GT-BE98 10G Gaming Port limits WAN upload to ~1 Gbps, while 2.5G LAN reaches full 2 Gbps upload

[Sundraw]

Hi everyone,

I’m trying to understand whether this is a firmware issue, hardware limitation, or compatibility problem involving the 10G Gaming port on my ASUS GT-BE98.

My internet line is 10 Gbps down / 2 Gbps up.

Setup
- Router: ASUS GT-BE98
- Tested firmware: latest official ASUS firmware, Gnuton 3006.102.6_1-gnuton1
- Mode: Wireless Router
- ONT: external ONT
- Motherboard: Z790
- CPU: Intel i7-14700K
- Tested NICs: Realtek RTL8127, Realtek RTL8125 (5 GbE)
- Ethernet cable: tested and working

Problem
If the PC is connected to the 10G Gaming port, WAN upload is limited to about 1 Gbps.
If I move the exact same PC / same cable / same NIC to a 2.5G LAN port, WAN upload goes up to the expected ~2 Gbps.

So the issue appears to be specific to the 10G Gaming port.

What I tested
I tested this with both RTL8127 and RTL8125, on Windows 11 and Linux.

On both NICs I disabled the usual troubleshooting settings:
- Flow Control
- Energy Efficient Ethernet / EEE
- Interrupt Moderation
- other commonly disabled advanced NIC options where applicable

The result is always the same:
- 10G Gaming port -> ~1 Gbps upload
- 2.5G LAN port -> ~2 Gbps upload

So this is reproducible across:
- two different Realtek NICs
- two different operating systems
- same PC
- same cable
- same WAN line

Router settings
- Adaptive QoS: disabled
- Game Boost: disabled
- Gaming Port Priority: disabled
- Open NAT / port forwarding: enabled, but I assume unrelated to raw WAN throughput
- Under LAN > Switch Control:
- Jumbo Frame: disabled
- Bonding / Link Aggregation: disabled
AiProtection / Traffic Analyzer / VPN / Parental Controls are also disabled

There is no visible NAT acceleration toggle in this firmware.

Important additional point
The issue happens on both:
- latest official ASUS firmware
- Gnuton 3006.102.6_1-gnuton1

So this does not look limited to a single firmware build.

Why I suspect the router
At this point, the only variable that changes the result is the router port:
- 10G Gaming port -> ~1 Gbps upload
- 2.5G LAN port -> ~2 Gbps upload

Everything else stays the same:
- PC
- cable
- WAN line
- NIC family
- OS

That makes me suspect either:
- a GT-BE98 10G port / switch path issue
- a 10G PHY compatibility issue
- some router-side limitation affecting WAN upload only

Questions
- Has anyone seen similar behavior on the GT-BE98 or BE98 Pro 10G ports?
- Could this be related to the internal switching / routing path of the 10G Gaming port?
- Has anyone compared WAN upload on the 10G Gaming port vs 2.5G LAN?
- Could this still be a PHY compatibility issue even though it happens with both RTL8127 and RTL8125?
- Would testing with an Intel X550-T1 or Marvell/Aquantia AQC113-based NIC be worthwhile, or does this look clearly router-side?

My goal is to get the full 2 Gbps upload from my 10/2 line while using the 10G LAN port.

Any suggestions for additional tests are welcome. Thanks.

 

[degrub]

What is the link rate detected/negotiated by the ASUS and PC while plugged into the 10 Gbit/s port ?

If you use the 10Gb REALTEK nic, what is the link rate negotiated between the PC and the ASUS 10Gb port ?
If it doesn't sync at 10Gb/s something is physically wrong or drivers.

2.5 Gb is an oddball for compatibility. You may be able to place a non managed 10 Gb switch between the two and solve the issue.

 

[Sundraw]

The link is correctly negotiated at 10/10 Gbps (RX/TX), so this does not appear to be a simple physical link or cabling issue.
I also tested with another ASUS router (BQ16) on its 10G WAN/LAN ports and I see the same behavior: WAN upload stays around 1.1-1.2 Gbps.
Example result while link is negotiated at 10/10 Gbps:
- Download: ~7079 Mbps
- Upload: ~1190 Mbps

So at this point the problem seems to be real throughput on ASUS 10G ports, not link negotiation.

