You need to run the iPerf server on a wired connection. Wifi spectrum is shared so when you have both the client and server on the same network you will be killing performance.
@@AZisk what if you used your old non wifi 6E router? would the new wifi modules on the new macbooks help get higher speeds on routers that only support older standards
When running ipref between two clients connected over WiFi both clients have to share the airtime. So while one is transmitting, the other can not share the spectrum (outside of MiMo -- but even so, that's still sharing the spectrum space with the other client.) So with one device downloading at a time, from an upstream wired connection, the whole WiFi spectrum is dedicated to serving that one client making it much faster. As a general rule, you have 1 lot of spectrum giving 1 one level of bandwith and you divide that number by the clients using it at the same time. So 1/1 is the full connection speed, 1/2 is half connection speed for each client, 1/3 is a 33% and I think you get the picture.
Thank you for all your DEV oriented test, please don't forget to make a comparition between M1 Max, M2 Max, M2 Pro in Machine Learning, it would be great if the comparition uses a BIG amount of data and RAM, the last test you made where very small using the clasic CIFAR-10 training that is a beginners model , maybe a VGG16 traning from scrutch could be a better way of testing, VGG16 needs alot of RAM and verely fits in 8GB of VRAM like my RTX-3070, this is were Apple products shine in comparison with NVIDIA GPUs, the RTX-4090 has 24GB of RAM, but with unified memory we can work with BIG models that requires more than 32B of RAM.
I think for software people, networking is a classic "if it works, gtfo!". My place had some networking issues and rather than fixing it for real, threw RaspAP on a Pi and my goodness - it just does the job! As long as it works, hey?
The Unify enterprise access point has a 2.5GB network connection, for better results between clients you should connected it to a 2.5GB switch like the Unify Switch Enterprise 8 Poe, that can also provide power to access point itself. With the right configuration and with all WIFI6E clients connected to the 6GHZ band you should get speeds well above 1GB between those clients, your connection to the internet will stay the same as you are caped by yours ISP at 1.GB regardless of a wired or wireless connection.
Don't forget that it depends on the wireless router your connecting to. A friend of mine lives in Ashburn VA and using an ASUS RT-AX92U works great. We have FIOS here and the new FIOS routers give us 325 to 370. If you have the FIOS app on your phone run a speed test from there and see what you get. As always OUTSTANDING video's.
Would be interesting to see the PyTorch benchmark between m1 max and m2 max. Other benchmarks showed massive gains in 3d gpu Rendering (Blender), let’s see if ML also has such improvements or not. Thanks for the reviews! Always great to see more of a developer focused reviewer!
I've got 10GBit to my nas over cat6 and its really fast, but the thunderbolt adapter is loud and hot, so use it only when I need it. 2,5 Gb is the cheap sweet spot for now. And wireless is just so convenient for most things
@mrcoffee laplace I would vouch for 32GB if you are dead set on doing Deep Learning on M chips. The memory compression for large models is nowhere as good as Nvidia implementation.
This very Monday we got 5G fibers (on a 550mb plan) and I'm getting around 310-330 download with my m1 mini on wireless, and around 100 for upload (don't worry, it was around 6-7 on 4G). But the thing is that most sites aren't capable of sending hundreds of megabytes per second for so many clients simultaneously...
I upgraded my router about a year and 8 months ago to a wifi 6 router. I don't remember what the previous router was. I didn't realize what a difference it would make. It was mind blowing 🤯
Hi, nice vídeo, I think the bigest problem is about usb port speed. cuz in theory we could use gigabit ethernet, but I don´t trust on that, in other side, you have Wifi 6E, it uses 2 pci express lane. this is why your speed test is better on wifi. Forgive my poor english.
please press option + wifi icon, it gives you more information about the wifi connection...you can see how strong and how much bandwidth your wifi connection have. Since you can't tell at this moment whether you MacBooks have a good or comparable wifi connection.
Shameful Network 🙂 but awesome Upload speeds (I have 1Gb cable modem so upload is slow). What are you using for a Firewall to protect yourself? Highly recommend a small Fortinet - 40F or maybe a 60F.
