Product Link(s): cat5e ethernet cable (15ft, 50ft, 100ft or 200ft options): amzn.to/3VY4ZEg cat5e ethernet cable (35ft or 300ft options): amzn.to/49TbePz cat5e ethernet cable (similar to one in video): amzn.to/41Ea6go cat6 ethernet cable (similar to one in video): amzn.to/3ZVNeI6 cat7 ethernet cable: amzn.to/3QIm3dT cat8 ethernet cable: amzn.to/4fiFTqs NETGEAR 12-Port 10G Plus Switch (XS512EM): amzn.to/3OH7o22 NETGEAR ORBI 970 Series (2 Pack): amzn.to/3S68oja As an Amazon Associate I earn from qualifying purchases
I find your result quite odd. I don't see how TCP can achieve 9.4Gbps if you have 20+% packet loss - 9.4Gbps is around 6% for overhead and packet loss. I would have like to see the result of the "netstat -s" command between each run to double check what iperf3 is indicating for packet loss. Also, if the NICs have negotiated 10Gbps, then they are always sending data at 10Gbps rate. The BW setting in iperf3 limits the time interval that data is transmitted so that the average throughput is less than 10Gbps. At the end of the day, TCP throughput will cover 95+% of your traffic and is what matters. If you get 9.4Gbps, then you are making full use of your 10Gbps link. The error rate doesn't matter much unless it lowers the throughput significantly. The protocols are designed to handle packet loss and will either compensate to prevent data loss or don't care. FYI, Microsoft doesn't recommend using iperf3 on Windows to validate network performance.
Would love to see this same packet loss test at various bandwidths using 100' cat6, cat6a, "cat7", and cat8 (from a reputable seller that shows MHz testing that the cable actually is to cat8 spec) cables. Just to see how much better (or the same) packet loss is with each type at 2.5gb, 5gb, 10gb speeds. Id be very curious to see how much lower the packet loss is than Cat5e at the higher spec cables. What you are seeing here looks far worse than what I have always heard was typical.
Looks like the listing titled "Cat8 Ethernet Cable 100Ft Round S/FTP Outdoor&Indoor Heavy Duty High Speed Cat8 LAN Network Cable 40Gbps 2000Mhz" by seller "LEKVKM" shows a picture with Fluke testing data that the cable meets Cat8 spec. That is a pretty good priced cable to use for a test
@@LordSaliss you'll really only see the difference at 200 to 300+ feet at that point. I'd really like to see if Cat6a, 7, and 8 keep 10Gig speeds and low packet loss at 300ft. That would be the test to do.
Thanks for the video. The results are as expected. Cat5e can handle 1 & 2.5Gb. So the thousands of homes and building with Cat5e will be good for a few years yet.
@ That’s a great question. There are standards and procedures which need specialized equipment to perform the test. One idea to mimic a situation one could encounter in a common household. For example to place the Ethernet cable next to a device and wiring that we know makes radio interference. Think about a long extension cable (plugged inn) cheap power supply and some household appliances with a motor not to forget a microwave oven. While totally unscientific test with an AM radio to hear the noise generated. Make sure to ground the network cable on one side to that the shielding can bleed the energy to ground. And do your practical test to show speed and packet loss.
Yes, this video somehow ruins the joy of the previous ones. I'm curious about the connection configuration. All lines are available in one direction 4x2 or 2x2 in both directions. The cables have 4 pairs and can be configured with connection protocols. Despite the loss, I'm curious about how much time difference there would be if we were to transfer 100GB in one direction with cables of different lengths. Because transfer programs have detection of lost packets and would resend them. The transfer time here is defined by the transfer speed despite the losses and whether it would be clearly visible.