 

[fryedchikin]

Not seeing this issue with a GT-BE98 Pro. I can easily cap out my 5 Gig symmetrical connection using a TP-Link TX401 NIC.

 

[Sundraw]

Quick update.
Given the suggestions and after testing two different Realtek NICs (RTL8127 and RTL8125) with the same result (~1.1–1.2 Gbps WAN upload on the ASUS 10G port), I decided to order a TP-Link TX401 (Marvell/Aquantia based NIC).
It should arrive tomorrow and I will test it directly connected to the 10G LAN port on the router.
If the upload goes up to ~2 Gbps (which is my line limit), that would strongly suggest the issue is related to the Realtek NICs interacting with the ASUS 10G port.
I'll report back with results once I install and test the TX401.
For reference, link negotiation is already confirmed at 10/10 Gbps and the issue persists across two ASUS routers (GT-BE98 and BQ16), two cables and both Windows and Linux.

 

[degrub]

Back in the early consumer days of multigig NICs, i vaguely recall that there were buffer size issues . So, could also be that sort of thing in the REALTEK driver versus what the ASUS is expecting. Maybe someone else will remember more details.

 

[Sundraw]

Small update after additional testing
I did more troubleshooting to rule out the possibility that the issue was caused by the Realtek NICs.
I installed a TP-Link TX401 (Marvell/Aquantia based NIC) and repeated the exact same tests.

Results

• With both Realtek NICs (RTL8127 and RTL8125) and the TX401, the behavior is identical when the PC is connected to the 10G LAN port of the router.
• When testing against very low latency servers (~1 ms RTT), I can reach the full ~2 Gbps upload of my line.
• When testing against servers with ~15–18 ms RTT, upload drops to around ~800–1000 Mbps.

However, when connecting the TX401 to a 2.5G LAN port on the same router, upload returns to ~1.8–2 Gbps even with ~15 ms RTT.

Summary of tests

10G LAN → WAN upload → RTT ~15 ms → ~800–1000 Mbps
2.5G LAN → WAN upload → RTT ~15 ms → ~1.8–2 Gbps

This behavior is consistent across:

• Realtek RTL8127
• Realtek RTL8125
• Marvell/Aquantia TX401

This suggests the issue is not related to the NIC, but possibly something in the 10G LAN → WAN path of the router when RTT increases.

Additional notes

• Link negotiation is confirmed at 10 Gbps full duplex
• Same behavior observed on two routers:
– ASUS GT-BE98
– ASUS BQ16
• Tested with different Ethernet cables (Cat6a)
• Tested on Windows 11 and Linux
• QoS, Game Boost and other traffic shaping features disabled

At the moment it looks like the throughput degradation happens only when traffic enters the router from the 10G LAN port and RTT increases, while the internal switch path used by the 2.5G ports does not show the same behavior.
If anyone with a GT-BE98 / BE98 Pro / BQ16 could try a similar test (10G LAN vs 2.5G LAN with a server >10 ms away) it would be very helpful to confirm whether this is platform-wide or specific to my setup.

 

[jzchen]

Can you connect the two routers as an AiMesh, and perform a speed test like iPerf through them?

 

[Sundraw]

Can you connect the two routers as an AiMesh, and perform a speed test like iPerf through them?

That's a good suggestion, but currently my BQ16 is already deployed as a wireless AiMesh node and reconfiguring everything just for this test would be a bit disruptive.
However, based on the tests I already performed:

• Two different routers (GT-BE98 and BQ16) show the same behavior
• Two different NIC chipsets (Realtek RTL8127 and Marvell/Aquantia TX401) behave identically
• The issue only appears when the PC is connected to the 10G LAN port, not when connected to the 2.5G LAN ports
Given that the same pattern appears across different hardware combinations, it seems more likely related to the 10G LAN → WAN path in the router rather than the NIC itself.
If I get the chance later I might try the AiMesh iperf test to isolate the LAN switching path as well.

 

[jzchen]

I have the GT-BE98 Pro and BQ16 Pro and wired 10 Gb backhaul. As an AiMesh the node is using a designated "WAN" port (as these are WAN/LAN). Of course the main router is connected through a LAN port to the node....

I have a somewhat similar issue with two ExpertWiFi EBM68. They are nodes to an EBG15. The EBM68 has 2.5 Gb WAN ports. ExpertWiFi have built in iPerf but even though my backhaul is 10 Gb the speed reported is roughly 300 Mbps. Between EBG15 and EBM68 the speed is 900s Mbps (expected speed factoring overread)....

Since my fiber is limited to 1 Gbps I can't test to the ISP side unfortunately....

 

[Sundraw]

I have the GT-BE98 Pro and BQ16 Pro and wired 10 Gb backhaul. As an AiMesh the node is using a designated "WAN" port (as these are WAN/LAN). Of course the main router is connected through a LAN port to the node....

I have a somewhat similar issue with two ExpertWiFi EBM68. They are nodes to an EBG15. The EBM68 has 2.5 Gb WAN ports. ExpertWiFi have built in iPerf but even though my backhaul is 10 Gb the speed reported is roughly 300 Mbps. Between EBG15 and EBM68 the speed is 900s Mbps (expected speed factoring overread)....

Since my fiber is limited to 1 Gbps I can't test to the ISP side unfortunately....

Thanks for the suggestion and the detailed explanation.
I did try the built-in speed test on the GT-BE98, but I'm not entirely confident in its accuracy since it runs on the router CPU and usually uses only a small number of TCP streams. Because of that I’m hesitant to draw conclusions from it.
What I can consistently reproduce from the PC side is this pattern:

• ~2 Gbps upload when testing against very low latency servers (~1 ms RTT)
• ~800–1000 Mbps upload when RTT increases to ~10–18 ms

This happens regardless of the NIC (tested both Realtek RTL8127 and Marvell/Aquantia TX401) and also across two different ASUS routers (GT-BE98 and BQ16).
At this point I’m starting to suspect there may be some kind of RTT-dependent limitation somewhere in the upstream path (possibly ISP side or TCP buffering related), but I’d like to confirm that with more controlled tests.
My next step will be running iperf3 tests against some high-capacity public servers to see whether the connection can sustain ~2 Gbps upload with multiple TCP streams even when RTT is higher.
If the line can reach ~2 Gbps with iperf but not with speedtest servers, that would point more toward server limitations rather than the router itself.
I'll report back once I have those results.

 

[Sundraw]

Update after running some iperf tests.
To rule out speedtest server limitations, I tested upload using iperf3 toward a public Scaleway server with multiple TCP streams.
Command used:

iperf3 -c iperf.scaleway.fr -p 5202 -P 16
Result:
~1.47 Gbps upload sustained.
This suggests that the connection itself is capable of significantly higher throughput than what some speedtest servers were reporting (~800–1000 Mbps when RTT was ~10–15 ms).
So at this point it seems the lower results I was seeing were likely related to speedtest server limitations and TCP stream behavior rather than the router, NIC, or the 10G LAN port.

Thanks again for the suggestions — iperf testing helped clarify things.

 

[BoostOver]

It's strange that the luminaries at fibraclick didn't give you a suitable answer

 

[Sundraw]

It's strange that the luminaries at fibraclick didn't give you a suitable answer

I haven’t really discussed this specific case elsewhere, as I preferred to approach it with more controlled testing rather than relying on quick assumptions (or immediately being told to just switch ISP ).
At first it really looked like a 10G LAN / hardware issue, which made it quite misleading.
After running more thorough tests (iperf with multiple TCP streams), it turns out the connection itself can sustain ~1.4–1.5 Gbps upstream to EU servers.
So the lower results appear to be related to TCP behavior, latency and speedtest server limitations, something that only becomes clear after more in-depth analysis.