Will A New AC Router Bring You Wi-Fi Joy?

[System Error Message]

i think you mis understand why it would be useful , say you have multiple transfers from a teamed nas or multiple streams the benefit of having twice as much supply is akin to tri band routers in that there is more supply available so each device will get more , think of it this way , 1 comp can transfer at 110MB/s from a nas to a giga ethernet connected pc , if 2 try and transfer at the same time they halve that eg 55MB/s each when used with a normal router without teaming

with teaming they share twice as much so 2 comps can both transfer at 110MB/s , so sure there are benefits to teaming a nas to a router

It depends on how wifi is connected to wire. If it goes through CPU than there is no point in teaming because the link between switch chip is 1Gb/s (2Gb/s has to go through 1Gb/s to go to 2x5Ghz). On the ASUS AC88U ports 1-4 is connected to ports 5-8 via a 1Gb/s link so teaming is still useless in this case if you have multiple wired clients.

 

[RMerlin]

On the ASUS AC3200 the wifi is connected to switch

Got any block diagram that confirms this? Because according to Broadcom, the BCM43602 has a PCI-Express interface. That means it's connected exactly the same way as the BCM4366 on the RT-AC88U - through the CPU's PCI-express interface.

 

[RMerlin]

Also, AFAIK, all 5 ports are on the CPU. The BCM4709 and BCM47094 do not use an external switch. So, Wifi should have full PCI-Express access to all five ports. The only exception to this are the four Realtek ports of the RT-AC88U, which are linked to the rest over RGMII.

EDIT: added links to BCM material.

 

[Nullity]

The advantage of not being tied to the switch means the WiFi can exceed Gigabit Ethernet speeds?

 

[scoopnog]

In the article, the R8500 records a 72% increase in downlink throughput over the R7000 baseline (on CH44) while the RT-AC3200 records a 4% improvement. I might be missing something here, but this difference in performance doesn't seem reflected in the recent review for the R8500 via the test chamber where they seem to share similar results.

Any thoughts on the discrepancy?

 

[System Error Message]

Also, AFAIK, all 5 ports are on the CPU. The BCM4709 and BCM47094 do not use an external switch. So, Wifi should have full PCI-Express access to all five ports. The only exception to this are the four Realtek ports of the RT-AC88U, which are linked to the rest over RGMII.

EDIT: added links to BCM material.

According to openwrt wiki the AC87U has a broadcom switch with 1 of the ports connected to CPU. https://wiki.openwrt.org/toh/asus/rt-ac87u It says it has an 8 port switch with 5 physical ports (1 to CPU).

 

[RMerlin]

According to openwrt wiki the AC87U has a broadcom switch with 1 of the ports connected to CPU. https://wiki.openwrt.org/toh/asus/rt-ac87u It says it has an 8 port switch with 5 physical ports (1 to CPU).

Has anyone actually spotted the switch IC on the pcboard? It would be odd for them to do that, since there's already a switch in the bcm4709 itself, as per Broadcom's specifications.

 

[thiggins]

In the article, the R8500 records a 72% increase in downlink throughput over the R7000 baseline (on CH44) while the RT-AC3200 records a 4% improvement. I might be missing something here, but this difference in performance doesn't seem reflected in the recent review for the R8500 via the test chamber where they seem to share similar results.
Any thoughts on the discrepancy?

Completely apples and oranges comparison.

Charts results use Channel 153 and AC1900 class STA.

You're referencing Channel 44 results, using an a completely different (N600) STA.

Channel 153 results are 23% LOSS for R8500 and 18% LOSS for the RT-AC3200

 

[System Error Message]

Has anyone actually spotted the switch IC on the pcboard? It would be odd for them to do that, since there's already a switch in the bcm4709 itself, as per Broadcom's specifications.

but within the bcm4709 itself the switch connects to the CPU with 1Gb/s

 

[RMerlin]

but within the bcm4709 itself the switch connects to the CPU with 1Gb/s

If that's true, then it means that even the Quantenna design was exactly identical as to every single Broadcom design of the past decade - 5GHz wifi having to go through a 1 Gbps bottleneck to reach the rest of the LAN. So, this can't explain any performance difference when moving from an N to an AC router (or moving to a newer AC router).

 

[scoopnog]

Completely apples and oranges comparison.

Charts results use Channel 153 and AC1900 class STA.

You're referencing Channel 44 results, using an a completely different (N600) STA.

Channel 153 results are 23% LOSS for R8500 and 18% LOSS for the RT-AC3200

http://www.smallnetbuilder.com/wire...8-smart-wifi-router-reviewed?showall=&start=3

On this page it mentions:

"For 5 GHz, I made an exception to our standard practice of using Channel 153. Since the 5 GHz-1 (low band) radio is connected to the external amplified antennas, I used it, set to its Channel 44 default. In fact, I initially tested using Channel 153, then retested using Channel 44 after discovering the low-band radio's external antenna connection."