Could you also test iperf3 using the -R flag? I found it somehow results in very different packet drop numbers when the data flow direction is reversed
Can someone explain to me why iperf3 caps out at 9.4 Gbps, while the speedometer or what was its name in the previous videos was able to do 9.7 - 9.8 Gbps ? Second thing I don't understand - how is that 9.4 Gbps computed if it's TCP and there's 20% packets lost ? Is it just a measure of packets sent, with no check of packets not arriving ? Isn't TCP forcing a resending, effectively reducing the bandwidth (that should be visible in the speed test) ?
I'm assuming it's processing overhead that causes the slightly lower speeds. i could be wrong but iperf3 is doing more than sending blank data like a normal speed test would.
I can't remember if it did, you can see the ping in the other video where I used openspeedtest server. th-cam.com/video/zACK7Aka4Us/w-d-xo.html It should be towards the end of the video
Probably not a bad guess that he’s using coiled up cables again and that’s why it took so long to connect. I thought I had made it pretty clear that coiled up cables act as a low-pass filter, meaning the signal is attenuated more and more the higher the frequency is. And this is precisely why in his earlier tests it would only negotiate 5 Gbps links, because when you start filtering out the signal that it needs to operate properly then it obviously cannot operate properly. I was very surprised to see that the same speed issues existed in this newer video since he should’ve learned from the last one that coiled up cables are not good. 🤷♂️
Can't remember what my results were but I needed to run iperf3 as multi threaded to get 10G with asus 10G rj45 cards. setup i9-12900k client and i7-6700k for server. I think 6700k was little bottleneck for single thread. I think I need to rerun this test and verify do I remember right.
You bought random cables off the Internet. Were these cables tested with a correctly calibrated device? Did they come with a certificate proving their quality?
10Gbps (while wasn't common) existed at the time, so they had to atleast support it. But all the optimizations and designs were made specifically for 1Gbps. The specification for 10Gbps doesn't support 5e or 6, it supports 6A minimum. (Because the specification requires 100m of distance)
I did the cat6 cables with open Speedtest server, no packet drops already. If there’s enough interest, I can do it with iPerf3 as well. I bought some long variants for cat7 and cat8, I will be doing a cat5e, cat6, cat6a, cat7 and cat8 comparison at the longer lengths.
Very amateurish testing experiment. On a direct cable connection between 2 LAN machines, not a single packet should be dropped.. 1% packet loss on the internet is pretty crappy in real life. Check your facts first, as you seem to spread totally invalid statements.
From an average user viewpoint, isn't that fair ? I ask because I know that wireless can drop packets like crazy in a busy area. I don't have concrete numbers, but I think that 1% is extremely good in this context. I'm sure that an average Joe watching TH-cam and downloading an update on Steam won't know the difference between 1% and 0% packet loss. You could say that having 1% on a cable is waay too much, but the video was saying from the perspective of using it.
Obviously you were not taught to be helpful and advise if you believe you know better. It's unfortunate whomever raised you couldn't figure out how to help you be respectful.
If your watching a movie or something similar a 1% drop isn't going to matter much. The whole point of UDP is that it's fast at the cost of not negotiating, it's designed to be so fast that some packet drops won't really effect anything. We're not talking about TCP/IP that goes through an entire handshake, hence why it could never be that fast.
Product Link(s):
cat5e ethernet cable (15ft, 50ft, 100ft or 200ft options): amzn.to/3VY4ZEg
cat5e ethernet cable (35ft or 300ft options): amzn.to/49TbePz
cat5e ethernet cable (similar to one in video): amzn.to/41Ea6go
cat6 ethernet cable (similar to one in video): amzn.to/3ZVNeI6
cat7 ethernet cable: amzn.to/3QIm3dT
cat8 ethernet cable: amzn.to/4fiFTqs
NETGEAR 12-Port 10G Plus Switch (XS512EM): amzn.to/3OH7o22
NETGEAR ORBI 970 Series (2 Pack): amzn.to/3S68oja
As an Amazon Associate I earn from qualifying purchases
Just wanted to let you know Netgear released a new firmware on 12/26 for the Orbi 970.
