Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Probably, but I wouldn't exclude that my troubles were just because of these Mac Pro slot 2 themselves.

Had half-speed on more than one of them, but other cards worked fine.

It might be something going wrong in the negotiation of the lanes? I am no expert in this regard though

My understanding is that PCIe 2.0 x4 SSD is fine, but PCIe 3.0 x4 SSD will be throttled down to PCIe 1.0 x4 speed in slot 2 for unknown reason. The Amfeltec card is not a ordinary x4 card but x16 card, that may be the reason why it works.

Since you said some of your HyperX Predator works fine, some other throttled down to 1.0 (half speed), I am thinking if some of the SSD are actually PCIe 3.0, but not 2.0.
 
...

Since you said some of your HyperX Predator works fine, some other throttled down to 1.0 (half speed), I am thinking if some of the SSD are actually PCIe 3.0, but not 2.0.

But only on Slot #2
Slot 3 or 4 worked perfectly.

Happened recently also with a Samsung EVO 960 1TB

750MB/s read and write on Slot 2 and 1550MB/S read and write on slot 3 - 4 o_O
 
But only on Slot #2
Slot 3 or 4 worked perfectly.

Happened recently also with a Samsung EVO 960 1TB

750MB/s read and write on Slot 2 and 1550MB/S read and write on slot 3 - 4 o_O

Yes, it's a known issue, only happen on x16 slot (I bet it happen in slot 1 as well, but of course no reason to put the SSD into that slot), but not the x4 slots.

960 Evo is a PCIe 3.0 x4 card, definitely fall into the affected group.
 
Yes, it's a known issue, only happen on x16 slot (I bet it happen in slot 1 as well, but of course no reason to put the SSD into that slot), but not the x4 slots.

960 Evo is a PCIe 3.0 x4 card, definitely fall into the affected group.


I think something else is happening,
because my Samsung 960 work perfectly on my Amfeltec in slot 2 (x16)
 
But only on Slot #2
Slot 3 or 4 worked perfectly.

Happened recently also with a Samsung EVO 960 1TB

750MB/s read and write on Slot 2 and 1550MB/S read and write on slot 3 - 4
It a known issue with the normal x4 adaptor + PCIe 3.0 x4 SSD combo, the Amfeltec card is NOT affect in any slot.
1) Slot 3 and 4 use a different PCIe controller than Slot 1 and 2.
2) Different PCIe 3.0 devices may have different behavior with different PCIe 2.0 controllers. They may skip PCIe 2.0 speed and drop to PCIe 1.0 speed.
3) Each NVMe drive is it's own PCIe device.
4) The amfeltec dual and quad cards use a PCIe switch between the slot and the NVMe drives. So the PCIe device connected to the slot's PCIe controller in that case is the switch. The Gen3 card may drop to Gen1 speed but only for the upstream port of the switch (connected to the PCIe slot) - the NVMe drives are separated from the PCIe slot by the switch and continue to work at Gen3 speed (downstream ports of the switch). The Gen2 card is more compatible, but the NVMe drives drop to Gen2 speed because the switch chip is Gen2. You can confirm this using the lspci command.
5) The switch chip on the amfeltec Gen3 dual and quad cards has the advantage of being able to convert PCIe 3.0 x4 to PCIe 1.0 x8 or x16 (or PCIe 2.0 x8 or x16 in a compatible slot - of which the Mac Pro's slots are not). A card supporting only one NVMe drive can only support x4.
The switch chip on the amfeltec Gen2 quad card has the advantage of being able to convert PCIe 2.0 x4 to PCIe 1.0 x8 or x16. A card supporting only one NVMe drive can only support x4.
 
I've received an amfeltec Gen3 quad M.2 card and did some tests. The lspci command (which I downloaded from https://github.com/pciutils/pciutils ) confirms that it starts at 2.5 GT/s x16 in slot 2 of my Mac Pro 2008.

The following command makes it go to 5 GT/s x16:
Code:
# Set Target Link Speed of MacPro3,1 slot 2 root port to 5GT/s
sudo setpci -s 00:1 CAP_EXP+30.w=2:F

This fix could be put into EFI firmware, or a kernel extension. This fix eliminates any reasons to get the Gen 2 card (except price?) unless there is some other issue?

I'm waiting for more NVMe drives to test the 5 GT/s x16 performance because there is no difference with only one NVMe drive for the Gen 3 card when it runs on a link that can do 4 GB/s (2.5 GT/s x 16, 5 GT/s x8, 8 GT/s x4).
 
Last edited:
  • Like
Reactions: RyanXM and Faraday
Hey, first post here. This thread is probably the biggest "community" of squid users.

I have a 5,1 2009 mac pro, Yosemite 10.10.5, a macvidcards flashed 6GB 980ti, 64GB RAM, pindelski upgrade to 2x 3.46GHz quad core xeon.

16x squid in 16x slot with 4x samsung AHCI 512GB sticks in a RAID 0 totalling 2TB.

Booting off the RAID 0, running off the RAID 0. Use case is video and audio and some CG. Why squid? Cus time is money and I only have so much time.

