News USB 4 Version 2.0 Announced With 80 Gbps of Bandwidth

[dimar]

Call it USBC80. I wonder if it's enough bandwidth for a 10 screen docking station.

 

[InvalidError]

Call it USBC80. I wonder if it's enough bandwidth for a 10 screen docking station.

Depends on the resolution and refresh rate. A single 8k60 monitor is already almost 80Gbps raw at 10bits color depth. If you break it down from there, that is 4xUHD, 16xFHD or ~10xQHD before DSC.

 

[jasonf2]

So USB just totally forked itself into normal USB and something akin to a Thunderbolt port. I can just hear the big block store salesman selling the clueless consumer a $60 USB Gen 4 V2 cable to charge their phone because their phone supports USB4. This really should have been released as USB 5 or maybe something like USB 4 ActiveDock to describe its use case. I understand that the bus bandwidth itself didn't double but the move to the active cable itself is enough to warrant the need to differentiate the product from the mainstream USB offerings. Running a mouse on this is pointless. Thunderbolt just had the performance bar set for version 5 though and my gut is that something just got bumped by a couple gigabits to put USB 4 v2 on an island.

 

[Krotow]

Fine. And now please release some real USB 4 hardware. And make at least 40 Gb cables at affordable price for general public. From what I see now, USB 4 2.0 will surpass Thunderbolt 3.0 where 40 Gb cable price is eye-watering and docking stations cost a fortune.

 

[jasonf2]

No, there is no point doing that. SATA provides more than simply connector for drives.

SATA is also compatible with SAS, this means you can plug SATA Drives into SAS controllers (not the other way round).

SATA controller is also provided by chipset so you do not need additional converters/controllers. Many ports too and with/without RAID too.

SATA is also an extremely mature standard. IF you look at ALL, yes ALL the portable HDDs today. They are ALL using SATA drives with an additional SATA-USB converter. No company is even trying to design a dedicated USB controller for their HDDs.

If you look at portable SSDs, they are also using nvme to USB converters (some SATA to USB). Again, no dedicated USB controllers.

SAS is very much alive in the server space because you can address up to 128 devices per controller, the controller typically integrates good RAID and it supports hotswap. USB is limited to 127 total. Rust drives are far more economical per terabyte than SSD as well. This makes them very viable for storage that doesn't need to be extremely fast. SATA/SAS are both custom designed for handling spin drives and as an interconnect is about as close to the iron as you can get. There are intrinsic design issues communicating with spin drives that SAS/SATA reduces. They are so much slower than the RAM/CPU linkup that provisions had to be made. Moving spin drives over to USB just adds an encode/decode layer into a system that is already pretty optimal that took its price race to the bottom long ago. Putting an SSD on USB is even worse. The marketing stuff you are getting here is showing total combined bandwidth of 80gbps, but they aren't showing the Giga-Transfers per second stats. When you figure out the the total turnaround latency here NVMe on PCIe 5 is going to be considerably faster in low latency storage workloads with dedicated communication lanes.

The major use case here is just like Thunderbolt. Docking and cases where case limitations make high speed NAS and custom external GPU boxes viable. Great for laptops trying to play workstations, not so much for anything big enough to use the PCI slots. The only thing this is trying to replace is Thunderbolt.

 

[LuxZg]

I like the general idea of clean desk AND simplification of internal disk space. But all I can think of is - just give us PCIe cables. X1 cable to internal HDD, x4 cable to SSD, x1 external cable for slower external devices, x4 cable to docks, high res displays, external GPUs. And since it's PCIe just scale it to x16 by using multiple cables/connectors.

We already have U.2, ribbon cables for risers, PCIe over USB4, etc, so just make it happen just make it USB-like, simple and standardized. Well, no, scratch that, make it what USB should've been, but failed miserably ;D

 

[JamesJones44]

I like the general idea of clean desk AND simplification of internal disk space. But all I can think of is - just give us PCIe cables. X1 cable to internal HDD, x4 cable to SSD, x1 external cable for slower external devices, x4 cable to docks, high res displays, external GPUs. And since it's PCIe just scale it to x16 by using multiple cables/connectors.

We already have U.2, ribbon cables for risers, PCIe over USB4, etc, so just make it happen just make it USB-like, simple and standardized. Well, no, scratch that, make it what USB should've been, but failed miserably ;D

That's really what Thunderbolt is. Thunderbolt ties directly into PCIe 4x.

