cancel
Showing results for 
Search instead for 
Did you mean: 

ASUS ROG Crosshair VIII Hero (Wi-Fi) network adapter problem?

t0rzzz
Level 7
Hi guys,
I'd like to have an advice for my current build.

I have an ASUS ROG Crosshair VIII Hero (WiFI model) and recently I have bought a 10Gbit network adapter (ASUS XG-C100C v2).

From what I can see, the adapter only takes 8Gbit out of 10Gbit when testing the connection locally, although it is (I guess?) perfectly configured and installed with the very latest drivers.

I believe it might be a PCI limitation, as it seems like it takes only 8x lanes, limiting the bandwidth to a total of 8GT (and also 8Gb, it won't go any little faster).

Following is my current build:

- CPU: AMD Ryzen 9 5950X
- Graphic card: NVIDIA GeForce GTX 1080Ti
- M.2 slots: 2 x SSD (Samsung 980 PRO and 960 EVO)
- Power supply: ASUS ROG Strix 850W
------------
BIOS version: 4304

Is there any way to confirm the network adapter is *actually* using the full bandwidth of the PCI-E?

Thanks in advance for your help!

t0r
975 Views
4 REPLIES 4

xeromist
Moderator
I'm not sure how you are testing? There is overhead to ethernet for things like error correction so you will never see "10gbps" sustained transfer regardless of how you test. It will always be lower than that. Also if you are transferring to or from storage (even RAID 0) there will be some reduction. Really the only way to get close to max transfer is with synthetic benchmarking tools that use memory buffers. But that's rather pointless as you'll never see the same speed in real world usage.
A bus station is where a bus stops. A train station is where a train stops. On my desk, I have a work station…

xeromist wrote:
I'm not sure how you are testing? There is overhead to ethernet for things like error correction so you will never see "10gbps" sustained transfer regardless of how you test. It will always be lower than that. Also if you are transferring to or from storage (even RAID 0) there will be some reduction. Really the only way to get close to max transfer is with synthetic benchmarking tools that use memory buffers. But that's rather pointless as you'll never see the same speed in real world usage.


I'm testing with iperf3 on the very same machine (iperf3 -s in a terminal, iperf3 -c 127.0.0.1 on another).
I'm not obviously complaining about the speed rate, I would like to understand just if it is limited for some reason to 8Gb.

Thanks for your reply.

xeromist
Moderator
OK, yes I think iperf would be the way to get the closest estimate. iperf is single threaded so maybe it's getting CPU bound. I would check the system monitor to see if one core is getting pegged at 100%. If so you might need to run parallel transfers. I think iperf has a -p option for that? Something like that. I've never actually used it before so I'm not an expert.
A bus station is where a bus stops. A train station is where a train stops. On my desk, I have a work station…

Murph_9000
Level 14
t0rzzz wrote:
I believe it might be a PCI limitation, as it seems like it takes only 8x lanes, limiting the bandwidth to a total of 8GT (and also 8Gb, it won't go any little faster).


Ok, two quite fundamental things there. First, the pictures I found of the XG-C100C show a card that has a x4 physical interface, so it can't possibly run as either x8 or x16. Even if it was a x16 card, you could only run it as x8 on a desktop platform with a GPU, as there's only a single x16 on mainstream desktops (which would be split as x8+x8 if you have two x16 cards installed). Second, 8GT is not 8Gbits. The transfer rate in GT/s is based on the PCIe version, not the lane count. PCIe 3.0 is 8GT/s and PCIe 4.0 is 16GT/s. The rate is then multiplied by the number of lanes for the total bandwidth. 8GT would be roughly 8Gbits per lane, so 32Gbits for a x4 card.

It's not immediately clear to me what PCIe version is implemented on the XG-C100C, but I'd guess it should be at least 3.0. The motherboard slot will slow down to the card's speed, for things like a 3.0 card in a 4.0 slot (and similarly for a 4.0 card in a 3.0 slot).

I'm honestly not 100% certain of exactly the rate you should be seeing, as I've not played around with iperf. For normal application usage, roughly 1Gbyte/s would be quite reasonable for a 10Gbit/s interface, to allow for protocol and encoding overheads. So, if you're calculating 1Gbyte x 8 bits = 8Gbits, and that's the application data rate, that would be roughly full performance on a 10Gbit interface. It's common practice in networking to use 10:1 as a rough estimate for bits:bytes (rather than 8:1), to allow for the overheads.

I can't say for certain, but I think you may not have a problem to solve.