It's teaser-time. Because why not.

[RMerlin]

You mean the 4916 right?

BCM4912 has two SXGMII ports, BCM4916 has three SXGMII ports.

That's why the GT-AXE16000 is able to provide two 10 Gbps ports.

 

[RMerlin]

I was having issues with the GT-AXE16000's 10Base-T ports which showed just barely 2.6 Gbps wired speeds as tested on Ookla with the full AiProtection suite turned on.

That's a NAT/CPU limitation, not Ethernet limitation.

Running iperf on a GT-AX11000_Pro and GT-AXE16000 Pro connected together over their 10 Gbps interfaces, I can get 3.2 Gbps - and the bottleneck is the CPU not being able to run iperf fast enough to generate more traffic.

 

[RMerlin]

Can those 10-Gb ports have hardware flow control configured?

I don't know. You could try installing ethtool from Entware and try it out, unless you can figure out the syntax for BCM's own switch control tool (I forgot its name).

 

[RMerlin]

Do I have to do a factory reset after loading Merlin firmware on a gt-axe16000?

No, but you might need to reconfigure/reset any VPN configuration, as our implementations are different.

Doing a factory default reset is probably a good idea tho. It seems that earlier builds of the stock firmware were leaving some large variables into kernel nvram, which is limited to 128 KB. Doing a factory default reset with the latest stock firmware or with 386.8 will probably move these to JFFS.

 

[Treadler]

No, but you might need to reconfigure/reset any VPN configuration, as our implementations are different.

Doing a factory default reset is probably a good idea tho. It seems that earlier builds of the stock firmware were leaving some large variables into kernel nvram, which is limited to 128 KB. Doing a factory default reset with the latest stock firmware or with 386.8 will probably move these to JFFS.

Is a reset indicated for AX88U with 386.8 too?

 

[RMerlin]

Is a reset indicated for AX88U with 386.8 too?

No.

 

[RTWZ80]

That's a NAT/CPU limitation, not Ethernet limitation.

Running iperf on a GT-AX11000_Pro and GT-AXE16000 Pro connected together over their 10 Gbps interfaces, I can get 3.2 Gbps - and the bottleneck is the CPU not being able to run iperf fast enough to generate more traffic.

Thanks for the insight. If the CPU is the bottleneck, I'm guessing no firmware revision's going to fix the lousy wired speeds for 10G. Or would a hardware revision by Asus do the trick? From what I've read on the RT-AX89X threads on the forum, there were at least 4 hardware revisions before the SFP+ and 10G-BaseT speed issues were ironed out. I'm holding onto my later version RT-AX89X for that very reason before deciding whether to pull the trigger on another AXE16000, after returning the first two.

 

[matthew johnson]

No, but you might need to reconfigure/reset any VPN configuration, as our implementations are different.

Doing a factory default reset is probably a good idea tho. It seems that earlier builds of the stock firmware were leaving some large variables into kernel nvram, which is limited to 128 KB. Doing a factory default reset with the latest stock firmware or with 386.8 will probably move these to JFFS.

Thank you

 

[RTWZ80]

Thanks for the insight. If the CPU is the bottleneck, I'm guessing no firmware revision's going to fix the lousy wired speeds for 10G. Or would a hardware revision by Asus do the trick? From what I've read on the RT-AX89X threads on the forum, there were at least 4 hardware revisions before the SFP+ and 10G-BaseT speed issues were ironed out. I'm holding onto my later version RT-AX89X for that very reason before deciding whether to pull the trigger on another AXE16000, after returning the first two.

Sorry wasn't fully awake when I posted this. I'm guessing that a firmware update by Asus developers/RMerlin could boost the CPU performance on the router similar to overclocking your Intel/AMD CPUs.

 

[sfx2000]

BCM4912 has two SXGMII ports, BCM4916 has three SXGMII ports.

That's why the GT-AXE16000 is able to provide two 10 Gbps ports.

Broadcom's own datasheet is saying 2.5Gbit on the internal PHY - which is implemented as USXGMII (which supports 10G/5G/2.5G/1G/100M/10M)

* Integrated 2.5G multi-Gig PHY for WAN or LAN
* 2x USXGMII Ethernet ports and 1x RGMII port

To get to 10G on runner, one has to do the upgrade to BCM4916, hence my question on the model number

Might explain why nobody is seeing 10G end to end on iperf...

 

[L&LD]

Yeah, but it's impressively fast LPDDR5 on the M2... and with Thunderbolt 4 and a 10Gbit adapter, would be a nice fit with this device

(my dell XPS13 also has soldered DDR, this is getting to be a thing in the thin and light laptop space)

For a home gateway (router/ap) - 2GB is plenty of space - and with the crazy supply chain these days, might be they got a really good deal on specific parts if they're not in high demand - the DRAM market right now is all over the place

My point about the 8GB RAM M2 Air is that the M2 chipset requires more RAM in and of itself (and/or more/faster swap file storage). As a 'new' for 2022 device, it is markedly below the original 8GB M1 Air performance. Particularly in multitasking use.