Thanks for doing and sharing all these tests. Its been really interesting and informative.
I'm glad you found it useful.
Would love to see you compare Cat5e to Cat8 for interference/signal integrity for different lengths.
Great video. Thanks for the upload.
No problem!
Thank you for adding the summary graphs at the end of your videos!
Great video!
Happy New Year!!!
Thanks!
Happy New Year!
I find your result quite odd. I don't see how TCP can achieve 9.4Gbps if you have 20+% packet loss - 9.4Gbps is around 6% for overhead and packet loss. I would have like to see the result of the "netstat -s" command between each run to double check what iperf3 is indicating for packet loss.
Also, if the NICs have negotiated 10Gbps, then they are always sending data at 10Gbps rate. The BW setting in iperf3 limits the time interval that data is transmitted so that the average throughput is less than 10Gbps.
At the end of the day, TCP throughput will cover 95+% of your traffic and is what matters. If you get 9.4Gbps, then you are making full use of your 10Gbps link. The error rate doesn't matter much unless it lowers the throughput significantly. The protocols are designed to handle packet loss and will either compensate to prevent data loss or don't care.
FYI, Microsoft doesn't recommend using iperf3 on Windows to validate network performance.
Would love to see this same packet loss test at various bandwidths using 100' cat6, cat6a, "cat7", and cat8 (from a reputable seller that shows MHz testing that the cable actually is to cat8 spec) cables. Just to see how much better (or the same) packet loss is with each type at 2.5gb, 5gb, 10gb speeds. Id be very curious to see how much lower the packet loss is than Cat5e at the higher spec cables. What you are seeing here looks far worse than what I have always heard was typical.
Looks like the listing titled "Cat8 Ethernet Cable 100Ft Round S/FTP Outdoor&Indoor Heavy Duty High Speed Cat8 LAN Network Cable 40Gbps 2000Mhz" by seller "LEKVKM" shows a picture with Fluke testing data that the cable meets Cat8 spec. That is a pretty good priced cable to use for a test
Also, what NICs are you using on both ends of the iPerf test?
@@LordSaliss you'll really only see the difference at 200 to 300+ feet at that point. I'd really like to see if Cat6a, 7, and 8 keep 10Gig speeds and low packet loss at 300ft. That would be the test to do.
Thanks for the video. The results are as expected. Cat5e can handle 1 & 2.5Gb. So the thousands of homes and building with Cat5e will be good for a few years yet.
Yup
Thanks for this really interesting test
no problem!
Very good Information! Appreciated🎉
Happy to help!
Would be cool if you could test shielded cable (one side grounded!) with some interference generated next to it.
What would some common or easy way to "make" interference ?
@ That’s a great question. There are standards and procedures which need specialized equipment to perform the test. One idea to mimic a situation one could encounter in a common household. For example to place the Ethernet cable next to a device and wiring that we know makes radio interference. Think about a long extension cable (plugged inn) cheap power supply and some household appliances with a motor not to forget a microwave oven. While totally unscientific test with an AM radio to hear the noise generated. Make sure to ground the network cable on one side to that the shielding can bleed the energy to ground. And do your practical test to show speed and packet loss.
Yes, this video somehow ruins the joy of the previous ones. I'm curious about the connection configuration. All lines are available in one direction 4x2 or 2x2 in both directions. The cables have 4 pairs and can be configured with connection protocols. Despite the loss, I'm curious about how much time difference there would be if we were to transfer 100GB in one direction with cables of different lengths. Because transfer programs have detection of lost packets and would resend them. The transfer time here is defined by the transfer speed despite the losses and whether it would be clearly visible.
Could you also test iperf3 using the -R flag? I found it somehow results in very different packet drop numbers when the data flow direction is reversed
Great video.Thanks for this wonderful content.
no problem!
Can someone explain to me why iperf3 caps out at 9.4 Gbps, while the speedometer or what was its name in the previous videos was able to do 9.7 - 9.8 Gbps ?