Have filled up the card several times and then archived off to disk.

When I initially got the card I benchmarked it and hit around 4500MB/s read/write. Low but I needed to work and went ahead with the "good enough" config. I recently noticed my mac getting laggy. Would click on things in various editors and things wouldn't respond right away. Various apps were taking longer to open. I ran the aja benchmark again.

Saw 1000MB/s read 4000MB/s write.

And then watched it drop down to 300MB/s.

Free space at this time was 300GB. Archived a bunch and freed up 1TB. Enabled TRIM and saw it drop down to 600MB/s this time.

WTF not cool. Currently making a fresh clone and will run some tests:

  • boot from USB clone, run aja on card
  • if crappy, then reinstall OSX in place (not fresh) on card, re-run aja
  • if crappy, delete raid 0 set, format, reconfigure raid 0 set, run aja on card from usb drive
  • if good then clone back onto the card, run aja
  • if crappy then ???????
I am hoping that TRIM just needed to be enabled with OSX 10.10 and that the Samsung S4LN058A01-8030 controller just wasn't trimming and garbage collecting properly without that.

Anyways if anyone has any ideas about what might have happened in my environment please let me know. I am also curious if any heavy users with several months under their belt can re-run aja and report their results.

Thanks
FF

There will probably be suggestions below to check the temp of your SM951s, in case of throttling. But this might also help. Check the "recondition" feature: https://diglloydtools.com/disktester.html.
 
  • Like
Reactions: r6mile
There will probably be suggestions below to check the temp of your SM951s, in case of throttling. But this might also help. Check the "recondition" feature: https://diglloydtools.com/disktester.html.
Apparently the writer of that article has never heard of TRIM. The advice in the article is very, very bad.

It says: "I would need to erase the file system on the drive, so that I could write to its entire capacity".

This is a horrible thing to do to an SSD - unless TRIM is active. (And if TRIM were active, they'd never see the slow writes that they're trying to fix.)

It's just bad advice.

Read https://en.wikipedia.org/wiki/Write_amplification to understand why it is so bad.
 
Last edited:
  • Like
Reactions: itdk92 and h9826790
I've received an amfeltec Gen3 quad M.2 card and did some tests. The lspci command (which I downloaded from https://github.com/pciutils/pciutils ) confirms that it starts at 2.5 GT/s x16 in slot 2 of my Mac Pro 2008.

The following command makes it go to 5 GT/s x16:
Code:
# Set Target Link Speed of MacPro3,1 slot 2 root port to 5GT/s
sudo setpci -s 00:1 CAP_EXP+30.w=2:F

This fix could be put into EFI firmware, or a kernel extension. This fix eliminates any reasons to get the Gen 2 card (except price?) unless there is some other issue?

I'm waiting for more NVMe drives to test the 5 GT/s x16 performance because there is no difference with only one NVMe drive for the Gen 3 card when it runs on a link that can do 4 GB/s (2.5 GT/s x 16, 5 GT/s x8, 8 GT/s x4).
I've received 4 Samsung 960 Pro 1TB SSDs and did some tests to confirm that 5 GT/s x16 works in the 2008 Mac Pro.

To change the link speed of a card:
1) Find the location of the card using "sudo lspci -nnvt". The following is the results of the amfeltec card in slot 1 of the 2008 Mac Pro.
Code:
-[0000:00]-+-00.0  Intel Corporation 5400 Chipset Memory Controller Hub [8086:4003]
           +-01.0-[01]--+-00.0  NVIDIA Corporation GK104 [GeForce GTX 680] [10de:1180]
           |            \-00.1  NVIDIA Corporation GK104 HDMI Audio Controller [10de:0e0a]
           +-05.0-[02-07]----00.0-[03-07]--+-08.0-[04]----00.0  Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961 [144d:a804]
           |                               +-09.0-[05]----00.0  Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961 [144d:a804]
           |                               +-0a.0-[06]----00.0  Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961 [144d:a804]
           |                               \-0b.0-[07]----00.0  Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961 [144d:a804]
2) The root port containing the amfeltec gen 3 card is at PCIe bus 00, device 05, function 0, or "00:5"
3) Use the following commands to change the link speed (set the speed and root port variables). In the Mac Pro, the possible values are 1 (2.5 GT/s 8b/10b) or 2 (5 GT/s 8b/10b). A Hackintosh may have PCIe 3.0 slots supporting speed 3 (8.0 GT/s 128b/130b).
Code:
# Set Target Link Speed to 2.5GT/s of MacPro3,1 slot 1 root port
speed=2
rootport=00:5
linkstatus=$(sudo setpci -s $rootport CAP_EXP+12.w); echo '# Before: PCIe' $(( 0x$linkstatus & 15)).0 x$(( 0x$linkstatus >> 4 & 31))
sudo setpci -s $rootport CAP_EXP+30.w=$speed:F # Set Link Speed
sudo setpci -s $rootport CAP_EXP+10.w=20:20 # Start Retrain
linkstatus=$(sudo setpci -s $rootport CAP_EXP+12.w); echo '#  After: PCIe' $(( 0x$linkstatus & 15)).0 x$(( 0x$linkstatus >> 4 & 31))

I've tested various link speeds of the gen 3 amfeltec card in both a 2008 Mac Pro and a Hackintosh (GA-Z170X Gaming 7). The 2008 Mac Pro seems to be able to get only slightly above 4000 MB/s using 5GT/s x16. This is much lower than the 2010 Mac Pro is capable of with the gen 2 amfeltec card, but it is still greater than what 2.5 GT/s x16 allows which proves that 5 GT/s x16 is working. It may be that the 2008 Mac Pro's CPU and RAM don't allow this benchmark to saturate the PCIe slot's bandwidth which means that a gen 2 amfeltec card would have the same result in the 2008 Mac Pro.

