I should probably clarify the distinction in my criticisms between what you might call the "line encoding overhead" vs "regular overhead" (Not official terms but you know what I mean). My main criticism about the naming scheme is *only* with the difference between: 1 The advertised speed (such as 5 Gbps), and 2 The raw signaling rate minus line encoding overhead (which would be 4 Gbps in the case of the 5 Gbps rating). And I criticize only this difference specifically because the line encoding overhead is constant and and could be accounted for if they wanted, like has been done with network speed ratings. Then there is also the "regular overhead". And while I of course still wanted to talk about that in order to explain the rest of the difference, my criticisms do not extend to accounting for the regular overhead in the naming scheme. I probably should have used a better word than "misleading" at 6:39, because while I'd say it's linguistically true, I think it wrongly implies fault by the USB-IF in not accounting for the entire 12% difference, while really only 3% is because of the line encoding overhead, and the rest is regular overhead. Probably "confusing" would make more sense. But again I more meant to describe the regular overhead difference in a purely informative sense.
Thank you for providing English subtitles. They are very important for UA-cam to be able to translate correctly. For example, I speak Portuguese. UA-cam's speech recognition doesn't generate very good subtitles in real time and the translation in this case is even worse. I'm always very happy when there are native English subtitles. As for the video, wonderful, I had no idea about all this USB confusion. 😅 Greetings from Brazil. 👏🇧🇷
Even though I am native born (new york city), I too appreciate the real English language subtitles. When I watch other videos, a lot of the hosts these days are adding background music to their narrations. Why, you ask? Beats me, as it is totally annoying and useless. So, I turn on the closed captioning and what do you get? Well, when the host is a pilot speaking English as a second or maybe third language, the captioning has a hard time, especially for aviation terms. For example, it will write 'four' when the speaker said 'fore' as in the front of the plane. Sure, I know what it meant, but I can certainly understand that others might not catch on right away. So, I am glad ThioJoe does not use any music in his videos and he has real captions.
8b/10b is actually amazing. For the extra data you get parity, DC balance, PLL synchronization, and out of band control characters. Simply amazing. 128/132 will most likely be well behaved but it is not guaranteed. Don't knock encoding schemes unless you know what problems they are trying to solve. Yup, electrical engineer here.
If I remember correctly, the reason that networking speeds are reported as actual speeds is a legal reason because in some countries including the US, if I'm remembering correctly, networking is mandated to deliver within something like 2% of advertised
When the advertising is "Up to 20Mbps" and you only get a max of 3 Mbps (suburban London on broadband, compared to 33Mbps on "fibre" and "full fibre" hasn't rolled out yet)
@@bootmii98 in the UK we only have Broadband (ADSL) and "fibre" (TTC). They're only just rolling out "full fibre" (TTP) now. To be fair I'm also a few miles from the phone exchange so that will affect things as well.
Bandwidth sharing is a major issue people come up against. Dedicated USB channels are expensive to produce, so a lot of motherboards share bandwidth instead of providing dedicated ports. It's why there's such price disparity with PCI-e USB cards, where some that have more ports are cheaper than those that have less ports but don't share bandwidth across each port.
You can't ignore flow control, framing, error correction and protocol overheads from one but count them all against another. PCIe 3.0 and later still have all of those overheads in addition to line encoding overhead, so the practical througput is only around 80% of the line rate. This means a PCIe 5.0 SSD won't get much faster than 100Gbps despite having a 4·32Gbps or 128Gbps connection. Even the TCP/IP commonly used with ethernet is only 94% efficient with the default packet size while 10GbE and faster can overwhelm less powerful systems. Ethernet routers and switches can also be limited by the total packet per second throughout, so very many small packets won't make full use of the link rate
@@killerb255 I find going to jumbo frames prematurely can mask other issues like interrupt scaling, excessive copying, inefficient ACLs, etc. Better to get it near the line rate on the standard MTU using iperf3, then maybe increase it for storage networks. It's also OK to go a little bigger when encapsulating with VPN or VxLAN so the inner frames don't fragment
Wrong, 80% (or even less) was with PCIe 1.0/2.0, actual 3.0/4.0/5.0 can max out at ~93% of the line rate. So the best Gen5-SSDs in the future could reach ~119 Gbps or ~14,9 GB/s. (and yes, my best PCIe3.0 result is 3740 MB/s, PCIe4.0 ~7480 MB/s, so the future Gen5-SSDs could probably reach ~14950 MB/s).
Super informative. And yeah, USB 2.0 equals 35 MB/s max. is what you see in practice. There are so many stupid ports, cables, adapters and enclosures which bottleneck your chain to 2.0 that I'm still painfully aware. Getting a full setup from Mainboard to drive with 3.1 speeds is a challenge.
I dont even get 35MBS on my USB 3 sandisk 32gb stick, its really weird, it goes into the blue usb 3 spots but I get terrible speeds for transfers. I would be more than happy with USB 2 speeds lol
@@memcmeepants2392 that's just cause of the really slow flash controller and/or flash memory chip. if you want speed, look for a usb3 to nvme adapter/enclosure, then slap a fast nvme ssd in it.
This video reminded me of the incredible consternation I have over usb cables and how the overwhelming majority of them have no kind of markings or product codes or anything that can be looked up and then tell you what it does and does not support. I don't understand how this problem, in 2023, after how many years of usb existing, still exists! Oh sure, the blue color coding on the end of a USB cable plug could tell you it may at least be usb 3.0 speed, but now, with USB-C being absolutely everywhere and basically being the standard expectation, you can't tell jack! The amount of USB-C cables that I've dealt with that were still only usb 2.0 OR could only support charging and not data or supports data but only slower charging, absolutely drove me crazy! The fact that every USB-C cable that exists doesn't at least minimally support usb 3.0 speeds is i think a crime, and/or just another piece of evidence of societies acceptance of low standards, and apathy.
You forgot the big difference between USB 2.0 and 3.0: half duplex vs full duplex. USB 2.0 shares the same wires for both directions, so traffic both from the host and from the device eat into the same 480 Mbps pool. With 3.0 there's a separate pair for each direction, so the host can be sending 5 Gbps in one direction at the same time as a device is sending 5 Gbps back.
I want to add another thing, the reason 8b/810 is done is because at the end of the data transmission (or while) the amount of zeroes and ones that pass through must to 50/50, if there are 10000 ones in a transmission, there MUST be 10000 zeroes in it.
I don't think that's true, because 8/10 encoding wouldn't be able to guarantee it in all cases. I think it's more to do with not having separate clock signals, and combining the clock and data. This is more efficient, because otherwise you would need an entire twisted pair carrying just alternating 1s and 0s for the clock, which would lose even more than 8/10 encoding. However, to achieve this, you do need the signal to change every so often, as pure ones or zeros would give no clock information. It doesn't need to change enough to give equal ones and zeroes, just enough to work out how fast the signal changes.
