.No, that's almost impossible. A properly installed Cat.6 cabling might /just/ work... Cat.5 is only certified for 100MHz operation, Cat.6 for 250MHz (and hence might work for shorter distances), Cat.6a for 500MHz. Proper installation is a hard requirement, so either a point-to-point connection with a single ready-made cable or a properly done fixed installation is a hard requirement; if you don't have a measurement protocol for your installation chances are you do not apply for 10GBase-T.
Duncan Macdonald - Wednesday, May 18, 2016 - link
Cat5e may not be certified for 10G - however for short distances it will work. (Just like an incorrect type of coax would work for a short thin Ethernet connection back in the early days of networking.)According to Intel's specs for their current 10Gbase-T NICs:
- For cable distances up to about 55m Cat6 is acceptable (YMMV)
- For cable distances up to about 20m Cat5e is acceptable (YMMV)
- For cable distances up to about 3m, plain old Cat5 is acceptable (YMMV)
(Info copied from https://hardforum.com/threads/cat-5e-vs-cat-6.1600... )
coastinthefog - Thursday, May 19, 2016 - link
I have the 1st gen 8 port version of one of these. It's running at 10G over cat5e without a problem. Relatively short run probably less than 50 feet.cm2187 - Tuesday, May 17, 2016 - link
And with SSDs getting cheaper, they will make their way to NAS pretty quickly and then we actually are going to use 10GbE regularly.I am not bullish on 100GbE though. That would be more data than a CPU can pump in and out and those aren't really going any faster. I don't see how we could use that. Not absurd to assume 10GbE is the last deployment at least for end points (I am not talking about inter router links).
bcronce - Tuesday, May 17, 2016 - link
Every so often I need to move a 60GiB file. Waiting 1min for a 10GB/s server is jarring to my work flow, but waiting 10 minutes will interrupt my work flow.SirNuke - Wednesday, May 18, 2016 - link
There is *literally* nothing productive that you can do for 10 minutes every once in a while? That sounds more like a time-management issue than a technology issue.bcronce - Tuesday, May 17, 2016 - link
Once chipsets reach 22nm, we should start to see 10Gb on laptops. Intel went from 32nm to 28nm on their NICs and went from 10Gb to 40Gb, reduced the price 30%, and reduced power consumption 50%. That was only a 12% reduction in size.Intel said that 22nm we should see 10Gb NICs in low end laptops, and 14nm we should see 40Gb. Major players are still too busy cranking out $500 CPUs, not $1 integrated NICs. Once 10nm makes it, we should see the 22nm manufacturing free up enough for low end parts to be made.
Death666Angel - Tuesday, May 17, 2016 - link
"That was only a 12% reduction in size."Isn't it more like a 23% reduction, since it is an area (32nm in one direction and the other, vs 28nm in one direction and the other) and not a line? :)
bcronce - Tuesday, May 17, 2016 - link
Correct, but I didn't want to dive into all of that, but it is a common mistake for many to make when talking about chip sizes.damianrobertjones - Tuesday, May 17, 2016 - link
I do not eat data.melgross - Tuesday, May 17, 2016 - link
When pricing is at the point where manufacturers think they can sell laptops, or desktops, with this technology, and, for laptops, when the power requirements drop enough, we will begin to see it.I remember back when Apple was the first to adopt 100Mbe for their entire lineup, and then later 1Gbe. That helped to kickstart Intel into supporting it on everything as well, back when they were still making about half of OEM motherboards.
WorBlux - Thursday, March 9, 2017 - link
WiGig comes in the newer premium Intel wireless chip-sets, and can hit about 7G over short-range. You may see a dock that support 10G given that, or a dedicated bridge.3DoubleD - Tuesday, May 17, 2016 - link
I'm just curious, what is it that makes 10GbE so expensive and power hungry? Is something radically different being done versus 1 GbE? At a cost of >$100 per port, doesn't fiber optics start to make more sense? I can certainly see one barrier to entry with fiber being the requirement that you re-cable your business/home.cm2187 - Tuesday, May 17, 2016 - link
My guess is that you are operating at speed close to the bandwidth of modern RAM, and that's per port.bcronce - Tuesday, May 17, 2016 - link
1Gb had the same issue until modern fabrication processes could be used. I remember 1Gb NICs with massive heatsinks and fans. Now they're integrated into a chipset and only consume 1mmx1mm, and passively cooled.22nm chipsets is when 10Gb will become a reality.