In the Hackintosh, I tested both x8 and x16 slots using AJA System Test Lite 12.4.3. The following are for the average Read speeds for 16 GB file size:

x8 slot:
  1. At speed 1, the x8 slot is saturated by one SSD (1610 MB/s).
  2. Full performance from one SSD requires speed 2 (2880 MB/s).
  3. At speed 2, the x8 slot is saturated by two SSDs (3099 MB/s).
  4. Full performance from two SSDs requires speed 3 (4998 MB/s).
  5. At speed 3, the x8 slot is saturated by three SSDs (5376 MB/s).
  6. In an x8 slot, full performance from three SSDs is not achievable.
  7. In an x8 slot, full performance from four SSDs is not achievable.

x16 slot:
  1. At speed 1, the x16 slot cannot be saturated by one SSD.
  2. Full performance from one SSD requires speed 1 (2864 MB/s).
  3. At speed 1, the x16 slot is saturated by two SSDs (3047 MB/s).
  4. Full performance from two SSDs requires speed 2 (4924 MB/s).
  5. At speed 2, the x16 slot is saturated by three SSDs (5304 MB/s).
  6. Full performance from three SSDs require speed 3 (6359 MB/s).
  7. Full performance from four SSDs requires speed 3 (7467 MB/s).
You can see that each x16 result is similar to the x8 result at double the speed.

The speed 2 x16 result is similar to the 2010 Mac Pro barefeats.com result for the gen 2 amfeltec card meaning that both cards and both computers are able to saturate the PCIe 2.0 x16 slot but the gen 3 card can do that with just 3 SSDs.
 
I've received 4 Samsung 960 Pro 1TB SSDs and did some tests to confirm that 5 GT/s x16 works in the 2008 Mac Pro.

To change the link speed of a card:
1) Find the location of the card using "sudo lspci -nnvt". The following is the results of the amfeltec card in slot 1 of the 2008 Mac Pro.
Code:
-[0000:00]-+-00.0  Intel Corporation 5400 Chipset Memory Controller Hub [8086:4003]
           +-01.0-[01]--+-00.0  NVIDIA Corporation GK104 [GeForce GTX 680] [10de:1180]
           |            \-00.1  NVIDIA Corporation GK104 HDMI Audio Controller [10de:0e0a]
           +-05.0-[02-07]----00.0-[03-07]--+-08.0-[04]----00.0  Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961 [144d:a804]
           |                               +-09.0-[05]----00.0  Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961 [144d:a804]
           |                               +-0a.0-[06]----00.0  Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961 [144d:a804]
           |                               \-0b.0-[07]----00.0  Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961 [144d:a804]
2) The root port containing the amfeltec gen 3 card is at PCIe bus 00, device 05, function 0, or "00:5"
3) Use the following commands to change the link speed (set the speed and root port variables). In the Mac Pro, the possible values are 1 (2.5 GT/s 8b/10b) or 2 (5 GT/s 8b/10b). A Hackintosh may have PCIe 3.0 slots supporting speed 3 (8.0 GT/s 128b/130b).
Code:
# Set Target Link Speed to 2.5GT/s of MacPro3,1 slot 1 root port
speed=2
rootport=00:5
linkstatus=$(sudo setpci -s $rootport CAP_EXP+12.w); echo '# Before: PCIe' $(( 0x$linkstatus & 15)).0 x$(( 0x$linkstatus >> 4 & 31))
sudo setpci -s $rootport CAP_EXP+30.w=$speed:F # Set Link Speed
sudo setpci -s $rootport CAP_EXP+10.w=20:20 # Start Retrain
linkstatus=$(sudo setpci -s $rootport CAP_EXP+12.w); echo '#  After: PCIe' $(( 0x$linkstatus & 15)).0 x$(( 0x$linkstatus >> 4 & 31))

I've tested various link speeds of the gen 3 amfeltec card in both a 2008 Mac Pro and a Hackintosh (GA-Z170X Gaming 7). The 2008 Mac Pro seems to be able to get only slightly above 4000 MB/s using 5GT/s x16. This is much lower than the 2010 Mac Pro is capable of with the gen 2 amfeltec card, but it is still greater than what 2.5 GT/s x16 allows which proves that 5 GT/s x16 is working. It may be that the 2008 Mac Pro's CPU and RAM don't allow this benchmark to saturate the PCIe slot's bandwidth which means that a gen 2 amfeltec card would have the same result in the 2008 Mac Pro.