What channel and channel bonding are you using? And you won’t get over 1Gibit over Wi-Fi to your iPerf server when your server is on 1gbit network. Need to use multi streams over iPerf
You cant measure properly wifi 6e if the whole connection to the server isn’t at least 2,5G. From Macbook to Macbook on the same AP isn’t a proper test either, because the AP has to share the Wifi time or antennas (depending if both macbooks are in a 2x2 configuration and the AP is in an 4x4 Configuration or not) Additionally the AP doesn’t have the CPU or Offloading power to allow full throughput for 2 devices. Otherwise the U6-Enterprise is actually the right AP to test with! So thats great at least, but you still have to do a proper test to a server with no switches in between that can’t do at least 2,5G… And yeah, for the unifi switch: USW-Enterprise-8-PoE for 500usd xD Thats actually in my opinion even the best switch out there, alternatively there is an Hasivo S600-5GT, which costs almost nothing, like 100usd, but there is no management, means no vlan config etc… And the next step up in Price is the Netgear MS108UP which uses the same realtek chip as the Hasivo, just costs almost 400usd… Netgear is just crap, was always and will never change, always the worst hardware for a price that looks okay, but if you check whats inside and you know a bit what similarily priced hardware uses, you’ll realize what netgear sells. Including that they have no knowledge of firmware programming or networking xD However in short, you have to do the unifi update for the sake of the video xD PS: If you find sth from Microtik or TP-Link, those 2 are my favorites actually. TP-Link was a long time ago a really cheap stuff crap company, but hell, they made it, i do like what they do at the moment. Cheers
Hi. Can you please test Stable Diffusion on DiffusionBee? I have an m1 macbook and it was more than enough for me for 2 years, but now it's choking and losing battery like crazy on an image generation tasks (3 hours of work vs 11 before, each image takes from 2 to 10 minutes to create). I want to know is M2 Max any better? AI art is trendy nowadays, I think many people are curious, you can get some views.
@@AZisk Ok, but when you do your "wired" iperf3 tests, the client AND the server should be wired, not just the client if you want a fair comparison. But yea, WiFi 6E seems quite fast, the problem with high frequency is at long distance and through many obstables or foreing noise, the speed will drop dramatically.
@@joezuu honestly can’t believe people like this exist. “There is a version with a bigger number ergo I need a bigger number”. Apple thrives on people like this running google docs and safari on 96gb ram
@@Philosophism995 it's really funny to read your comment, you don't event know me lol. I am actually a software engineer running many local kubernetes clusters inside many devcontainers on vscode, and also each of of them have completely isolated db's running in a docker compose "docker-in-docker", Also I have several apps on ios simulator running with react native to connect to my different graphql api's being port forwarded so they are available in my localhost to my simulators and real in network ios device. I hope I read your answer to this real soon! probably something like: "ThIs Is N0T true" or "Wh4tEvEr" :p but if you know what I am talking about (without searching) you have my respect.
a question please i really want to buy a m2 pro 16" im a front end deveoper, it would be good to buy the extra 16 GB RAM? please , because i will have this laptop like for 5 years (for the price)
+1 for the more, the better. Who knows what you will do in 5 years. You may switch to some more resource-demanding tech stack one day, and RAM will become the bottleneck.
Wow, makes me feel better about my 7 year old galaxy 5 phone getting 400 MB/s on my gigabit Ethernet connection. I think it's running Wi-Fi 4 or 5. Lol!
It's not about absolute values. It's about comparing old vs new models. Alex mentioned that somewhere at the beginning of the video. There is a chance that M2 Macbooks would score more than 900MBps in your network.
You need to run the iPerf server on a wired connection. Wifi spectrum is shared so when you have both the client and server on the same network you will be killing performance.
yes, tried that after the vid and got same result
@@AZisk what if you used your old non wifi 6E router? would the new wifi modules on the new macbooks help get higher speeds on routers that only support older standards
When running ipref between two clients connected over WiFi both clients have to share the airtime. So while one is transmitting, the other can not share the spectrum (outside of MiMo -- but even so, that's still sharing the spectrum space with the other client.) So with one device downloading at a time, from an upstream wired connection, the whole WiFi spectrum is dedicated to serving that one client making it much faster. As a general rule, you have 1 lot of spectrum giving 1 one level of bandwith and you divide that number by the clients using it at the same time. So 1/1 is the full connection speed, 1/2 is half connection speed for each client, 1/3 is a 33% and I think you get the picture.
Thank you for all your DEV oriented test, please don't forget to make a comparition between M1 Max, M2 Max, M2 Pro in Machine Learning, it would be great if the comparition uses a BIG amount of data and RAM, the last test you made where very small using the clasic CIFAR-10 training that is a beginners model , maybe a VGG16 traning from scrutch could be a better way of testing, VGG16 needs alot of RAM and verely fits in 8GB of VRAM like my RTX-3070, this is were Apple products shine in comparison with NVIDIA GPUs, the RTX-4090 has 24GB of RAM, but with unified memory we can work with BIG models that requires more than 32B of RAM.