BurntMyBacon - Tuesday, May 17, 2016 - link
@3DoubleD: "I'm just curious, what is it that makes 10GbE so expensive and power hungry?"The power consumption of transistors can be divided into two groups: Switching power, and leakage power. Leakage power is power loss when the transistor is holding a state. It is generally a much smaller quantity than switching power, but has become a much larger quantity as transistor sizes have shrunk. FinFETs can do a lot to combat this rising leakage. Switching power is the power spend to "switch" the transistor from one state to the other. This is generally the dominant power usage. Assuming the exact same operations, switching the transistors 10 times as often should result in larger part of an order of magnitude more power draw. Of course, that's all before you even get off of the silicon. Practicalities in copper quality, construction tolerances, electromagnetic interferences, etc. force the NICs to use more complex algorithms to reach these speeds, causing even more transistor switches.
The good news is that, like processors, NICs benefit from process size reduction. This lowers the switching threshold of the transistors requiring less supply voltage to operate. Lower supply voltages reduce power for both switching and leakage.
Daniel Egger - Tuesday, May 17, 2016 - link
> I'm just curious, what is it that makes 10GbE so expensive and power hungry? Is something radically different being done versus 1 GbE?Yes, the amount of signal processing required to fish a stable link out of the "noise" coming out of the cable is just insane. After all existing cable (qualified for 500MHz operating frequency) was supposed to be able to carry the signal so they had to crank up the signal encoding quite a bit to make that possible.
> At a cost of >$100 per port, doesn't fiber optics start to make more sense?
Yes, it does but who cares?
> I can certainly see one barrier to entry with fiber being the requirement that you re-cable your business/home.
For 10GBase-T you'll at least need properly(!) installed Cat.6 (short distance only) or Cat.6a cabling anyway and who does already happen to have that anyway?
azazel1024 - Tuesday, May 17, 2016 - link
No, actually Cat5e supports 10GbE.Spec is 45 meters for 5e, 55 meters for 6 and 100 meters for 6a.
The issue of course is that in a high cross talk or outside EMI environment, that 5e and 6 will fall apart at much shorter distances than 6a will. So if you have a bundle of cables, 5e might only realistically support 10GbE at 10-15 meters, but under ideal conditions can carry it to 45 meters.
So your average power user at home can probably use old 5e or 6 cables in their walls with no problems, but in a data center, it would fall apart.
2.5 and 5GbE are supposed to support Cat5e to 100 meters, which is one of the reasons it is being pushed towards a ratified standard, along with the lower port cost and lower power dissipation. There isn't any real need in consumer market for 10GbE ANY year soon, other than we need something faster than 1GbE in the next couple of years and in the SMB market, most don't really need 10GbE. However, having 2.5GbE or even 5GbE and suddenly you have a product to address next generation wireless networking backhauls for consumers as well as SMB's that need >1GbE, but don't need 10GbE. All in a smaller, less expensive package.
Daniel Egger - Tuesday, May 17, 2016 - link
> No, actually Cat5e supports 10GbE.> Spec is 45 meters for 5e, 55 meters for 6 and 100 meters for 6a.
I unfortunately don't have access to IEEE documents but all other sources of information disagree strongly with what you're claiming, let's take e.g. Wikipedia for example (https://en.wikipedia.org/wiki/10_Gigabit_Ethernet)...
"10GBASE-T, or IEEE 802.3an-2006, is a standard released in 2006 to provide 10 Gbit/s connections over unshielded or shielded twisted pair cables, over distances up to 100 metres (330 ft).[35] Category 6a is required to reach the full distance of 100 metres (330 ft) and category 6 may reach a distance of 55 metres (180 ft) depending on the quality of installation, determined only after re-testing to 500 MHz. "
rahvin - Tuesday, May 17, 2016 - link
You just validated two of his 3 numbers. Cat 5e isn't listed as compatible but in a low noise environment it's capable of 10G at shorter distances. The number I remember is 30m not 45m. The difference between Cat 5e and Cat 6 is just the number of twists per inch. I doubt anyones tested it but even Cat 5 might work at 10Gbe over a foot or two. The better the cable the further it goes. If you go up to shielded Cat7 you can get (IIRC) 200m at 10G.Technically if a cable can't do 100M it doesn't meet the specification so even Cat6 isn't "approved" for 10G because it can only do 55m.
Daniel Egger - Tuesday, May 17, 2016 - link
> You just validated two of his 3 numbers.As far as I am aware Cat.5e is not even specified for use with 10GBase-T and I don't think the onus is on me to prove that it is.
> Technically if a cable can't do 100M it doesn't meet the specification so even Cat6 isn't "approved" for 10G because it can only do 55m.
Well, I'm not sure at all whether 10GBase-T is actually specified to work on Cat.6 (at a lower maximum length) at all but what I do know is that is it specified to work with Cat.6a. And to ensure this is is exactly why performance specifications by TIA and ISO exist which need to be met (and proven with a test protocol). If an installation (meaning *all* of it) does not meet the requirements for Cat.6a you're not allowed to call it a Cat.6a cabling, in that case 10GBase-T might work, or might not work but who cares... my whole point is: Despite contrary claims 10GBase-T has very stringent requirements -- which I know for a fact many installations will not fulfil -- and thus in many cases it will not be the drop-in replacement many are hoping for. That's a big difference to the introduction of 1000Base-T where people simply needed to swap NICs and switches and everything was go...