In the Hackintosh, I tested both x8 and x16 slots using AJA System Test Lite 12.4.3. The following are for the average Read speeds for 16 GB file size:

x8 slot:
  1. At speed 1, the x8 slot is saturated by one SSD (1610 MB/s).
  2. Full performance from one SSD requires speed 2 (2880 MB/s).
  3. At speed 2, the x8 slot is saturated by two SSDs (3099 MB/s).
  4. Full performance from two SSDs requires speed 3 (4998 MB/s).
  5. At speed 3, the x8 slot is saturated by three SSDs (5376 MB/s).
  6. In an x8 slot, full performance from three SSDs is not achievable.
  7. In an x8 slot, full performance from four SSDs is not achievable.

x16 slot:
  1. At speed 1, the x16 slot cannot be saturated by one SSD.
  2. Full performance from one SSD requires speed 1 (2864 MB/s).
  3. At speed 1, the x16 slot is saturated by two SSDs (3047 MB/s).
  4. Full performance from two SSDs requires speed 2 (4924 MB/s).
  5. At speed 2, the x16 slot is saturated by three SSDs (5304 MB/s).
  6. Full performance from three SSDs require speed 3 (6359 MB/s).
  7. Full performance from four SSDs requires speed 3 (7467 MB/s).
You can see that each x16 result is similar to the x8 result at double the speed.

The speed 2 x16 result is similar to the 2010 Mac Pro barefeats.com result for the gen 2 amfeltec card meaning that both cards and both computers are able to saturate the PCIe 2.0 x16 slot but the gen 3 card can do that with just 3 SSDs.

Is your command useful only for a 2008 Mac Pro? Would it be needed if using the same Amfeltec card on a mid 2012 Mac Pro?
 
Is your command useful only for a 2008 Mac Pro? Would it be needed if using the same Amfeltec card on a mid 2012 Mac Pro?
If the 2012 Mac Pro boots with the gen 3 amfeltec card set to gen 1 speed then you need to use the command to get gen 2 speed. I suppose a kernel extension would be preferable, or a EFI patch which can affect all OSs that you choose to run on the Mac (macOS, Linux, Windows).

The commands should work on any computer. I also used it on a Hackintosh with PCIe 3.0 slots to test PCIe 2.0 and PCIe 1.0 speeds.

I redid the tests using SoftRaid 5.6.3 (64K stripe size – optimized for workstation) and AJA System Test Lite 12.4.3 3840x2160 4K RED HD, 16 GB file size, 16bit RGB. My previous results used Disk Utility.

Results:

MacPro3,1 x16 slot 2:
  1. At speed 1, the x16 slot cannot be saturated by one SSD.
  2. Full performance from one SSD requires speed 1 (2942 MB/s).
  3. At speed 1, the x16 slot is saturated by two SSDs (3045 MB/s).
  4. Full performance from two SSDs requires speed 2 (5086 MB/s).
  5. At speed 2, the x16 slot is saturated by three SSDs (5226 MB/s).
  6. Full performance from three SSDs is not achievable.
  7. Full performance from four SSDs is not achievable (performance drops to 4985 MB/s).

Hackintosh x8 slot:
  1. At speed 1, the x8 slot is saturated by one SSD (1658 MB/s).
  2. Full performance from one SSD requires speed 2 (3140 MB/s) or speed 3 (3194 MB/s).
  3. At speed 2, the x8 slot is saturated by one (3140 MB/s) or two SSDs (3270 MB/s).
  4. Full performance from two SSDs requires speed 3 (6032 MB/s).
  5. At speed 3, the x8 slot is saturated by three SSDs (6199 MB/s).
  6. In an x8 slot, full performance from three SSDs is not achievable.
  7. In an x8 slot, full performance from four SSDs is not achievable.

Hackintosh x16 slot:
  1. At speed 1, the x16 slot cannot be saturated by one SSD.
  2. Full performance from one SSD requires speed 1 (3171 MB/s) or speed 2 (3198 MB/s).
  3. At speed 1, the x16 slot is saturated by one (3171 MB/s) or two SSDs (3277 MB/s).
  4. Full performance from two SSDs requires speed 2 (6091 MB/s) or speed 3 (6165 MB/s).
  5. At speed 2, the x16 slot is saturated by three SSDs (6315 MB/s).
  6. Full performance from three SSDs requires speed 3 (8733 MB/s).
  7. Full performance from four SSDs requires speed 3 (11141 MB/s).

Summary:

The MacPro3,1 slot 2 results are much more reasonable now. It’s probably still not as good as the 2010 Mac Pro but much closer than before.