Second thing I don't understand - how is that 9.4 Gbps computed if it's TCP and there's 20% packets lost ? Is it just a measure of packets sent, with no check of packets not arriving ? Isn't TCP forcing a resending, effectively reducing the bandwidth (that should be visible in the speed test) ?
I'm assuming it's processing overhead that causes the slightly lower speeds. i could be wrong but iperf3 is doing more than sending blank data like a normal speed test would.
Is there a difference in ping with the long cable, since you said it took 45 seconds to connect?
I can't remember if it did, you can see the ping in the other video where I used openspeedtest server. th-cam.com/video/zACK7Aka4Us/w-d-xo.html
It should be towards the end of the video
Probably not a bad guess that he’s using coiled up cables again and that’s why it took so long to connect. I thought I had made it pretty clear that coiled up cables act as a low-pass filter, meaning the signal is attenuated more and more the higher the frequency is. And this is precisely why in his earlier tests it would only negotiate 5 Gbps links, because when you start filtering out the signal that it needs to operate properly then it obviously cannot operate properly. I was very surprised to see that the same speed issues existed in this newer video since he should’ve learned from the last one that coiled up cables are not good. 🤷♂️
Can't remember what my results were but I needed to run iperf3 as multi threaded to get 10G with asus 10G rj45 cards. setup i9-12900k client and i7-6700k for server. I think 6700k was little bottleneck for single thread. I think I need to rerun this test and verify do I remember right.
Interesting
Are expecting cat 6 and 7 to drop less?
If the cables are of better quality then yes.
You bought random cables off the Internet. Were these cables tested with a correctly calibrated device? Did they come with a certificate proving their quality?
I’ve posted the links to the cables, they are advertised as cat5e cables
how is this possible if Cat 5e is only designed for 1Gbps?
10Gbps (while wasn't common) existed at the time, so they had to atleast support it.
But all the optimizations and designs were made specifically for 1Gbps.
The specification for 10Gbps doesn't support 5e or 6, it supports 6A minimum. (Because the specification requires 100m of distance)
That was my thought as well but it seems to be rated for that at the 100m distance.
Oh where are all the peeps that were saying you only need cat5e and cat6 etc were scams
For long distances, definitely need something better
Can you test Cat 7 vs 8? Maybe it’s too expensive lol
First the various Cat6 cables
I did the cat6 cables with open Speedtest server, no packet drops already.
If there’s enough interest, I can do it with iPerf3 as well.
I bought some long variants for cat7 and cat8, I will be doing a cat5e, cat6, cat6a, cat7 and cat8 comparison at the longer lengths.
@@landpet I would like to see iperf3 tests for the cat6 as well :) Well, for all cables :D
I'd tell you a UDP joke, but I don't know if you'll get it.
lol, I got the joke cause I didn't get the packet :)
Next time tell me a TCP joke so I'm sure to get it.
Very amateurish testing experiment. On a direct cable connection between 2 LAN machines, not a single packet should be dropped.. 1% packet loss on the internet is pretty crappy in real life. Check your facts first, as you seem to spread totally invalid statements.
From an average user viewpoint, isn't that fair ? I ask because I know that wireless can drop packets like crazy in a busy area. I don't have concrete numbers, but I think that 1% is extremely good in this context. I'm sure that an average Joe watching TH-cam and downloading an update on Steam won't know the difference between 1% and 0% packet loss.
You could say that having 1% on a cable is waay too much, but the video was saying from the perspective of using it.
Obviously you were not taught to be helpful and advise if you believe you know better. It's unfortunate whomever raised you couldn't figure out how to help you be respectful.
If your watching a movie or something similar a 1% drop isn't going to matter much. The whole point of UDP is that it's fast at the cost of not negotiating, it's designed to be so fast that some packet drops won't really effect anything. We're not talking about TCP/IP that goes through an entire handshake, hence why it could never be that fast.