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.
Mac Pro 6,1 Cylinder - Fact Gathering.
- Thread starter Tesselator
- Start date
- Sort by reaction score
With regards to Thunderbolt expanders and using PCIe graphics cards, there's something called Thunderbolterizer from people at mediaPros.co.uk.
They have successfully tested Sonnet Technology Echo Express Pro with a Quadro 4000 over Thunderbolt.
A script/patcher is available for download from their site, and what it appears to be doing is patching the Info.plist of the NVIDIA kexts by adding IOPCITunnelCompatible boolean true.
So if you have a TB expander and a PCIe graphics card that fits in it, you could try the above and report back.
They have successfully tested Sonnet Technology Echo Express Pro with a Quadro 4000 over Thunderbolt.
A script/patcher is available for download from their site, and what it appears to be doing is patching the Info.plist of the NVIDIA kexts by adding IOPCITunnelCompatible boolean true.
So if you have a TB expander and a PCIe graphics card that fits in it, you could try the above and report back.
My guess is that Apple has deliberately held back working TB GPU drivers.
Possibly to keep test results like the ones I posted from getting out before TB could get a foot hold.
It reminds me of a PowerBook I bought new in '97 or so.
It was a 2300c I think. Right after I bought it they came out with G3 PowerBooks. G3 PowerBooks could use Cardbus PCMCIA adapters and have FireWire. Mine would work sometimes and not others. This was early days of Internet and I finally discovered that the capability had always been in the chipset. Apple had added an extra wire to disable Cardbus. To fix it you just had to cut the wire.
They wanted the moment that this feature was available to be a moment of their choosing.
I am guessing that either at nMP launch or shortly thereafter they will unveil the Miracle of PCIE GPUs over TB and possibly have an actual enclosure themselves. The results I showed pretty much guarantee that it will hold either one or 2 cards as that seems the most that could be used with obvious and significant throttling.
They will act as if they have descended from heaven and are handing out free gold bars. Meanwhile they will hope that test results can be swept under the rug while they extoll the wonders and virtues of their benevolence.
If it really just requires patching that kext, I'll bet the smart people who already sell enclosures know all about it and are having their tongues held via ND edicts from Cupertino. I'll be there later today and will see if I can get Tim to admit as much. (Just have to get by security first)
Possibly to keep test results like the ones I posted from getting out before TB could get a foot hold.
It reminds me of a PowerBook I bought new in '97 or so.
It was a 2300c I think. Right after I bought it they came out with G3 PowerBooks. G3 PowerBooks could use Cardbus PCMCIA adapters and have FireWire. Mine would work sometimes and not others. This was early days of Internet and I finally discovered that the capability had always been in the chipset. Apple had added an extra wire to disable Cardbus. To fix it you just had to cut the wire.
They wanted the moment that this feature was available to be a moment of their choosing.
I am guessing that either at nMP launch or shortly thereafter they will unveil the Miracle of PCIE GPUs over TB and possibly have an actual enclosure themselves. The results I showed pretty much guarantee that it will hold either one or 2 cards as that seems the most that could be used with obvious and significant throttling.
They will act as if they have descended from heaven and are handing out free gold bars. Meanwhile they will hope that test results can be swept under the rug while they extoll the wonders and virtues of their benevolence.
If it really just requires patching that kext, I'll bet the smart people who already sell enclosures know all about it and are having their tongues held via ND edicts from Cupertino. I'll be there later today and will see if I can get Tim to admit as much. (Just have to get by security first)
I am not sure how will TB ever gain any foothold when Apple is holding all the keys to all the doors. When a TB-TB 0.5m cable costs $30 and is made and sold by Apple you know that's just plain wrong.
The pathetic state of TB peripheral availability is a good indication of TB not going anywhere, if ever.
We have half a dozen or so TB HDD enclosures that are generally prohibitively expensive compared to the USB ones; maybe a couple of TB hubs that don't work well at all and cost arm and a leg; and a few TB expanders that work with some PCIe cards but not with all/any. Generally, everything is glitchy and people are having all kinds of problems.
On the other side we have USB3(.1) that's dirt cheap, hundreds and hundreds of devices and peripherals. Cables are given out for free with any device, and can be found at almost any corner store, for a few bucks.
Hell, I can buy a microcontroller for a few dollars and make my own device that would send data over a USB without a glitch. I'd love to see TB match that.
The pathetic state of TB peripheral availability is a good indication of TB not going anywhere, if ever.