@@martineylesclock recovery is one thing. SATA, USB 3.x and SFPs are using series capacitors at receiver ends. If you feed only ones - capacitor will charge up and no signal would be passed after that. Having signal that cross bias voltage at frequent interval is required for such interfaces to work
@@martineyles the description of 8b/10b in the video is rather bad; it doesn't simply insert a few extra bits randomly, it actually maps each 8-bit data byte onto one or two 10-bit codewords. The encoding is specifically designed so that the number of 1s and 0s are always exactly equal, and there are no runs of 1s or 0s longer than I think 6, with the only runs of 6 being in comma characters that are used for frame sync. On the other hand, 64b/66b, 128b/130b, and 128b/132b use an LFSR scrambler to put a probabilistic bound on the disparity between 1s and 0s. The sync headers that are added contain a transition, so the max run length is guaranteed to be no longer than the size of the encoded symbol.
I'm pretty sure that I've read that the newest version of USB-C is 80/80 Gbit by default and the 120/40 configuration is meant for video signaling. I could be remembering wrong, though. Either way, I doubt we'll see anything use that in some time. Thunderbolt 80Gb meanwhile could be very useful for external GPUs.
@@b4ttlemast0rUSB4 is based on Thunderbolt 3 (many though not all TB3 devices are compatible with USB4), but Thunderbolt 4 is still proprietary to Intel.
@@b4ttlemast0r While that's the case mostof the time there some things that thunderbolt supports that USB 4 dosn't and the same goes the other way around two, so not every thunderbolt device will work with every usb 4 port even at the same speed
I really enjoy your content and the way you deliver it. Excellent balance between not too noob and not too expert. As an IT support person, i am still learning stuff every other videos and I thank you for that.
During the california tropical storm earlier this year, I tried backing up my dad's entire computer to an external hdd, but it was a custom built budget system from 2011 and had no usb 5 Gbps ports, only USB 2.0 ports. Took nearly 15 hours to back the entire thing up, and all the time I was wondering why I was only getting 30 MB/s instead of the full 480 mbps, when I knew the 8 tb hdd was capable of at least 100 MB/s. Maybe I could've backed up over the network, but we don’t have a nas and I didn’t have the time nor knowledge to set that up. Didn't know if macrium would support that either.
"Up to" and "as low as" are weasel phrases that are technically accurate, but misleading most of the time. I previously thought that the reason actual transfer speeds were lower was because of capacitance in the USB cable. This is why I watch your videos
I think USB has retransmission on data loss (just like TCP), so capacitance or other forms of interference which could cause data loss, will also lower your speed, in addition to the slowdowns discussed on the video.
The same thing also happens with SATA. SATA 3 "6 Gbit/s" is actually 4.8 Gbit/s or 600 MB/s (and also called SATA-600 apparently). In practice I get about 560 MB/s on my internal SATA drive and only about 450 MB/s on my external USB 3 drive. But I don't think I've really seen SATA being advertised as 6 Gbit, I was aware that it's about 500-600 MB/s, but I did believe that USB was actually the speeds they are, and they definitely are heavily advertised with those speeds. Edit: I wrote this before finishing the video
Ok, so the discrepancy is mostly the same as with CDs which use eight-to-fourteen modulation and have a max capacity of 846,720,000 bytes... but if the eight-to-fourteen modulation were negated, 1,481,760,000 bytes could be stored. Double the advetised capacity by using XA mode and not using 8-14 modulation. Is it viable to do so? No. But I think something similar may have been how Dreamcast managed to pack 1GB onto what was essentially just a CD. Hm... this... gives me a LOT to think about. Much information here.
I recall a discussion with some techie friends about USB 2.0 vs 2nd gen firewire Even though USB has a higher bit rate, firewire was a little faster transferring data
It's basically like saying that all cars can reach 500 km/h or 300 miles/h... just drop them out of a plane. And whoever came up with the USB naming scheme must be the smartest person(s) on the planet... of the apes.
Great review, I've watched it a few times and always go back for reference. Could you please let me know the maker/link of your cable tester to check is the resistor is present in a USB-C/USB-A cable? Thanks
TL;DW: it's basically that old marketing trick of bits vs bytes, where the lowest value is chosen to give the highest perceived number, effectively meaning you'll actually only get 1/10th the advertised speed in real use.
Love your explanation of every person in you audience and no wonder everyone loves you, almost there is always a few out there that I don't understand.
Curious if you can explain this.... I have a very old SSD, it is IDE, I put it into an IDE to USB2.0 enclosure... I tested it's data transfer, and actually got OVER 480mbps. The flash memory in the SSD is of course faster than BOTH the IDE bus and the USB2 bus, but... how is it able to exceed the speed of BOTH? And yes... is, I still have it, it still works. Well, doesn't work on macs as Apple apparently removed the ability to talk to IDE drives even over usb.
For example , i use a Kingston Datatraveler 80M , the label say on the package 200mb/s read mode , i really do those 200mb/s but the write speed is really poor like 15mb/s ........ on usb-c 3.2 interface i tried a few usb 3.2 gen 1 from usb-a and usb-c and most of the time the read speed advertise is good but the write speed is bad. From what i've tested , Teamgroup c155 was the best so far , 125mb/s read and 40mb/s write for 9$ . I cant wait to test usb-c 3.2 gen 2x2 Scam advertised at 20gb/s ( it should be around 2,5GB/s on a usb stick .....) Good video by the way :)
That explains why some things are slower than they appear. But, is it possible for them to bring more efficient coding to current and older USB/SATA and whatnot?
Not for older devices; the encoding is baked into the hardware at a very low level. Newer standards do use lower overhead line codes like 64/66 and 128/130. SATA has the added problem if being half duplex at the protocol level for backwards compatibility with parallel ATA. But there will be no more versions of the SATA spec, newer stuff will just use NVMe instead, including hard drives.
Maybe the ISPs are delivering the advertised bandwidth in the USA, but not in Germany, that's 100% sure. They actually advertise much higher speeds, then they must deliver. For example an advertised 100 Mbit/s connection according to the contract must deliver minimum 54 Mbit/s at almost every available ISP. They also give you a typical average speed, which is not binding them anyhow, but I must say I newer saw a connection delivering the true advertised speed in Germany, in most cases it's around 80% or less, or at least that's what I ever experienced.
I remember LTE in German countryside being unable to do 720p livestream. At home I have 1G GPON and this actually works per spec, but both locations I have tested against are directly connected to exchange routers
@@brylozketrzyn It's not rare, that in German countryside you don't even have 3G network or any cellphone service at all. I live in a small town in Germany, without any mountain, hill or a smallest natural obstacle in a 10km radius, however in most cases I have a maximum of EDGE (2G) speed at home (2nd floor) and sometimes a very weak 4G. It's a shame for Germany. Oh, and 3G is already completely shut down, because it's "old-school", plus Vodafone is going to shut down it's 2G network in 2025 too. Probably a great part of Germany is going to stay in the cellphone dark ages in a few years, because nobody really wants to invest money outside the largest cities.
@@Mikaci_the_Grand_Duke by "countryside" I meant "tourist destinations" like Hammeln etc. Aside from finance there is one social thing that is convenient excuse but I think you know it
Even more fun facts: every single one of eight wires in ethernet cabe is rated for 250MHz! How can you get 10GBit? Pack more bits into a single "sample", encode the stream so that statistically it looks more like a sine wave (no sharp voltage transitions), and pack lots of error correction.