 

[XIII]

As a 'new' for 2022 device, it is markedly below the original 8GB M1 Air performance.

I have only seen reports that the M2 Air is slower than the M1 Air because it now uses a single chip in the 256 GB SSD of the base model.

Where can I read more about this CPU/memory issue?

 

[L&LD]

Note in the first few minutes of the first video how much more swap file is used (objective).

 

[RMerlin]

Thanks for the insight. If the CPU is the bottleneck, I'm guessing no firmware revision's going to fix the lousy wired speeds for 10G. Or would a hardware revision by Asus do the trick? From what I've read on the RT-AX89X threads on the forum, there were at least 4 hardware revisions before the SFP+ and 10G-BaseT speed issues were ironed out. I'm holding onto my later version RT-AX89X for that very reason before deciding whether to pull the trigger on another AXE16000, after returning the first two.

If you are referring to NAT throughput, I don't expect this to be changed by a hardware revision. It's simply a hardware limitation. At those speeds, you'd need either a business-class router, or at the very least IPv6 to get rid of the NAT overhead.

Might explain why nobody is seeing 10G end to end on iperf...

Keep in mind that my test was done by running iperf on the router itself (without NAT - I was doing LAN to LAN test, with the "client" being a second router connected in AP mode)). CPU usage was showing all three cores (I ran three threads) were 100% occupied by iperf itself, and the scaling was about right when compared to running iperf in a single thread.

For NAT throughput, I have no idea what can be expected. We'd need an expert with the necessary equipment to do a review/benchmark.

To get to 10G on runner, one has to do the upgrade to BCM4916, hence my question on the model number

BCM4916 is designed for Wifi7. Maybe it could still be coupled with a BCM6715 to provide a faster Wifi 6 product (since that's connects over PCI-Express), but I doubt Broadcom is already mass producing the BCM4916 at this point. I don't know either if their SDK would support that kind of hybrid design (Wifi 6 and Wifi 7 generation products).

The fact that Asus is launching a series of Pro models using a BCM4912 (there is a third Pro model yet to be announced that is expected to launch this autumn) makes be believe that BCM might be slowing down production of the BCM4908, but they haven't fully ramped up production of BCM4916 yet - or that product cannot be used in a Wifi 6 design.

 

[RMerlin]

I have only seen reports that the M2 Air is slower than the M1 Air because it now uses a single chip in the 256 GB SSD of the base model.

Where can I read more about this CPU/memory issue?

Two unrelated issues. Disk I/O is slower on the M2 using the 256 GB version due to the use of a single NAND chip rather than two chips in dual channel. This isn't related to RAM however.

 

[XIII]

Two unrelated issues. Disk I/O is slower on the M2 using the 256 GB version due to the use of a single NAND chip rather than two chips in dual channel. This isn't related to RAM however.

Indeed.

However, I only read about the SSD issue, hence my question.

I don’t understand how these videos show the RAM issue mentioned by L&LD, since they don’t compare an 8 GB M1 with an 8 GB M2, but an 8 GB M2 with a 16 GB M2?

PS: I’m “sometimes“ a bad reader; what am I missing here / this time?

 

[RMerlin]

PS: I’m “sometimes“ a bad reader; what am I missing here / this time?

The core issue is probably the fact that RAM isn't upgradable, and since the RAM is shared by the CPU and the GPU, 8 GB can quickly become a problem.

 

[Morris]

Ookla

The Okla speed test unless proven with other hardware to measure that type of speed is probably not going to cut it. You need a direct link between two capable machines

 

[Morris]

Thanks for the insight. If the CPU is the bottleneck, I'm guessing no firmware revision's going to fix the lousy wired speeds for 10G. Or would a hardware revision by Asus do the trick? From what I've read on the RT-AX89X threads on the forum, there were at least 4 hardware revisions before the SFP+ and 10G-BaseT speed issues were ironed out. I'm holding onto my later version RT-AX89X for that very reason before deciding whether to pull the trigger on another AXE16000, after returning the first two.

If the chipset dose fast switching, then the CPU only needs to tell the chipset where the buffer is in memory and the chipset moves the data out the port

 

[RTWZ80]

The Okla speed test unless proven with other hardware to measure that type of speed is probably not going to cut it. You need a direct link between two capable machines

I've just heard back from Asus Tech Support regarding my crap download/upload wired speeds on 10Gbps broadband and this was what they said verbatim:

'Dear Valued Customer,

If you recently encounter limited speed acceleration, or unable to reach maximum connection speed, please try to disable the following features: AiProtection, Traffic Analyzer, Web History, Adaptive QoS, Game boost, Web & apps filters, bandwidth monitor, Mobile Game Mode to maintain the optimal internet speed.

Sincerely,
ASUS Team'

Seems to tally with what RMerlin said of it being a NAT/CPU limitation throttling my wired speeds.