We have half a dozen or so TB HDD enclosures that are generally prohibitively expensive compared to the USB ones; maybe a couple of TB hubs that don't work well at all and cost arm and a leg; and a few TB expanders that work with some PCIe cards but not with all/any. Generally, everything is glitchy and people are having all kinds of problems.
On the other side we have USB3(.1) that's dirt cheap, hundreds and hundreds of devices and peripherals. Cables are given out for free with any device, and can be found at almost any corner store, for a few bucks.
Hell, I can buy a microcontroller for a few dollars and make my own device that would send data over a USB without a glitch. I'd love to see TB match that.
likewise, I hope they'll magically unveil the ability to connect multiple computers (IP.. not target disk) via thunderbolt.. you can do this with firewire.. not sure why thunderbolt isn't doing it.
#
Yeah! Remote Desktop, VNC, and DIY NAS would be awesome over TB2 - compared to what we can do natively now.
Yeah, 1.5m is fine. KVM is probably going to be less than that. And with RDT or VNC I can control a huge number of machines. KVM alone is an entirely different animal and doesn't offer communication.
Also there are optical cables available where I live in lengths up to 15m (that I've seen) and probably longer ones if I hunt around. Still, I think 25cm to 30cm is about right for my needs and for people wanting to remote control and manage cluster nodes.
Still a silly thing to do. Because you could doesn't mean that you should...
Beside why would you buy an overprice workstation to use it as a RPU/Server?
Again, and if it's business/work related and not for show off, you would be better served with a different box.
As a RPU (remote processing unit), the nMP is innadequate since it lacks the expandability that a more conventional PC brings.
Probably not all the machines in the cluster would be nMP's tho. The nodes would likely be a Wintel, Linux or Hackintosh based on a MB which spots TB.
Using a workstation like the nMP and having it control and manage many headless nodes is a pretty great thing and examples using the same concept but with an alternative interconnect to TB, are already around in great abundance. I have one. At one point I had 48 (EV6 21264) nodes. Currently I have 8 multi-core Xeon nodes. And the whole thing started with a stack of 12 Amiga 2000's back in 1987. I've always used copper-wire ethernet for my interconnects IIRC - Currently 1000BaseTX.
Using a workstation like the nMP and having it control and manage many headless nodes is a pretty great thing and examples using the same concept but with an alternative interconnect to TB, are already around in great abundance. I have one.
At one point I had 48 (EV6 21264) nodes. Currently I have 8 multi-core Xeon nodes. And the whole thing started with a stack of 12 Amiga 2000's back in 1987. I've always used copper-wire ethernet for my interconnects IIRC - Currently 1000BaseTX.
Last edited:
That would be one hell of an expensive node controller/manager.
The thing is, nowadays you can even use a toaster as a node manager and it would work, so investing $5,000+ into Trashcan Pro to manage nodes makes little sense to me.
That would be one hell of an expensive node controller/manager.
The thing is, nowadays you can even use a toaster as a node manager and it would work, so investing $5,000+ into Trashcan Pro to manage nodes makes little sense to me.
to me, buying mac pros as nodes is over the top price wise ..but to drive them? dunno, sounds great to me.. (but I'm sure at least someone is going to hook 20 of these things together.. will be a neat picture but not something I'd even consider investing in
)and besides, I was just talking about using things I already own and only on occasion (a couple weeks per year)... as is- plugging in a couple of laptops when rendering hi-res or multiple images.. or if the render is set up on a laptop in the first place, pluging in to the Mac Pro as a node.
the new wifi will probably be totally fine for doing this but my laptops aren't ac yet and it'd be nice to have the hard wired option anyway
It probably doesn't particularly matter much where the controllers sit for the new Mac Pro. Most likely the controllers are on both ( hub's and CPU's PCIe lanes ) since there is more than one. ( most likely three).
The Mac Pro 2013 doesn't have any competing SATA drives or USB 2.0 connections. The USB 3.0 is likely limited to 1x PCIe lane while one of the TB controllers is sitting on 4x. Except for a corner case of the Mac Pro being used as a router ( to get both Ethernet and Wifi going full tilt) it is is going to be tough to swamp the C602's bandwidth to the CPU package. The TB controller has the biggest ( x4 ) pipe to the C602 and everything else is capped at x1 PCIe v2. In most use cases, not all of the C602's other four lanes is going to get used at max capacity. The audio not going through isn't a big bandwidth factor.
Long term, why would someone with a Mac Pro 2013 be pressed about a 3rd card? There are two inside the box. After Effects is a bit behind Premiere on evolution. At this point there is relatively little CUDA only aspects to Premiere Pro.
".. According to adobe here are the only 4 effects not supported by OpenCL cards. ..."
http://www.dslrfilmnoob.com/2013/07/09/adobe-cc-official-opencl-support-update/
".. Finally, for customers using configurations containing multiple GPUs, Premiere Pro CC can use all of them during export ... "
http://blogs.adobe.com/premierepro/2013/05/improved-gpu-support-in-adobe-premiere-pro-cc.html
In 2012 there were lots of "world going to end" comments because PPro didn't have OpenCL support right then and there. 2013 and it is rolling out and the world didn't end.
If GPGPU is a major issue, the Mac Pro 2013 directly addresses that issue by putting two GPUs inside the system. Thunderbolt was never intended to be a primary contributor to addressing that specific problem.
The primary root cause problem that After Effects has right now is proprietary single source dependency on CUDA. Thunderbolt is just a diversion from that issue.
Last edited:
Apple put the foundation for "Fusion Drive" functionality into CoreStorage the version before they announced it. There isn't much of a historical track record for "big bang" low level updates from Apple. They tend to roll core services things out slowly (e.g., incremental transitions of OpenGL, file system updates , etc.)
Especially in graphics low level service software it takes years to deploy is pretty much standard procedure.
Probably not. Extremely more likely that they want a solution that works well in most conditions. Most of these hacks to get a GPU card into an enclosure go bad is suddenly disconnect the GPU. Apple isn't going to be looking for a solution that appears to work. It will be a solution that does work.
Doubtful. Thunderbolt's 'wheelhouse' is not external GPU cards. Apple knows that and probably will not be trying to push that aspect with a product of their own (since it would highly unlikley have reasonable, for Apple, sales volume). External card enclosures is likely to remain strictly a 3rd part opportunity along with printers, external drives , and monitors (not docking stations) .
When the 2011-2012 Mac laptops start to age and the GPUs become a significant sore spot then there might be a market for "add some incremental" life with an external GPU. But brand new, not even released yet systems being a huge market for external GPUs????? That's the pipe dream here.
It remains to be seen if USB 3.1 cables will be dirt cheap or if dirt cheap actually function well (e.g., limited interference with other devices, achieve rated speeds, etc. ).
Hell, I can buy a microcontroller for a few dollars and make my own device that would send data over a USB without a glitch. I'd love to see TB match that.
Why? What amount of bandwidth does your $1-2 microcontroller need ? That is up there with making a keyboard/mouse with a Thunderbolt connector. The TB controllers cost more than the microcontroller. ( $5-14 http://ark.intel.com/products/series/67021 )
Why does TB have to be a cost and functionality redundant solution with USB ?
Just because you can isn't a particularly good design practice.
The question
The answer
Most nMP will ship with low power entry level cards. Aren't going to age much better than anything else. There will need to be a solution or the iCans will lose value very quickly.
"Now with last year's hottest GPU" isn't a good marketing slogan.
Current MP still has GPUs from 2010, if they don't update faster than that they will be an even bigger laughingstock once they are an integral proprietary part of machine.
And not everyone is in love with or using OpenCl.
The answer
Most nMP will ship with low power entry level cards. Aren't going to age much better than anything else. There will need to be a solution or the iCans will lose value very quickly.
"Now with last year's hottest GPU" isn't a good marketing slogan.
Current MP still has GPUs from 2010, if they don't update faster than that they will be an even bigger laughingstock once they are an integral proprietary part of machine.
And not everyone is in love with or using OpenCl.
Last edited:
Facts....?
Lets me see:
1-Will be black
2-Will be cylindrical
3-Will have Mavericks as OS
4-Heavily based in TB
Forget fact gathering or guessing....wait for the real thing....
Lets me see:
1-Will be black
2-Will be cylindrical
3-Will have Mavericks as OS
4-Heavily based in TB
Forget fact gathering or guessing....wait for the real thing....
Not particularly.
What the low end of the MP 2013 GPU configuration is somewhat open to question. It is extremely likely not a W9000 equivalent specs that Apple has talked about. It also seems doubtful that it will be W5000 equivalents since those couldn't drive 7 displays (according to AMD ). So seems likely will start off around W7000 class. AMD labels that as mid range in the FirePro line up. If you want to spin those as "low power entry cards" you can but it is bit premature to throw that classification at the configs.
It isn't so much how they age but rather where they start off ( how much "headroom" do users have at the start) and how long the device is in service all in combination with user workload.
A laptop with a single GPU is going to run out of headroom (expecially in a GPGPU context ) alot faster if there is just one GPU in the system than a two GPU system. They simply not in the same class in terms of "aging" impact.
The other far more relevant issue is that the number of limited Macbook's is order of magnitude higher than these new Mac Pro's ever will be. To create a market for the solutions there needs to be a sizable number of folks to sell to. The laptop users will be moving from "fast enough" to "fast enough". These 10-25% drop offs from peak performance aren't particularly necessary.
The set of users who are primarily buying either a PCI-e GPU card or CPU package(s) wrapped in box the new Mac Pro design isn't going to be a fit. Those who do more balanced system upgrades every 3-6 years the "external GPU" isn't going to be a factor new GPU typically due about the same time as new CPU.
If you buy a Mac Pro with the primary intent to sell it later ( and not to generate money now) then you are primarily missing the point of buying one IMHO. And frankly even used Mac Pro often don't get sold with bleeding edge video cards in them. Typically users strip parts that could be used in an updated Mac Pro ( or equivalent ) and sell the Mac Pro with the "old stuff" that was put aside... or just stripped.
Folks who buy used are looking for "good enough" performance. It isn't maximum available performance. If they wanted max they'd be buying new. If the aged Mac Pro 2013 is substantially faster than what they are currently sitting on then the future Mac Pro won't particularly use loose value much more than a current design Mac Pro would.
And Apple was going to use this one ? They are also highly unlikely to use the "Now with the most bleeding edge spec porn available this month" . That isn't a particularly good marketing slogan either.
In what way does the Mac Pro 2013 represent a continuation of the 2008-2010 Mac Pro strategy? Sure if Apple continued with their 2010-2012 strategy with the new Mac Pro it would be a failure .... but why would they commit to an abandon/transition strategy for their post transition product?
Frankly, the base Xeon E5 updates are going to be slower than mainstream CPU package updates. Likewise if Apple sticks to the FirePro/Quadro class custom GPU cards this to will be slower than the mainstream GPU card updates. So Mac Pro's probably aren't going to iterate as fast a MacBook's or iMacs. or minis. That has relatively little to do with whether PCIe slots are included or not.
Apple gets custom GPU boards out on time for MBP and now for the iMac on a regular basis. It is puzzling why that would be a problem for the Mac Pro going forward as long as there is a sufficient customer interest to support assigning the R&D resources to moving the product forward over time.
Not everyone is in love with CUDA either. It far from being a critical factor of whether the Mac Pro 2013 will be successful or not.
Frankly, I am excited about this new machine. It is the perfect logical step up from the Mac Mini. This new Mac Mini Pro has a rich history of Mini from where it comes from.
As for Mac Pro owners, you have great machines and hope you can make them keep working for you as long as possible. Just sorry Apple chooses not to offer TB cards (and yes it can be done), their own drivers for USB3 or the upcoming USB 3.1. So we have to close this chapter on the Mac Pro and welcome in the new Mac Mini Pro. I suppose it would look good next to the D-Link AC wifi router given its shape or next to the new AE that is rectangular upright and hope someone then makes a triangular shape upright unit so we can have our kid's building blocks and hope not to make a round 'peg' fit into a rectangular hole that Apple is creating.
As for Mac Pro owners, you have great machines and hope you can make them keep working for you as long as possible. Just sorry Apple chooses not to offer TB cards (and yes it can be done), their own drivers for USB3 or the upcoming USB 3.1. So we have to close this chapter on the Mac Pro and welcome in the new Mac Mini Pro. I suppose it would look good next to the D-Link AC wifi router given its shape or next to the new AE that is rectangular upright and hope someone then makes a triangular shape upright unit so we can have our kid's building blocks and hope not to make a round 'peg' fit into a rectangular hole that Apple is creating.
Or simply the GPU's are replaceable by Apple. From the pictures they look like the regular GPU without the enclosure. So I don't see why one wouldn't be able to buy one from Apple and replace them at home, or Apple will do that for you.
Being workstation cards, and adding in that they'll be "made" for Apple, even if they're just normal ones are going to get extremely expensive. Especially if they're only replaceable by Apple.
Looking at the nMP photos I don't see a normal PCIe connector for them , so they are proprietary to some extent, which will just increases costs more.
Never mind the down time you'll suffer if you're sending off the nMP to get its GPU's upgraded or replaced.
I agree, although after having to send off my rMBP and going nearly bonkers without a decent machine, I decided to order a refurbed 2010/12 Mac Pro.
At least now if something goes wrong I can just change parts myself.
I'm sure the nMP is going to be powerful, especially if the GPUs aren't under powered to fit that form factor (the w9000 can nearly match the 7970 in benchmarks).