The Hackintosh results are much improved. SoftRaid’s RAID 0 performance scaling is much more linear. SoftRaid’s performance for a single SSD is so good that you can get slightly more performance by doubling the link bandwidth of the upstream port. Poor performance from the gen 3 x4 NVMe SSD would see no benefit from a better upstream link (the links are like pipes, and small items going through the pipe are not hindered by misaligned pipes). When the performance of the SSD nears the limit of the gen 3 x4 M.2 slot, then an upstream link of the same bandwidth (gen 1 x16 or gen 2 x8 or gen 3 x4 - which is not tested here) would have a slight negative impact because it has a similar limit (like two pipes of the same size but not perfectly aligned). This is why you can get slightly more performance (between 30 and 100 MB/s) from one fast SSD (gen 3 x4) when the upstream port has more bandwidth (gen 2 x16 or gen 3 x8) (the little pipe fits in the big pipe even if it’s not centered).
 
  • Like
Reactions: splifingate
Has anyone used the Amfeltec carrier board with a Thunderbolt PCIe expansion chassis? Barefeats has an article where they used a TB3 patch, but I was wondering if this is still necessary for a Sonnet Echo Express SE III TB3 chassis.

http://barefeats.com/tricktb3.html

I would love to have these speeds on an iMac or MBP without any hacks.

I just did this Amfeltec upgrade. First I got the 2-slot card and two SM951's. But I was so happy with the performance I wanted more (Isn't that how it always goes?). I wanted to dump the rest of my SATA SSD's and spinners. So I bought the 4-slot card and two more SM951's. The whole thing was pretty pricey and Amfeltec gives no choice on shipping. So on top of the initial price it cost $44.00 for shipping Priority FedEx and another $16.50 for a paypal processing fee.

The pain of the purchase will fade eventually =).

But I now have (4) 512GB smoking fast drives that can boot to macOS Sierra or Win 10 with no hacks or issues.

(1) SQUID PCIe Gen2 Carrier Board for up to 4 M.2 PCIe SSD modules - $427.00

(4) Samsung SM951 512GB PCIe AHCI M.2 SSD MZHPV512HDGL - $300.00 each = $1,200.00

Total Cost: $1,627.00 (OUCH!)
 
I would like to understand the difference in performance between the
Squid PCI Express Gen 2 Carrier Board for M.2 SSD modules
and the
Squid PCI Express Gen 3 Carrier Board for 4 M.2 SSD modules
in the classic Mac Pro.

The barefeats.com article at http://barefeats.com/hard220.html concluded that the Gen 3 version has less performance than the Gen 2 version.
gen 2 is for recommended for Mac because the Mac pros are pci-e 2.0 and you will get better performance using Squid PCI Express Gen 2 Carrier Board on a Mac pro but if you have a hackintosh that use pci-e 3.0 then you get the maximum speed using Squid PCI Express Gen 3 Carrier Board

is not that Squid PCI Express Gen 2 is faster than Squid PCI Express Gen 3
is just that version 2 runs better on a Mac Pro
version 3 is definitely faster but it won't work as good as version 2 on a mac pro because the pci-e 2.0 bus speed

but on a hackintosh a version 3 will smoke and blow a version 2 out of the water
as long as that hackintosh has a 16X pci-e 3.0 slot

this is the new card but it is for Mac Pro 2013
http://amfeltec.com/products/mac-pro-late-2013-carrier-board-for-m-2-pcie-ssd-modules/
 

Now this is interesting...it makes me wonder what the "engineers" at Apple are thinking at times. There is plenty of space for more slots, but they'll decided it wasn't worth it.

2_top.jpg


It even has a heatsink....

1.jpg


Fully installed....

0.jpg
 
Last edited:
  • Like
Reactions: iamtheonlyone4ever
Hi Carpsafari,

Thank you for your response.

I think there's a lot of confusion about what devices work and how well they work, in part because this threaded forum layout is not a good organization for a body of knowledge but also there's misleading or contradictory information across multiple threads.

I can see how you might of thought that I had not read the last 194 messages in this thread because I asked a question about compatibility with the Aplicata adapter. But I assure you, I had already read through this entire thread as well as much of the 'SATA Express meets the '09 Mac Pro - Bootable NGFF PCIE SSD' thread as well, and reading these threads only served to increase my confusion.

Let me ask a different, more definitive question, that once answered may also serve to clear the confusion for other users: Does the Mac Pro (4,1 & 5,1) support PCI Bifurcation, and if so, in passive or only active forms?

Thank you for your thoughtful reply,
Your most humble servant,
Eksu.
good question , I know you didn't asked me but I found out about this, some cards do requiere a compatible motherboard that supports that option "bifurcation" 4x4x4x4x = 16X ,but if I'm not mistaken and I understood things correctly there are some card that doesn't need or requiere "bifurcation" because they have some special chip on the m.2 adapter board that takes care of that and it will see the card as a single 16x card, I can't confirmed this because I just read about it in some page but I will order my squid gen 3 tomorrow , then I will know for sure, I really like that card but man that green pub color is killing me but that card is the best option, is bootable on Mac OS , I have a hackintosh so gen 3 is the option for me but now the problem is High Sierra, I will only be able to raid using HFS+ that is if I want to use the card as bootable device, if I'm going to use the card as storage only then I can make the raid on APFS.
if any of those 2 Macs don't have support for bifurcation then you need to Get a card that doesn't requiere that feature or option.
[doublepost=1512055395][/doublepost]
Now this is interesting...it makes me wonder what the "engineers" at Apple are thinking at times. There is plenty of space for more slots, but they'll decided it wasn't worth it.
what I really like is the pcb color that is brown , while the squid gen 3 is green , I really will like a brown card but the ones that are brown do no work for Mac or hackintosh, they are windows only and some of them are not even bootable, they are storage only like the new good looking high point card, that one looks beautiful but no Mac support and they are storage only. I'm so upset haha
 
Last edited:
gen 2 is for recommended for Mac because the Mac pros are pci-e 2.0 and you will get better performance using Squid PCI Express Gen 2 Carrier Board on a Mac pro but if you have a hackintosh that use pci-e 3.0 then you get the maximum speed using Squid PCI Express Gen 3 Carrier Board

is not that Squid PCI Express Gen 2 is faster than Squid PCI Express Gen 3
is just that version 2 runs better on a Mac Pro
version 3 is definitely faster but it won't work as good as version 2 on a mac pro because the pci-e 2.0 bus speed

but on a hackintosh a version 3 will smoke and blow a version 2 out of the water
as long as that hackintosh has a 16X pci-e 3.0 slot
I guess you didn't read my later post #214 where I show that the Squid Gen 3 is faster than the Gen 2 for one SSD even when it runs at Gen 1 speed.

I also showed how to get the Gen 3 card to work at Gen 2 speed, which gives better performance for two SSDs in raid 0 than the Gen 2 card. The Gen 3 card can saturate the Gen 2 slot with three SSDs in Raid 0, but the Gen 2 card requires four SSDs in Raid 0 to do that.

Amfeltec support said the following:
"We are planning to include in the next Carrier board Gen 3 revision hardware option for downgrade connection speed to Gen 2. The new rev we are planning to release in end of Q4 / beginning Q1."


good question , I know you didn't asked me but I found out about this, some cards do requiere a compatible motherboard that supports that option "bifurcation" 4x4x4x4x = 16X
Macs don't support bifurcation.
but if I'm not mistaken and I understood things correctly there are some card that doesn't need or requiere "bifurcation" because they have some special chip on the m.2 adapter board that takes care of that and it will see the card as a single 16x card, I can't confirmed this because I just read about it in some page but I will order my squid gen 3 tomorrow , then I will know for sure.
The amfeltec cards use a PCI Express Switch chip. The lspci command shows that the Gen 3 card uses the following:
Code:
PCI bridge [0604] (class code): PLX Technology, Inc. PEX 8732 32-lane, 8-Port PCI Express Gen 3 (8.0 GT/s) Switch [10b5:8732] ([vendor id:device id]) (rev ca)
A switch chip has an upstream port and a one or more downstream ports. The Squid Gen 3 uses 5 ports, and all 32 lanes of the PEX 8732. The upstream port (connected to the PCIe slot) is x16. The four downstream ports (connected to the NVMe devices) are each x4. The upstream port can run at Gen 1, Gen 2, or Gen 3 speed (The classic Mac Pro PCIe slots support Gen 1 and Gen 2 speeds). The downstream ports match the speed of the connected NVMe device (usually Gen 3).
 
I guess you didn't read my later post #214 where I show that the Squid Gen 3 is faster than the Gen 2 for one SSD even when it runs at Gen 1 speed.

I also showed how to get the Gen 3 card to work at Gen 2 speed, which gives better performance for two SSDs in raid 0 than the Gen 2 card. The Gen 3 card can saturate the Gen 2 slot with three SSDs in Raid 0, but the Gen 2 card requires four SSDs in Raid 0 to do that.

Amfeltec support said the following:
"We are planning to include in the next Carrier board Gen 3 revision hardware option for downgrade connection speed to Gen 2. The new rev we are planning to release in end of Q4 / beginning Q1."



Macs don't support bifurcation.

The amfeltec cards use a PCI Express Switch chip. The lspci command shows that the Gen 3 card uses the following:
Code:
PCI bridge [0604] (class code): PLX Technology, Inc. PEX 8732 32-lane, 8-Port PCI Express Gen 3 (8.0 GT/s) Switch [10b5:8732] ([vendor id:device id]) (rev ca)
A switch chip has an upstream port and a one or more downstream ports. The Squid Gen 3 uses 5 ports, and all 32 lanes of the PEX 8732. The upstream port (connected to the PCIe slot) is x16. The four downstream ports (connected to the NVMe devices) are each x4. The upstream port can run at Gen 1, Gen 2, or Gen 3 speed (The classic Mac Pro PCIe slots support Gen 1 and Gen 2 speeds). The downstream ports match the speed of the connected NVMe device (usually Gen 3).
yes most likely I missed that part, I don't know if the squid card supports windows on a PC because the lack of documentation in Amfeltec web site, but they do work on a hackintosh PC to install Mac OS, at this point I don't know if windows is supported. another important part that is missing is how high can the card really go, let's just say with 4 samsung pro or evo, no info is available either, the problem that I have is with the 4th m.2 disk , I have no where to install it because I have a Nvidia 1080 card but with just 3 regular 4x pci-e cards I get over 8,000 in read speed and over 4,000 in write , don't remember the actual number right now, that's why I need a 16x card but I'm not going to pay 580 including shipping and PayPal expenses if the card can only go up t 8000 or 9000, the high point card is certified to reach at least up to 12,000 or even a little bit more depending on what drives are installed, keep in mind that you need a powerful CPU to be able to push the card and the M.2 drive to archive the maximum speed. the only problem with the high point card is that it doesn't support Mac OS, if anybody has any info about if the squid card can go over 10,000 please , I will appreciate, thank you
[doublepost=1512080103][/doublepost]about what you told me, there are 2 articles in bare feet
this is one is for gen 2
http://barefeats.com/hard210.html
and this one is for gen 3
http://barefeats.com/hard220.html
there you will be able to see and understand the differences between the 2 with different test
 
Last edited:
What version of macOS are you installing the PCIUTILS onto? High Sierra doesn't seem to be supported.
I'm using High Sierra. I use pciutils from https://github.com/pciutils/pciutils
Download the zip then use Terminal.app to execute these commands:
Code:
cd ~/Downloads/pciutils-master
sudo make install
Then append /usr/local/sbin to the file /etc/paths
Then in a new Terminal.app window, execute:
Code:
sudo update-pciids
sudo nvram boot-args="debug=0x144"
Restart the computer. Now you can run some pciutils commands like this:
Code:
sudo lspci -nnvt
I've attached a script that does the same thing as "lspci -nnvt" except it also lists the PCIe switches, and the link width and link speed capability (max values) and status (current values). It includes example output from a Z170X Gaming 7 with a OWC Thunderbolt 2 dock connected to a OWC Thunderbolt 3 dock.
Make sure the script has execute privilege.
Code:
chmod 744 pcitree.sh
Then execute it like this:
Code:
sudo pcitree.sh
[doublepost=1512119444][/doublepost]
yes most likely I missed that part, I don't know if the squid card supports windows on a PC because the lack of documentation in Amfeltec web site, but they do work on a hackintosh PC to install Mac OS, at this point I don't know if windows is supported.
Windows has drivers for PCI switches, SATA drives, and NVMe drives. The amfeltec card does nothing special. I'm not sure there's a way that it couldn't be supported in Windows...

another important part that is missing is how high can the card really go, let's just say with 4 samsung pro or evo, no info is available either, the problem that I have is with the 4th m.2 disk , I have no where to install it because I have a Nvidia 1080 card but with just 3 regular 4x pci-e cards I get over 8,000 in read speed and over 4,000 in write , don't remember the actual number right now, that's why I need a 16x card but I'm not going to pay 580 including shipping and PayPal expenses if the card can only go up t 8000 or 9000, the high point card is certified to reach at least up to 12,000 or even a little bit more depending on what drives are installed, keep in mind that you need a powerful CPU to be able to push the card and the M.2 drive to archive the maximum speed. the only problem with the high point card is that it doesn't support Mac OS, if anybody has any info about if the squid card can go over 10,000 please , I will appreciate, thank you
You need to reread my post a third time. It includes a test where I got over 11GB/s in macOS using an x16 slot in a Hackintosh. The Hackintosh I used only has one x16 slot. You need to check with your motherboard documentation for the options available to you. The benchmark and OS I used may produce different results than the benchmark and OS used for that High Point card. The point is that you can achieve greater than the maximum theoretical performance of a PCIe 3.0 x8 slot (or a PCIe 2.0 x16) slot when you use a PCIe 3.0 x16 slot.

[doublepost=1512080103][/doublepost]about what you told me, there are 2 articles in bare feet
this is one is for gen 2
http://barefeats.com/hard210.html
and this one is for gen 3
http://barefeats.com/hard220.html
there you will be able to see and understand the differences between the 2 with different test
I've emailed rob-ART to update/correct those articles with the information from my tests. To reiterate, the problem with the gen 3 card in the Mac Pro is that the upstream port starts up at gen 1 speed instead of gen 2 speed. The down stream ports start up correctly at gen 3 speed, making the gen 3 card faster than the gen 2 card anyway for individual M.2 devices. You can set the gen 3 card's upstream port to gen 2 speed after startup. You could create some code to do that in EFI, so it works with any OS, before the OS starts.
 

Attachments

  • pcitree.zip
    4 KB · Views: 354
I'm using High Sierra. I use pciutils from https://github.com/pciutils/pciutils
Download the zip then use Terminal.app to execute these commands:
Code:
cd ~/Downloads/pciutils-master
sudo make install
Then append /usr/local/sbin to the file /etc/paths
Then in a new Terminal.app window, execute:
Code:
sudo update-pciids
sudo nvram boot-args="debug=0x144"
Restart the computer. Now you can run some pciutils commands like this:
Code:
sudo lspci -nnvt
I've attached a script that does the same thing as "lspci -nnvt" except it also lists the PCIe switches, and the link width and link speed capability (max values) and status (current values). It includes example output from a Z170X Gaming 7 with a OWC Thunderbolt 2 dock connected to a OWC Thunderbolt 3 dock.
Make sure the script has execute privilege.
Code:
chmod 744 pcitree.sh
Then execute it like this:
Code:
sudo pcitree.sh
[doublepost=1512119444][/doublepost]
Windows has drivers for PCI switches, SATA drives, and NVMe drives. The amfeltec card does nothing special. I'm not sure there's a way that it couldn't be supported in Windows...


