What is this?!? You are mixing things that don't apply here. You can't share lanes from a PCIe slot without a switch in a Mac Pro and only post X299 chipsets can share lanes without a switch. PCIe is full duplex.
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.
USB 3.x PCIe Cards for Classic Mac Pro
- Thread starter ActionableMango
- WikiPost WikiPost
- Start date
- Sort by reaction score
- Status
Re-reading, I can agree it was badly worded, I've edited a lot of posts to be closer to what I've meant. I came here to try and help but I can see it hasn't changed much, but the controller the in the Sonnet Allegro card has lane negotiation in the MP PCIe 2.0 slot.
"...Sonnet’s Allegro USB-C 4-Port PCIe computer card supports data transfers from a single USB 3.1 Gen 2 SSD RAID device at up to 800 MB/s— compared to using a USB 3.1 Gen 1 or USB 3.0 adapter card, transfers complete in half the time!"
http://www.sonnettech.com/product/allegro-usbc-4port-pcie.html
I'm confused as to how anyone wouldn't see how I stated the only way to get full speed is to dedicate a single controller to a single port-and if I'm wrong on the overhead, then that just means increased speeds are even more likely to be attainable.
EDITED for clarity.
Last edited:
I agree, a two port card is generally less expensive (but if you can name the two port card, then that would make it more clear as it will allow comparing prices).
I would read on their website if you tell me who "they" are and the name of the product you are referring to. I don't think Sonnet uses a PCIe 3.0 USB chip yet so I don't know what product you are talking about. x4/2 doesn't mean anything without context.
In your example, a PCIe 3.0 x2 chip (like the ASM2142) can send data through a PCIe 2.0 x4 link using a PCIe switch chip without much loss in performance, as PCIe 3.0 x2 < PCIe 2.0 x4. And since USB 10 Gbps < PCIe 3.0 x2, then you can send USB 3.1 gen 2 data also without much loss in performance (but only if you're using a single port). Nowhere is there a divide by two (/2) operation.
The slot's bandwidth is shared between the controllers the moment transfer something onto the second controller or the shared port with the first controller. We are literally saying the same thing-I'm just suggesting since they would likely only benefit from the new Gen 2 enclosure they might only need one 10 Gbps controller.
EDIT: Corrected to sharing bandwidth and added more info for clarity.
Last edited:
Definitely not. That's not how PCIe switch chips work. There is no splitting of bandwidth unless you try to send or receive data to both controllers at the same time.
--------------
Ok, I see you are talking about the Allegro USB-C 4 Port PCIe card.
It uses a PCIe 2.0 switch with x4 upstream link and two x2 downstream links connected to two ASM1142.
They say one port can do 800 MB/s. That's higher than I've usually seen (750 MB/s). This is limited by the PCIe 2.0x2 of the ASM1142 controller (1000 MB/s). An ASM2142 controller (PCIe 3.0x2) would allow full USB 3.1 gen 2 speed for a single port (with a PCIe 3.0 switch).
They say two devices, each connected to a different controller, can do 660 MB/s each. It makes sense that there is a drop in performance from that of a single device (more overhead when communicating with more devices). 660x2 = 1320 MB/s is significantly under the PCIe 2.0 x4 (2000 MB/s) limit so that is not a problem (you could expect something like 1600 MB/s from a PCIe 2.0 x4 link).
They say four devices, each connected to a different port, can do 300 MB/s each. It makes sense that this is less than half the performance of the two device result (more overhead when communicating with more devices). Again, this is limited by the PCIe 2.0x2 link of a single ASM1142 controller.
PCIe switches don't act like you think, the bandwidth is SHARED, not split. Even if both controllers are active, you can't guarantee 50% each.
Yes shared is what I've meant (but see how I needed clarification)-but to get full throughput only one controller can be actively transferring data (not to the secondary controller).
EDITED for clarity.
Last edited:
You should learn about how this really work, look how many times we pointed incorrect info on your posts, you are spreading some incorrect and some plain wrong information.
Yes. Think of the PCIe bus like an Ethernet network. Both use switches for connecting multiple devices and to transmit data between devices. Every PCIe device is like a computer on the Ethernet network.
Connecting a USB device to a port of the controller does not make the controller active, or making a controller active does not reduce bandwidth to another controller.
To get full throughput, only one port of one controller can be sending or receiving data because doing the same to a second port on the same link exceeds a bandwidth limit or adds overhead. Overhead cannot be eliminated. Links can be widened or made faster (The ASM2142 uses a faster link, the GC-TITAN RIDGE uses a faster and wider link).