This suggests the chart on the linked page for the R8500 represents the use of Channel 44.

 

[System Error Message]

If that's true, then it means that even the Quantenna design was exactly identical as to every single Broadcom design of the past decade - 5GHz wifi having to go through a 1 Gbps bottleneck to reach the rest of the LAN. So, this can't explain any performance difference when moving from an N to an AC router (or moving to a newer AC router).

Seeing as i have been able to get up to 90% wifi utilisation and that i have an AC3200 i could test this using a few wifi clients and some wired clients. I cant seem to find a bridge option for WAN.

 

[pete y testing]

Channel 153 results are 23% LOSS for R8500 and 18% LOSS for the RT-AC3200

hi tim

i think suggesting its a loss is maybe a confusing way of putting it and may confuse some folks , the results would be better referred to as less that or more that the benchmark , you prob would have been better of using the lowest performer as the benchmark to then show everything else as a gain in comparison as you said in the article there is a lot of data there to consume but understanding the reference point and why you use it as the base line in the first point would help

you could have used any of the routers tested as the reference lime and ended up with different looking charts but would also be exactly the same results for those that know how to read them but for those that dont they way you have gone about the whole test procedure could end up being quite confusing

pete

 

[htismaqe]

The problem is that he's trying to show whether or not upgrading to MU-MIMO or Tri-band offers significant improvements over AC1900. The only way the data makes sense is to use the AC1900 as the reference point and show all other routers as above or below that line, which is precisely what he did.

 

[pete y testing]

hi

as i said for those that know what they are looking at its pretty easy to read but referring to it as a loss give the wrong impression , its not so much a loss but more the fact its not as strong , having a mass of data makes ots always hard to give a fair and accurate representation

i would also like to see time use samba and transfer from devices within the network and graph the same using the average transfer time as the basis , as this would be a better real world representation that other owners can expect on their devices as i dont think the use of icharriot and or ipref test under the same protocols as samba , its a bit like testing using ftp as this will give different results as it uses different protocols , seeing as most of us transfer over out lan / wlan using the samba / sma protocols its a better representation of what end users can expect

pete

 

[Deleted member 22229]

The problem is that he's trying to show whether or not upgrading to MU-MIMO or Tri-band offers significant improvements over AC1900. The only way the data makes sense is to use the AC1900 as the reference point and show all other routers as above or below that line, which is precisely what he did.

Down the near road there is no question these new routers are going to out preform the older 1900 class routers it's just a matter of firmware the hardware is already there much faster processer and much more memory. Once the firmware is nailed down people wont be buying 3 year old routers thinking they are better then the new tech routers. And normally most vendors stop caring about routers after about three years and support gets less and less or stopped all together. I must say though Asus is better at long term support then most. Would i buy a bran new 68U in very late 2015 NO it just dont make since. Just my opinion. Not to mention the AC-3100 and 88U work great and will only get better. So for those that have the means then go for it.

 

[sfx2000]

If that's true, then it means that even the Quantenna design was exactly identical as to every single Broadcom design of the past decade - 5GHz wifi having to go through a 1 Gbps bottleneck to reach the rest of the LAN. So, this can't explain any performance difference when moving from an N to an AC router (or moving to a newer AC router).

SEM is correct - the internal Ethernet switch fabric that BRCM uses is 1GB full duplex, not a big deal...

 

[thiggins]

http://www.smallnetbuilder.com/wire...8-smart-wifi-router-reviewed?showall=&start=3

On this page it mentions:

"For 5 GHz, I made an exception to our standard practice of using Channel 153. Since the 5 GHz-1 (low band) radio is connected to the external amplified antennas, I used it, set to its Channel 44 default. In fact, I initially tested using Channel 153, then retested using Channel 44 after discovering the low-band radio's external antenna connection."

This suggests the chart on the linked page for the R8500 represents the use of Channel 44.

SORRY! You're correct. I need to remember what I write!
The other comments about different STAs still apply.

 

[thiggins]

Pete, I could have done a lot of things. But I chose to use an AC1900 router as the baseline for the reasons htismage pointed out. The reference and technique are explained in detail.

 

[thiggins]

KAL-el. More CPU power and memory help wireless performance, sure. Are they the cure for all Wi-Fi performance problems? No. They mainly push the throughput vs. path loss curve up and to the right in the low and mid-signal areas. Once signals start to drop, MCS rates fall back and it's more up to the RF design to save the day. I could be missing something, but technology gains there are going to me minor and slow to come.

That said, MU-MIMO is going to need all the CPU it can get, once devices really come on line and start mixing with SU, N and legacy devices. The beamforming calculations are a b*tch and moreso for MU.

You're going to see our testing move to include load testing in the coming months. Things are going to get interesting for vendors as we show what their expensive routers can and can't do.