Article: USB 4.0 Is A No-Go, For Now

[Bob Stern]

On 8/16/2021 at 5:35 AM, AudioDoctor said:

the latest USB schemes are very confusing.

 

Here's what I recently figured out (corrections appreciated, of course):

 

USB 3.1 is identical to USB 3.2!  The speed is specified by the Gen suffix, not by the .1 or .2 suffix.  Weird.

 

USB 3.1/3.2 Gen 1 = USB 3.0 = 625 MB/s = 5 Gbps.

USB 3.1/3.2 Gen 2 = 1250 MB/s = 10 Gbps.

USB 3.2 Gen 2x2 = 2500 MB/s = 20 Gbps.  (I think this is same as USB 3.1 Gen 3.)

USB 4.0 = 5000 MB/s = 40 Gbps.

 

Thunderbolt 1 = 1250 MB/s = 10 Gbps.

Thunderbolt 2 = 2500 MB/s = 20 Gbps.

Thunderbolt 3 = 5000 MB/s = 40 Gbps.

Thunderbolt 4 = ??

 

Thunderbolt 1 & 2 use identical Apple-only connector.

 

Thunderbolt 3 & 4 use same USB-C connector as USB 3.1/3.2, but Thunderbolt 3 and USB protocols are different.  Thunderbolt 3 hosts on Mac can communicate with USB Gen 1 and Gen 2 peripherals but probably not Gen 2x2.

 

M1 Mac Mini has USB-C connectors with USB 4.0 and Thunderbolt 3.

[AudioDoctor]

10 minutes ago, Bob Stern said:

 

Here's what I recently figured out (corrections appreciated, of course):

 

USB 3.1 is identical to USB 3.2!  The speed is specified by the Gen suffix, not by the .1 or .2 suffix.  Weird.

 

USB 3.1/3.2 Gen 1 = USB 3.0 = 625 MB/s = 5 Gbps.

USB 3.1/3.2 Gen 2 = 1250 MB/s = 10 Gbps.

USB 3.2 Gen 2x2 = 2500 MB/s = 20 Gbps.  (I think this is same as USB 3.1 Gen 3.)

USB 4.0 = 5000 MB/s = 40 Gbps.

 

Thunderbolt 1 = 1250 MB/s = 10 Gbps.

Thunderbolt 2 = 2500 MB/s = 20 Gbps.

Thunderbolt 3 = 5000 MB/s = 40 Gbps.

Thunderbolt 4 = ??

 

Thunderbolt 1 & 2 use identical Apple-only connector.

 

Thunderbolt 3 & 4 use same USB-C connector as USB 3.1/3.2, but Thunderbolt 3 and USB protocols are different.  Thunderbolt 3 hosts on Mac can communicate with USB Gen 1 and Gen 2 peripherals but probably not Gen 2x2.

 

M1 Mac Mini has USB-C connectors with USB 4.0 and Thunderbolt 3.

 

Thanks, I will print this out and reference it when needed. ;-)

[SoundSparks]

Dear experts,😀

Could you please help me understand what real-world throughput I can possibly get for the following setup (please note DASD employs Intel JHL7440 Thunderbolt 3 controller Chris discussed in his article)?❓

iMac <--> Thunderbolt™ 3 Optical Cable <--> OWC ThunderBay Flex 8😀

Does the throughput depend on the direction of the data flow?❓

Cable length?❓

HDD / SSD / RAID configuration employed? Any optimum (performance / price) configuration you would recommend?❓

Any Thunderbolt controller with native Thunderbolt 4 or USB 4.0 support (at least in the specs), and DASD's employing it?❓

[billg]

I'm more interested in the sound of the drive than just its' speed. I use an internal sata hdd on quiet external power to avoid electrical noise. As it's the only sata device in the box there's little or no contention/jitter. I also read the thread on audiophile nvme ssds with external power which try to address these issues (but too small capacity for me). I wonder whether anyone has done any testing of these aspects of the very fast drive/interface options?

 

 

 

[EccentricSage]

First of all, you require specific USB 4 cables (thunderbolt 40gigabits rated cables and power delivery standards rated and such of amps and voltages for USB 4.0) 

 

Second of all when i first upgraded to pci Express 3.0 a few years ago the nvme drives were all the rage. Though older sata 3 6GB/sec were lower latency and better at gaming and accessing small files and for loading the OS and the bios on the mainboards used to not allow installation of OS onto NVME drives. At the time i purchased one of the fastest nvme drives for affordable price then wished i bought a samsung.. because again lower latency and they invented the SSD/nvme storage technologies literally.. anyway the PCI express lanes on the board and bus were far more bandwidth and higher number than the intel boards as i went with a ryzen. Then i noticed i only had one M.2 ultra slot 40gigabits pcie gen 3 lanes x 4? when my NVME drives max bandwidth required 38.5gigabits of bandwidth in the NVME slot.. i bought 2 hoping to RAID them up! as they're smaller storage sizes than i would have liked. it seems the second m.2 slot was only 10gigabit and not ultra.. no 40gigabit.. so you see i looked around and saw more expensive models of similar boards with 2x m.2 ultra .. but realised pci E gen 4 was around the corner and just waited for a X570 board a short while after the X470. my life would have been way worse if i had an intel system as they had far less PCI express lanes and zero ultra m.2 slots 10gigabits per sec was max on everything under the sun i could see and for a while they didnt even have PCI express 4.0 and when they did it was super limited.. (like that 10gigabits example) while AMD currently has EPYC PCI express 5.0 96 core CPUs and stuff.. anyway. the point is, even you got the NVME drive installed say the samsung 980 pro in the RIGHT board from AMD and straight into the board none of those adaptors slowing things down, no old USB 2 cables or 3.1 cables in your USB 4 port? you need to enable AHCI mode rather than RAID mode.. and use something called RAPID STORAGE MODE where it buffers all your drive access with 2GB of system memory as a RAM disk swap file for something like "drive tiering" which is actually native to the ryzen platform you can find the drivers for it on the AMD website where you get your RAID drivers from and you can format your fastest drive to buffer your old spinning disk drives or slower disks of larger capacity. 

 

Anyway point being thats not even the start of the problem. previously windows 10 and 8 and 7 back to 98 all support floppy disk drives and CD ROM drives still. so they have stupid lame slow disk access commands from the 1970s and 50's.. and this garbage obsolete stuff called C++ thats like 2bit binary on your 64bit CPU with 16bit software code and doubled in a memory curtain to 32bit.. so they sorta fake scale it up via software.. instead of like sextoquadrary compilers in modern post 90's programming languages.. and NO not 1991 python scripting LLVM turd layer caked over the top of C++ that cripples maths and memory usage to pretend its going faster by using less and doing less via "simplification" you can use a short few lines of code to test if an app is true 64bit if you search. windows powershell used to have a true 64bit mode which wouldnt execute anything not 64bit.. but were court ordered to remove the feature i heard it was in windows 10 previously but ya know its all just so criminals could crime all over us some more.

 

Anyway you need something like VULKAN or directx 12.5 like windows 11.. and probably on mac you will need to be running METAL to start with.. dont ask.. Also USB 4 ports need USB 4 drivers. You will notice USB 3 has SCSI over USB in the standard.. and external storage drives supported via SCSI or external sata.. but theres IAstore.ah or what it .sys file in your drivers.. which is the maybe older way of doing it which has limited speeds and older standards but is "compatible" then theres EHstore.sys or whatever.. so one of these is wayyy newer. If you got like an old sandybridge intel system board and 6700k from asus I used to own few years back.. the ASUS board i had had USB USB ports with their own SCSI USB driver literally. if you used the generic asmedia 3.1 USB drivers they sort of worked but occasionally disconnect and reconnect you hear the notification sound, and some devices werent fully powered and transfer speeds were abysmal. But you could download the drivers for USB from the website and have it be still entirely not working as it should and the same broken faulty junk. GRAB THE DISC CD THAT CAME WITH THE ASUS MAINBOARD and search a bunch of directories for a hidden USB driver under some unassuming name of their SCSI USB technology.. then install that one and it gets rocket fast speeds and works as the USB 3.1 intended! YAY!! now that everybody doesnt have a CD ROM or floppy disc in their PC and thinks those discs are just coasters.. i think you may need to seriously find and locate yours and give that a try. and No i have no idea what these windows 10 OS files are called on Mac. Or what versions or filenames they are in windows 11 currently. but yes it is a bit like trying to buy a fast charger quick charge 4.0 PD 3.0 or whatever thunderbolt 3.0 charger cable for charging your android type C mobile phone just going vaguely by "but this one says 5 amps on it! why isnt the super fast charge icon showing?! did i not spend enough money on this 45 watts or 65 watts fast charger wall plug? is it the wrong brand?" when you can plug your entire laptop into your PC's USB 3.2 gen 2 port on the back of your AMD mainboard and PD quick charge it or thunderbolt whatever.. im guessing i dont own a laptop..