Moral of the story. Reject USB. Embrace Thunderbolt. Coherent naming scheme and the cool features that USB leaves off as optional Thunderbolt requires.
Hey Thio, speaking of SATA... my old networking teacher once told me that if you have a SATA cable with one 90° ending it DOES matter which side you will plug into your SSD and which to your motherboard. I've been arguing with him that it's not true and I even tested it in my home plugging the cable both ways and measuring the speed, it was the same. But he said it could be different due to some "denoiser", "decoder" or whatever on the 90° angle side. So I've been wondering, could there be at least some amount of truth in his statement?
The explanation of how the 8b/10b line code works is rather misleading. It's not simply adding some extra bits here and there, the data that gets sent on the wire is completely different from the original data bits. It's more of a lookup table, where each of the 256 possible data bytes are encoded as specific 10-bit symbols, and there are some additional 10-bit symbols that are used for control. For 64/66, 128/130, and 128/132, the line code basically adds 2 (or 4 bits) every 64 or 128 data bits, but the data bits are also scrambled (inverted in a predictable/reversible way). Edit: Another thing that's a bit tricky to deal with is protocol overhead, which can vary quite a bit depending on the specifics of what is being sent over the link. I don't think any protocol would specify the expected throughput including protocol overheads due to the variability. For example, a TCP connection over 10G Ethernet with MTU 1500 is only going to run at 9.4 Gbps, with the difference coming from the packet headers that have to be sent. With PCIe in particular, the situation is rather bad as the maximum packet size is rather small, generally 256 or 512 bytes, so you take an even bigger hit to the throughput of actual payload data. It might be interesting to see a nice breakdown of various protocols, the advertised rate, and the actual rates before/after encoding. Another important consideration is duplex, in other words can data be send in both directions at the same time, or does the link have to be turned around to switch between sending and receiving? USB 2 only has one pair and operates in half duplex, which definitely doesn't help when doing mixed reads and writes on an SSD. Interestingly, SATA is also half duplex at the protocol level, despite having separate TX and RX lanes, for backwards compatibility with parallel ATA, which is half duplex. PCIe and USB 3 are full-duplex. Seems like USB 4 may actually support some sort of asymmetric mode, which is full duplex but with a different number of upstream and downstream lanes, which is useful when you have to send a bunch of video data in one direction. Old 10 Mbps Ethernet over coax is half duplex, but I think just about everything newer is full duplex. Interestingly, Gigabit Ethernet uses all four pairs for both TX and RX at the same time, using a combination of analog and digital techniques to separate the incoming and outgoing signals. I should also note that PAM-n is not exactly an encoding in the same sense of 8b/10b. PAM-n means you have n voltage levels, so PAM-3 has 3 levels, PAM-4 has 4, etc. With 4 levels, you can encode two bits per symbol. PAM-3 gives you 1.5 bits per symbol, which necessitates a more complex encoding scheme. PAM-n encodings usually also require the use of forward error correction (FEC), which consumes some additional overhead in the form of parity bits.
i've had 10gbe for about 11 years, rarely use anything less once it is on the nvme drives or ssd. i bring it in, with the fastest possible devices. last time i was concerned with speed was backing up blu rays with a bluray drive to hard drive, about 13 years ago. i still back up those blu rays, but just set and forget, not in a hurry.
For consumer it is like that in most cases. When you have datacenter (even single rack) it is essential to have higher speeds. My case: I got two datacenters 300kms apart connected via dedicated fiber (not so expensive these days). Then inside I have clusters using 25G interconnects to sync StorageSpaces Direct and 10G for backup purposes. I cannot risk scenario, when my sync queue gets too long, because that may cause loss of data. Backups have to be done overnight, it is typically 20TBs per cluster. Full backup gets done in around 6hrs. If I had 1Gbps I would be unable to perform them. And some use cases require even more than that
Why do flash drives slow down over time? I have a SanDisk Extreme Pro 3.0 that I bought when it was the speed champ, and I don't want to complain about the fact that it couldn't match its rated speed. I transfer mostly multi-gigabyte video files, and when new it's write transfer rate was somewhere between 125-200 MB/sec. Today, I'm lucky if it reaches 20 MB/sec. These are USB 2.0 speeds. I've tried formatting, changing the format (exFAT vs NTFS). I've since replaced this drive with a new Extreme Pro and it screams. What happens during a flash drive's life?
The speed drops with late versions of USB data transfers are all caused by bad cables. The spec is not on the cable but on the box which contains the cable. I don't buy any kind of Apple device or accessories because I strongly disagree with their "walled garden" approach. But I must admit that the best USB 4 cable on the market, 3 metres long and capable of the full 40 Gbps data transfer speed, is their Thunderbolt cable. It costs a small fortune, but it has 3 main chips at each end of the cable, and use triple shields wires. Actually it is monetarily convenient in a speed / price comparison, or the ability to work correctly in a very electromagnetically noisy environment, as it can be a recording studio or a testing lab. And, since I play a lot with power devices, the latest Apple 100+ Watt brick is the only one to be superior to the Ugreen brand - but it doesn't have the continuously variable voltage protocol (PPS) used by Samsung, which is in most Ugreen bricks. The only disadvantage of Apple bricks is their fuse/PTC input protection, so a slight input overvoltage blow the fuse for fire safety and the $ 100 power brick is gone forever, even if all the components are still good. Ugreen uses components and designs which are superior to those used by Anker; but Apple goes even higher for the USB4 cable and their last power brick. Thank you for all of your hard work Greetings Anthony
very informative, thank you for your efforts enlightening the people to be informed in this very dynamic yet standard tier of tech knowledge! Not only that, the clear explanation as to why this topic is as is, much appreciated
@Thio Should I install DOWNFALL mitigations on my 11900K & kill its performance like I did with my Kaby Lake years ago? Its a custom build used mainly for gaming, benching, surfing, music & video etc And why havent you done a "downfall" video yet?
For some people signalling rate and raw data rate can be misleading, for other it is just convenience. Personally I do have network equipment rated for 28 and 56GBps signalling but I know that it will push my precious ethernet frames at 25 and 50Gbps respectively. I also know the protocol overhead and other factors limiting raw data rate. Should it state that it is capable of "just" 3 and 6 GB/s?
I have been trying to figure out where that 28/56/112 sequence comes from. 25G Ethernet is 25.78125 Gbps, which is definitely not 28 Gbps. What protocols actually run at 28 Gbps?
Yeah, using external hard drives or flash drives over usb 2.0, i usually get around 30MB/s but on really rare occasions I see around 42MB/s. I wanted to know what are the conditions that make it go as high as 42MB/s an I could never find it, i thought it could be due to cable interference but my testings showed that that's not the case cause i would get around 30MB/s with good quality cables and even flash drives (that have no cables at all). It really seems to be completely random
Well, there is a difference - both technical/theoretical and practical - between signalling rate/signal transfer rate; raw throughput rate, and effective bandwidth..........
The theoretical max speed of a SATA SSD is up to 600Mbps. When connected directly to the computers SATA port, you could expect to see speeds closer to 550Mbps due to the SATA data transfer protocol. But going through a SATA to USB adaptor, it's very normal to see the speeds drop quite a bit more due to the added overhead of changing interface protocols. So, it doesn't matter how fast you make the USB protocol, when the limiting factor is the SSD drive itself.
Are there any native USB4 devices out there? Most I see are using USB 3.2 Gen 1 / Gen 2 and tunneling over a USB4 connection just like thunderbolt 3 / 4
Hmm I kinda figured, because I don't really see any difference in speed when dealing with files on a storage device. Of course SSD's are faster than HDD's, I don't expect I/O speeds to be ultra fast.
Don't even get people started by trying to explain why a 1TB HDD only has about 939~GB of storage. If only storage makers kept the normal Base 2 storage calculations the OS uses people wouldn't be so quick to get angry and leave bad reviews when they think HDD makers are lying to them about the size of their HDD. Windows (and other OSes except macOS since 10.6) use Base 2 where HDD makers use Base 10 so a 1000GB Base 10 HDD will always be displayed as about 939GB Base 2. When Apple released Mac OSX 10.6 and switched to Base 10 storage calculations people actually thought they were getting more storage thanks to Apple magic. To me the worst thing about USB is Type-C but not for the reasons you might think. My issue is that not all USB-C cables are USB 3.0. Even USB-C to C cables may not be USB 3.0. I have ones that are only USB 2.0. Another issue I have are Power Only cables that many times have no labels on the cable warning you of that.
Fascinating and thanks for laying all of this out. As an engineer who works with low production USB peripherals, I have had plenty of bad experiences. In my case, I really don't notice speeds, but reliability has been terrible. In fact, we used to call USB "USeless Bus" because of all the freezes, reboots, and corrupted data.
@5:52 and @10:30 - "network speed ratings like gigabit ethernet are true speed" Uh... Unless you have some magic ethernet that we don't know about, overhead applies here too, like the critisisms being leveled at the USB speeds. It might be rated higher than spec but the end result is the same as USB, you will never get the full speed that is advertised. People wondering why they are only able to get 940 mbit/s on their gigabit connection is still a regular question.
That's protocol overhead rather than signalling overhead. Ethernet advertise the speed after signalling losses are taken into account, and USB/SATA advertise the raw bits before the signalling losses.
@@itskdog Yeah, but that is splitting hairs as far as the end result. "Has one kind of loss vs two (or more)" doesn't change the fact that the stated rated isn't the rate customers can experience. Actual gig speeds are 6% less than stated. Here he is saying usb 4 10gig is 3% less so ethernet is worse that usb 4 (though he doesn't actually test anything to show what the max real-world transfer rate / what other losses might add in).
Unlike the line code which has a fixed overhead, protocol overheads are variable. And it also depends on exactly how you measure it - do you count the whole ethernet frame, or just the payload that's buried several headers deep?
interesting , I recently bought a 20 Gbps external drive, the speeds were very very slow so I bought a thunderbolt 4 cable to match my TB4 port on my surface pro 9 I found that if I attach it before starting the machine it doubles the speed over plugging it in when machine is on. just an interesting occurrence , but maybe some truth in there?
was actually wondering about it, as i get onl 40MBps on usb 2.0, while expecting 60.... (113Mbps on usb 3.1 g1), yet to try with samsung T7 1TB, this was a sandisk ultra 64GB
USB and HDMI naming schemes vs actual performance and features is atrocious. Particularly on USB naming, pretty much everyone is confused by it, in particular "regular" people that are not tech savvy which leads to immense waste when people buy devices and cables that don't actually work for their intended use case because they were confused when they made their purchase. It's so frustrating....
If Ethernet can do that, why can't other ports do the same? Counting only the final output that matters for the customer. I mean it doesn't matter if it will be 4 vs 3 or 3.2 vs 2.7. People would want to buy what's better anyway. So why lying? Misleading people is dumb, because people who like to make a drama would go and hate the company for that. It's already common for people and their Internet connection speed. They use some communion PC and are mad that they can't get 30Mbps speed. Well, of course they can't if their disk can't allow that. Buy a new PC and it would. The device capability does mater. If You are buying Porsche to ride in the city, You won't blame the car dealer that You can't drive 300km/s, because it's Your fault for using it where You have to drive much slower than that. This is exactly same situation.
I should probably clarify the distinction in my criticisms between what you might call the "line encoding overhead" vs "regular overhead" (Not official terms but you know what I mean). My main criticism about the naming scheme is *only* with the difference between: 1 The advertised speed (such as 5 Gbps), and 2 The raw signaling rate minus line encoding overhead (which would be 4 Gbps in the case of the 5 Gbps rating). And I criticize only this difference specifically because the line encoding overhead is constant and and could be accounted for if they wanted, like has been done with network speed ratings.
Then there is also the "regular overhead". And while I of course still wanted to talk about that in order to explain the rest of the difference, my criticisms do not extend to accounting for the regular overhead in the naming scheme. I probably should have used a better word than "misleading" at 6:39, because while I'd say it's linguistically true, I think it wrongly implies fault by the USB-IF in not accounting for the entire 12% difference, while really only 3% is because of the line encoding overhead, and the rest is regular overhead. Probably "confusing" would make more sense. But again I more meant to describe the regular overhead difference in a purely informative sense.
Thank you for providing English subtitles. They are very important for UA-cam to be able to translate correctly. For example, I speak Portuguese. UA-cam's speech recognition doesn't generate very good subtitles in real time and the translation in this case is even worse. I'm always very happy when there are native English subtitles. As for the video, wonderful, I had no idea about all this USB confusion. 😅 Greetings from Brazil. 👏🇧🇷
Even though I am native born (new york city), I too appreciate the real English language subtitles. When I watch other videos, a lot of the hosts these days are adding background music to their narrations. Why, you ask? Beats me, as it is totally annoying and useless. So, I turn on the closed captioning and what do you get? Well, when the host is a pilot speaking English as a second or maybe third language, the captioning has a hard time, especially for aviation terms. For example, it will write 'four' when the speaker said 'fore' as in the front of the plane. Sure, I know what it meant, but I can certainly understand that others might not catch on right away.
So, I am glad ThioJoe does not use any music in his videos and he has real captions.
I know youtube is worst
I am not native language and I apreciate the video using subtitles too.
@@VictorCampos87 very me
It's good that the USB branding all around is so consistent and never misleading😎👌
I can feel the sarcasm all the way to up here in the north. Oooooooh boy.
(I mean, I agree.)
The cool thing about standards?
So many to choose from!
After USB 3.2 Gen 2x2 we also get USB4, which does not have a space in between.
8b/10b is actually amazing. For the extra data you get parity, DC balance, PLL synchronization, and out of band control characters. Simply amazing. 128/132 will most likely be well behaved but it is not guaranteed.
Don't knock encoding schemes unless you know what problems they are trying to solve. Yup, electrical engineer here.
It kindof reminds me of gcr encoding as was used at some point by some manufacturors for disk storage.
He isn't knocking on encoding, he is knocking on the presentation of speeds by USB.
If I remember correctly, the reason that networking speeds are reported as actual speeds is a legal reason because in some countries including the US, if I'm remembering correctly, networking is mandated to deliver within something like 2% of advertised
Plus it makes sense
When the advertising is "Up to 20Mbps" and you only get a max of 3 Mbps (suburban London on broadband, compared to 33Mbps on "fibre" and "full fibre" hasn't rolled out yet)
@@itskdog wait only 33? on fibre? I've seen cable and VDSL do better, c'mon
@@bootmii98 in the UK we only have Broadband (ADSL) and "fibre" (TTC). They're only just rolling out "full fibre" (TTP) now.
To be fair I'm also a few miles from the phone exchange so that will affect things as well.
@@itskdogi wonder how would they name ftth then
Bandwidth sharing is a major issue people come up against. Dedicated USB channels are expensive to produce, so a lot of motherboards share bandwidth instead of providing dedicated ports. It's why there's such price disparity with PCI-e USB cards, where some that have more ports are cheaper than those that have less ports but don't share bandwidth across each port.
You can't ignore flow control, framing, error correction and protocol overheads from one but count them all against another. PCIe 3.0 and later still have all of those overheads in addition to line encoding overhead, so the practical througput is only around 80% of the line rate. This means a PCIe 5.0 SSD won't get much faster than 100Gbps despite having a 4·32Gbps or 128Gbps connection. Even the TCP/IP commonly used with ethernet is only 94% efficient with the default packet size while 10GbE and faster can overwhelm less powerful systems. Ethernet routers and switches can also be limited by the total packet per second throughout, so very many small packets won't make full use of the link rate
…and that’s where jumbo frames come in, but that’s a whole other can of worms there.
@@killerb255 I find going to jumbo frames prematurely can mask other issues like interrupt scaling, excessive copying, inefficient ACLs, etc. Better to get it near the line rate on the standard MTU using iperf3, then maybe increase it for storage networks. It's also OK to go a little bigger when encapsulating with VPN or VxLAN so the inner frames don't fragment
Wrong, 80% (or even less) was with PCIe 1.0/2.0, actual 3.0/4.0/5.0 can max out at ~93% of the line rate. So the best Gen5-SSDs in the future could reach ~119 Gbps or ~14,9 GB/s.
(and yes, my best PCIe3.0 result is 3740 MB/s, PCIe4.0 ~7480 MB/s, so the future Gen5-SSDs could probably reach ~14950 MB/s).
Super informative. And yeah, USB 2.0 equals 35 MB/s max. is what you see in practice.
There are so many stupid ports, cables, adapters and enclosures which bottleneck your chain to 2.0 that I'm still painfully aware. Getting a full setup from Mainboard to drive with 3.1 speeds is a challenge.
I dont even get 35MBS on my USB 3 sandisk 32gb stick, its really weird, it goes into the blue usb 3 spots but I get terrible speeds for transfers. I would be more than happy with USB 2 speeds lol
@@memcmeepants2392 that's just cause of the really slow flash controller and/or flash memory chip. if you want speed, look for a usb3 to nvme adapter/enclosure, then slap a fast nvme ssd in it.
This video reminded me of the incredible consternation I have over usb cables and how the overwhelming majority of them have no kind of markings or product codes or anything that can be looked up and then tell you what it does and does not support. I don't understand how this problem, in 2023, after how many years of usb existing, still exists! Oh sure, the blue color coding on the end of a USB cable plug could tell you it may at least be usb 3.0 speed, but now, with USB-C being absolutely everywhere and basically being the standard expectation, you can't tell jack! The amount of USB-C cables that I've dealt with that were still only usb 2.0 OR could only support charging and not data or supports data but only slower charging, absolutely drove me crazy!
The fact that every USB-C cable that exists doesn't at least minimally support usb 3.0 speeds is i think a crime, and/or just another piece of evidence of societies acceptance of low standards, and apathy.
You forgot the big difference between USB 2.0 and 3.0: half duplex vs full duplex. USB 2.0 shares the same wires for both directions, so traffic both from the host and from the device eat into the same 480 Mbps pool. With 3.0 there's a separate pair for each direction, so the host can be sending 5 Gbps in one direction at the same time as a device is sending 5 Gbps back.
On a related note, SATA is half duplex at the protocol level, despite having separate RX and TX pairs, for backwards compatibility with parallel ATA.
I want to add another thing, the reason 8b/810 is done is because at the end of the data transmission (or while) the amount of zeroes and ones that pass through must to 50/50, if there are 10000 ones in a transmission, there MUST be 10000 zeroes in it.
I don't think that's true, because 8/10 encoding wouldn't be able to guarantee it in all cases. I think it's more to do with not having separate clock signals, and combining the clock and data. This is more efficient, because otherwise you would need an entire twisted pair carrying just alternating 1s and 0s for the clock, which would lose even more than 8/10 encoding. However, to achieve this, you do need the signal to change every so often, as pure ones or zeros would give no clock information. It doesn't need to change enough to give equal ones and zeroes, just enough to work out how fast the signal changes.
@@martineylesclock recovery is one thing. SATA, USB 3.x and SFPs are using series capacitors at receiver ends. If you feed only ones - capacitor will charge up and no signal would be passed after that. Having signal that cross bias voltage at frequent interval is required for such interfaces to work
@@martineyles the description of 8b/10b in the video is rather bad; it doesn't simply insert a few extra bits randomly, it actually maps each 8-bit data byte onto one or two 10-bit codewords. The encoding is specifically designed so that the number of 1s and 0s are always exactly equal, and there are no runs of 1s or 0s longer than I think 6, with the only runs of 6 being in comma characters that are used for frame sync. On the other hand, 64b/66b, 128b/130b, and 128b/132b use an LFSR scrambler to put a probabilistic bound on the disparity between 1s and 0s. The sync headers that are added contain a transition, so the max run length is guaranteed to be no longer than the size of the encoded symbol.
I'm pretty sure that I've read that the newest version of USB-C is 80/80 Gbit by default and the 120/40 configuration is meant for video signaling. I could be remembering wrong, though. Either way, I doubt we'll see anything use that in some time. Thunderbolt 80Gb meanwhile could be very useful for external GPUs.
USB 4 80 Gbit and Thunderbolt 80 Gbit is basically the same. Anything that's USB4 is also Thunderbolt I think.
@@b4ttlemast0rUSB4 is based on Thunderbolt 3 (many though not all TB3 devices are compatible with USB4), but Thunderbolt 4 is still proprietary to Intel.
@@b4ttlemast0r While that's the case mostof the time there some things that thunderbolt supports that USB 4 dosn't and the same goes the other way around two, so not every thunderbolt device will work with every usb 4 port even at the same speed
A good clear explanation of the difference between signalling speed vs data speed. Excellent video.
I really enjoy your content and the way you deliver it. Excellent balance between not too noob and not too expert. As an IT support person, i am still learning stuff every other videos and I thank you for that.
That was a seamless transition to the sponsor. Well done!
During the california tropical storm earlier this year, I tried backing up my dad's entire computer to an external hdd, but it was a custom built budget system from 2011 and had no usb 5 Gbps ports, only USB 2.0 ports. Took nearly 15 hours to back the entire thing up, and all the time I was wondering why I was only getting 30 MB/s instead of the full 480 mbps, when I knew the 8 tb hdd was capable of at least 100 MB/s. Maybe I could've backed up over the network, but we don’t have a nas and I didn’t have the time nor knowledge to set that up. Didn't know if macrium would support that either.
I watched this video in 12.4 minutes but accounting for coffee overhead, scratching my nose, etc. my true watch time was 11:47.
"Up to" and "as low as" are weasel phrases that are technically accurate, but misleading most of the time. I previously thought that the reason actual transfer speeds were lower was because of capacitance in the USB cable. This is why I watch your videos
IMO Up to means that it somehow could be achieved.
I think USB has retransmission on data loss (just like TCP), so capacitance or other forms of interference which could cause data loss, will also lower your speed, in addition to the slowdowns discussed on the video.
The same thing also happens with SATA. SATA 3 "6 Gbit/s" is actually 4.8 Gbit/s or 600 MB/s (and also called SATA-600 apparently). In practice I get about 560 MB/s on my internal SATA drive and only about 450 MB/s on my external USB 3 drive. But I don't think I've really seen SATA being advertised as 6 Gbit, I was aware that it's about 500-600 MB/s, but I did believe that USB was actually the speeds they are, and they definitely are heavily advertised with those speeds. Edit: I wrote this before finishing the video
Yeah he said that in the video.
Ok, so the discrepancy is mostly the same as with CDs which use eight-to-fourteen modulation and have a max capacity of 846,720,000 bytes... but if the eight-to-fourteen modulation were negated, 1,481,760,000 bytes could be stored. Double the advetised capacity by using XA mode and not using 8-14 modulation. Is it viable to do so? No. But I think something similar may have been how Dreamcast managed to pack 1GB onto what was essentially just a CD. Hm... this... gives me a LOT to think about. Much information here.
I recall a discussion with some techie friends about USB 2.0 vs 2nd gen firewire
Even though USB has a higher bit rate, firewire was a little faster transferring data
I liked the ad for the charger, wish more of that popped up in more channels, instead of SquareSpace for example
My table of thumb:
Usb 2.0 ~= 20MB
Usb 3.1 5gbps ~= 200MB
Usb 3.2 ~= 750MB
It's basically like saying that all cars can reach 500 km/h or 300 miles/h... just drop them out of a plane.
And whoever came up with the USB naming scheme must be the smartest person(s) on the planet... of the apes.
Your sponsor is actually really useful.
Great review, I've watched it a few times and always go back for reference.
Could you please let me know the maker/link of your cable tester to check is the resistor is present in a USB-C/USB-A cable?
Thanks
Absolute unit of a thumbs up 👍🏻
Of course USBs wouldn't be at the speed of light 😂😂😂
yo fax
Speed of light is not time it’s a distance
@@Thegoal2.P yea ik
@@Thegoal2.PIt's distance over time.
@@Egg.Of.Glory999fax isn't either.
TL;DW: it's basically that old marketing trick of bits vs bytes, where the lowest value is chosen to give the highest perceived number, effectively meaning you'll actually only get 1/10th the advertised speed in real use.
You clearly haven't watched the video lol
TL:DW = Too long; Didn't watch :/@@BusAlexey
@@BusAlexey Except I did, and it was a very long-winded way of saying what I commented, hence TL;DW.
@@Abigblueworld "tldw" is meant as a summary of the video for those who don't want to watch. Not as an excuse
@@FlyboyHelosim thiojoe didn't mention what you have said even once, what are even talking about
Hello Thio, I think you missread NRZI 7:12 and 80 here 8:41 I love your videos Im a big fan
Love your explanation of every person in you audience and no wonder everyone loves you, almost there is always a few out there that I don't understand.
Curious if you can explain this.... I have a very old SSD, it is IDE, I put it into an IDE to USB2.0 enclosure... I tested it's data transfer, and actually got OVER 480mbps.
The flash memory in the SSD is of course faster than BOTH the IDE bus and the USB2 bus, but... how is it able to exceed the speed of BOTH? And yes... is, I still have it, it still works.
Well, doesn't work on macs as Apple apparently removed the ability to talk to IDE drives even over usb.
As someone who works low level with these protocols, this is a pretty good explanation
my 10 dollar USB claiming to have 500Gb per second and 1 petabyte of storage
They are pretty petty with that petabyte.
I like how this channel became Linus 2.0 in no time.
Uh, not quite.
the good part or the bad one ? 😂
@@mahdi9064 The good part, damn it. The good part. :))
you know the usb 3.0 flash drives
that have speeds of 10mb/s😂
I think I'm going to have to pay more attention to these kinds of protocols because I always assumed they were rated accurately.
For example , i use a Kingston Datatraveler 80M , the label say on the package 200mb/s read mode , i really do those 200mb/s but the write speed is really poor like 15mb/s ........ on usb-c 3.2 interface i tried a few usb 3.2 gen 1 from usb-a and usb-c and most of the time the read speed advertise is good but the write speed is bad. From what i've tested , Teamgroup c155 was the best so far , 125mb/s read and 40mb/s write for 9$ . I cant wait to test usb-c 3.2 gen 2x2 Scam advertised at 20gb/s ( it should be around 2,5GB/s on a usb stick .....) Good video by the way :)
That explains why some things are slower than they appear. But, is it possible for them to bring more efficient coding to current and older USB/SATA and whatnot?
Both devices need to have the same coding without any handshakes, so not really without violating the spec.
Not for older devices; the encoding is baked into the hardware at a very low level. Newer standards do use lower overhead line codes like 64/66 and 128/130. SATA has the added problem if being half duplex at the protocol level for backwards compatibility with parallel ATA. But there will be no more versions of the SATA spec, newer stuff will just use NVMe instead, including hard drives.
In the case of USB4, only gen 4 links use PAM-3, with an oscilloscope you wouldn't see 40 billions bits but rather about 25,6 billion trits
and also USB4 gen2x1, which is 10Gbps, use 64b/66b
Still hate that we still have to divide numbers by 8 to get the useful transfer speed
Maybe the ISPs are delivering the advertised bandwidth in the USA, but not in Germany, that's 100% sure. They actually advertise much higher speeds, then they must deliver. For example an advertised 100 Mbit/s connection according to the contract must deliver minimum 54 Mbit/s at almost every available ISP. They also give you a typical average speed, which is not binding them anyhow, but I must say I newer saw a connection delivering the true advertised speed in Germany, in most cases it's around 80% or less, or at least that's what I ever experienced.
I remember LTE in German countryside being unable to do 720p livestream. At home I have 1G GPON and this actually works per spec, but both locations I have tested against are directly connected to exchange routers
@@brylozketrzyn It's not rare, that in German countryside you don't even have 3G network or any cellphone service at all. I live in a small town in Germany, without any mountain, hill or a smallest natural obstacle in a 10km radius, however in most cases I have a maximum of EDGE (2G) speed at home (2nd floor) and sometimes a very weak 4G. It's a shame for Germany.
Oh, and 3G is already completely shut down, because it's "old-school", plus Vodafone is going to shut down it's 2G network in 2025 too. Probably a great part of Germany is going to stay in the cellphone dark ages in a few years, because nobody really wants to invest money outside the largest cities.
@@Mikaci_the_Grand_Duke by "countryside" I meant "tourist destinations" like Hammeln etc. Aside from finance there is one social thing that is convenient excuse but I think you know it
The signaling rate is basically relevant to cable and hardware designers, who need to know the bandwidth of the signal.
usb 2 at 40MB that im using is slow tbh, but usb 3 at 450MB is enogh
Can you also do a video about WiFi standards with explaination of its coding rate/line encoding rate?
Even more fun facts: every single one of eight wires in ethernet cabe is rated for 250MHz! How can you get 10GBit?
Pack more bits into a single "sample", encode the stream so that statistically it looks more like a sine wave (no sharp voltage transitions), and pack lots of error correction.
Moral of the story. Reject USB. Embrace Thunderbolt. Coherent naming scheme and the cool features that USB leaves off as optional Thunderbolt requires.
Hey Thio, speaking of SATA... my old networking teacher once told me that if you have a SATA cable with one 90° ending it DOES matter which side you will plug into your SSD and which to your motherboard. I've been arguing with him that it's not true and I even tested it in my home plugging the cable both ways and measuring the speed, it was the same. But he said it could be different due to some "denoiser", "decoder" or whatever on the 90° angle side. So I've been wondering, could there be at least some amount of truth in his statement?
USB/SATA to devices:
> It's not you, it's me.
Us about their relationship:
> It's complicated.
The explanation of how the 8b/10b line code works is rather misleading. It's not simply adding some extra bits here and there, the data that gets sent on the wire is completely different from the original data bits. It's more of a lookup table, where each of the 256 possible data bytes are encoded as specific 10-bit symbols, and there are some additional 10-bit symbols that are used for control. For 64/66, 128/130, and 128/132, the line code basically adds 2 (or 4 bits) every 64 or 128 data bits, but the data bits are also scrambled (inverted in a predictable/reversible way).
Edit: Another thing that's a bit tricky to deal with is protocol overhead, which can vary quite a bit depending on the specifics of what is being sent over the link. I don't think any protocol would specify the expected throughput including protocol overheads due to the variability. For example, a TCP connection over 10G Ethernet with MTU 1500 is only going to run at 9.4 Gbps, with the difference coming from the packet headers that have to be sent. With PCIe in particular, the situation is rather bad as the maximum packet size is rather small, generally 256 or 512 bytes, so you take an even bigger hit to the throughput of actual payload data. It might be interesting to see a nice breakdown of various protocols, the advertised rate, and the actual rates before/after encoding.
Another important consideration is duplex, in other words can data be send in both directions at the same time, or does the link have to be turned around to switch between sending and receiving? USB 2 only has one pair and operates in half duplex, which definitely doesn't help when doing mixed reads and writes on an SSD. Interestingly, SATA is also half duplex at the protocol level, despite having separate TX and RX lanes, for backwards compatibility with parallel ATA, which is half duplex. PCIe and USB 3 are full-duplex. Seems like USB 4 may actually support some sort of asymmetric mode, which is full duplex but with a different number of upstream and downstream lanes, which is useful when you have to send a bunch of video data in one direction. Old 10 Mbps Ethernet over coax is half duplex, but I think just about everything newer is full duplex. Interestingly, Gigabit Ethernet uses all four pairs for both TX and RX at the same time, using a combination of analog and digital techniques to separate the incoming and outgoing signals.
I should also note that PAM-n is not exactly an encoding in the same sense of 8b/10b. PAM-n means you have n voltage levels, so PAM-3 has 3 levels, PAM-4 has 4, etc. With 4 levels, you can encode two bits per symbol. PAM-3 gives you 1.5 bits per symbol, which necessitates a more complex encoding scheme. PAM-n encodings usually also require the use of forward error correction (FEC), which consumes some additional overhead in the form of parity bits.
i've had 10gbe for about 11 years, rarely use anything less once it is on the nvme drives or ssd. i bring it in, with the fastest possible devices. last time i was concerned with speed was backing up blu rays with a bluray drive to hard drive, about 13 years ago. i still back up those blu rays, but just set and forget, not in a hurry.
For consumer it is like that in most cases. When you have datacenter (even single rack) it is essential to have higher speeds. My case: I got two datacenters 300kms apart connected via dedicated fiber (not so expensive these days). Then inside I have clusters using 25G interconnects to sync StorageSpaces Direct and 10G for backup purposes. I cannot risk scenario, when my sync queue gets too long, because that may cause loss of data. Backups have to be done overnight, it is typically 20TBs per cluster. Full backup gets done in around 6hrs. If I had 1Gbps I would be unable to perform them. And some use cases require even more than that
Why do flash drives slow down over time? I have a SanDisk Extreme Pro 3.0 that I bought when it was the speed champ, and I don't want to complain about the fact that it couldn't match its rated speed. I transfer mostly multi-gigabyte video files, and when new it's write transfer rate was somewhere between 125-200 MB/sec. Today, I'm lucky if it reaches 20 MB/sec. These are USB 2.0 speeds. I've tried formatting, changing the format (exFAT vs NTFS). I've since replaced this drive with a new Extreme Pro and it screams. What happens during a flash drive's life?
The speed drops with late versions of USB data transfers are all caused by bad cables. The spec is not on the cable but on the box which contains the cable.
I don't buy any kind of Apple device or accessories because I strongly disagree with their "walled garden" approach. But I must admit that the best USB 4 cable on the market, 3 metres long and capable of the full 40 Gbps data transfer speed, is their Thunderbolt cable. It costs a small fortune, but it has 3 main chips at each end of the cable, and use triple shields wires. Actually it is monetarily convenient in a speed / price comparison, or the ability to work correctly in a very electromagnetically noisy environment, as it can be a recording studio or a testing lab.
And, since I play a lot with power devices, the latest Apple 100+ Watt brick is the only one to be superior to the Ugreen brand - but it doesn't have the continuously variable voltage protocol (PPS) used by Samsung, which is in most Ugreen bricks. The only disadvantage of Apple bricks is their fuse/PTC input protection, so a slight input overvoltage blow the fuse for fire safety and the $ 100 power brick is gone forever, even if all the components are still good.
Ugreen uses components and designs which are superior to those used by Anker; but Apple goes even higher for the USB4 cable and their last power brick.
Thank you for all of your hard work
Greetings
Anthony
very informative, thank you for your efforts enlightening the people to be informed in this very dynamic yet standard tier of tech knowledge! Not only that, the clear explanation as to why this topic is as is, much appreciated
@Thio Should I install DOWNFALL mitigations on my 11900K & kill its performance like I did with my Kaby Lake years ago?
Its a custom build used mainly for gaming, benching, surfing, music & video etc
And why havent you done a "downfall" video yet?
For some people signalling rate and raw data rate can be misleading, for other it is just convenience. Personally I do have network equipment rated for 28 and 56GBps signalling but I know that it will push my precious ethernet frames at 25 and 50Gbps respectively. I also know the protocol overhead and other factors limiting raw data rate. Should it state that it is capable of "just" 3 and 6 GB/s?
I have been trying to figure out where that 28/56/112 sequence comes from. 25G Ethernet is 25.78125 Gbps, which is definitely not 28 Gbps. What protocols actually run at 28 Gbps?
Hey Joe, I have an idea for a video: Tips and tricks for Windows.
Also has parity bits for error check and verification while data transfer
Note that linerates are specified in powers of 10, not in powers of 2.
