USB 3

[Paul Bruening]

I had heard buzz about the ultra-speed USB 3. It seems cards are coming out that allow my slightly older boards to run USB 3 ports. This is really good news since my render farm is severely choked by the Gig-LAN connections. What I can't find is if any USB 3 routers are out yet that can replace my LAN system. Any news or buzz?

[Paul Bruening]

Could any of our Japanese posters tell me if the Buffalo 4-port USB 3.0 hub has hit the shelves yet? If so, what's the price on it? Can the salesman tell you if it is stackable? Does it run multiple channels with no bandwidth loss?

[Paul Bruening]

Could John, Hal or any takers give me an idea if I could use my existing CAT 5e cables with adapters to push USB 3.0 data around my boiler room since they are both 8-wire systems? Could I chop the ends and solder in USB 3.0 ends?

[Paul Bruening]

C'mon, buddies. Hazard even a guess on the cables. If I can make this work it could be the greatest leap in my farm's productivity for the fewest bucks that I've ever come across.

[Ben Syverson]

Paul,

 

I'm have no real EE experience, and I haven't read all that much about USB 3, but my initial reaction would be "hey, 8 wires is 8 wires!" There's nothing mystical about wire, despite what audiophiles say.

 

That said, I would keep the cable runs short -- the longer the cable, the more resistance, the more you're taxing the signal. Of course, for a render farm, you may only need to go about 12-18 inches at a time!

 

Can you catch me up on why you're running a render farm? I'm guessing it has something to do with processing RAW frames from your scanning rig? I'd be curious what kind of processing you're doing... I've done a little work on GPU-accelerated film tools. Things like dust/scratch removal and color correction. On my laptop, they're realtime at 2K and near-realtime at 4K...

[Paul Bruening]

Paul,

 

I'm have no real EE experience, and I haven't read all that much about USB 3, but my initial reaction would be "hey, 8 wires is 8 wires!" There's nothing mystical about wire, despite what audiophiles say.

 

That said, I would keep the cable runs short -- the longer the cable, the more resistance, the more you're taxing the signal. Of course, for a render farm, you may only need to go about 12-18 inches at a time!

 

Can you catch me up on why you're running a render farm? I'm guessing it has something to do with processing RAW frames from your scanning rig? I'd be curious what kind of processing you're doing... I've done a little work on GPU-accelerated film tools. Things like dust/scratch removal and color correction. On my laptop, they're realtime at 2K and near-realtime at 4K...

 

Some of the runs are 30 ft. But, the power levels on USB 3.0 are really high. I'm hoping it can make the push. I could relocate the routing point and carve the lengths in about half if I have to. It's this way because Gig-LAN could handle the lengths and that let me distribute the stations to serve both as workstations and render farm.

 

Yes, it's because of the massive data loads for my 4K TrueRAW scanning. It's not just that 24Mb per frame but the repetitions for each station including Maya and Adobe AE stations. Fortunately, Adobe doesn't license its render engine. You can put one copy on as many PCs as you want. That alone makes the render farm idea a rockin' prospect. I'm running 6, 2core boards at 32bit and 2, 2core boards at 64bit. I've got 2, 21" trueflat, 2K CRTs on each station with some backups on the shelf. I've got the DG5 for color timing.

 

I'd rather be running those new Asus boards that I linked to in the "badass PC" thread. But drooling over technology never ceases.

 

If I can pull off this USB 3.0 thing I'll smoke anything within 300 miles with my productivity.

 

EDIT: ...and about 70% of the rest of the post production country.

[Hal Smith]

I haven't given USB3 any thought yet. CAT 5/6 cable consist of four twisted pairs with nominally 100 ohms impedance. The impedance does vary slightly from pair to pair due to the different twist pitch per pair (to minimize crosstalk) It's designed for balanced drivers and loads. If USB3 uses one to four balanced pairs AND at 100 ohms it should work. At the frequencies modern digital communication busses are running at its absolutely necessary to think of cable in RF terms and use cable designed specifically for that service, or other very similar services.

[Paul Bruening]

I forgot to mention, Maya has a free render farm utility as well.

[Chris Bowman]

Paul,

I did some lazy man's research on wikipedia, and noticed two possible restrictions on your server farm USB 3.0 over CAT5 bliss. First, SuperSpeed requires a grounded shield, which CAT5/6 doesn't have, so runs of more than a very short length will probably suffer from serious interference and signal degradation. Secondly, although not an official limitation, it has been estimated that the real world max length of a run is about 3 meters (say about 10 ft.)

