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.
- Status
When the Samsung 960 Pro is in the System the MacPro don´t Start. There is only a grey Screen. Sometime it works. 1 of 10 trys. With the SM951 NVMe it works every time.
Well, this NVMe-stuff in cMP are to bee "uncharted territory". Experimentally, stuff. Unsupported from Apple, so it is trial and error, for everybody. So why Your "blade" does not work, well....!!
/Per
/Per
=> "if everyone can get off the mining craze so prices go back to normal! I just can't justify spending more on a card today that what it retailed for 2 years ago.. "
we need an AMEN here.
Hi Zozomester,
I wonder if you can help me achieve what you just did, in other words :
- Booting natively high sierra from 4 NVME drives in RAID0 (in my case 3 drives)
- Having this RAID array perform better than a single NVME drive.
What I did so far is add the NVME driver to the Macpro 5,1 rom and reflash it.
The result was that I was able to boot natively from a single NVME drive (Samsung EVO 970)
Happy with this success, I bought two more 970 EVO's but unfortunately did not get the results I expected :
- The macOS high sierra installer cannot load preboot fies on the RAID array (no matter if formated in APFS or HFS) and won't let me get past that so no way to boot from this raid so far.
- I booted from a hard disk to test the performance of The 970 EVO x3 RAID array and the result is only a 10% increase from a single drive. Far from the score you got.
- I tried to cheat the "preboot files" bug by cloning a working macOS system to the RAID array but the computer reboots during startup.
Any help would be appreciated, nice work by the way !
Fab
I use cheap, basic pcie-nvme adapter cards
Hi,
I use this basic, cheap pcie adapter card https://www.aliexpress.com/item/PCI...900.html?spm=a2g0s.9042311.0.0.51594c4dQzFnCY
I have 3 of them on slot 2-3-4 ,.
I'm aware that performance on slot 2 is cut in half but even a raid0 with 2 drives on slot 3 and 4 won't give me a performance increase or a bootable system.
Yes, each one of the SSD is detected. The one on slot is only working at 2.5GT/s though.
Nvram reset is one of the things I tried yes.
Only the one in slot 2 but the problem has more to do with the RAID0 not performing better than a single drive (true even if I use only 2 drives on slot 3 and 4) and not being bootable.
I have. I just got the Samsung 970 Pro in the mail today, installed it using a lycom dt-120 I had laying around, downloaded the public beta and ran it from High Sierra installed on a different SSD. It worked fine.
Now I just have to figure out how to get my microcode back since I went to 87 previously by accident. =)
Congrats. I had a successful install with the HighPoint 7101a AND THE 970 pro of the public beta as well. FWIW.. The developer beta refuses to install.
Not true. If it where incompatible, it wouldn't show up at all.
The 2.5GT/s link speed is a MacPro issue for slots 1 & 2 explained here: 5GT/s for AMD card without resistor mod
The 2.5GT/s is perfectly normal.
http://amfeltec.com/products/pci-express-carrier-board-for-m-2-ssd-modules/
[doublepost=1530092450][/doublepost]
No, normal is 5Gt/s the all slot....
That IS half speed per slot. The AVERAGE performance for each slot will be half--on average.
If you want to use slots 3 and 4 for high bandwidth cards, you probably should think about the fact they SHARE bandwidth. With two PCIe SSD drives, you would expect 3000 MB/s combined. You will only get 1500 MB/sec. The math there suggests HALF the expected data rate.
[doublepost=1530285014][/doublepost]
You can get 5GT/s instead of 2.5GT/s.
I've attached an imperfect, but close, document about using pciutils from the command line. You can change the PCIe rate on the fly. With this, you can put PCIe v2 cards in any slot in a cMP and change the speed from v1 to v2. It is useful on an Xserve, too.
I disagree. I already know these are shared slots with a maximum 1500 MB/s, whether using 1 card or 2. It may be 1/2 the potential speed That each card could bring to the table on independent Slots, but the slots didn’t go any slower, and thes slots are not independent. Potential speed is not speed, the combination of 2 slots getting 1500 MB/s is not the same as slowing down. They are on the same channel and max speed has been & always will be 1500 MB/s combined or not. The bottleneck didn't change sizes or get any worse.
Total bandwidth is the same, not reduced, they spread out the transfer load to the maximum of 1500 MB/s across 2 Slots as designed.
It’s not half the expected data rate at all. All that should ever be expected is 1500 MB/s unless you are an uninformed upgrader.
It’s a matter of perspective. With a hard speed limit of 1500 MB/s, why would anyone expect it to go faster just because you stuck another one in an adjoining slot? You wouldn’t...
It’s not that I don’t understand YOUR point of view, I just don’t share it.
This is my perspective and we don't have to see it the same way. It's really semantics and changes nothing.
Last edited:
I agree with your point that there is a difference between shared and "halved", also. It is only halved if both slots are trying to use the full bandwidth. I agree also that it is partly semantic. Worst case performance, though, is the full rate of slot 3 plus slot 4 divided by 2.
A naive user who configures a RAID0 across those two slots and will see half the maximum expected throughput. What it changes is that forum readers, including new folks to cMP modding, will be aware of the unexpected characteristics of slots 3 and 4. They won't have to burn hours trying to figure out why they aren't seeing expected behavior.
- Status