I'm a MBP using dev and his content rules
I think for software people, networking is a classic "if it works, gtfo!". My place had some networking issues and rather than fixing it for real, threw RaspAP on a Pi and my goodness - it just does the job! As long as it works, hey?
yep - i see doubling in my wifi speed, i’m happy
network debugging is a pain the soul that we avoid as much as possible
The Unify enterprise access point has a 2.5GB network connection, for better results between clients you should connected it to a 2.5GB switch like the Unify Switch Enterprise 8 Poe, that can also provide power to access point itself. With the right configuration and with all WIFI6E clients connected to the 6GHZ band you should get speeds well above 1GB between those clients, your connection to the internet will stay the same as you are caped by yours ISP at 1.GB regardless of a wired or wireless connection.
Don't forget that it depends on the wireless router your connecting to. A friend of mine lives in Ashburn VA and using an ASUS RT-AX92U works great. We have FIOS here and the new FIOS routers give us 325 to 370. If you have the FIOS app on your phone run a speed test from there and see what you get. As always OUTSTANDING video's.
Would be interesting to see the PyTorch benchmark between m1 max and m2 max. Other benchmarks showed massive gains in 3d gpu Rendering (Blender), let’s see if ML also has such improvements or not. Thanks for the reviews! Always great to see more of a developer focused reviewer!
I've got 10GBit to my nas over cat6 and its really fast, but the thunderbolt adapter is loud and hot, so use it only when I need it. 2,5 Gb is the cheap sweet spot for now. And wireless is just so convenient for most things
Hi Alex
How are you?
I’m a regular visitor on this channel
Keep doing the great work, this world needs more of it.
Amazing video as usual :) can you please do a M2 max vs RTX for ML training?
i’m planning an ML video shortly
@mrcoffee laplace I would vouch for 32GB if you are dead set on doing Deep Learning on M chips. The memory compression for large models is nowhere as good as Nvidia implementation.
Thank you for your tests!
This very Monday we got 5G fibers (on a 550mb plan) and I'm getting around 310-330 download with my m1 mini on wireless, and around 100 for upload (don't worry, it was around 6-7 on 4G). But the thing is that most sites aren't capable of sending hundreds of megabytes per second for so many clients simultaneously...
Best to the point comparisons, thanks a lot! Looking forward to unity
Glad you enjoyed it!
Great as always! Would love to see M2 Max compared to Nvidia too with Unity. That would be very interesting!
Can you run some docker build tests on the new M2 Pros?
coming soon
I upgraded my router about a year and 8 months ago to a wifi 6 router. I don't remember what the previous router was. I didn't realize what a difference it would make. It was mind blowing 🤯
Hi, nice vídeo, I think the bigest problem is about usb port speed. cuz in theory we could use gigabit ethernet, but I don´t trust on that, in other side, you have Wifi 6E, it uses 2 pci express lane. this is why your speed test is better on wifi. Forgive my poor english.
please press option + wifi icon, it gives you more information about the wifi connection...you can see how strong and how much bandwidth your wifi connection have. Since you can't tell at this moment whether you MacBooks have a good or comparable wifi connection.
omg shame on you for that networking!!! but for real i feel admiration sr! you look really real and no BS please keep doing your videos like this!!
I appreciate that
Alex to Enterprise... - Energize!
that’s so lazy, just walk up a few decks
Wow there's so many macbooks around. How easy or difficult is it to resell those? assuming you're not keeping them all 😅
I think he returns them😂
some are easier to sell than others. :)
But Alex! Cable management! XD
yeah, not the best
Shameful Network 🙂 but awesome Upload speeds (I have 1Gb cable modem so upload is slow). What are you using for a Firewall to protect yourself? Highly recommend a small Fortinet - 40F or maybe a 60F.
Thanks for the video. Can you also do a stable diffusion test using apple's core ml library? It's supposed to be faster than normal stable diffusion
What channel and channel bonding are you using? And you won’t get over 1Gibit over Wi-Fi to your iPerf server when your server is on 1gbit network. Need to use multi streams over iPerf
Did you tested M1 Max and M2 Max on only 5Ghz to see if there is improvement without needing Wifi6E?
