I have a hard time believing anything the Red Rocket is doing actually requires PCI-E 16x.
Video is actually a fairly low bandwidth operation. It's 3D graphics that are very high bandwidth.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Thunderbolt 2.0: capable of PCIE 3.0 16x cards in external enclosures?
- Thread starter darkgoob
- Start date
- Sort by reaction score
I have a hard time believing anything the Red Rocket is doing actually requires PCI-E 16x.
Video is actually a fairly low bandwidth operation. It's 3D graphics that are very high bandwidth.
It is a data converter as much as a video output. There is tons to data to drag in. The RAW data format isn't pragmatically useful until converted.
----------
All the benchmarks in the previous posts are for TB v1. There are released TB v2 benchmarks yet. Latency is probably down, but bandwidth isn't going to show a big bump. TB v2 doesn't improve full TB backbone aggregate bandwidth at all.
Netstor have several Thunderbolt products including PCIe chassis.
http://www.netstor.com.tw/_03/products_Thunderbolt_TB.htm
Tess, this might be a stupid question, but do you think it's within the realm of possibilities that some sort of expansion chassis could combine the throughput of 2 or more TB ports?
Last edited:
Thunderbolt ports are no more additive than USB 3.0 ports are.
Two ports from the same TB controllers does not get you a single bit of increased bandwidth. None.
One port each from independent controllers isn't going to get you an additive result in the remote box. So two TB backbone networks each delivering 2 x4 PCI-e lanes of bandwidth won't magically turn into an x8 lane bundle in the remote box. Furthermore the remote box would now also require two TB controllers .... driving up costs on it also.
Could do put in two TB controllers so that each socket in a 2 socket box each supplied x4 PCI-e worth of bandwidth. Only big upside is covering both slots with probably a single longer power supply. The number of cables , cost and complexity are up though.
But even uncompressed video is fairly low bandwidth comparatively. You're going to run into drive constraints way before you even hit a fraction of Thunderbolt's possible speeds, much less PCIe x16.
Yes, the circuitry is pretty simple to accomplish this. So it's entirely possible yes. Will it happen? I doubt it, no. I mean at least not for a single slot PCIe expansion. Such a device still needs to be developed, debugged, fabricated, marketed, and so forth - and the demand for such a thing (IMHO) is minuscule.
Now, in a slightly different way you can do this directly yourself in the area of storage for example. The following in equation form are absolutely true:
(TB2 --> PCIe Expansion + RAID Card + SSD drives = 20Gb/s) x 2 = 40Gb/s throughput
(TB2 --> 4-Drive RAID Enclosure + SSD drives = 20Gb/s) x 2 = 40Gb/s throughput
Of course given 2 controllers. The new MacPro6,1 has three controllers so you could also triple up like so:
(TB2 --> PCIe Expansion + RAID Card + SSD drives = 20Gb/s) x 3 = 60Gb/s throughput
(TB2 --> 4-Drive RAID Enclosure + SSD drives = 20Gb/s) x 3 = 60Gb/s throughput
And the later is about 6GB/s tho there is some overhead. A highly tuned configuration could probably get somewhere around 5.8 to 5.9 GB/s sustained.
Last edited:
So 2 cars going 60 MPH is same as......1 car going 120 MPH? Or isn't it still 2 going 60 MPH?
I agree, we need benchmarks! My hunch is that those cards will absolutely be impacted when using them for display. I guess still very usable but quite squelched at the same time. OTOH when used as compute cores there's some very good evidence which suggests nearly no slow-down at all.
I said it somewhere else previously but I guess it's worth repeating here. I just can't see any rationale for adding a GPU card to a TB2 connector for use as a display when considering the MP6,1.
Considering TB is a method of transporting large amounts of data quickly, two cars going 60MPH is indeed equivalent to one car doing 120MPH in terms of transfer speed.
However, the analogy still holds with it not being the same with latency, or cost.
That said, worst case scenario (if combining 2 TB2 controllers is impossible) I would think it would still be viable to use two TB controllers & cables to connect to two separate mid-range GPUs in the same box, with a Crossfire/SLI bridge. Far from a good solution, but as far as I can tell, it should work.
yes, 120... You mean your RAID0 array isn't faster than one of the single drives which compose the array?
Consider the formula:
X trucks delivering 1ton of goods 100 miles away.
Increasing X increases the delivery speed.
This works for USB2.0 RAID0 sets, over FW, SATA, USB3, you name it. TB/TB2 is no different in that regard.
Last edited:
Only reason I can see is if Apple insists on charging thousands each for their FirePro cards. An enthusiast GPU in an external TB enclosure, as shown in above graphs, can still perform extremely well; at least in some/most games. And it's at least possible it will be much cheaper than Apple's equivalent W9000.
Although I admit, even that reason would pretty much evaporate if Apple can squeeze a CrossFire bridge somewhere in there, and similar performance can then be achieved through 2x W7000's, or something similar, and cheaper.
yes, 120... You mean your RAID0 array isn't faster than one of the single drives which compose the array?
This works for USB2.0 RAID0 sets, over FW, SATA, USB3, you name it. TB/TB2 is no different in that regard.
I still don't follow your math.
If I set up all 3 TB2 connectors on 3 different enclosures and need to copy a single 500GB file, how fast can I move it?
Or, I have 1 King of Prussia. I need to move him 500 miles. Would I be better off with 1 car that does 120 MPH or 2 cars that go 60?
Last edited:
Gb = Gigabits
GB = Gigabytes
at about 5.8 to 5.9 gigabytes per second that would be what? But let's make it a round 1GB file size and a copy operation instead of move - which is slightly different:
A 1GB file would read into RAM (which apple said is 60GB/s) at 1/6th of a second and need another 1/6th of a second to write it from RAM to the new file (on the same volume). So totally 1/3rd of a second ± Alpha.
Keep in mind of course the above example is using all the drives in a single striped volume. If you created 2 striped sets or "volumes" (one per controller to make it easy to follow) it would read and write from volume1 to volume2 at the same time at the speed of TB2 (about 2GB/s) and in that case a 1GB file would transfer in 1/2 a second. (again ± alpha).
PS: if you had 1 king of Prussia might I suggest just cutting off his head and not worry about transporting him anywhere.
GB = Gigabytes
at about 5.8 to 5.9 gigabytes per second that would be what? But let's make it a round 1GB file size and a copy operation instead of move - which is slightly different:
A 1GB file would read into RAM (which apple said is 60GB/s) at 1/6th of a second and need another 1/6th of a second to write it from RAM to the new file (on the same volume). So totally 1/3rd of a second ± Alpha.
Keep in mind of course the above example is using all the drives in a single striped volume. If you created 2 striped sets or "volumes" (one per controller to make it easy to follow) it would read and write from volume1 to volume2 at the same time at the speed of TB2 (about 2GB/s) and in that case a 1GB file would transfer in 1/2 a second. (again ± alpha).
PS: if you had 1 king of Prussia might I suggest just cutting off his head and not worry about transporting him anywhere.
Last edited:
Not sure why this is so difficult.
I have a 500GB file on my new 6,1 on it's SSD. I want to move it to my external storage via TB2.
You claimed that using 3 @ TB2 I can move things at 60GB/s.
Can I move a single 500GB file at that speed or not?
(Note: Not 60GB/s but rather 60Gb/s or about 6GB/s - yes)
I guess the MP6,1 SSD is slower than 6GB/s so it would copy at whatever the speed of the of the slowest volume is. Apple says the internal MP6,1 SSD is 2.5 times as fast as any other SATA SSD so that would be approximately between 1.2GB/s and 1.5GB/s.
Last edited:
Did you just take a bath in oil?
You have gone to a great deal of trouble to NOT answer the question.
Can you move a file from a 6,1 onto an external drive 3 times faster by using 3 TB channels or not?
Why use a 60 Gb/s figure if nobody will ever see more than 20 Gb/s during a file move?
This reminds me of a G4 DP system I saw on Craigslist. The guy said it was 2.4 Ghz. He came to this number by saying there were 2 CPUs both running at 1.2 Gh/z, therefore he had a 2.4 Ghz G4.
I don't think it applies to CPUs anymore than you can claim 60 Gb/s for RAID speed by saying you can run three of them at 20 Gb/s.
If you actually answer the question instead of endlessly finding new ways to NOT answer, I'll let it drop.
But that's your specific question and has little or nothing to do with the topic I was speaking to. So it's like I say there is a dragster that can do 200mph and you wanna know if I can legally drive on a public road that fast? What?
Well, now you're getting closer! Because reads from the 6GB/s volume can feed applications like 4K video editors and/or process (load/save/generate thumbnails) 36mpx RAW images from the D800 at break-neck speeds and/or feed 10 to 30 layers of 1080 video into your compositing application or video editor with ease and/or backup multiple terabytes from one stripe set to another lickety-split and/or screen-grab multiple monitor 4K video without any significant slowdown, etc. etc.
Such stripes also dramatically increase small file I/O (like 4K and etc.) so if you place your OS and apps on one of them your mac will start up in like 3 seconds or something ridiculous. PS will load and open 20 large images in 2 seconds (arbitrary example).
Additionally if you use such a stripe as system or application cache then all of that is dramatically sped up as well.
It's extremely useful to have fast volumes available! I would have thought you of all folks knew this very well. You can ask any content creator or digital artist and they'll tell you the same things. Why do you think so many people use RAID0 stripes in the MP1,1 2,1 3,1 ,4,1 and 5,1? It's awesome bro! Useful!
I seriously doubt users will spend much time file copying from the internal SSD to one of these arrays. That's not what it's for.
Well CPU is a little different. Some apps use only one CPU at a time - so the core speed it very important to specify. With storage I/O the potential speed is always available. With a CPU only a few programs (such as render engines) use all cores at full speed - for those apps it would be OK to say "this machine has 2.4GHz of horsepower".
Anyway, I don't mind if you let it drop or not. You should keep at it until you understand it. Understanding should be your desired result. Personally I doubt I would create a stripe set of 12 SSD drives across three TB controllers unless they became a LOT cheaper! Probably I would like to have two stripe sets: one on a single controller capable of about 2GB/s and one which spans two controllers capable of about 4GB/s. Of course SSDs are still too price prohibitive for me at this point but I guess that will change by this time next year. <shrug>
Last edited:
Yes, the circuitry is pretty simple to accomplish this. So it's entirely possible yes.
Thought so. My non-engineering logic was like this...
If you have a 40 gallon barrel that you want to fill with water, and 1 water hose that delivers 20 GPH, it will take 2 hours to fill the barrel.
If you have a 40 gallon barrel that you want to fill with water, and 2 water hoses that delivers 20 GPH each, it will take 1 hour to fill the barrel.
Water=Data
Hose=Dedicated ThunderBolt port
Anyway, that's why it thought it may be POSSIBLE.
Last edited:
Yup, exactly correct. But you should specify:
Hose=Dedicated ThunderBolt port.
Dedicated of course meaning a discrete physical TB channel (I guess there is one channel per controller but this may be in error - it maybe two dedicated channels per controller given some of the specifics discussed on this site today and yesterday.)
So just like USB. If you have 20MB/s from dedicated USB port1 and 20MB/s from dedicated USB port2 then combining them (via RAID0 or whatever) you can have a total of 40MB/s. However if you place a USB hub (similar to port multiplying - i.e. virtual or "logical" channels) on a 20MB/s dedicated USB port1 and then connect 8 HHDs to that hub (prot1a, port1b, etc.) you will never get more than 20MB/s from that configuration.
I guess you already knew this but just for others reading on who may not.
Last edited: