Has nothing to do with Windows but with the x86 architecture in general.
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.
MP All Models PCIe SSDs - NVMe & AHCI
- Thread starter MisterAndrew
- WikiPost WikiPost
- Start date
- Sort by reaction score
- Status
So, if it helps anybody - I know there are people that it did work for but the 980Pro most certainly did not work for me.
I ended up with slow boots, (up to 8 mins), beach balling, and corruption culminating in no boot at all. Good job I had a back up.
It has the latest firmware.
I was running Sonoma 14.2.1 in a 7,1
It's now a data only drive as I probs can't sell.
I ended up with slow boots, (up to 8 mins), beach balling, and corruption culminating in no boot at all. Good job I had a back up.
It has the latest firmware.
I was running Sonoma 14.2.1 in a 7,1
It's now a data only drive as I probs can't sell.
Yes, 980 PROs need a switched card to work properly in Macs. And even with that, the Trim-caused slow booting persists. But, once booted, the two i have work fine for me.
Ah, gotcha. I missed that part. My bad. Need to read more carefully next time.
Just as a matter of interest, what is the card you have?
Last edited:
Hi All. I'm looking at getting a Sonnet Silent m.2 4x4 card for my Mac pro 5.1. I want to run WD black 850 4TB SSD's. My concern is that I'm going to have a "Disk Not Ejected Properly" every time I wake the computer from sleep (I rarely turn it off).
Do any of you PCI SSD users have a) this issue and/or b) a solution?
I contacted Sonnet and their solution was to keep a text doc open from the attached drive which is supposed to keep the computer from sleeping the card/drives (even when the computer is asleep) because there is an "open" document. I tried this but, upon waking the computer, I still got the "Disk eject" warning.
Any advice before I buy the card would be appreciated. Thanks,
Do any of you PCI SSD users have a) this issue and/or b) a solution?
I contacted Sonnet and their solution was to keep a text doc open from the attached drive which is supposed to keep the computer from sleeping the card/drives (even when the computer is asleep) because there is an "open" document. I tried this but, upon waking the computer, I still got the "Disk eject" warning.
Any advice before I buy the card would be appreciated. Thanks,
It's some kind of HighPoint 7101a . At least it has identical PCB-layout and the same PLX8747 chipset.
They were on Aliexpress for 190 Euros tax included some two years ago. Actually there are similary parts around even cheaper and eqipped with the successor 8748.
This is something I can't get my head around. Accrodig to the Samsung website, (https://www.samsung.com/uk/memory-storage/nvme-ssd/980-pro-pcle-4-0-nvme-m-2-ssd-1tb-mz-v8p1t0bw/), it states;
Unleash the power of the Samsung PCIe 4.0 NVMe™ SSD 980 PRO for your next-level computing. Leveraging the PCIe 4.0 interface, the 980 PRO delivers double the data transfer rate of PCIe 3.0 while being backward compatible for PCIe 3.0 for added versatility.
The 2019 Mac Pro has a PCIe 3.0 interface, so I can't see how an additional card suddenly makes it suitable. I can confirm that in at least some instances operating without the card isn't good thing to do. PCIe 2.0 shouldn't even factor in with the 7,1
Samsung is surely having firmware troubles with 990 line, they released several firmware updates this year.
Some people lost data, some blades had endurance problems and needed RMA. I'd avoid buying any 990 blades.
Well that's annoying, the 980 / pro doesn't come in a 4TB flavour :/
Edit: *ouch* and the WD Red SSD is twice the price in 4TB.
Last edited:
970 Plus all the way for me. Never had a problem with any of them buying several at different times.
So does anybody have any clue, what kind of speeds a 4.0 blade like the 980 PRO on a given card (i. e. HighPoint 7101) delivers on the 7,1's PCIe 3.0?
But performance should be just the same. And with the newer "cheap" cards linked in some earlier post, which has a 8748 on it, even slightly better.
The 2019 Mac Pro has PCIe 3.0 slots. Without a PCIe switch, a PCIe 4.0 NVMe will be limited to PCIe 3.0 x4 speed which is less than 3.9 GB/s probably closer to 3500 MB/s.
To use a PCIe 4.0 NVMe at gen 4 speed, you need a PCIe card with a PCIe gen 4 PCIe switch with at least x8 lanes upstream for PCIe 3.0 slots and x16 lanes for PCIe 2.0 slots. I've tried the HighPoint 7505 and was able to get 5 GB/s from a Sabrent Rocket 4.0 2 TB in a PCIe 2.0 slot of a MacPro3,1.
https://forums.macrumors.com/threads/pcie-ssds-nvme-ahci.2146725/post-29483973
Hi Joevt Yes I see that but a lot of my confusion comes form statements like this;
- 4.0-capable motherboard and M.2 heatsink are recommended, but the drive is backward compatible with PCIe 3.0 and 2.0 M.2 sockets.
Thanks again for your very detailed and valuable answer!
In this case question targeted switched cards. I have this card from Ali, which was even more affordable some time ago. It's equipped with a PLX8747, just like the 7101.
Sadly the card i have only has 3.0 on it. So question is, if, with this limitation, will the 3.0 sockets of the 7,1 show any advantage vs. the 2.0 of the 5,1 with a single blade. My guess would be to see the saturation of 4 lanes PCIe 3.0, resulting in about 4 GB/s vs. the 2.5 read 2.7 write i have now. Not exactly what the SSD has to offer, but a nice little step in performance.
Update for you.
It would appear that the third party NVMe SSDs I'd chosen do in fact work without a switched card. It just depends on the flavour. The OWC Accelsior seems to be just the right recipe.
If you want more speed then yes the PLX8747 based card will work too, (which I will very likely still buy).
It would appear that the third party NVMe SSDs I'd chosen do in fact work without a switched card. It just depends on the flavour. The OWC Accelsior seems to be just the right recipe.
If you want more speed then yes the PLX8747 based card will work too, (which I will very likely still buy).
Looking at Highpoint options as examples, is this the correct understanding of the maths with conversions between PCI generations and lanes?:
If I install a PCIe 3 card (PEX8747 switch) in a PCIe 3 x8 slot, that card's switch could provide 2 PCIe 3 x4 SSDs with full saturation at once.
If I install a PCI 4 card (PEX88048 switch) in a PCI 3 x8 slot, that card's switch could provide 1 PCIe 4 x4 SSD with full saturation. Installed in a PCIe 3 x16 slot, it could saturate 2 PCIe 4 x4 SSDs at once?
7,1: CPU -> 96 lane switch -> x lane switch -> NVMe
5.1: CPU -> x lane switch -> NVMe
I don't know. the 7,1 has an extra switch to go through but has a faster CPU.
4 GB/s is impossible. 8GT/s/lane * 4 lanes * 128b/130T * 1byte/8b = 3.9 GB/s. You'll get between 3500 and 3800 MB/s. Need to find benchmarks to get some real numbers.
I have a 1 TB Samsung 960 Pro (gen 3) in my Skylake hackintosh and it only does ≈3200 MB/s.
The Sabrent (gen 4) can do 3300 MB/s in my MacPro3,1 in the x16 slot running at gen 1 speed instead of gen 2 speed.
Multiple disks (using ATTO Disk Benchmark, not RAID) can do 6.3 GB/s in the MacPro3,1 gen 2 x16 slot.
Multiple disks (using ATTO Disk Benchmark, not RAID) can do 3.4 GB/s in the MacPro3,1 x16 slot at gen 1 speed.
Samsung 960 Pro 1TB is 3160 MB/s in the MacPro3,1.
XPGSX8200 Pro 2TB is 3229 GB/s in the MacPro3,1.
MacPro3,1 tests were done using HighPoint SSD7505 in slot 2 running with gen 1 or gen 2 speed for the upstream and max speed for the downstream.
Benchmark is AmorphousDiskMark or ATTO Disk Benchmark.
Yes.
Yes! At least the latter is my understanding of „about 4 GB/s“. 😉
This is quite similar to what i get with a Raid0 of two 970 EVO plus on the 8747 card in the 5,1.
My 3800 MB/s estimate for gen 3 x4 might be high. I didn't achieve that with any of the NVMe's I have. For example, the Sabrent which is gen 4 and can do ≈5 GB/s only did 3293 MB/s with gen 1 x16. Even with multiple disks, gen 1 x16 wouldn't do more than 3449 MB/s.
gen 1 x16 is slightly faster then gen 3 x4.
2.5 GT/s/lane * 16 lanes * 8b/10T * 1 byte/8 bits = 4000 MB/s
8 GT/s/lane * 4 lanes * 128b/130T * 1 byte/8 bits = 3938 GB/s.
I think ATTO Disk Benchmark is able to do more bandwidth than Raid0 because Raid0 has to wait for all disks to finish each transaction which is why it's important for all disks to be the same for Raid0 - so they all finish at about the same time.
ATTO Disk Benchmark's multi-disk tests just send data as fast as possible to any of the selected disks that are ready to do more work. So it's a good test for getting max bandwidth for a slot or a group of slots. I used it to test the max bandwidth of all SSDs (up to 6) connected to MacPro3,1 built-in SATA ports: ≈894 MB/s which is probably the limit of PCIe gen 1 x4. Each SATA port can do ≈286 MB/s.
So in case of my question, 7,1 with PCIe 3.0, PLX8747 gen 3 card, single SAMSUNG 980 PRO (gen 4), the limiting factor is the 4 lane gen 3 socket on the card, while in case of the HighPoint 7505 (gen 4 ), gen4 SSD, machine with gen 1 board, it will be the board's gen 1 x16 slot?
I didn't even know ATTO Disk Benchmark offering a multi-disk test. Just checked my version, which is 1.01.0 (2), and still can not find such an option. Is this a feature of a later version? Would really like to test this and compare to the Raid-results.
Edit: Just found out! "show results for all disks" has to be selected in the Graphs-window. Right?
Last edited: