Has anyone ever tried this with an iMac (Late 2013 and later)? Thanks, MagnusUpdate: Beginning of thread is mostly NVME related. Work for replacing internal GPU with a 3rd-party GPU starts later in the thread (in progress). The thread is still active. Questions and comments for both NVME and GPU are welcome.
Update: Unedited feedback from Kris Kelvin
- Not all NVMe drives are compatible. Unfortunately, there's no definitive list. Samsung's 960 series seems to work for everyone. Here is a report of someone using a Toshiba XG3.
- The speed will be limited to how many PCIe lanes your Mac supports. Here's an overview over different Models.
- High Sierra must previously have been installed on this Mac (on an AHCI drive). Installing High Sierra upgrades the Mac's EFI Firmware / Boot ROM, allowing for NVMe drives to work. The latest Boot ROM for MacPro6,1 is MP61.0120.B00.
- Your adapter looks to be the Chenyang/CableCC one. Several people reported problems with it. The ones from Sintech and PC PARTS 239 seem to be working properly (including sleep, reboot etc.).
- Make sure the adapter is properly seated and aligned. If it isn't, it might work with fever PCIe lanes (= slower) or not at all.
- Taping the adapter with EM shielding (e.g. Kapton tape) is recommended to prevent EM leaks / shorts.
- High Sierra (or later) is required when using a NVMe SSD. If you're using Internet Recovery, make sure you boot with Option-Command-R (and not just Command-R), so the latest compatible version of macOS is loaded. Bootable installers on USB also work.
- A heat sink for the SSD is not required, but might help sustain performance when stressing it over a long time. This one fits: EK-M.2
- High Sierra's Disk Utility may not show uninitialized devices. If your SSD isn't shown, try this:
- Open Disk Utility
- Click the top left button in the toolbar ("View") and select "Show All Devices"
- Relaunch Disk Utility
Update: Unedited uncut video proof
https://www.dropbox.com/s/emb558u38whfdmm/20171202_000342.mp4?dl=0
I have two questions. I see no thread about NVMe on the trash can Mac Pro, so I made a new thread.
1 - Does Mac Pro 6, 1 (2013) support NVMe drive type and if not is there a firmware hack? I know iMac Pro coming in December is supporting NVMe, but that support may not trickle down to the trash can Mac Pro. Update; it does work. Internal drive in the Mac Pro 2013. Booting even!
2 - Apple's proprietary connector for M.2 is double height with an extra row that is two solid pads on one side. Does anyone know where to get an M.2 adapter dongle that supports this double-height connector format? Update; the two solid pads are connected to ground.
I have Mac Pro 2013 and want to upgrade my M.2 from 1 TB to 2 TB. The only 2 TB M.2 stick on the market besides the OWC custom solution is Samsung MZ-V6P2T0BW. So I got this stick and a dongle to convert from standard M.2 NVMe connector type to Apple's proprietary M.2 connector type. But the dongle connector does not have the second row of pins like my Mac Pro's built-in SSD connector. I've tested it, and it currently does not work in the slot, I'm not sure if the problem is a missing driver or because of the connector difference. Update; it does work. But don't forget to sand down jagged edges using a metal file or sandpaper. And don't forget the tape!
SSD:
https://www.newegg.com/Product/Product.aspx?Item=N82E16820147598
http://www.samsung.com/semiconductor/minisite/ssd/product/consumer/960pro.html
Adapter (USA-only):
Update - PC Parts 239 is discontinued
https://www.amazon.com/gp/product/B01LXUH921
http://www.pcparts239.com
Adapter (inside or outside the USA):
ST-NGFF2013-C (after extensive corroboration it must be the 'C' version; do not get the non-C version)
http://eshop.sintech.cn/ngff-m2-pcie-ssd-card-as-2013-2014-2015-macbook-ssd-p-1229.html
Heat Sink (optional):
https://www.amazon.com/gp/product/B073RHHYCM
https://www.ekwb.com/shop/ek-m-2-nvme-heatsink-black
http://www.samsung.com/semiconducto...document/SAMSUNG_Memory_NVMe_Brochure_web.pdf
Original 1 TB Drive:
View attachment 739747
View attachment 739748
New Drive:
View attachment 739749
View attachment 739751
View attachment 739752
Adapter:
View attachment 739745
Adapter with Tape:
View attachment 739746
Adapter Sanded Down:
View attachment 739755
View attachment 739756
View attachment 739757
(previously titled "NVMe is working; 3rd-Party Internal GPU; Mac Pro 2013 (6,1)", "NVMe is working; Mac Pro 2013 (6,1)", "2 TB NVMe M.2 SSD in Trash Can Mac Pro")
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.