NB: A real Cat.7 installation is undesirable to have in many cases because you MAY NOT use the common RJ45 sockets and connectors anymore and thus it can be a very expensive experience with little gain.
coastinthefog - Thursday, May 19, 2016 - link
I have 10G working at home on cat5e wired through my walls. Less than 50 ft run.colonelclaw - Tuesday, May 17, 2016 - link
Has anyone here got real-world experience of a 10GBase-T PCI card that isn't an Intel X540T1/2? I'd like to find something cheaper for our workstations.alexdi - Tuesday, May 17, 2016 - link
There are piles of older models that'll suffice if you don't mind 20W+ TDP and little screamer fans. The X540 series is among the few that's cooled passively, though I also have an Intel AT2 card here from 2010 that's the same.azazel1024 - Tuesday, May 17, 2016 - link
What distances? There are a number of GBIC cards out there. If you just need 10GbE cheap and low power, 10GbE Twinax GBIC modules, plus 10 meter cables is about $130 for two modules and the cable, and you can get GBIC 10Gbps PCI-e cards for like $30-40 each for the modules to go in.Perfect for rack to rack or small distance work areas. Twinax 10GbE also only consumes about 1-1.5w a port, instead of about 4-5w a port of the newest and best 10GBASE-T stuff.
Just a thought.
azazel1024 - Tuesday, May 17, 2016 - link
To clarify, the downside is, new cables and the maximum supported distance of Twinax 10GbE is 10 meters.Daniel Egger - Tuesday, May 17, 2016 - link
> Perfect for rack to rack or small distance work areas. Twinax 10GbE also only consumes about 1-1.5w a port, instead of about 4-5w a port of the newest and best 10GBASE-T stuff.> Just a thought.
Or use fibre instead... just a thought...
hechacker1 - Tuesday, May 17, 2016 - link
I've seen X520's with SFP+ (no optics included) go for $230. No fan, but it probably requires case fans blowing over it. Then buy some cheap SFP+ twinax cables and you're good to go.coastinthefog - Thursday, May 19, 2016 - link
I'm using one by startech. Works fine.http://www.amazon.com/StarTech-com-Express-Gigabit...
hechacker1 - Tuesday, May 17, 2016 - link
Those prices aren't really great at the high end. I've just looked up some Cisco quotes from our sales guy, and after discounts it's less than $7000 for a Catalyst 3850 4x10 SFP, with 36 multi-gig POE ports, including redundant power supplies. And it's modular so you can swap things or stack them.Granted we buy in bulk so we're getting a good deal.
hechacker1 - Tuesday, May 17, 2016 - link
Nevermind, I didn't read it correctly that these are all 10Gb ethernet. So it would be more apt to compare them to Nexus switches, which do cost more.Lucky Stripes 99 - Tuesday, May 17, 2016 - link
Call me when you release a hardened/industrial model with +12 to +48VDC inputs, all SFP+ 10Gbps ports and a pair of 40Gbps uplinks that supports Ethernet rings (G.8032).zodiacfml - Wednesday, May 18, 2016 - link
Very small niche. It is probably useful for a highly dense office floor with high bandwidth requirements for media workstation exchanging video files to one another.TomWomack - Wednesday, May 18, 2016 - link
At $1600 this 8-port switch is the same price as a Mellanox QDR Infiniband switch; if you're building a cluster you would be daft not to use Infiniband. $100 more for a QDR HCA versus the cheapest Intel 10GbaseT network card, but four times the speed!zodiacfml - Wednesday, May 18, 2016 - link
Right. This Netgear is pricey.abufrejoval - Wednesday, May 18, 2016 - link
Just looking at those fan cut-outs on the left hand side makes we want to cover my ears!I've used cross wired dual 10Gbit Ethernet NICs to interconnect VMware ESX systems running on high-end PCs to play around with vMotion and VSAN and I can certainly see the benefit of a desktop 10Gbit switch in my home-office.
But only if it is no louder than the PCs themselves, which run nearly silent on idle.
It can't be all that difficult to put a nice big PC cooler on those merchant silicon chips and a few more holes into the chassis to support desktop use at least on the entry model.
I even considered a DIY when somebody did a teardown of the previous model and it seems straightforward enough. But since it's four digits even for the smallest model, hesitation prevailed so far.
At work I remember doing an inital setup of some 48port 10Gbit HP switches in my office: When I started that switch, it was like an airplane starting up on my table.
At the price they are asking, I expect them to do better than lots of noisy small fans and nothing in terms of heat conduction, when even a €250 notebook has heat-pipes and thermal management.