 

[jasonf2]

I like the general idea of clean desk AND simplification of internal disk space. But all I can think of is - just give us PCIe cables. X1 cable to internal HDD, x4 cable to SSD, x1 external cable for slower external devices, x4 cable to docks, high res displays, external GPUs. And since it's PCIe just scale it to x16 by using multiple cables/connectors.

We already have U.2, ribbon cables for risers, PCIe over USB4, etc, so just make it happen just make it USB-like, simple and standardized. Well, no, scratch that, make it what USB should've been, but failed miserably ;D

Having a wide open PCIe interface that anyone can hook to creates a really nasty on premise security issue. PCI is really good at close to the iron fast transfers, but that low level access is also a wide open hacking vector due to PCIe's memory access. It also would be incredibly wasteful to dedicate a pci lane to a mouse or keyboard. Cable length is going to be too short as well with cross talk issues. Thunderbolt is really what you are describing. I think intel and apple thought it would replace usb but it never really caught on that way. If you ever had to use rs232 or a parallel printer port you would understand why USB became as popular of a device interface as it has. Think printers, mice and keyboards. I think Apple and intel thought that Thunderbolt was going to supersede USB just like PCIe replaced ISA, AGP and PCI. The way Intel played early licensing stalled the whole thing though by making it too expensive and somewhat of an exclusive. The other thing is while every computer needs a mouse and keyboard most computers don't really need a bus that can dock everything including your monitor and hard drives when other mature specialized protocols do it cheaper, and better.

 

[jasonf2]

Fine. And now please release some real USB 4 hardware. And make at least 40 Gb cables at affordable price for general public. From what I see now, USB 4 2.0 will surpass Thunderbolt 3.0 where 40 Gb cable price is eye-watering and docking stations cost a fortune.

Any good active cable is going to run $60-$100 right now. This will not be cheap.

 

[InvalidError]

Any good active cable is going to run $60-$100 right now. This will not be cheap.

USB was originally created to be a simple and cheap replacement for a myriad of legacy ports. Now it has become perversely complex, expensive and non-universal with only certain type-C ports on certain devices supporting any given subset of USB capabilities and you need to match devices to the correct port to get them to work as expected assuming the two devices support at least one common protocol they can use.

 

[brandonjclark]

Well, it's about damn time. Finally, my days-long robocopies will finish quicker!

Oh wait, they won't because NTFS is still slow? Dang it!

 

[LuxZg]

That's really what Thunderbolt is. Thunderbolt ties directly into PCIe 4x.

Thing is that Thunderbolt is in it's own mess.

Having a wide open PCIe interface that anyone can hook to creates a really nasty on premise security issue. PCI is really good at close to the iron fast transfers, but that low level access is also a wide open hacking vector due to PCIe's memory access. It also would be incredibly wasteful to dedicate a pci lane to a mouse or keyboard. Cable length is going to be too short as well with cross talk issues. Thunderbolt is really what you are describing. I think intel and apple thought it would replace usb but it never really caught on that way. If you ever had to use rs232 or a parallel printer port you would understand why USB became as popular of a device interface as it has. Think printers, mice and keyboards. I think Apple and intel thought that Thunderbolt was going to supersede USB just like PCIe replaced ISA, AGP and PCI. The way Intel played early licensing stalled the whole thing though by making it too expensive and somewhat of an exclusive. The other thing is while every computer needs a mouse and keyboard most computers don't really need a bus that can dock everything including your monitor and hard drives when other mature specialized protocols do it cheaper, and better.

Security is always an issue, that's why certification exists. USB is no less of an issue with malformed devices. So I wouldn't say "open" PCIe would be much worse.

Thunderbolt is, as you say yourself, in it's own world, and unfortunately Intel made it what it is.

USB is cheap and works ok mostly, but I started to despise it, their management over the years is going down the toilet.

PCIe is last of the "modern" interconnects that has good roadmap, good execution, trust of consumers AND manufacturers, healthy ecosystem, etc.

Yes, I agree that single lane to mouse would be a waste, and that cable length would be an issue. But there are very smart people in that organisation and whole PCI-SIG community. Is PCIe 5.0 overkill for mouse and can't get to 2m length for mouse? Ok, why not PCIe 1.0. There are 7 meter PCIe cables (!). Sure, expensive. But I am sure at millions of pieces prices would drop fast. Just sic some clever people at it.

I mean ok, keep USB 1.0 / 2.0 for mouse and keyboard and printers. Those don't need better speeds anyway. Everything else doesn't require more than 50cm cable, and would do great with PCIe.