NVMe with ST-NGFF2013-C; Vega Internal GPU; Mac Pro 2013 (6,1)
- Thread starter CodeJingle
- Start date
- Sort by reaction score
ok setting up external drive for apple stock ssd (mac pro (late 2013)). still have the nvme as internal drive because i don't yet have an external enclosure that supports nvme. the stock ssd (oem equivalent) is by sandisk and most of the actual chips on it are samsung. so hopefully it works with this enclosure. here are some pics of the enclosure.
[doublepost=1531896007][/doublepost]the external enclosure is working with this specific oem equivalent of stock sdd. i haven't tested this before but tested it now, i can boot from external drive with no internal drive in the gpu! a win in my book with all the limitations of the internal ssd slot going through the ancient chipset.
the funny thing, even when booting from stock ssd which is non-nvme, i am still constantly getting ionvme crashes, but only when the nvme drive is installed. the crashes go away when i remove the nvme drive. again regardless of which drive i am booting from. so this is more evidence to suggest macos or at least the current mojave beta is unable to cope with a severely corrupt nvme drive installed (or severely corrupt apfs partition) even if not the boot drive.
or could also be that there is actually an issue with using an nvme in the internal ssd slot because it goes through the ancient chipset which is prejudice against protocols newer than ahci.
[doublepost=1531896007][/doublepost]the external enclosure is working with this specific oem equivalent of stock sdd. i haven't tested this before but tested it now, i can boot from external drive with no internal drive in the gpu! a win in my book with all the limitations of the internal ssd slot going through the ancient chipset.
the funny thing, even when booting from stock ssd which is non-nvme, i am still constantly getting ionvme crashes, but only when the nvme drive is installed. the crashes go away when i remove the nvme drive. again regardless of which drive i am booting from. so this is more evidence to suggest macos or at least the current mojave beta is unable to cope with a severely corrupt nvme drive installed (or severely corrupt apfs partition) even if not the boot drive.
or could also be that there is actually an issue with using an nvme in the internal ssd slot because it goes through the ancient chipset which is prejudice against protocols newer than ahci.
Last edited:
chipset pinout c602j; this one was in the spec i didn't make it. intel's palette and patterns for this color-coded pinout was designed in partnership with imaginary cartoon character strawberry shortcake.
Last edited:
Just to get a proper update: the new Sintech Adapter is trouble free and provides the Mac Pro 6,1 with a fully functional, bootable NVME SSD (e.g. Samsung 960, 970) which behaves like an Apple SSD (except perhaps for a longer grace period at (cold) start)?
It's what I use, although it still doesn't physically fit as a perfect replacement for the Mac Pro's SSD. With the full heatsink and two layers of thermal pads applied, I needed at least one piece of kapton tape to hold the mess together.
You might be able to avoid that by only applying one of the thermal pads that comes with this kit. Specifically, don't apply the thermal pad that would be placed behind the SSD, between the SSD and the adapter. Just the one that goes between the SSD and the heatsink itself.
In my experience you'll want to have some manner of heatsink on it; not only does the original SSD have a heatsink that the MBP/MBA SSDs didn't, the Samsung EVOs run hotter than the PROs.
EDIT: Just filling out more deets as I go...
Wait - yes, but there appear to be indications that the release of Mojave may cause corruption/kernel panics and you will not be able to update the firmware of your machine using anything other than an AHCI m.2 drive. At lease thats what I gather from reading the last couple of weeks of this thread.
it works for some with zero issue, for others it works mostly except kernel panics when the computer is entering or exiting sleep mode. only rarely does it not work at all. you have to temporarily boot from a stock apple drive or oem equivalent in order to update the firmware if you switch to nvme and afterwards apple updates the firmware (you can allegedly boot stock from usb or thunderbolt to update firmware not just internal). if you are running mojave beta don’t run boot camp assistant on the nvme drive, unless you are willing to sacrifice for the greater good.
[doublepost=1532308680][/doublepost]
booting from an nvme drive, try causing a kernel panic in the middle of your nvme partition/volume being resized by boot camp assistant in mojave. then tell me how stable it is after to boot from the same nvme.
[doublepost=1532309642][/doublepost]
it should fit perfectly in the slot. it doesn’t line up with the screw hole?! can you send a pic? here is what mine looks like, it fits just fine both to the screw hole and the nvme heat sink is firmly secure to the m.2 board with the supplied clips.
Last edited:
i'm still doing my quasi phd-level deep dive. the ancient chipset might be limiting the hard drive functionality but i'm not completely sure.
and when i call it ancient i don't just mean c602j was launched q1 2012. i mean the documentation is putting down other storage protocols as if nothing will ever succeed sata or ahci. it is an ancient mindset at intel that engineered the chipset. like an american university in the 1970s still struggling with desegregation.
since the mac pro (late 2013) will boot without an internal drive connected i would suggest nvme is external only, and perhaps keeping the internal drive slot stock ssd only or even unpopulated. you can gaffer-tape the external drive to the outside of your mac pro cylinder lol.
internal nvme on mac pro (late 2013) is working no problem for some but i still cannot recommend it 100% not at this moment anyway. hopefully i'll figure out more over time or apple can officially say whether nvme is fully supported as internal drive for mac pro (late 2013).
and when i call it ancient i don't just mean c602j was launched q1 2012. i mean the documentation is putting down other storage protocols as if nothing will ever succeed sata or ahci. it is an ancient mindset at intel that engineered the chipset. like an american university in the 1970s still struggling with desegregation.
since the mac pro (late 2013) will boot without an internal drive connected i would suggest nvme is external only, and perhaps keeping the internal drive slot stock ssd only or even unpopulated. you can gaffer-tape the external drive to the outside of your mac pro cylinder lol.
internal nvme on mac pro (late 2013) is working no problem for some but i still cannot recommend it 100% not at this moment anyway. hopefully i'll figure out more over time or apple can officially say whether nvme is fully supported as internal drive for mac pro (late 2013).
Last edited:
my understanding of pcie is that there can only be one device connected per root. coming straight from the cpu there are 40 pcie 3.0 lanes, split between three roots - 16, 16, 8. i’m not partial to switches so other people can worry about that. i already know the computer boots fine without the i/o wall and without the internal ssd. i bet if i don’t need thunderbolt or usb and can directly bypass the safety interlock that i can also boot the computer without the i/o board (or instead of bypassing the interlock, boot with stock pc power supply and fudge the 11 volt rails).
initially, disconnect gpu a (gpu a is the one without ssd), replace it with a 3rd party gpu. if that works, disconnect gpu b, replace with an nvme drive (wasting 12 lanes i know).
later on, after figuring out how to boot without i/o board, replace gpu b with a 2nd 3rd party gpu. if that works, connect the nvme drive to the 8 lanes that were previously going to the i/o board (now wasting only 4 lanes).
later on split the 8 lanes to 2x4 with a switch and have two nvme, manually set them to raid 0 to simulate the type of setup shipping in the imac pro and 2018 macbook pro (if apfs directly supports raid 0 with two ssd now, or *when* it does, if that support comes later).
the 4 pcie 2.0 lanes can continue going to the c602j, which splits the one root into five. mouse and keyboard can still connect to bluetooth. internet can still connect to wifi or ethernet.
it looks like everything is here. 44 pcie lanes are clearly routed from the cpu. the path they all take to the two gpus, i/o board, and chipset are all there in front of me. all the lanes coming out of the chipset are also all clearly there and easy to follow. a few more weeks of labeling and color coding and i’ll be ready to test a single third party gpu with full-throughput connections straight to the cpu (then nvme).
i would like to be sure of all the pins because i don’t want to fry more boards. i want to get it right and the money i save not replacing more fried boards can go towards buying up a few random mac pro (late 2013) machines that no one wants. at least if it works with vega and samsung nvme. maybe work on a solution that doesn’t require turning the mac pro (late 2013) inside out. if both internal gpus are driven outside of the enclosure that should considerably reduce the thermal load inside the trash can.
