CPU Core 1 Spiking 100% GT-AX6000 other cores seems no utilized. Merlin 388.1

[TheLostSwede]

No. That page has the wrong info.

Actually, that was the device that was meant to launch as the GT-AX6000, see:

Gt-AX6000 manufacturing year

Put a low ball offer on a 'new' GT-AX6000 on ebay, as there is a live cashback promotion in the UK for Asus products from retailers. Guy said price was way too low as it was the 2024 model and had improved 'firmware updates, performance improvements, port configuration and WiFi 6 enhancements'...

www.snbforums.com

 

[RMerlin]

Actually, that was the device that was meant to launch as the GT-AX6000, see:

Gt-AX6000 manufacturing year

Put a low ball offer on a 'new' GT-AX6000 on ebay, as there is a live cashback promotion in the UK for Asus products from retailers. Guy said price was way too low as it was the 2024 model and had improved 'firmware updates, performance improvements, port configuration and WiFi 6 enhancements'...

www.snbforums.com

That device was the GT-AX6000N. Different from the GT-AX6000. Snippet from its build profile:

BOOT_FLASH_TYPE=NAND WIFI_CHIP=QCN50X4 SWITCH_CHIP=QCA8075_PHY_AQR107
DTB="qcom-ipq807x-hk01.dtb"

 

[TheLostSwede]

That device was the GT-AX6000N. Different from the GT-AX6000. Snippet from its build profile:

BOOT_FLASH_TYPE=NAND WIFI_CHIP=QCN50X4 SWITCH_CHIP=QCA8075_PHY_AQR107
DTB="qcom-ipq807x-hk01.dtb"

Well, I guess that's the thing that's being confused all the time with the current GT-AX6000, despite never having launched.

 

[sfx2000]

That device was the GT-AX6000N. Different from the GT-AX6000. Snippet from its build profile:

Well, the QCA solution likely would have been the better one

IMHO - perhaps Broadcom is getting left behind, much like Quantenna - they've grown too big and they're basically a holding company now days... they sold off half of their WIFi client portfolio to Cypress, and the rest is basically hanging on - I wonder how many hardware/firmware engineers are still there - thru layoffs and attrition - and I've heard from peers in the industry, Broadom/Avago is not a nice place to work unless one is on the high-bandwidth switching chipsets focused on the AI and Cloud.

Qualcomm and MediaTek - they're very focused on wireless, and fierce competitors in the mobile 4G/5G space, and much of that tech is trickling down into WiFi...

 

[hasuboy]

Year old thread, but..



This is not true. QoS is completely independent from how fast your network is.

QoS applies prioritization. Without QoS, if you max out your connection with a transfer, all other data gets traffic jammed. This is disastrous with latency sensitive data like games, VoIP and streaming services. Latency goes through the roof, they can hang for multiple seconds at a time.

With QoS, you can still do all these things with minimal degradation. There is still some, yes, but it's orders of magnitude lower (30-50ms increase) than without any QoS at all (7000ms).

When I upgraded from a 200/35 cable connection to 1G symmetrical fiber, I ran into the same issue as posted in this thread: Terrible network performance with QoS enabled. It didn't matter what QoS I tried using, the transfer speed was crippled. The only 'solution' was to disable QoS and run into the awful problem of not having it anymore, or deal with erratic 20-60% lower transfer speeds than I'm paying for. This is ignoring bufferbloat issues too.

Been using QoS since my first "gaming" router (D-Link DGL-4100) from 2005. I even invested into Bigfoot's Killer NIC because it provided tangible benefits over generic onboard crap (especially for UDP). I've been a competitive gamer and I also play some lockstep games, and consistent low latency is extremely important in both these cases. I also live with a family who's become more and more digitally oriented. QoS is a godsend for people like me.

Surely there's a router out there that either has a CPU powerful enough for at minimum 1G QoS, or is multi-threaded? This can't still be a bottleneck in 2024.

With how many ASUS routers have had glaring flaws (former 87U user), I'm close to not caring anymore and will even consider paying for enterprise hardware...

The problem of QoS causing Core 1 to be fully loaded has always existed in ASUS routers. Actually, I think this issue should be a problem of Broadcom，because his QoS is implemented at the software level。If you really want to use QoS on an ASUS router, based on my testing, only adaptive QoS can be used. This mode will not cause one core to be fully loaded, but I have not conducted in-depth testing on the effectiveness of QoS in this mode. Most of Broadcom's SOC products are based on the home market, and QoS belongs to enterprise level functions in a sense。In recent years, MTK's new SOC has hardware level QoS, but it still can not be separated from the fact that it is not an enterprise level SOC. Moreover, MTK support hardware QoS is one thing, and ultimately depends on the strength of the router manufacturer in developing firmware for the finished product. Enterprise level products are a viable option, but from a cost perspective, I believe building software router using idle x86 hardware is also a viable option.

 

[sfx2000]

The problem of QoS causing Core 1 to be fully loaded has always existed in ASUS routers. Actually, I think this issue should be a problem of Broadcom，because his QoS is implemented at the software level。If you really want to use QoS on an ASUS router, based on my testing, only adaptive QoS can be used.

Could be upper layer stuff, but QoS and connections are single threaded - has to be due to how linux networking actually works on the WAN side - Getting a loaded/hot core isn't that big of a deal - that means it's working as designed, and there are more cores available for other tasks...

It's much like the high mem usage in linux - folks complain here many times as well - and from a dev perspective, any memory not used, is memory wasted...