The you all should contact Sonnet Allegro and tell them how wrong they are as well. Best of luck-in your future ventures.
http://www.sonnettech.com/product/allegro-usbc-4port-pcie.html
A connected device consume a slight amount of overhead to stay connected; thats how it stays connected to my knowledge.
http://www.sonnettech.com/product/allegro-usbc-4port-pcie.html
A connected device consume a slight amount of overhead to stay connected; thats how it stays connected to my knowledge.
Last edited:
Do you know what is overhead? Do you really think that a device impinge overhead to others limiting bandwidth?
https://www.xilinx.com/support/documentation/white_papers/wp350.pdf
This started because a member wanted speeds that are achievable when it was initially suggested that it was not-I did my best to push joevt towards the info they were making an assumption on, which led to be them being able to explain what I could not.
Bravo joevt.
I've stated many times-I'm just a user. I would think a hard RAID enclosure would gain some benefit over a SoftRAID array though (but is dependent on the drives and array arrangement).
EDITED for clarity.
Last edited:
So, if you really don't know how things work and like you said you are just an user, it's not better to not spread misinformation? People who can't judge your posts will take it as correct and a lot of things that you posted are just wrong. @joevt explained a lot of how things work down to the protocol level, use his posts as a starting point to learn.
You don't help anyone spreading wrong info.
I attempted to simply point out the spec's of the card first.
The original bandwidth commented on were the SATA II bays speed-in a 4x4TB SoftRAID array running 280 MBps/390 MBps from the bays, being moved to an external enclosure; I hope you realize that, and if it doesn't see an increase I'd be surprised (dependent on original drive speed, array setup, and enclosure controller).
My statement were based on Sonnet Allegro Literature-which states that with a 3.1 Gen 2 enclosure could result in an array speed increase in-comparison to mavots first reported speeds when used with a what I believe to be a dedicated controller. Additional consideration that should be accounted for: Sonnet Allegro is testing with SSD's.
SoftRAID reroutes the SATA bays via slot 4, and if the current USB card is sitting there-it could be bottlenecking an internal softRAID slightly. Rerouting will allow the new USB to potentially circumvent such a potential bottleneck; this is only a possibility if the current card is sitting in slot 4. But keep in mind-joevt's point that these are both ~10 Gbps routes, and IMO a better RAID controller will have a chance of providing better RAID performance, even though the bandwidth is approximately the same.
Be real about expectations of speed improvements though considering you are using HDD's-it probably would not be mind blowing jump, but you could possibly see a bump in performance in best case scenario; best realized by SSD's.
EDITED for clarity.
Last edited:
I did not think my original question would elicit such a discussion. Just now getting back to my Mac and am trying to digest all this.
To clear any confusion, I am using 4x4TB WD Caviar Black HHDs in SoftRAID RAID5 using the internal 3gb/sec SATA III ports.
Aja-280/390 MB/s
I also have a 2x6TB WD Caviar Black HDD backup in RAID 0 in a USB 3.0 enclosure attached to a Sonnet Allegro Pro 4 port USB 3.0 PCIe card. (USB 3.0 enclosure/card combo)
Aja-190/240 MB/s <<<Apples to apples. This is may be the speed indicator that would increase with the new USB C 3.1 Gen 2 enclosure/card combo.
New Sonnet Allegro Pro 4 port USB C 3.1 Gen 2 card would use these ports:
#1-new External USB C 3.1 Gen 2 enclosure w/ the current 4x 4TB RAID 5-What speed increase to expect by moving from Internal 3gb/s bays to USB-C 3.1 Gen 2 enclosure/card combo?
#2-empty
#3-Current 2x6TB USB 3 RAID 0 Backup-no speed increase expected as the enclosure is only USB 3.0. (Maybe a little boost?)
#4-Anker 10-port USB 3.0 hub for peripherals.
Other USB 2 peripherals on the original MP USB 2 ports. Printers, etc.
My understanding is that the Allegro Pro USB-C slots #1 & 2 are independent from #3 & 4. I am hoping that using just 1 of the first two slots will maximize throughput speeds on the card. I'm not concerned with the 2x6TB backup drive. It may actually go back into the internal bays.
bsbeamer - I actually have a KT4004 (it was my original USB PCIe card!) along with the Allegro Pro 4 port USB 3.0 card One had to go when I added a NVME PCIe.
I am selling my 2017 27" iMac i7 4.2. I used a Thunderbay 4 TB3 enclosure with the same 4x4TB WD RAID5 connected to the iMac's TB3 port and speed was about 400/400. I was surprised and thought it would be a lot higher than that. Maybe the issue is the WD Caviar Black HDDs have a max speed. (I also had a NEC wide gamut monitor attached to the second TB3 port. The TB3 ports share one bus. Why Apple, why?)
I'm not happy with the iMac form factor and will use my maxed out Mac Pro for a few years and see what's happening with future Mac Minis and the new modular Mac Pro.
