SATA 6Gb is 750MB/s. 8 bits to a byte.
But another good point is why would you need a fake Thunderbolt PCIe bus when you have a real one on the machine. Instead of putting two Thunderbolt controllers in the away, just put it in as a PCIe card.
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.
Internal Thunderbolt on new MacPro
- Thread starter ekwipt
- Start date
- Sort by reaction score
SATA 6Gb is 750MB/s. 8 bits to a byte.
But another good point is why would you need a fake Thunderbolt PCIe bus when you have a real one on the machine. Instead of putting two Thunderbolt controllers in the away, just put it in as a PCIe card.
8b/10b encoding drops the maximum to 600MB/s. Plus additional overheads because SATA 3Gb/s can't achieve 300MB/s either, more like ~280MB/s in real world.
is there a limitation to how many pci-lanes you can have on a similar motherboard to what the current mac pro has? Couldn't apple just add 10-20 lanes (pcie 3.0 of course) and use these extra lanes for all expansions?
2 TB-ports on the graphics cards (for displays etc)
4 extra ports for expansion replacing all ports (but one of the TB ports is used internally for 3-5 usb-ports on the chassis itself, don't want to be forced to use converters to get usb). This way you could buy a "TB-hub" if you want firewire, extra usb, extra ethernet etc.
This way all communications inside the mac pro could be through the PCI interface, beginning the transformation to a truly modular computer
(a bit OT: imagine a "cpu and memory" controller box with lots of PCI-lanes to which you add gpu-TB-boxes and TB-storage boxes.. no more SATA, no more Firewire, no more usb-drives.. yummy)
2 TB-ports on the graphics cards (for displays etc)
4 extra ports for expansion replacing all ports (but one of the TB ports is used internally for 3-5 usb-ports on the chassis itself, don't want to be forced to use converters to get usb). This way you could buy a "TB-hub" if you want firewire, extra usb, extra ethernet etc.
This way all communications inside the mac pro could be through the PCI interface, beginning the transformation to a truly modular computer
(a bit OT: imagine a "cpu and memory" controller box with lots of PCI-lanes to which you add gpu-TB-boxes and TB-storage boxes.. no more SATA, no more Firewire, no more usb-drives.. yummy)
The number of real PCI-e lanes is set by the core chipset and/or the CPU package. For Sandy Bridge and Ivy Bridge the number of PCI-e v3.0 lanes is set solely by the CPU package.
The system vendors can perpetrate that there are more by adding switches. This is usually where the PC vendor boards with oodles of slots implement that. Slots share lanes so if wanted to put two high bandwidth cards into a slot they'd actually get 1/2 or 1/3 or some other fraction of what a it looks like they get.
TB ports on discrete PCI-e cards doesn't make much sense.
a. The TB controllers need for 4x PCI-e lanes competes with the GPU's need for bandwidth. So putting the two on the same card just means they directly compete with each other.
b. The TB controller is in part a PCI-e switch. Dual GPU cards have a PCI-e switch so that work gets sent to the appropriate GPU. Can imitate that by replacing a TB controller for a GPU. There are two problems. In a Dual GPU card only half the work is sent to the "other" GPU (that includes splitting the PCI-e bandwidth). Here the workload is not being split. Second, have also introduced a switch immediately behind a switch. That should be a clue that something is amiss with the design.
4 more external TB ports??? That will take two more TB controllers and another x8 PCI-e lanes. At this point have ripped another slot worth of lanes bandwidth out of a box that only had 4 PCI-e slots in it in the first place. With the new core chipset may be able to snag 4x PCI-e v2 lanes from it (if any left after ethernete, FW, audio , etc. ), but two controllers would be impacting slot availability and bandwidth.
TB controller used internally? Loopy. PCI-e lanes are used internally. It is cheaper and less complicated. Routing TB singals inside the box to another TB controller is gratuitous encoding of the data. It just adds complexity and cost with no performance improvements and adds latency.
When a Mac Pro needs a docking station.... that again should be a clue that something is amiss with the design.
This way all communications inside the mac pro could be through the PCI interface, beginning the transformation to a truly modular computer
All comm between major components inside the Mac Pro is already in PCI-e interface. TB adds nothing new to the situation.
TB is a somewhat dubious interconnect for a modular computer. It is too slow for anything that requires more than 4x PCI-e v2.0 worth of bandwidth.
How many gallons of TB marketing kool-aid did you drink???
SATA is not going away. eSATA may drop substantially over time, but TB is not going to displace SATA. Nor USB.
TB's modular traction is going to come largely from systems that are heavily space constrained (e.g., the ever thinner laptops) that have historically lacked PCI-e slot like flexibity. For physical system volumes the size of the Mac Pro, that doesn't make much sense. You could shrink the Mac Pro so that it was space constrained so then have to add space back with modules. However, that is an awfully circular rationale. "Shrink it so you can grow it". Stripping away the PCI-e slots to add them back in as TB ports isn't likely to be as effective in bandwidth or cost.
Compatibility and uniformity with what? You are choosing TB only as the solution when that is not necessarily true. I think there are some benefits with putting TB on a Mac Pro. I just don't think is a "must" or hard requirement for majority of Mac Pro users.
For example, let's says users currently have an MBP with ExpressCard with eSATA (or getting by with a dual interface FW800/eSATA0 drive in the field and an eSATA/SAS card in their Mac Pro. A solution involving TB could be to add a eSATA hub to the TB equipped non Mac Pro.
https://forums.macrumors.com/threads/1305703/
In this context, the solution maintains compatibility and uniformity with equipment you already have and a large fraction of the overall PC market which has eSATA solutions. If get a new Mac Pro can move the PCI-e SATA card already have to the new box. Again uniformity and compatibility achieved.
TB doesn't make SATA drives faster than they are. Moving SATA over a wire gives you the same throughput as moving SATA encoded into TB protocol over a different (more expensive) wire.
For example the Seagate GoFlex system which mates a basic USM SATA drive with adapters to mutlinterfae (USB/FW) or TB.
http://www.anandtech.com/show/5345/seagates-goflex-thunderbolt-adapters
That's for single boxes but illustrates TB is a interface transform and not necessarily the solution in all cases. GoFlex is the only drive solution following the USM SATA standard right now but that isn't all that much different than Apple pretty much being the only system vendor following quasi standard TB.
In the Mac Pro space TB in many cases involves throwing out PCI-e card solutions you already have for exact same functionality wrapped in a TB container. For example, The Promise SANLink FC adapter (in the comments section):
"... Different firmware same LSI card. ... "
http://www.xsanity.com/article.php/20110922214420170#comments
I can see the benefit where don't already have the card. In the case, purchase once and use in multiple places. For primarily non Mac Pro users there would that would be a larger fraction because don't have any PCI-e cards. I think the ratio is substantially different if confine to only Mac Pro users.
What this ignores is that the majority of TB peripherals have stuff that plugs into them. The TB device is then plugged into the more constrained Mac that lacks PCI-e slots. With the Mac Pro you can often skip the TB "middle man".
There is only a subset of TB devices that make the additional plug in disappear. PCI-e card + drive container combos is one of those. It is the example people bring up but it is not the most prevalent TB device. Since it is a restricted subset, it is leap to flip that into a "must have" requirement for all Mac Pros.
Unless designers can drop the defacto requirement that PCs (including the Mac Pro) integrate video into the TB output, putting TB on a Mac Pro has almost as many downsides (cost , complexity) as it has upsides. If there was a "data only" TB port (smaller and not display port compatible) it would be a simpler call. Might loose a slot (go from 4 down to 3 ) or just keep the 4th on a PCI-e switch (and pretend there was the enough bandwidth for all 4).
You can add PCIe lanes but you cannot increase the bandwidth. E.g. NF200 chip provided 16 extra PCIe 2.0 lanes, but these 32 lanes shared the bandwidth with all the other lanes. For example, you get 16 lanes from the CPU and then you add NF200 and get a total of 32 lanes. However, the bandwidth to the CPU is only 64Gb/s (PCIe 2.0), so if each lane is in use, the bandwidth is 2Gb/s per lane. The gain is that if one of the PCIe devices is not in use (example setup: GPU and eSATA card), the other gets full x16 bandwidth. Without NF200, you would have two x8 slots, i.e. only x8 bandwidth even if only one device is in use.
The PCIe controller is integrated into the CPU die in Sandy Bridge architecture. In current Mac Pros, the PCIe controller is part of the I/O hub (i.e. chipset). Hence you can't add bandwidth. PCIe 3.0 is twice as fast as PCIe 2.0 though, so we should see a healthy boost because of that. Also, as SNB has the PCIe controller on-die, you actually have two PCIe controllers in dual CPU setups. I'm not 100% sure but in theory, dual CPU setups should have a total of 80 PCIe 3.0 lanes.
Got slammed by my un-educated thoughts again
Oh well it;s been a good read from the more knowledgable folk. Carry on
Sharing your thoughts and asking questions is the whole point here. No problems
I Thunderbolt is already changing the market. Soon it will only be iMac,Mac Mini, and Macbook Pro.
check this out: http://www.sonnettech.com/product/thunderbolt/
I work at a post production commercial company and only could imagine how it would change our workflow. The possibility of being able to connect to our In house unity by a single thunderbolt wire would be awesome. As of right now we can hook up from a laptop through wireless (super slow) or use ethernet but that is bottlenecked. Only our Mac Pro's are hooked via Fibre.
There is talk that the future holds a small box(mac mini) as the CPU and all the peripherals are hook via Thunderbolt. Apple seems to be focused on making their products simpler every refresh, and using third parties to produce the icing on the cake. That is what they are doing with Final Cut X. You can't capture from tape(AJA software), no EDL,AAF,OMF (Auto Duck), etc.
Just my thoughts.
check this out: http://www.sonnettech.com/product/thunderbolt/
I work at a post production commercial company and only could imagine how it would change our workflow. The possibility of being able to connect to our In house unity by a single thunderbolt wire would be awesome. As of right now we can hook up from a laptop through wireless (super slow) or use ethernet but that is bottlenecked. Only our Mac Pro's are hooked via Fibre.
There is talk that the future holds a small box(mac mini) as the CPU and all the peripherals are hook via Thunderbolt. Apple seems to be focused on making their products simpler every refresh, and using third parties to produce the icing on the cake. That is what they are doing with Final Cut X. You can't capture from tape(AJA software), no EDL,AAF,OMF (Auto Duck), etc.
Just my thoughts.