Yeah, using external hard drives or flash drives over usb 2.0, i usually get around 30MB/s but on really rare occasions I see around 42MB/s. I wanted to know what are the conditions that make it go as high as 42MB/s an I could never find it, i thought it could be due to cable interference but my testings showed that that's not the case cause i would get around 30MB/s with good quality cables and even flash drives (that have no cables at all). It really seems to be completely random
Well, there is a difference - both technical/theoretical and practical - between signalling rate/signal transfer rate; raw throughput rate, and effective bandwidth..........
I'm just waiting for the new SATA-n standard to come out
The theoretical max speed of a SATA SSD is up to 600Mbps. When connected directly to the computers SATA port, you could expect to see speeds closer to 550Mbps due to the SATA data transfer protocol. But going through a SATA to USB adaptor, it's very normal to see the speeds drop quite a bit more due to the added overhead of changing interface protocols. So, it doesn't matter how fast you make the USB protocol, when the limiting factor is the SSD drive itself.
Are there any native USB4 devices out there? Most I see are using USB 3.2 Gen 1 / Gen 2 and tunneling over a USB4 connection just like thunderbolt 3 / 4
USB4 is literally Thunderbolt 3 with slight changes. Anything Thunderbolt 3 is already a USB4 device
Hmm I kinda figured, because I don't really see any difference in speed when dealing with files on a storage device. Of course SSD's are faster than HDD's, I don't expect I/O speeds to be ultra fast.
Yeah. Wendell lvl1tech talk about rabbit hole on USB standard
Don't even get people started by trying to explain why a 1TB HDD only has about 939~GB of storage. If only storage makers kept the normal Base 2 storage calculations the OS uses people wouldn't be so quick to get angry and leave bad reviews when they think HDD makers are lying to them about the size of their HDD. Windows (and other OSes except macOS since 10.6) use Base 2 where HDD makers use Base 10 so a 1000GB Base 10 HDD will always be displayed as about 939GB Base 2. When Apple released Mac OSX 10.6 and switched to Base 10 storage calculations people actually thought they were getting more storage thanks to Apple magic.
To me the worst thing about USB is Type-C but not for the reasons you might think. My issue is that not all USB-C cables are USB 3.0. Even USB-C to C cables may not be USB 3.0. I have ones that are only USB 2.0. Another issue I have are Power Only cables that many times have no labels on the cable warning you of that.
Tell Microsoft to fix Windows 11 .. the USB file transfer rate sucks. Windows 10 is fine.
Absolute unit of a thumps up given!
Wish they would just drop the brand naming crap and name it based on the speed, would be far less confusing for everyone.
Thanks for the video, very informative 🙂
Fascinating and thanks for laying all of this out. As an engineer who works with low production USB peripherals, I have had plenty of bad experiences. In my case, I really don't notice speeds, but reliability has been terrible. In fact, we used to call USB "USeless Bus" because of all the freezes, reboots, and corrupted data.
How does thunderbolt 4 and the new usb4.0 work what are those real speeds?
@5:52 and @10:30 - "network speed ratings like gigabit ethernet are true speed" Uh... Unless you have some magic ethernet that we don't know about, overhead applies here too, like the critisisms being leveled at the USB speeds. It might be rated higher than spec but the end result is the same as USB, you will never get the full speed that is advertised. People wondering why they are only able to get 940 mbit/s on their gigabit connection is still a regular question.
That's protocol overhead rather than signalling overhead. Ethernet advertise the speed after signalling losses are taken into account, and USB/SATA advertise the raw bits before the signalling losses.
@@itskdog Yeah, but that is splitting hairs as far as the end result. "Has one kind of loss vs two (or more)" doesn't change the fact that the stated rated isn't the rate customers can experience. Actual gig speeds are 6% less than stated. Here he is saying usb 4 10gig is 3% less so ethernet is worse that usb 4 (though he doesn't actually test anything to show what the max real-world transfer rate / what other losses might add in).
Unlike the line code which has a fixed overhead, protocol overheads are variable. And it also depends on exactly how you measure it - do you count the whole ethernet frame, or just the payload that's buried several headers deep?
0:25 why are you displaying the stock search term on the screen?
Just as a goof
interesting , I recently bought a 20 Gbps external drive, the speeds were very very slow so I bought a thunderbolt 4 cable to match my TB4 port on my surface pro 9 I found that if I attach it before starting the machine it doubles the speed over plugging it in when machine is on. just an interesting occurrence , but maybe some truth in there?
was actually wondering about it, as i get onl 40MBps on usb 2.0, while expecting 60.... (113Mbps on usb 3.1 g1), yet to try with samsung T7 1TB, this was a sandisk ultra 64GB
I wash my hands totally off of advertised USB versions & speeds, and just plug&pray&wait for the FINISHING_BELL!
Speaking from experience tuf gaming x570 motherboard gigabit LAN is advertised 1gbps but it barely pushes 500mbps
I wonder how is is? Doesn't USB have a start bit and a end bit?
10:47 Does this also count for germany and its internet speeds?
Is a 2.0 USB compatible with a 3.0 USB port on my laptop?
USB and HDMI naming schemes vs actual performance and features is atrocious. Particularly on USB naming, pretty much everyone is confused by it, in particular "regular" people that are not tech savvy which leads to immense waste when people buy devices and cables that don't actually work for their intended use case because they were confused when they made their purchase. It's so frustrating....
The namings should've been simple like this:
USB 3, USB 3.1, USB 3.1x2, USB 3.1x4
this is just as bad
@@BusAlexey at least it's more understandable that gen 1 and gen 2
i love u ur so good i love watching ur videos i am subscribe to u from late
2022
Oh no, USB is confusing everyone again! How could anyone expect that?
USB-IF - We make sure data transfer *stays being hell.*
Maybe Ugreen should read the USB specs themself because they put the USB A port upside down
for the next video you should call yourself Jhio Toe
It's funny how these people say each USB runs faster than the other yet most USB cables run the same, no matter how big or small they are.
It should seriously be illegal for these standards groups to misrepresent their protocols like they do.
this channel is a gem!
Thanks. You da man. Feels weird to say that. You were just a boy when I started watching.
If Ethernet can do that, why can't other ports do the same? Counting only the final output that matters for the customer. I mean it doesn't matter if it will be 4 vs 3 or 3.2 vs 2.7. People would want to buy what's better anyway. So why lying? Misleading people is dumb, because people who like to make a drama would go and hate the company for that. It's already common for people and their Internet connection speed. They use some communion PC and are mad that they can't get 30Mbps speed. Well, of course they can't if their disk can't allow that. Buy a new PC and it would. The device capability does mater. If You are buying Porsche to ride in the city, You won't blame the car dealer that You can't drive 300km/s, because it's Your fault for using it where You have to drive much slower than that. This is exactly same situation.
7:10 Little nit pick, he says NZRI evem though the screen says NRZI
You should do the same for WiFi since it gets abused all the time too
Hey Theo are win32 pua files are considered as virus plz reply sir
I don’t know
@@ThioJoe ok
really helpful, thank you
But why do they market their drives with RAW capacity instead of the actual formatted capacity
Another great video !!
And SATA speeds have the same faults, NVMe.2 4 gen does get to about 90%-95% of what it advertises. 😮