TB1 vs TB2 Supply & Demand

[Average Pro]

Not sure if this thread is placed in the correct forum, but here goes.

I've been waiting for the nMP release to exponentially expand/replace/update my data storage. Long before the nMP was announced, my intent was to start purchasing TB1 products and connect them to my MP with an adapter. This way when the nMP was released, I could just swap it on over. Then they announce TB2. Now the plan is to buy maybe one or two TB1s to hold me over until TB2 is released.

My impression is that several consumers were waiting for Apple to release the nMP before jumping on the TB(1). I can't count the number of times a store representative or magazine article stated, "the prices for TB(1) will drop once they get mass produced." Now that Apple announced TB2, doesn't that kind of kill TB1?

 

[wonderspark]

It sounds like you're learning a lesson in living in the present, as opposed to worrying about tomorrow. There's nothing wrong with trying to maximize efficiency, but if you need technology, the devices available today matter more than what may or may NOT come tomorrow. What if TB(x) dies off completely for reasons unforeseen?

Magazines and store reps have the same insight into the future as the rest of us, which is a guess at best. Some are luckier than others with their guesses.

I conducted a lot of research when I needed to expand my data storage. I chose an Areca 1880ix-12 RAID card, since it can connect 16 disks without any bottlenecks in speed, and that satisfied my needs from then until some years from now. A few months later, Areca released the 1882 series, which was an improvement, yet my card still runs faster than my needs to date. I may need to expand my 8-disk RAID set to 12 or 16 disks, or I may find I never need more than what I have today, but I chose a system available at the time that could grow well into the future. So far, Thunderbolt hasn't really shown me that it can do that as well as PCIe and mini-SAS.

On the other hand, when all desktop manufacturers drop PCIe and leave no other option but TB(x) or whatever the next big thing is, I'll find myself looking to catch up with the present day once again. I'll do the same as before, and get something robust and expandable in the current and foreseeable near-future.

 

[Average Pro]

Guilty

Thank you Transformer dude. I am guilty of doing that which annoys me most - worrying about something that isn't even available when a present solution exists.

Not only that, but here I am reaching for something, only cuz it's going to be faster. Especially when the TB1 will do everything that's required. I spoke to a major vendor of external HDs and inquired when they would be offering TB2. I kid you not, he started laughing. That made me think that some manufacturers may rush to get a product on the market when the MP is released.

Can't wait for the MP to hit the shelves. As for the oMP, not sure what will happen with it.

 

[jrlcopy]

Some companies are offering free TB1 > TB2 Upgrades.
http://www.sonnettech.com/product/echoexpressse2.html?tab=3

 

[haravikk]

TB1 will do everything that's required.

This. Unless you're looking to combine together a heap of SSDs in RAID-0 array or something then you probably don't need Thunderbolt 2 at all.


One related question I'm wondering if someone can answer is; since Thunderbolt 2 is actually just Thunderbolt 1 using two channels at once, then can you connect two Thunderbolt 1 devices to a single Thunderbolt 2 port (i.e - daisy chain) with no loss of speed? That way you could at least benefit from Thunderbolt 2 simply by using less ports with no loss of performance, but I don't know if the way devices are daisy chained will support that using Thunderbolt 1 devices?

 

[thekev]

On the other hand, when all desktop manufacturers drop PCIe and leave no other option but TB(x) or whatever the next big thing is, I'll find myself looking to catch up with the present day once again. I'll do the same as before, and get something robust and expandable in the current and foreseeable near-future.

I'm not sure whether that will actually happen. Consider that workstation parts are shared with some server components. So far they have no integrated graphics chip, and Apple went the custom route there. Given that the thunderbolt connector isn't really designed for data center or server room use (non-locking for one), it's unlikely that intel is going to somehow cram this down the throats of other oems who purchase Xeon EP chips.

 

[deconstruct60]

...
I've been waiting for the nMP release to exponentially expand/replace/update my data storage. Long before the nMP was announced, my intent was to start purchasing TB1 products and connect them to my MP with an adapter.

There are no such adapters. For the pre-2013 MP models there never will be (at least certified ones ... which at this point is the same thing as any.). A limited subset of iGPU less Xeon E5 ( and equivalent Core i7 49xx ) systems may get Thunderbolt cards in the future, but most are going to be in the same boast as the older Mac Pros. [ Appropriate GPIO in addition to PCIe connection was/is going to be a requirement. Most systems don't have this. ]

So this "plan" never was going to work anyway.

This way when the nMP was released, I could just swap it on over. Then they announce TB2. Now the plan is to buy maybe one or two TB1s to hold me over until TB2 is released.

If you currently have TBv1 systems then that is a reasonable plan. If you have no current TBv1 systems is a fundamentally flawed plan.

I can't count the number of times a store representative or magazine article stated, "the prices for TB(1) will drop once they get mass produced." Now that Apple announced TB2, doesn't that kind of kill TB1?

Some of the reps and magazine folks are engaging in wishful thinking on for a wide spectrum of device price drops. They'll drop simply because they are used/old rather than primarily for TBv2.