 

you get the point. I also noticed i needed to purchase some expensive samsung Bdie RAM and lower all my memory timings and secondary timings to lowest latency possible and command rate 1T and so on.. 1866mhz and then i adjusted all my secondary timings lower. i loosely based it off the DRAM timing info lowest values listed in the RAM bios information for 1866mhz on my CL14 3700mhz teamgroup extreeme ARGB 7.8ns (tho i maybe got the wrong one somehow cant seem to boot below 9? point is in the L/S values columns theres one that governs delay timing for reads and writes and also for storage access. TWRRdS or whatever the heck was called i aint going to reboot and check. on my system i believe it was the 3rd set of S/L timings before the refresh cycles? depending on your brand of NVME/SSD you can tighten or adjust those timings and have a snappier more responsive time of everything. actually samsung should be able to set to 2 based on Bdie RAM from manufacturer specs. and for drive storage im not too sure as i've only had one brand of Adata nvme to test but 2 works and is snappy quick while sometimes has a sort of time out lag 3's for both is kinda suck.. my RAM timing info says 5,5 which is fast at 1866 and great 4,4 is slower somehow.. and 6,6 seems rather fast.. you can set all your S/L's to all 6's for me.. and most everything else basically.. and it might go pretty quick but you may get some storage errors corruption on a massive game installation you would need to verify it again i dont know if its the internet downloads or the RAM or what but setting to 5,5 fixes it for me but its a bit less fast. your TMAC maybe i set it to 10? i think it was called the 16 one.. or uhh TRAS and TRC should be like what it says in the timing info supplied but i found lowering them a little and TFAW raised can be interesting. it goes quite a bit faster and some game characters hair flutters wildly like wriggly worms if you go loopy with the TFAW. i dunno man.. unless i could somehow hex edit the bios and undo the sabotage and limitations on it i cant fix much of anything but i dont wanna try that im dirt poor and cant afford to replace anything. but yeah change your TWRRd L and S and your hard drive benchmarks go through the roof into outer space! you will be scoring the full 7.5GB/sec or whatever when it should be about 5GB/second. you see all that XMP fake bullshit from fake computer makers is your board has a fixed bus bandwidth.. and stacking the mhz higher and overclocking .. simply means it does multiple passes or takes longer to do each trip! its slowing things down and increasing latency. in very few situations is this actually beneficial. All super computers are ZERO LATENCY (as in below billions of a second is costly or tricky to measure so we dont even bother we just go "thats super".

 

so in the same amount of bus bandwidth a much much lower latency faster 1866mhz can happen many times more in shorter time than anything over 3200mhz with command rate 2 where it sends everything twice. I also like to disable gear down mode and other such things. But the 980 pro samsungs PCI E Gen 4.. tend to perform similar to pci e Gen 3 devices as we dont have the software drivers and "hardware usage" the whole OS supports android and DOS and 16bit junk like every nvidia intel driver or software or whatever. just having a single line of that old code means the whole system must run that slowly because a chain is a strong as its weakest link. a system is fast as its slowest component. AMD CPU and GPU hardware hasnt been used to play back video for hardware video decoding. your headphone amplifier hasnt been used to play back audio.. Which is why even though WASAPI and dozens of other hardware access methods existed they literally had to build ASIO standards as a way of accessing via lower latency.. when it barely shaves one ms off the existing windows drivers at the time and was more of a way of milking money and OEMS refusing to code drivers and software that use the hardware. our CPU's have been 64bit since the 90's and multicore.. but 99% of all everything is not even 64bit.. let alone multicore.. dont you think its strange your CPU cores have a tiny 3% usage on this core then 3% usage on that core.. and its all controlled by software in these thin threads.. instead of pouring all the software into it like a garbage disposal and ALL CORES ALL THE TIME! with this hardware scheduling stuff.. if it ever gets too hot it throttles back or ramps the fan curve up. None of this puny lets put this in this corner and use 3% of that core.. and 2% on this .. its DISGUSTING! modern code and software is millions of times faster in execution but when used correctly modern hardware can be probably many many times faster, maybe dozens or thousands of times depending what you're doing. if your cheapest mobile devices can stream 4k youtube and netflix and your most expensive samsung mobile phones and anything with a qualcomm chip can 8k VR with an AMD graphics card in it and it uses barely any watts of power and charges in 30 minute to last you 3-4 days.

 

Well imagine how much more powerful your 7000-1000 watts power supply AMD computer is when microsoft and intel and nvidia let you actually use them for once and game devs stop prying code and software from their grandpa's high school text books and grabbing dated 80's and 90's stuff for free in C++ commonly called a linux distro and repurposing it pretending they've reinvented the wheel. open source doesnt technically mean free to steal.. but man new software and languages and API's can cost a lot of money like licencing a game engine.. so they get the cheapest fakest nvidia and intel junk and pretend its the latest and greatest when its built on 90's linux C++ librarys for the same 15 year old intel CPU core design on the 13th gen now? whats it at rocket lake? alder lake? its probably 20 years old with out a change to the compiler.. and they pretend flags and libraries fix a decades old DOS prompt junk language. not new API's and new code and "DIRECT TRANSPORT" or "direct output" commands in the code. they gotta pass it through memory and overlays and virtualize instance it and shadow volume copy it and spy and record and capture everything.. cant "copy files to the audio device hardware directly like DSD or AC3, how would we spy on them then?" the graphics card hardware needs correct file format and formatting and something like HDMI passthrough or SPDIF too. and your USB ports are a tragic victim in that intel is pretending they made them when we all know its microsoft who did all the work they basically just repurposed a cheap ethernet chip.. like a $5 10/100 or 1000gigabit USB ethernet adaptor you can buy off ebay with all of microsofts OS to detect the device ID and such with a USB port device.. 

so yeah i suggest you having a samsung drive install the SAMSUNG magician application. ensure you are in AHCI mode.. and uhh.. try disabling a bunch of unnecessary services and functions that are drive related. win windows 11/10 you could try disable drive indexing see if that helps, disable or enable the write cache, in the uhh task scheduler services and group policy editor you may be able to disable drive paging and page file.. or enable direct storage access somehow.. i uhh know whats wrong but sadly i dont know how to fix it.. your microsoft windows OS probably has a storage controller with a year 2003 dated driver on it.. how the heck do i fix that? same with your SCSI over USB scsi standards.. you can see they released many different SCSI standards over the years.. theres even SCSI over PCI express obviously with different speeds and standards.. the PCI express gen 4 SCSI standards are something to look into i guess. just like how games claim dolby atmos compatability but not use AC3 files i bet and probably arent 19-23 or whatever speakers for true home theater experience im guessing i just own a cheap 5.1.2 atmos system.

 

tweak your bios and your RAm ensure your bios is updated.. give windows 11 a try and whatever latest drivers you can find. and check for a USB driver disk and ensure you have a fast enough USB 4 cable. then benchmark away with samsung magician software! oooh also your system BIOS may be in gen 3 mode or auto.. try force to gen 4 or gen 5. ensure you've got secure boot and TPM 2.0 and such enabled all that stuff, but if your memory is encrypted by default maybe turn that off.. uhh your RAM interleaving.. things like that! 

 

to put it in terms of things audio forum users can understand.. you purchased a 21 channel atmos receiver or whatever for like $3-5,000 USD.. you then use 90's software like winamp and lame mp3 and DOS prompt garbage like hand brake.. to encode all your stereo 128kbit mp3 16bit 44khzsoftware.. and play it back on 2 of your speakers via software decoding.. instead of say inserting a 9-13 channel AC3 lossless master quality 32bit 384khz bluray disc and using passthrough to your receiver to have the DOLBY ATMOS light blink on and engage and you are then hardware decoding it and you now are playing back in 21 channels of surround sound. Your video player is using windows media video renderer 7 in software mode.. you need minimum i mean minimum d3d11 renderer hardware decoding passthrough to the GPU for industry standard 10bit HEVC 8k or 4k decoding with hundreds of FPS in the highest quality you ever saw in your software video player.. mpc BE and HC all fail.. cant do it. Mad VR had a something that sort of allowed it to passthrough to a TV or device.. like how you can copy a HEVC HDR file to the cheapest of smart phones and smart TVs but your thousands of dollar AMD graphics card cant do it as well as nvidia and cheap wrist watches and mobile phones can for $80 -200.. because its a decicated graphics device with GDDR 6X or HBM RAM and massive power draw billions of transistors and monstrous floating point.. it doesnt have a cheap $2 Via chipset to hardware accelerate and allow playback of netflix and youtube without artifacts in 4k HEVC. its not like AMD invented HEVC in the 70's and licensed it for your personal use with their GPU's or anything?

or you cant purchase a HEVC extensions for $1 from microsoft app store so you have the correct codec to play stuff back with for a start? no you need a cheap nasty nvidia card to type stuff into a 90's DOS prompt for some real NVENC action!