ASUS RT-AC5300

[RMerlin]

I wish they would just get on with releasing the newer models
so i can replace my rt-ac66u with a faster cpu

I just want a 1.4ghz duel core with 3x3 or better on a budget
no need for duel 5ghz

Then this router isn't for you, since it has two 5 GHz radios.

 

[pege63]

How many amplifiers are ASUS uses at 2.4GHz and at 5.0GHz and what strength in mW , and on 2-Stage amplifiers, and advanced, low noise amplifiers on there ROUTERS?

 

[RMerlin]

How many amplifiers are ASUS uses at 2.4GHz and at 5.0GHz and what strength in mW , and on 2-Stage amplifiers, and advanced, low noise amplifiers on there ROUTERS?

The RT-AC5300 isn't on the market yet, so nobody knows what are the actual hardware details.

 

[pege63]

And in General on the Routers AC1900, AC2600, AC3200 from ASUS?

 

[System Error Message]

No ones mentioned how many ethernet ports that thing would have or if it would have 10G. Would need to do LAGG just to support that wifi.

 

[pege63]

Just take a look in the manual https://www.asus.com/Networking/RT-AC5300/HelpDesk_Manual/

 

[RMerlin]

And in General on the Routers AC1900, AC2600, AC3200 from ASUS?

Block diagrams are under a seal at the FCC. The closest you might get would be to dig through the CFE source code - sometimes the PA model is revealed.

Tim's reviews also often include bits of that information. But I don't think anyone publicly knows all those details.

 

[sfx2000]

No ones mentioned how many ethernet ports that thing would have or if it would have 10G. Would need to do LAGG just to support that wifi.

Nah, if this follows what Broadcom has done with their other XStream deployments, it's two AC2600 class radios on the 5GHz side, so no bonding... will be fine with GiGE, and no LAGG needed..

 

[madhatter01]

What are we thinking the ETA is on this?

 

[snakebite3]

What are we thinking the ETA is on this?

Probably Nov or Dec.

 

[System Error Message]

Nah, if this follows what Broadcom has done with their other XStream deployments, it's two AC2600 class radios on the 5GHz side, so no bonding... will be fine with GiGE, and no LAGG needed..

But the practical wifi rates. Thiggins say you can get 50% practically while i say 30%. but even 30% of that 5 or even 4Gb/s (talking about maximum achievable performance) will exceed 1Gb/s so having at least 2 ports teamed up would prevent ethernet from behind the bottleneck.

So the router will have a better working MU-MIMO and 2 5Ghz radios (8 streams). That means i could have 4 wifi clients each with dual channel MU-MIMO AC wifi downloading/uploading from a file server all at the same time and a single 1Gb/s ethernet port cannot keep up. Many embedded NAS now have multiple Gbe ports or even 10Gbe so LAGG would actually be very useful with at least 2 ports. Even so i still think that wired router + switch design is very poor because taking a typical wifi router:
[Wifi chip]------(bus/PCIe/similar)-----CPU-----1Gb/s-----[5 port switch with 4 ethernet ports and 1 port to the CPU]
Also connected to the CPU is a 1Gb/s WAN port.
On a hardware accelerated router
[Wifi chip]-----[Accelerator chip]---RGMII (1Gb/s)-------[5 port switch with 4 ethernet ports and 1 port to CPU with 3 RGMII interfaces]----1Gb/s----CPU
Also connected to the CPU is a WAN port or accelerator chip + WAN port. Wifi to wire traffic bottlenecked by the 1Gb/s RGMII interface which means that if i properly had 3 or 4 streams connected and they were all MU-MIMO including the wifi chip itself it will not go beyond 1Gb/s. You can test it yourself by bridging 2 AC87U together very closely, combining 2 ethernet ports at each end to a computer with 2 ethernet ports each and start stress testing to the point where the CPU is maxed out and the NICs are filled with packet queues trying to send and recieve at wirespeed. I recommend disabling windows network autotuning if you wish to test this.