In many cases, vendors are simply going to straight retire their TBv1 devices completely. Temporarily may get some 'close out" pricing on "last year's model" but the same drivers of "new tech , tweaked drivers , new R&D" are likely going to apply to TBv2 devices about as much as they applied to TBv1 ones.

Price drops will occur more so where there starts to be a higher multiple of vendors in vendors in a product category. For example going from 1-3 dock vendors to 3-8 dock vendors is likely to drive down pricing. As long as the majority of peripheral vendors don't get into the game the prices will remain relatively higher than those markets where there is higher competition.

TB dongles and cables will probably get cheaper with the TB and associated tranciever process shrinks that will probably arrive in 2014 but those are a minor subset of the TB device market that is at issue here in a Mac Pro context.

 

[deconstruct60]

One related question I'm wondering if someone can answer is; since Thunderbolt 2 is actually just Thunderbolt 1 using two channels at once, then can you connect two Thunderbolt 1 devices to a single Thunderbolt 2 port (i.e - daisy chain) with no loss of speed?

There will be no loss because the TB 1 devices will operate at TB 1 speeds. There is no gain. Loss isn't an issue. Once you plug a TBv1 device into a TB daisy chain everything from that device down rest of the chain is limited to TBv1 bandwidth restrictions.

The TBv2 controller in the system isn't going to "tunnel through" a TBv1 device to deliver TBv2 bandwidth allocations to something downstream. Once you have hit the controller internal switch that only understands TBv1 you are limited to TBv.1 ( even if some of the downstream switch might understand TBv2 protocols you sill have that throttling choke point. )

That way you could at least benefit from Thunderbolt 2 simply by using less ports with no loss of performance,

There is no gain. The TBv2 system will perform no differently than if had a TBv1 controller in your host system.

Each physical port has two channels. Those two channels are STILL bound to the physical ports in TBv2. If using just one physical port then have exact same limitations v1 or v2.


The only relative "loss" that would occur is that if there are two TBv2 ports and devices on the other daisy chain are running at full TBv2 bandwidth speeds. In that case the "loss" due to increased sharing will likely go up. There are more heavyweight bandwidth consumers on the overall Thunderbolt network and still capped at the same x4 PCIe v2 host bus and in most system designs very similar limited DisplayPort host bandwidth too.

 

[deconstruct60]

...
On the other hand, when all desktop manufacturers drop PCIe and leave no other option but TB(x) or whatever the next big thing is, ...

This is same goofy notion that TB is somehow are replacement for for PCIe. It isn't. It displaces PCIe to another phsyical box but does not particularly remove/replace PCIe. You sill have controlllers that want to sit on PCIe bus in the external TB devices. So it is still present.

What you are possibly trying to talk about is PCIe standard slots. Slots aren't PCIe. Even in todays system most PCIe connections (in terms of quantiy of connections ) are not done through standard slots.

I'm not sure whether that will actually happen.

Have to look at what the long term motivating trend is. Highly increased integration of functionality drives down the need for discrete components. If 'all' (really should be vast majority.... there will always be fringe vendors no matter what) desktops drop "slots" because components are all integrated then it can happen.

At one point in time computers all had individual transistors wired into the logic board. Over time people gave that up for increased cost and efficiency of far more integrated components. As long as fab processes keep marching along integration is going to generally go higher over a larger number of products.

Consider that workstation parts are shared with some server components. So far they have no integrated graphics chip,

Doesn't have to be integrated down to just one chip die. Computers with just 4-6 major die packages doesn't really require slots. All of the workstation GPU vendors are driving toward GPU+eDRAM packages. Same general integration forces coupled to ever higher clock/bandwidth speeds.

Even in the server space

" ... He's talking about the heavy footprint of the internet, and how HP's enterprise customers are complaining that they don't have the room or power to meet the needs of the future. HP's solution, of course, ..."
http://www.engadget.com/2013/10/29/HP-Moonshot-servers-announced/


Nobody wants to run physically larger data centers. Well, other than vendors who charge for running larger data centers. "I want my computers smaller" is a general overall long term trend. Can debate whether moving too fast to that mark with specific products at specific points in time, but if opening up to very long term trends. It has been present for quite a while now.

 

[MacPrometheus]

Look at the current Thunderbolt display. It has a TB1 port on the back. Will TB1 significantly impact the speed of a 7200 RPM HDD over at TB2? My guess is Apple is waiting to update the Thunderbolt displays after Christmas. Why else have a TB1 port in them? Maybe because for most items TB2 doesn't offer vast improvement for existing TB1 accessories. My $0.02.

 

[deconstruct60]

Look at the current Thunderbolt display. It has a TB1 port on the back. Will TB1 significantly impact the speed of a 7200 RPM HDD over at TB2?

A single HDD? No. Even single streaming sequential files TBv1 is substnatially faster. For any kind of mixed access that approaches random access it TBv1 blows it out of the water. TBv2 isn't particuarly adding anything in that context.

My guess is Apple is waiting to update the Thunderbolt displays after Christmas. Why else have a TB1 port in them?

Perhaps because not alot of folks are buying them. The iPod Classic hasn't been touched in a very long time. Apple added another color to the iPod Touch line up. That is primarily because they are seriously non-growth products. Enough folks buy them to keep selling, but not enough to put tons of effort into the product.

But yes, there is an aspect to TBv2 that is about timing. Intel's objective always was that 2014 would be the context in which TBv2 would ship in volume.

http://www.anandtech.com/show/7049/intel-thunderbolt-2-everything-you-need-to-know

Not enough TBv2 controllers to go around until 2014 is a good reason to wait till "after Christmas" .... since it will essentially be 2014 by then.

Maybe because for most items TB2 doesn't offer vast improvement for existing TB1 accessories. My $0.02.

For a docking stations like the TB display/docking station it does make a difference. Pragmatically these devices typically are placed closest to the host computer. Because the dock's TB controller had impact on the throughput of the downstream device it isn't just the single device that is important.

For example folks using a "sneaker net" HDD/SSD drive array for moving files from place to place. That typically will go at the end of a chain. It is there temporarily. Sure you can unplug chains and juggle "most bandwidth closest" for every single set of TB devices that show up.... but that can turn into a pain if happens regularly.

For external dongle ( TB -> Ethernet, TB -> FW ) or devices which extremely overlap with classic USB 2.0 territory ( single external bus powered HDD ) there is not much that to be gained with TBv2. But those also tend to be "chain enders" so don't necessarily need to be placed close to the host system on the daisy chain.

 

[nateo200]

You think it'll be possible to couple two thunderbolt 1 ports to form a bigger pipe so to speak? I know thunderbolt 2 is just thunderbolt 1 coupled but I'm imaging say hooking an external GPU up and instead of daisy chaining it plugging in two thunderbolt 1 cables into both ports (if it has two) for 20Gbps one way and another 20Gbps another which would exceed even TB2. Honestly it seams like TB1 could be converted to TB2 and I think Intel and Apple gouged allot of loyal paying customers by not just releasing TB1 with the 20Gbps option like TB2. The whole thing is just weird to me.

 

[deconstruct60]

You think it'll be possible to couple two thunderbolt 1 ports to form a bigger pipe so to speak? I know thunderbolt 2 is just thunderbolt 1 coupled

Thunderbolt 2 is not Thunderbolt 1 coupled the way Ethernet ports are bonded. Thunderbolt 2 is the same Thunderbolt v1 bandwidth just allocated a different way. The channels are physically just as coupled in the wire in v1 as v2. The only thing changing is the logical overlay , better data management controls, and some small behind-the-scenes protocol adjustment. (along with matching the evolution of DisplayPort to v1.2) The external facing physical ports are exactly the same as there were.

Trying to bond two physical TB ports that are provided by the same Thunderbolt controller doesn't buy you anything. Both physical ports share the same "back end" so "bonded"/"coupled" together isn't going to make throughput to the underlying host system go any faster.

Bonding two ports that are each provided by a separate TB controller doesn't really help all that much because because destroying the opaqueness of the PCIe data mapping. Could huff-and-puff to offset the complexity by adding another layer on top but for cases like "storage throughput" it is far simpler just to use the "virutal bonding" mechanisms that already exist at a higher level and just overlay them onto the independent ports ( RAID 0 of storage on port A1 and on port B1 ).

but I'm imaging say hooking an external GPU up and instead of daisy chaining it plugging in two thunderbolt 1 cables into both ports (if it has two) for 20Gbps one way and another 20Gbps another which would exceed even TB2.

How many GPU cards have you seen that plug into two PCIe slots? Exact same impediment issues. Data arrival sync issues because of the independent two paths. Gross mismatch with how the PCIe controller is mapping things out at the lower level. etc.

Rube Goldberg solutions trying to make GPUs work aren't going to be effective because Thunderbolt wasn't primarily designed for that.

Honestly it seams like TB1 could be converted to TB2

It is the exact same set of wires. Essentially that all TBv2 is.

and I think Intel and Apple gouged allot of loyal paying customers by not just releasing TB1 with the 20Gbps option like TB2.

Not segregating the data is actually harder to do with mixed data workloads. If not implemented correctly folks would be gouged a whole lot more. Just doing the logical remapping requires a more complicated controller be built. More complicated , more likely have bugs. Hardware with bugs is just broken. ( there are no 'patches' can distribute ).

Similarly it is cheaper to implement larger logic designs on newer, smaller fab processes. Even if Intel threw manpower at earler design and got out all the bugs it would have cost even more to implement TBv2 on 1-2 year old processes that TB1 went out on.

It is far better just to implement something simpler and then incrementally move to more complex solutions.

The whole thing is just weird to me.

It is normal. USB 1.0 was followed by USB 2.0 which was followed by USB3.0. The gaps are 4-8 years long, but same general technology progression. It is just a slightly different mechanism for doing the progression ( not necessary primarily just clocking the signals on the wire faster. Like much of 1.0 -> 2.0 ).