You linked the wrong model of the hub, do you have the right link? I'm searching for a good dock to buy.
check the link called my gear
You cant measure properly wifi 6e if the whole connection to the server isn’t at least 2,5G. From Macbook to Macbook on the same AP isn’t a proper test either, because the AP has to share the Wifi time or antennas (depending if both macbooks are in a 2x2 configuration and the AP is in an 4x4 Configuration or not)
Additionally the AP doesn’t have the CPU or Offloading power to allow full throughput for 2 devices.
Otherwise the U6-Enterprise is actually the right AP to test with! So thats great at least, but you still have to do a proper test to a server with no switches in between that can’t do at least 2,5G…
And yeah, for the unifi switch: USW-Enterprise-8-PoE for 500usd xD
Thats actually in my opinion even the best switch out there, alternatively there is an Hasivo S600-5GT, which costs almost nothing, like 100usd, but there is no management, means no vlan config etc…
And the next step up in Price is the Netgear MS108UP which uses the same realtek chip as the Hasivo, just costs almost 400usd…
Netgear is just crap, was always and will never change, always the worst hardware for a price that looks okay, but if you check whats inside and you know a bit what similarily priced hardware uses, you’ll realize what netgear sells. Including that they have no knowledge of firmware programming or networking xD
However in short, you have to do the unifi update for the sake of the video xD
PS: If you find sth from Microtik or TP-Link, those 2 are my favorites actually. TP-Link was a long time ago a really cheap stuff crap company, but hell, they made it, i do like what they do at the moment.
Cheers
did you try Ethernet on the M2s? ;-)
4:55 using different server in Speedtest?
Enjoy the speed.
Hello can you review m2 pro with low power mode like how much watt it use and performance it decrease also battery life compared to m1 pro? Thanks
Hi. Can you please test Stable Diffusion on DiffusionBee? I have an m1 macbook and it was more than enough for me for 2 years, but now it's choking and losing battery like crazy on an image generation tasks (3 hours of work vs 11 before, each image takes from 2 to 10 minutes to create). I want to know is M2 Max any better? AI art is trendy nowadays, I think many people are curious, you can get some views.
Isn’t WiFi 7 Coming Out soon ?
You wired only one laptop (the client) but not laptop that acts as the server for iperf3 test. So how can you get that bump in speed?
the server is connected via wifi6e to the 6Ghz network, so should be faster
@@AZisk Ok, but when you do your "wired" iperf3 tests, the client AND the server should be wired, not just the client if you want a fair comparison.
But yea, WiFi 6E seems quite fast, the problem with high frequency is at long distance and through many obstables or foreing noise, the speed will drop dramatically.
So should I buy it because of the 96gb ram or not?
LOL! If you don't know that you need it, you don't need it. But if you have the money, why not?
@@joezuu honestly can’t believe people like this exist. “There is a version with a bigger number ergo I need a bigger number”. Apple thrives on people like this running google docs and safari on 96gb ram
@@Philosophism995 it's really funny to read your comment, you don't event know me lol. I am actually a software engineer running many local kubernetes clusters inside many devcontainers on vscode, and also each of of them have completely isolated db's running in a docker compose "docker-in-docker", Also I have several apps on ios simulator running with react native to connect to my different graphql api's being port forwarded so they are available in my localhost to my simulators and real in network ios device. I hope I read your answer to this real soon! probably something like: "ThIs Is N0T true" or "Wh4tEvEr" :p but if you know what I am talking about (without searching) you have my respect.
@@joezuu I have the money.
a question please i really want to buy a m2 pro 16" im a front end deveoper, it would be good to buy the extra 16 GB RAM? please , because i will have this laptop like for 5 years (for the price)
imo get as much ram as you can
@@AZisk thanks for your reply !!! i Will
+1 for the more, the better.
Who knows what you will do in 5 years. You may switch to some more resource-demanding tech stack one day, and RAM will become the bottleneck.
Alex, I think you posted your hub, not the dock.
Wow, makes me feel better about my 7 year old galaxy 5 phone getting 400 MB/s on my gigabit Ethernet connection. I think it's running Wi-Fi 4 or 5. Lol!
I dunno what's wrong with your "old" machine but my M1 pro gets 800-900MBps via WiFi....
nothing wrong with the machines. likely the network
It's not about absolute values. It's about comparing old vs new models. Alex mentioned that somewhere at the beginning of the video.
There is a chance that M2 Macbooks would score more than 900MBps in your network.
@@valentinkhomutenko6308 exactly this. 😊
It's time to renew Airports
If you don't need all of these macbooks, I would be happy to help you :'D
if you dont change ur service order, you are most likely still testing on wifi even if you have ur usbc/eth plugged in 4:50 (i think)
When one Mac pro is not enough 😅