[Noah Yuan-Vogel]

You are talking about running a network over USB3.0? I wasnt aware that was possible, I guess I imagined it is possible to do that over USB but IP networking certainly doesnt seem like what USB was designed for. I'm sure there would be cable length issues. Most USB uses I am familiar with use 1-to-1 host-device connections and I doubt standard USB drivers would support much else. I could be wrong though. Seems like if you really need extra speed for a renderfarm the old routes of trunked GigE or 10GigE or Infiniband, or maybe a Fibre SAN would be the way to go. It would be cool to be able to harness new superfast super cheap buses like USB3 to speed things up instead, though, but I think it could be more trouble than its worth since they were really meant for connecting one fast device 3ft away not 100 computers 100' away.

[Michael Collier]

from the quick research I was able to dig up it looks like USB3 is nominally 90 ohm impedance. Tolerance is measured as +/- 13.5 ohm. If Hal is correct that Cat-5 is nominally 100 ohms this would appear to work, but I assume that's a target goal that can be affected by real world conditions (the different winds, standard deviation in construction, a 100 ohm wire mated with a 90 ohm plug, etc, etc) you might be really pushing the levels of what is possible, especially with a theoretical limit of +/-3.5 ohm.

 

My research did indicate its an 8 wire, 4 group twisted pair, like Cat-5. I didn't find anything saying weather or not it requires a shielded cable.

 

But impedance mismatch could cause big problems for packetized data at high speeds. A mismatch could be viewed roughly as the equivalent to shooting through glass. Most of the light that strikes the glass goes through and lights your subject, but some bounces off and back into the lens, obscuring the image. That bounce can truncate packets easily, causing their checksum to not line up. Then the receiver asks for a resend. Assuming the bounce doesn't degrade the packet this time and the checksum checks out the packet is accepted. But you just spent 3x the time sending that one packet. You might end up with a system that works on the surface, but whose throughput in reality is much less than stated, and maybe even lower than the current throughput?

 

From what I can tell it wouldn't be worth it to make up all those wires only to find there is little to no boost in productivity. Or worse that it doesn't work at all.

[Paul Bruening]

I just went to check my Cat 5e cables to see if they were shielded and discovered that they are actually Cat 6 cables. I recall that I bought them as shielded cables. This means that they are rated at 250MHz. They are not Cat 6a (augmented) that are rated at 500 MHz. Since they are well under the length limits for Cat 6 cable running 10GbE and are rated to handle loads twice that for USB 3.0 wouldn't I be within the cables' tolerances for a USB 3.0 application? This is assuming that I soldered and shielded the new ends and there was enough voltage to drive the 30 feet or less lengths. As well, they are patch cables and are made with stranded wires instead of solid core wires.

 

You see, a USB 3.0 card is only $40 at Tiger Direct. Whereas 10GbE NIC's seem to be insanely expensive and almost impossible to find on consumer sales sights. Does anyone know where 10GbE NIC's can be had for $100 or less? 10GbE is about twice as fast in specification as USB 3.0. But, the cost is what's bugging me.

[Paul Bruening]

From what I can tell it wouldn't be worth it to make up all those wires only to find there is little to no boost in productivity. Or worse that it doesn't work at all.

 

I might could do a test run between two computers. It would only put me out $80 in cards. If it didn't work, I could just crimp 8P8C plugs and put it back to Gig LAN. The cards would still come in useful later as the 3.0 technology on devices starts coming out. The software driver would handle the mobo cross on one computer. Whether it can bridge two to move data between their HDDs is the part I don't know. The driver will be looking for a device on one end, won't it?

[Paul Bruening]

You are talking about running a network over USB3.0? I wasnt aware that was possible, I guess I imagined it is possible to do that over USB but IP networking certainly doesnt seem like what USB was designed for. I'm sure there would be cable length issues. Most USB uses I am familiar with use 1-to-1 host-device connections and I doubt standard USB drivers would support much else. I could be wrong though. Seems like if you really need extra speed for a renderfarm the old routes of trunked GigE or 10GigE or Infiniband, or maybe a Fibre SAN would be the way to go. It would be cool to be able to harness new superfast super cheap buses like USB3 to speed things up instead, though, but I think it could be more trouble than its worth since they were really meant for connecting one fast device 3ft away not 100 computers 100' away.

 

I only need it to move data from drive to drive between stations. I don't need all the chat/email/et al junk of a regular network. If I can make the stations act like a device on one end of any link, I'll be in good shape.