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 1,1-5,1 NVMe working in 3,1
- Thread starter christiann
- Start date
- Sort by reaction score
x4 slot is PCIe gen 1 (≈750 MB/s).
x16 slot is PCIe gen 2 (≈1500 MB/s). But the NVMe may boot as PCIe gen 1. I made an EFI driver to fix that.
https://forums.macrumors.com/threads/mac-pro-3-1-nvme-support-upgrade-guide-questions.2194878/
If you want PCIe gen 3 speed (≈3400 MB/s), then you need a PCIe card with a PCIe gen 3 switch chip.
https://forums.macrumors.com/threads/pcie-ssds-nvme-ahci.2146725/
x16 slot is PCIe gen 2 (≈1500 MB/s). But the NVMe may boot as PCIe gen 1. I made an EFI driver to fix that.
https://forums.macrumors.com/threads/mac-pro-3-1-nvme-support-upgrade-guide-questions.2194878/
If you want PCIe gen 3 speed (≈3400 MB/s), then you need a PCIe card with a PCIe gen 3 switch chip.
https://forums.macrumors.com/threads/pcie-ssds-nvme-ahci.2146725/
@joevt favours the Driver#### approach for most things, but I don't think it is the best way for most of us mere mortals that do not have the same high level of needs he has. I am not sure it survives an NVRAM reset without needing to be redone for instance.
Separately, while you indeed do not have to use OpenCore, you will ultimately be better off doing so.
If you use MyBootMgr linked in my signature to set RefindPlus|OpenCore up, RefindPlus will:
Separately, while you indeed do not have to use OpenCore, you will ultimately be better off doing so.
If you use MyBootMgr linked in my signature to set RefindPlus|OpenCore up, RefindPlus will:
- Supply APFS support on MP31
- Supply NVME support on MP31
- Supply and load Joe's PCIE driver
It doesn't. Which is a good thing in case it causes a problem.
If you need to NVRAM reset your computer more than once a year then you got problems.
Executing a reset is one way, if not having a reconstructed BootROM for instance, to try to free up space in the NVRAM when updating currrent MacOS to reduce the chances of problems.
That's likely to generate more than one reset a year but yeah, you do have problems, brick avoidance, to manage when running/updating current MacOS on cMP indeed!
See:
Mac Pro 3,1 NVMe support + Upgrade Guide + Questions
Adding NVMe and apfs support to MacPro3,1 without flashing ROM or using a EFI boot loader or Shell: You can place the EFI drivers on a EFI or FAT or HFS+ partition, then set Driver#### and DriverOrder NVRAM variables using the commands discussed here. In my example below, I chose a folder...
forums.macrumors.com
Hmm, the MP31 does not have a second VSS to overrun/corrupt in the same way as those examples indeed. I doubt that makes it immune to bad stuff happening if a large lump of data hits the chip at a critical point in the NVRAM space cycle though.
we've run across an anomaly (I think) with NVMe..
So, this is for a server in a business application, managing large backup files (the smallest file is around 450GB, though some are near 1TB).
We're finding something for which we don't have any explanation.
on certain NVMe cards (such as say,Crucial, WD and Kingston, as well as some other lesser known drives), data throughput drops thru the floor after a certain data load is on the drive..
so, we can copy a ~500GB file onto a blank/empty drive, and it completes the entire duplication without dropping below ~1GB/sec data speed.
if we then copy a 2nd file to the same drive, it performs comparably, until 200-300GB into the file.. then data speed drops thru the floor... down to 30-40MB/second..
Has anyone else seen this sort of activity?
it's one thing to copy a large file to a naked/empty drive..
it's operating entirely differently when the drive is already 1/2 full..
ideas??
So, this is for a server in a business application, managing large backup files (the smallest file is around 450GB, though some are near 1TB).
We're finding something for which we don't have any explanation.
on certain NVMe cards (such as say,Crucial, WD and Kingston, as well as some other lesser known drives), data throughput drops thru the floor after a certain data load is on the drive..
so, we can copy a ~500GB file onto a blank/empty drive, and it completes the entire duplication without dropping below ~1GB/sec data speed.
if we then copy a 2nd file to the same drive, it performs comparably, until 200-300GB into the file.. then data speed drops thru the floor... down to 30-40MB/second..
Has anyone else seen this sort of activity?
it's one thing to copy a large file to a naked/empty drive..
it's operating entirely differently when the drive is already 1/2 full..
ideas??
Probably your write throughput slowdown is related to the SLC cache is full while TRIM is being done at the same time.
could be..
we just got in some Samsung 970s, and tried a pair of them.
ZERO ISSUES AT ALL.
in fact, we duplicated 3x 450gb files simultaneously to the 970's..
NOT A SINGLE HICCUP.
Apparently it really is about flash chip/drive quality.
The crucials hiccuped at around 300gb
The Kingstons at around 250gb
The WD at around 350gb...
TeamGroup at around 300gb
Samsung 970 EVO Plus - clear winner.. no issues..
this test set was done on a MacPro 2010 5,1, using 5 different NVME adaptors:
2 by ASUS (both were crap)
1 - AbleConn 8x16x (dual card) - https://www.amazon.com/gp/product/B07PRN2QCV
1 - RIITOP 16x - dual card - amazon - https://www.amazon.com/gp/product/B08P57G1JW
1 - 'no brand' 8x single card from amazon - https://www.amazon.com/gp/product/B098JJTYBR
The 16x was minimally faster in the Mac's single available 16x slot.
The other cards were all equally consistent in the 4x and 16x slots, with no significant performance difference..
understand: these are serious sequential writes - we're doing 4x/day backups of 450-550GB per backup, via SQL dumps which are fed from cache... so it's dumping just as fast as the drive can write.
The odd thing to me is that none of the problem drives (WD/Crucial/Kingston/TeamGroup) ever 'overheated'... sure, they all were warm.. but none of them really got hot..
In fact, we added some heat sinks to them just as an attempt to help minimize thermal issues, but there was no real impact.
I've played with hundreds or even thousands of SSDs and have never once seen this level of impact on large file handling. this was quite a surprise..
It's not a over temperature issue, it's a controller/cache one. You need a high-end blade for this workload, something like 970EVOPlus/970PRO/980PRO, anything less and you will have issues after some cycles of erase. When the SLC cache is full and waiting for the sectors to be TRIMmed, you will have serious throughput slowdowns, since the NAND cell erase operation is the slowest one.
Writing consistency is what informed my search and ultimate choice of the Western Digital SN700 drives. The complete review is on the STH website [https://www.servethehome.com/wd-red-sn700-1tb-nvme-ssd-review-vastly-improved/3/]
Attachments
Yep, it's consistently bad when installed in a MacPro5,1
Jokes apart, most people only get write throughput at ~1100MB/s when installing directly to the PCIe v2.0 slots.Btw, all drives shine at benchmarks when brand new, what really matters is after some TB written/erased.
More data related to the 3 links you've provided:
The top link for the AbleConn product uses an ASM2824 chip, so it can drive a x8 connection to the motherboard. I had one of these in March of this year, and returned it because it seemed to mis-behave in my 2010 Mac. Since then I've learned that perhaps I was too harsh in my assessment.
The middle one, the RIITOP 16x card apparently uses an ASM2812 chip, which only has enough PCIe channels to provide a x4 connection to the motherboard. Therefore, no matter what slot it's in, it'll give 4-lane performance.
The bottom one, is a 4-lane passive card, so again, it'll only offer 4-lane performance no matter what slot.