Using the same enclosure will likely result in the similar speeds (maybe a p/u of 220/230), but I will direct you to joevt's earlier post which may explain more technical jargon than I am capable of explaining. Hopefully joevt may be willing to offer additional comment after their exceptional write-up because this is how I understand it as well:
From my understanding of both joevt's breakdown and your scenario you are correct, Controller #1 (1/2) would enjoy full speed while Controller #2 (3/4) would operate at a 5 Gbps bandwidth per port because of your USB 3.0 hardware.
Last edited:
joevt,
My proposed USB-C set up would (see full description above):
New Sonnet Allegro Pro 4 port USB C 3.1 Gen 2 card would use these ports:
#1-new External USB C 3.1 Gen 2 enclosure w/ the current 4x 4TB RAID 5 (WD Caviar Black SATA III 6GB/s HDDs)
#2-empty
#3-Current 2x6TB USB 3 RAID 0 Backup (Same type of WD HDDs)-no speed increase expected as the enclosure is only USB 3.0. (Maybe a little boost?)
#4-Anker 10-port USB 3.0 hub for peripherals.
I'm hoping that using just 1 slot of the first 2 will result in the highest throughput on the new card. My question is how will the RAID5 in a USB-C 3.1 Gen 2 enclosure/card combo perform compared to the 280/390 MB/s I am getting with the internal 3gb/sec bays. Using WD Caviar Black HDDs.
** Perhaps its worth waiting for a new Sonnet card with the ASM3142 chip if this will boost the speed somewhat further. OTOH, maybe it will make no difference. Maybe the WD HDDs just reach a speed limit**
This post from joevt is an excellent foundation to learn what these very commonly used USB 3.1 gen 2 controllers from ASMedia are capable of doing in both PCIe 2.0x4 & PCIe 3.0x2 links. I have learned so much from this post and others like it on this thread. Anyone in the market for a new USB card can refer to this post to gain an understanding of what kind of performance they can expect from similar cards using these controllers.
As crjackson2134 said, it's not the new BootROM update that is making the card undetectable. However, if I were in your position, I would first try a SMC reset then PRAM/NVRAM reset. You may need to reset the PRAM/NVRAM a few times (3x) if it doesn't work right away. Give that a shot.
What card is it?
Still works with Sierra or stopped even with Sierra?
Did you tried moving it to another slot?
I don't think Sonnet said anything wrong in their product page. The numbers they mention make sense as I described at #2280. The bandwidth is shared between each port of a controller (adding up to PCIe 2.0x2) and each controller (adding up to PCIe 2.0x4). A connected device consumes very slight amount of overhead. Very slight. Almost nothing.
Barefeats did some Mac testing at https://barefeats.com/allegro-usb3-gen2-pcie-boards.html
Their single port result (674 MB/s) is much lower than I expected (700+ MB/s) and lower than the 800MB/s suggested by Sonnet considering their iMac Pro Thunderbolt 3 result (PCIe 3.0x4) was 940 MB/s (the highest benchmark result I've seen). They also show a USB 3.0 result of 365 MB/s which seems reasonable (5 Gbps * 8b/10b = 4 Gb/s = 500 MB/s). They mention the x2 PCIe link problem, but did not mention the PCIe 2.0 speed, and did not mention the ASM2142 or ASM3142 alternatives which use a PCIe 3.0 speed.
You mean SATA II (3Gb/s = 300 MB/s; SATA III is 6Gb/s = 600 MB/s). The speed of the WD Black HDs are slower (202 MB/s for 4TB drivers, 227 MB/s for 6TB, according to their specification sheet pdf) than what I measured for my MacPro3,1 SATA II ports (268 MB/s). So, like I said in #2270, the drives are not currently limited by your MacPro5,1, so you probably won't see an improvement in a different enclosure.
240 MB/s seems low when compared to the Barefeats USB 3.0 measurement of 365 MB/s. Maybe there's a problem with the USB 3.0 enclosure since Sonnet says the USB 3.0 PCIe card can do 450 MB/s per port. I don't think you can expect an improvement with a different card.
Makes sense assuming you're correct about the slots #1 & 2 being from a different controller than slots #3 and #4. Keep the slow devices separate from the fast device.
Yup. The WD drives are being maxed in the MacPro5,1 so moving them isn't going to improve things much. Read increased by 10 MB/s which is insignificant (newer/faster computer?). Strange that write increased by 120 MB/s. Maybe a write increase may also occur with the new USB 3.1 gen 2 enclosure.
Yes.
You can test what happens when you connect the 2x6TB to port #2, to see if it affects benchmarks of #1. I don't think it should.
Maybe if you were thinking of making a new faster RAID.