This really limits throughput on WAN and in some cases also on LAN. If they keep to these traditional designs they are going to ruin the hardware performance potential since the bottleneck would be in the busses between the chips. Sure the accelerator chip on wifi with connection to the switch chip does reduce CPU load but than again only 1 Gb/s to the CPU? Even with wifi practical speeds being low but i managed to get 90% of the theoretical wifi bandwidth by stress testing in a nonsensical way to the point of dropped packets. If i see dropped packets and a low wifi utilisation that means that there is a bottleneck within the router which i think this AC5300 will have if you truely stress test it. I'd be happy to actually prove it if they release it with the traditional hardware design and lend me one using one of my nonsensical stress test by pairing up devices in such a way that the transfers will actually exceed the router's LAN capabilities.

My suggestion for the AC5300 would be if ASUS ever read this is
[Wifi Chips]-----[Accelerator with same arch as CPU such as with the AC3200]-----5x1Gb/s RGMII interfaces teamed up----managed switch chip with at least 8 RGMII and at least 8 ethernet ports----at least 2 RGMII to the CPU----- 2x WAN ports. This would be a significant upgrade obviously.

Another solution would be [wifi Chips]-----PCIe 2.0 at 2x at least------CPU----PCIe 2.0 at 4x at least or 8 RGMII interfaces-------managed switch chip with PCIe or 8 RGMII and 8 ethernet ports, 2 WAN ports connected to CPU as usual.

Perhaps some gadgetry like esata and a much faster CPU, perhaps a SoDIMM RAM slot so you can use laptop RAM with it and get as much memory as you want/need. Maybe the CPU could be swappable so it can be upgraded to a faster CPU if there isnt one that properly supports this. From a consumer point of view for that much wifi bandwidth there will be LAN media transfers, file transfers, gaming and ofcourse lots of internet stuff for the same reasons which means having 2 dedicated WAN ports would fare a lot better in throughput with a much faster CPU and a better internal design. It means being able to have 2 gigabit ISPs at the same time properly on a consumer router with enough LAN ports and users to actually utilise it.

 

[L&LD]

SEM, I agree that we (already) need more than 1GBe ports on future routers such as the RT-AC5300.

But I don't care about stress tests up to dropped packet scenarios that we'll never see in real life.

I hope Asus or another manufacturer (because then all will be forced) sees the benefits of including at least one or more 10GBe ports on the switch side. Going down the slow and arduous road of LAGG, 2GBe and 5GBe solutions is just prolonging the inevitable.

When MU-MIMO with more than 4 antennae and 4 streams is actually achieved, 1GBe ports are effectively obsolete.

Especially for anyone with a Gbps ISP. Of course, this will also push the router to be able to process packets at equal or greater than 2Gbps up / down to at that point.

I too like to eliminate bottlenecks. But only to the point of real world use.

And please, no LAGG, no 2Gbps / 5Gbps ports. Just do it right (from the start).

 

[System Error Message]

SEM, I agree that we (already) need more than 1GBe ports on future routers such as the RT-AC5300.

But I don't care about stress tests up to dropped packet scenarios that we'll never see in real life.

I hope Asus or another manufacturer (because then all will be forced) sees the benefits of including at least one or more 10GBe ports on the switch side. Going down the slow and arduous road of LAGG, 2GBe and 5GBe solutions is just prolonging the inevitable.

When MU-MIMO with more than 4 antennae and 4 streams is actually achieved, 1GBe ports are effectively obsolete.

Especially for anyone with a Gbps ISP. Of course, this will also push the router to be able to process packets at equal or greater than 2Gbps up / down to at that point.

I too like to eliminate bottlenecks. But only to the point of real world use.

And please, no LAGG, no 2Gbps / 5Gbps ports. Just do it right (from the start).

Currently 10Gbe is just too expensive to implement so they cant go for that but using a fast CPU and the PCIe busses is a much better way for internal design than RGMII and cheaper than 10Gbe interface implementations. With the PCIe bus you can have multiple lanes and each lanes gives 2.5Gb/s. On the ethernet side it means a switch chip that actually has a PCIe interface and/or a very fast CPU that will do all that at wirespeed with PCIe interfaces. A fast CPU without a switch chip will than be able to do wirespeed L2 without the restrictions of a switch chip and using the layer 2 bridge instead means more options to configure easily. People underestimate the wirespeed L2 capabilities of networked CPUs such as with MIPS or even the commonly used broadcom dualcore ARM A9.

One thing that you do not see in routers is that you could have a router that is essentially a CPU with lots of PCIe lanes which you can than use HDMI/mini HDMI as a form of networking. The PCIe would be 2.5Gb/s per lane and is very easy to combine multiple lanes to get more. This would be a cheap and simpler way to get more bandwidth and lower latency. PCIe uses a switch chip but for PCIe and not ethernet that has far less overheads and protocols to use and has no issues transferring data between 2 Tesla GPUs that each also have their own PCIe switch chip and the latest CUDA running on it without going through the CPU and with direct access to RAM. Think of the performance increase and cost reduction you could achieve using PCIe. Expresscard is a hotswappable variant of PCIe and compute servers have hot swappable PCIe x16 slots.

Than theres USB3 or 3.1 networking for speeds that will support that wifi bandwidth.

 

[L&LD]

10GBe is only expensive because it is such a low volume part. Putting it in consumer routers would change that overnight.

Please don't mention USB (3 or 3.1) vs. a proper network protocol. Bandwidth requirements are only part of the story. USB and storage simply don't mix when performance is of high importance.

 

[System Error Message]

10GBe is only expensive because it is such a low volume part. Putting it in consumer routers would change that overnight.

Please don't mention USB (3 or 3.1) vs. a proper network protocol. Bandwidth requirements are only part of the story. USB and storage simply don't mix when performance is of high importance.

Used to be when people manufactured networking hardware that they will make sure that their hardware is not the bottleneck. They dont make them like they used to.

 

[tim]

But the practical wifi rates. Thiggins say you can get 50% practically while i say 30%. but even 30% of that 5 or even 4Gb/s (talking about maximum achievable performance) will exceed 1Gb/s so having at least 2 ports teamed up would prevent ethernet from behind the bottleneck.

So the router will have a better working MU-MIMO and 2 5Ghz radios (8 streams). That means i could have 4 wifi clients each with dual channel MU-MIMO AC wifi downloading/uploading from a file server all at the same time and a single 1Gb/s ethernet port cannot keep up. Many embedded NAS now have multiple Gbe ports or even 10Gbe so LAGG would actually be very useful with at least 2 ports. Even so i still think that wired router + switch design is very poor because taking a typical wifi router:
[Wifi chip]------(bus/PCIe/similar)-----CPU-----1Gb/s-----[5 port switch with 4 ethernet ports and 1 port to the CPU]
Also connected to the CPU is a 1Gb/s WAN port.
On a hardware accelerated router
[Wifi chip]-----[Accelerator chip]---RGMII (1Gb/s)-------[5 port switch with 4 ethernet ports and 1 port to CPU with 3 RGMII interfaces]----1Gb/s----CPU
Also connected to the CPU is a WAN port or accelerator chip + WAN port. Wifi to wire traffic bottlenecked by the 1Gb/s RGMII interface which means that if i properly had 3 or 4 streams connected and they were all MU-MIMO including the wifi chip itself it will not go beyond 1Gb/s. You can test it yourself by bridging 2 AC87U together very closely, combining 2 ethernet ports at each end to a computer with 2 ethernet ports each and start stress testing to the point where the CPU is maxed out and the NICs are filled with packet queues trying to send and recieve at wirespeed. I recommend disabling windows network autotuning if you wish to test this.

This really limits throughput on WAN and in some cases also on LAN. If they keep to these traditional designs they are going to ruin the hardware performance potential since the bottleneck would be in the busses between the chips. Sure the accelerator chip on wifi with connection to the switch chip does reduce CPU load but than again only 1 Gb/s to the CPU? Even with wifi practical speeds being low but i managed to get 90% of the theoretical wifi bandwidth by stress testing in a nonsensical way to the point of dropped packets. If i see dropped packets and a low wifi utilisation that means that there is a bottleneck within the router which i think this AC5300 will have if you truely stress test it. I'd be happy to actually prove it if they release it with the traditional hardware design and lend me one using one of my nonsensical stress test by pairing up devices in such a way that the transfers will actually exceed the router's LAN capabilities.

My suggestion for the AC5300 would be if ASUS ever read this is
[Wifi Chips]-----[Accelerator with same arch as CPU such as with the AC3200]-----5x1Gb/s RGMII interfaces teamed up----managed switch chip with at least 8 RGMII and at least 8 ethernet ports----at least 2 RGMII to the CPU----- 2x WAN ports. This would be a significant upgrade obviously.

Another solution would be [wifi Chips]-----PCIe 2.0 at 2x at least------CPU----PCIe 2.0 at 4x at least or 8 RGMII interfaces-------managed switch chip with PCIe or 8 RGMII and 8 ethernet ports, 2 WAN ports connected to CPU as usual.

Perhaps some gadgetry like esata and a much faster CPU, perhaps a SoDIMM RAM slot so you can use laptop RAM with it and get as much memory as you want/need. Maybe the CPU could be swappable so it can be upgraded to a faster CPU if there isnt one that properly supports this. From a consumer point of view for that much wifi bandwidth there will be LAN media transfers, file transfers, gaming and ofcourse lots of internet stuff for the same reasons which means having 2 dedicated WAN ports would fare a lot better in throughput with a much faster CPU and a better internal design. It means being able to have 2 gigabit ISPs at the same time properly on a consumer router with enough LAN ports and users to actually utilise it.

Very interesting, thanks for that.

Just last week I found myself wondering about the interconnects within my RT-AC3200.

The logical diagram is here http://www.smallnetbuilder.com/wire...m-the-coming-battle-for-wi-fi-airtime?start=1

You can see 3 x radios connected to PCIe 2.0 straight over to the CPU. But, unless I'm being dumb, this is not reflected in the hardware.
The 3 radios will take up the 3 ports on the "PLX Technology PEX8603 3-lane, 3-port PCIe switch". So then how does the PCIe switch connect to the CPU?

Yeah yeah I think too much sometimes

 

[sfx2000]

So the router will have a better working MU-MIMO and 2 5Ghz radios (8 streams)

Correction - two 4-stream radios in the 5GHz, like I mentioned, you cannot bond them... and while yes, it's MU-MIMO and 4*4:4, in MU mode, it won't be any faster than AC1300 (3*3:3 with one sounding/training stream)

BTW - Broadcom on their 11AC chips tends to favor PCIe over RGMII/SGMII, as PCIe is a faster bus..

Anyways, GigE is more than sufficient for LAN/WAN...

 

[sfx2000]

You can see 3 x radios connected to PCIe 2.0 straight over to the CPU. But, unless I'm being dumb, this is not reflected in the hardware.
The 3 radios will take up the 3 ports on the "PLX Technology PEX8603 3-lane, 3-port PCIe switch". So then how does the PCIe switch connect to the CPU?

The Broadcom SoC has native PCIe - the PLX breaks that out - common approach seen on other devices as well

 

[tim]

The Broadcom SoC has native PCIe - the PLX breaks that out - common approach seen on other devices as well

I think I see it now. The BCM4709 has 2 x PCIe interfaces (according to this https://wikidevi.com/wiki/Broadcom).
So 1 radio will go straight to CPU via PCIe, 2 radios go to the PCIe switch with 3rd lane uplink to CPU.

 

[sfx2000]

I think I see it now. The BCM4709 has 2 x PCIe interfaces (according to this https://wikidevi.com/wiki/Broadcom).
So 1 radio will go straight to CPU via PCIe, 2 radios go to the PCIe switch with 3rd lane uplink to CPU.

Bingo!