You need to reread my post a third time. It includes a test where I got over 11GB/s in macOS using an x16 slot in a Hackintosh. The Hackintosh I used only has one x16 slot. You need to check with your motherboard documentation for the options available to you. The benchmark and OS I used may produce different results than the benchmark and OS used for that High Point card. The point is that you can achieve greater than the maximum theoretical performance of a PCIe 3.0 x8 slot (or a PCIe 2.0 x16) slot when you use a PCIe 3.0 x16 slot.


I've emailed rob-ART to update/correct those articles with the information from my tests. To reiterate, the problem with the gen 3 card in the Mac Pro is that the upstream port starts up at gen 1 speed instead of gen 2 speed. The down stream ports start up correctly at gen 3 speed, making the gen 3 card faster than the gen 2 card anyway for individual M.2 devices. You can set the gen 3 card's upstream port to gen 2 speed after startup. You could create some code to do that in EFI, so it works with any OS, before the OS starts.
I only read the post where you replied to me, but now I will read that post that you mentioned, thanks
about the barefeet article when they did the test with the NVME m.2
at that time the only driver available was the one from macvidcards
and it was a buggy driver then a better solution came out
finally apple implemented native support for 3rd party nvme drives in HS
I'm sure if they make a new test today they will have better results than in the past
when they used the first buggy driver to test the nvme m.2 drives

finally just read your post very interesting post , this part does it for me
Hackintosh x16 slot:
7. Full performance from four SSDs requires speed 3 (11141 MB/s).

that was exactly was I was looking for
thank you very much

just in case , what drives you have, I'm sure with the speed result they has to be samsung 960
most likely, I got 4 evos but my goal is at least 10,000 I'll be more than happy with that.
but I'm sure than I can reach 11,000 with 4 evos because I was hitting around 8400 or 8600 with just 3
once again thanks and sorry for missing your post before
[doublepost=1512136589][/doublepost]I ordered the card

DONE.png
 
Last edited:
I only read the post where you replied to me, but now I will read that post that you mentioned, thanks
about the barefeet article when they did the test with the NVME m.2
at that time the only driver available was the one from macvidcards
and it was a buggy driver then a better solution came out
finally apple implemented native support for 3rd party nvme drives in HS
I'm sure if they make a new test today they will have better results than in the past
when they used the first buggy driver to test the nvme m.2 drives
The first two Barefeats articles used 512 MB SM951 AHCI M.2 drives.

The first article dealing with the gen 2 card concluded that NVMe would not be faster than AHCI because the Gen 2 card is limited to PCIe 2.0 x4 1500 MB/s. It also concluded that you wouldn't be able to boot from NVMe. That might not be entirely correct: it may be possible to copy an NVMe EFI driver to make that possible (as some have done on Hackintosh).

The second article dealing with the gen 3 card concluded that you would get best performance in a classic Mac Pro using the gen 2 card. That is not entirely correct because you can get full/better performance in the classic Mac Pro from individual M.2 devices (no raid) using the gen 3 card. It also stated that the speed advantage of NVMe flash is nullified on the Mac Pro tower. That is referring to the gen 2 card though. The tests in the article show that even an AHCI drive can exceed 2GB/s in the gen 3 card. The article states "the PCIe 3.0 Carrier Board's controller gets 'lost in translation" which is a terribly uninformative description of the problem. There is a solution to that problem but it is not mentioned.

There is a third article http://barefeats.com/hard225.html using 512 MB 960 Pro NVMe drives in High Sierra. It seems to downplay the speed advantage of the gen 3 card for individual M.2 devices. The tests show the speed advantage but the conclusion is not very enthusiastic about it - for people that don't plan to raid.

finally just read your post very interesting post , this part does it for me
Hackintosh x16 slot:
7. Full performance from four SSDs requires speed 3 (11141 MB/s).

just in case , what drives you have, I'm sure with the speed result they has to be samsung 960
most likely, I got 4 evos but my goal is at least 10,000 I'll be more than happy with that.
but I'm sure than I can reach 11,000 with 4 evos because I was hitting around 8400 or 8600 with just 3
For my tests, I used four 1TB 960 Pro NVMe drivers as stated in my first tests in #212 using Apple Raid (Disk Utility.app). The updated tests in #214 used SoftRaid which gave much better results.


I ordered the card
I was quoted $465.00 USD + $59.66 USD taxes for the card SKU-086-34 (I'm in Canada) on Oct 2. The card is expensive because of the PEX 8732 switch chip (about $170) I'm not sure what the other $300 is for. Cards that require PCIe bifurcation don't use a switch chip (they use the motherboard's switching capabilities for the slot) and are therefore much less expensive ($200?).
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.