 

[jasonf2]

Thing is that Thunderbolt is in it's own mess.



Security is always an issue, that's why certification exists. USB is no less of an issue with malformed devices. So I wouldn't say "open" PCIe would be much worse.

Thunderbolt is, as you say yourself, in it's own world, and unfortunately Intel made it what it is.

USB is cheap and works ok mostly, but I started to despise it, their management over the years is going down the toilet.

PCIe is last of the "modern" interconnects that has good roadmap, good execution, trust of consumers AND manufacturers, healthy ecosystem, etc.

Yes, I agree that single lane to mouse would be a waste, and that cable length would be an issue. But there are very smart people in that organisation and whole PCI-SIG community. Is PCIe 5.0 overkill for mouse and can't get to 2m length for mouse? Ok, why not PCIe 1.0. There are 7 meter PCIe cables (!). Sure, expensive. But I am sure at millions of pieces prices would drop fast. Just sic some clever people at it.

I mean ok, keep USB 1.0 / 2.0 for mouse and keyboard and printers. Those don't need better speeds anyway. Everything else doesn't require more than 50cm cable, and would do great with PCIe.

I would still argue that the use case of most peripherals warrants the existence of USB. I just think that sometimes in their attempt to be all inclusive standards like this forget that the reason they exist is that they are a simple low cost solution that fills a need. I just don't really see this as being anything more than a single obscure port on the back of a $700 motherboard with no real hardware support until this moves into the full standard, ie USB 5. Even then, just as now, most devices on the standard will stay with usb 2 because there is no need for 80gbps and a $60 cable for a mouse that costs 20 bucks. So if you use this you will still be running a dock, with a USB hub, a displayport/hdmi and a power input that your laptop supplier provided for a premium. This doesn't really change much because there isn't really a need to be filled with it. Other than content creators most of those machines are lucky to have a 1080 monitor, a mouse and a keyboard plugged into them for a work at home setup. This is easily filled with the current standard. In a few years with low price this may replace thunderbolt, but probably not. Thunderbolt's next revision will be faster and people in that ecosystem are already good with paying for the cabling solutions and high device costs for the fastest on high end devices.

Open PCIe would at current level be a significant security hole. Please look up PCI RAM screamer. PCI is so close to the metal on the hardware stack that it has Direct Memory Access. Leaving a port like the one you describe on the outside of a machine leaves it wide open to a master key device that could bypass and record any software level security on the computer. It would be a matter of weeks before you would be seeing those devices on amazon. By the time you redesigned the whole standard to include the necessary security provisions, hotswapping, and power delivery it would be uselessly bloated for its purpose. There is a real reason that case alarms exist on server systems. USB is bad enough on its own, but at least it is sandboxed enough that it can't just directly read RAM and you can selectively shut it off for the most part.

PCIe needs to get faster, not slower. The fact that we have to use multiple concurrent lanes for our high speed components proves this.

 

[InvalidError]

Even then, just as now, most devices on the standard will stay with usb 2 because there is no need for 80gbps and a $60 cable for a mouse that costs 20 bucks.

Most mice and keyboards likely still run 1.2Mbps or 12Mbps USB1.

The single biggest flaw in USB's original design IMO was the lack of separate RX and TX pairs: the clunky 3.0-A/B/micro-B connectors to remedy this shortcoming paved the way for type-C's "lets throw everything in, including the kitchen sink" feature creep. Now we have type-C ports with 100+ different possible permutations of supported and unsupported capabilities which may differ between ports on the same device because supporting everything on every port is prohibitively expensive.

 

[jasonf2]

Most mice and keyboards likely still run 1.2Mbps or 12Mbps USB1.

The single biggest flaw in USB's original design IMO was the lack of separate RX and TX pairs: the clunky 3.0-A/B/micro-B connectors to remedy this shortcoming paved the way for type-C's "lets throw everything in, including the kitchen sink" feature creep. Now we have type-C ports with 100+ different possible permutations of supported and unsupported capabilities which may differ between ports on the same device because supporting everything on every port is prohibitively expensive.

Agreed. It is pretty apparent that most devices didn't need to go beyond USB 2 and still operate in that space today. USBc was a good move to try to consolidate the connectors, but if we are still forked into half a dozen cable connection combinations and just as many potential cable speeds confusion sets in pretty quickly.

The RX TX issue comes from the RS232c days. RS 485 and 422 have full differential signal pairs for cable length but I don't think anyone that designed USB back in the day expected the data rates they are trying to push today or the throw in the kitchen sink designs. They were just trying to make a good plug and play method of connecting peripherals on a serial bus that didn't require jumper setup and manual baud/packet configuration. Everyone here can complain, but I for one don't really miss using a copy con or edit command on the config.sys file and setting jumpers.

 

[InvalidError]

The RX TX issue comes from the RS232c days. RS 485 and 422 have full differential signal pairs for cable length but I don't think anyone that designed USB back in the day expected the data rates they are trying to push today or the throw in the kitchen sink designs.

You don't need a "throw the kitchen sink" design philosophy to foresee the likely need for USB to reach higher speeds: back when USB was introduced, FireWire was already demonstrating needs to go beyond 400Mbps for external storage, video cameras, networking, high-quality audio, etc. USB1 was a gross failure of imagination and the abortion called USB2 succeeded well enough to bury FireWire in most cases.

 

[jasonf2]

You don't need a "throw the kitchen sink" design philosophy to foresee the likely need for USB to reach higher speeds: back when USB was introduced, FireWire was already demonstrating needs to go beyond 400Mbps for external storage, video cameras, networking, high-quality audio, etc. USB1 was a gross failure of imagination and the abortion called USB2 succeeded well enough to bury FireWire in most cases.

It didn't matter. USBs success was due largely in part to its licensing and cheap cables. Firewire was plagued with a near exclusivity to Apple, ridiculously over priced cables and poor device availability, that isn't to mention that it was really developed to be a SCSI replacement not a peripheral bus . 1996-99 was a grossly different market place as well. While the devices you mentioned were emerging in that time period they only being used by early adopters and professionals. Both of them were paying through the nose for them. On top of that most people were using hard drives that topped out at 30MBps. The widespread adoption of SSDs have dramatically expanded the need for higher speed busses today, but that wasn't even in play for almost a decade and hindsight and the knowledge of what actually came to be is often grossly different from what the inventors vision was . I had a couple of computers that had IEEE 1394. They were all super high end motherboards and I think in all of the years of it being around I only had one device (a very expensive early digital video camera) that I ever used it on. I remember it because I liked to fell over when I had to buy the cable. USB didn't really replace Firewire, Firewire never made enough market penetration in the PC world to be replaced. Firewire just faded out of existence due to almost nonexistent uptake in consumer devices. The higher USB speeds were great but every computer had USB so that is what the device manufactures were putting on as their go to transfer bus. It didn't really matter which was better, only one of them was really available.

 

[InvalidError]

Firewire was plagued with a near exclusivity to Apple, ridiculously over priced cables and poor device availability

I paid only $10 or so for my FireWire cables, got FireWire for effectively free with my $100 Audigy sound card in my P3 which I carried over to my P4 and my ~$100 Core2 motherboard had FireWire on-board. My external HDD enclosures with FireWire cost only $10 extra over their USB2-only counterparts, not that much of an added cost there either for the ability to do sustained transfer between two external HDDs at ~35MB/s vs struggling to get to 30MB/s either way with a single drive over USB2 which I didn't have in my P3. My $1000 laptop from 2004 had a mini-1394 port too. Even to this day most external USB2 drives struggle to go much beyond 25MB/s due to the half-duplex polling-based architecture introducing heaps of dead time.

1394b (800Mbps) and faster stuff was far less common and far more expensive with the added headaches of new connectors to handle the extra lane in each direction. I never got any of that. Had it gone 800Mbps per pair first to reuse existing cables with the caveat of having to put 800Mbps devices first on either side of the chain, it would likely have fared better.

 

[Krotow]

I paid only $10 or so for my FireWire cables, got FireWire for effectively free with my $100 Audigy sound card in my P3 which I carried over to my P4 and my ~$100 Core2 motherboard had FireWire on-board. My external HDD enclosures with FireWire cost only $10 extra over their USB2-only counterparts, not that much of an added cost there either for the ability to do sustained transfer between two external HDDs at ~35MB/s vs struggling to get to 30MB/s either way with a single drive over USB2 which I didn't have in my P3. My $1000 laptop from 2004 had a mini-1394 port too. Even to this day most external USB2 drives struggle to go much beyond 25MB/s due to the half-duplex polling-based architecture introducing heaps of dead time.

1394b (800Mbps) and faster stuff was far less common and far more expensive with the added headaches of new connectors to handle the extra lane in each direction. I never got any of that. Had it gone 800Mbps per pair first to reuse existing cables with the caveat of having to put 800Mbps devices first on either side of the chain, it would likely have fared better.

USB 2.0 hardware was still cheaper and it matter when someone must produce it in million batches. I believe it was the real reason. In consumer space USB 2.0 speeds was still Ok and for archival purpose faster and more reliable than CD/DVD. Usual data size moved by external storage was relatively smalll then (under 1 GB). So world went in USB direction. Firewire was de facto replaced by Thunderbolt later. Now Thunderbolt is tied with USB 3+ by USB-C connector and I believe it will remain like that in foreseeable future.

 

[Colif]

I wonder how long before the fake drives start advertising self as USB 4. I can't see any yet but as soon as its out, I expect them to all miraculously changing overnight

 

[jasonf2]

I paid only $10 or so for my FireWire cables, got FireWire for effectively free with my $100 Audigy sound card in my P3 which I carried over to my P4 and my ~$100 Core2 motherboard had FireWire on-board. My external HDD enclosures with FireWire cost only $10 extra over their USB2-only counterparts, not that much of an added cost there either for the ability to do sustained transfer between two external HDDs at ~35MB/s vs struggling to get to 30MB/s either way with a single drive over USB2 which I didn't have in my P3. My $1000 laptop from 2004 had a mini-1394 port too. Even to this day most external USB2 drives struggle to go much beyond 25MB/s due to the half-duplex polling-based architecture introducing heaps of dead time.

1394b (800Mbps) and faster stuff was far less common and far more expensive with the added headaches of new connectors to handle the extra lane in each direction. I never got any of that. Had it gone 800Mbps per pair first to reuse existing cables with the caveat of having to put 800Mbps devices first on either side of the chain, it would likely have fared better.

I am agreeing that firewire in its day was technically superior. But up against your deal buying the $10 cable I am pretty sure if you were anything like me you had a box of unused USB cables that you had acquired for free from every device that was replaced that didn't need a new cable. That was also a sweet audio card, not some cheap integrated junk. As I recall that thing had a full drive bay front facing assembly that reminded me of a swiss army knife because of everything they had stuck in it. Also a $100 motherboard in the core 2 era wasn't junk like it would be today. Boards back then were $100-150 for the good ones. It sounds like we were pretty much in the same place. I used USB in that era because it was impossible to build a machine without USB. I only used firewire because I wanted to and had to go to effort to do so.

 

[InvalidError]

That was also a sweet audio card, not some cheap integrated junk. As I recall that thing had a full drive bay front facing assembly that reminded me of a swiss army knife because of everything they had stuck in it.

The standard Audigy (the one I got) was just the sound card. The model with the external drive bay break-out box is the Platinum edition.

Also a $100 motherboard in the core 2 era wasn't junk like it would be today. Boards back then were $100-150 for the good ones.

$100 was for one of the cheapest ATX boards I could get at the time. Could have saved $20 by going mATX instead, just wanted to retain the option of PCI slots just in case and ended up never using them.

 

[JamesJones44]

Thing is that Thunderbolt is in it's own mess.

I don't really agree with that. It's expensive which has caused low adoption, but anything that is guaranteed to be fast is going to cost more. Look at the cost of Motherboards since the introduction of PCIe 5. Now that Intel has dropped royalties I expect we'll see a small uptick in adoption, at least from MB makers. Laptop makers on the Windows side might be a different story.

 

[Albert.Thomas]

Specifically, why?

Benchmark drive speeds are not the be all and end all of performance.
People have a hard time telling the diff between current SATA III and PCIe 4.0 SSDs.

SATA is old and outdated. Why use old, outdated, slots when you can replace them with a newer standard that supports much faster performance and is backwards compatible with the old standard?

 

[InvalidError]

SATA is old and outdated. Why use old, outdated, slots when you can replace them with a newer standard that supports much faster performance and is backwards compatible with the old standard?

Typical SATA cables may have trouble running at 12+Gbps.

Personally, I'd say future internal interfaces should focus on 12VO PCIe. If you want backward compatibility, you can slap a PCIe-to-whatever bridge in-line somewhere along with the necessary DC-DC converters. Then you could get a 4x SATA4 (12Gbps) backplane to plug into that to take old SATA cables out of the equation along with the clunky SATA power cables using a single x1/x2/x4 cable. Same could be done for USB, NVMe or creating PCIe devices in different form factors that don't necessarily need to occupy a rear IO slot.