initially, disconnect gpu a (gpu a is the one without ssd), replace it with a 3rd party gpu. if that works, disconnect gpu b, replace with an nvme drive (wasting 12 lanes i know).
later on, after figuring out how to boot without i/o board, replace gpu b with a 2nd 3rd party gpu. if that works, connect the nvme drive to the 8 lanes that were previously going to the i/o board (now wasting only 4 lanes).
later on split the 8 lanes to 2x4 with a switch and have two nvme, manually set them to raid 0 to simulate the type of setup shipping in the imac pro and 2018 macbook pro (if apfs directly supports raid 0 with two ssd now, or *when* it does, if that support comes later).
the 4 pcie 2.0 lanes can continue going to the c602j, which splits the one root into five. mouse and keyboard can still connect to bluetooth. internet can still connect to wifi or ethernet.
it looks like everything is here. 44 pcie lanes are clearly routed from the cpu. the path they all take to the two gpus, i/o board, and chipset are all there in front of me. all the lanes coming out of the chipset are also all clearly there and easy to follow. a few more weeks of labeling and color coding and i’ll be ready to test a single third party gpu with full-throughput connections straight to the cpu (then nvme).
i would like to be sure of all the pins because i don’t want to fry more boards. i want to get it right and the money i save not replacing more fried boards can go towards buying up a few random mac pro (late 2013) machines that no one wants. at least if it works with vega and samsung nvme. maybe work on a solution that doesn’t require turning the mac pro (late 2013) inside out. if both internal gpus are driven outside of the enclosure that should considerably reduce the thermal load inside the trash can.
Last edited:
apple did reply to the bug i filed (mojave crashing when i have my nvme with corrupt apfs volume/partition connected as internal ssd). the reply was under nda so can't really talk about it. i'm not exactly satisfied with the content of the reply but i am still happy they replied at all, especially considering beta bugs are supposed to go through the beta feedback tool not through the developer bug report tool. i will still update the bug if i make progress on the issue. though i don't think i can make progress until i get an external enclosure for nvme.
a coworker seemed unhappy i was working on this onsite even though it was after hours. though i think it was more because they needed a nap and have no free time because they have two young children at home (my kids are 8 and 11 so i've passed that milestone a while back). like me they sometimes work till late in the night. i put in my 60 to 80 hour weeks for a long time at the company, i need to take a break from that. i helped the boss on july 4th. when they are in china i'm helping them in the middle of the night. so i've earned the right to work on my pet project after hours. it is also how i keep my sanity. i'm an aspiring hardware engineer in addition to established software engineer, the work i am doing on this project will help me in all my future work on and off the job. it has been hot at home and at least part of onsite has air conditioning. maybe i should invest in an air conditioner at home.
a coworker seemed unhappy i was working on this onsite even though it was after hours. though i think it was more because they needed a nap and have no free time because they have two young children at home (my kids are 8 and 11 so i've passed that milestone a while back). like me they sometimes work till late in the night. i put in my 60 to 80 hour weeks for a long time at the company, i need to take a break from that. i helped the boss on july 4th. when they are in china i'm helping them in the middle of the night. so i've earned the right to work on my pet project after hours. it is also how i keep my sanity. i'm an aspiring hardware engineer in addition to established software engineer, the work i am doing on this project will help me in all my future work on and off the job. it has been hot at home and at least part of onsite has air conditioning. maybe i should invest in an air conditioner at home.
Last edited:
i've been spending some more time on the topic of pcie debuggers. there is clearly a built-in framework at the hardware-level for pass-thru debugging of pcie, at least on the xeon e5 v2 line of processors. so why is a hardware debugger, even a simple hardware pass-thru capture tool, so expensive?
at a minimum, even for an unregulated and unlicensed device, one which the person who made the device did not buy into the pci-sig, and whom did not pay for pcie certification, and did not even pay for fcc licensing.
i'm thinking most operating systems don't let you change hardware register values. no way to do that. i would have to modify the kernel. which means i'd have to fork my own operating system. that is alot of work. macos isn't open source. so i can't ship pcie capture hardware running macos even if the guts of the hardware is a mac. maybe running the open source darwin kernel i can modify it to enable pcie pass-thru but that still isn't going to be the same as running macos mojave. i think that alone explains most of the cost for a hardware pcie capture tool. it has to be server grade hardware with a custom built operation system and a minimum of 32 pcie 3.0 lanes that are unused except for as pass-thru lanes.
at a minimum, even for an unregulated and unlicensed device, one which the person who made the device did not buy into the pci-sig, and whom did not pay for pcie certification, and did not even pay for fcc licensing.
i'm thinking most operating systems don't let you change hardware register values. no way to do that. i would have to modify the kernel. which means i'd have to fork my own operating system. that is alot of work. macos isn't open source. so i can't ship pcie capture hardware running macos even if the guts of the hardware is a mac. maybe running the open source darwin kernel i can modify it to enable pcie pass-thru but that still isn't going to be the same as running macos mojave. i think that alone explains most of the cost for a hardware pcie capture tool. it has to be server grade hardware with a custom built operation system and a minimum of 32 pcie 3.0 lanes that are unused except for as pass-thru lanes.
is there any way to modify the registers, like with a custom kext?
i haven't done much mac hacking, at least not in 10+ years.
i haven't done much mac hacking, at least not in 10+ years.
Last edited:
this is the cheapest nvme enclosure shipping out of usa. there is oem equivalent for half the price but i do not want to wait for shipping from china. it is usb which people hate but this is only a temporary solution i don't need to spend $300+ usd on an m2 thunderbolt solution if my eventual goal for the mac pro is to disconnect the i/o board in which case there will be no thunderbolt support.
this drive might not work at all so if you are thinking of getting one wait till i test it.
shipping out of usa
https://www.newegg.com/Product/Product.aspx?Item=9SIA6V87KE3807
shipping straight from china, allegedly oem equivalent
https://www.newegg.com/Product/Product.aspx?Item=9SIA2RP7K19337
when i get it i'll test with a usb-c to usb-a cable (the cable is specifically usb 3.0 compatible)
https://www.newegg.com/Product/Product.aspx?Item=N82E16812119884
this drive might not work at all so if you are thinking of getting one wait till i test it.
shipping out of usa
https://www.newegg.com/Product/Product.aspx?Item=9SIA6V87KE3807
shipping straight from china, allegedly oem equivalent
https://www.newegg.com/Product/Product.aspx?Item=9SIA2RP7K19337
when i get it i'll test with a usb-c to usb-a cable (the cable is specifically usb 3.0 compatible)
https://www.newegg.com/Product/Product.aspx?Item=N82E16812119884
Last edited:
the nvme enclosure i just linked to is usb c at 10 gigabit. 100 usd. usb 3.1 10 gigabit is not natively supported by mac pro (late 2013) so the connection will downgrade to usb 3.0 5 gigabit (if it works at all). here is the link again.
https://www.newegg.com/Product/Product.aspx?Item=9SIA6V87KE3807
this one is thunderbolt 3 enclosure for nvme, you can get a thunderbolt 3 to thunderbolt 2 adapter/cable to use it in mac pro (late 2013). 350 usd.
https://eshop.macsales.com/item/OWC/TB3EX4M2SL/
bi-directional adapter, thunderbolt 3 <==> thunderbolt 2
https://www.apple.com/shop/product/MMEL2AM/A/thunderbolt-3-usb-c-to-thunderbolt-2-adapter?fnode=8b
Last edited:
Thank you, I realised it when I was looking for a fast external drive that works with my cMP with USB 3.1 and my nMP with thunderbolt and is kind of mobile.
But its either TB like the jet drive or USB c enclosures.
Well thanks anyways for the OWC link, might be a solution.
And now I don't wanna disturb your thread any longer have been reading and enjoying it since day one! Thank you!
i bought istat menus 6 full version by bjango for monitoring thermals to do some r&d on whatever is causing the machine to do an emergency shutdown during the asd os thermal interface test. it is my way of obtaining a second opinion on whether the shutdown is actually due to thermals or rather because of a limitation of sustained power delivery over time.
https://bjango.com/mac/istatmenus/
i don’t think i can run the thermal interface test during a normal macos boot. hopefully i can come up with a mojave repro for the emergency shutdown.
it is my hypothesis that the reason the mac pro (late 2013) only supports processor tdp up to ~130 is not only related to heat but also related to the cpu riser card's ability to properly deliver and dissipate sustained sufficient power for a processor with a higher tdp (like ~150) not under typical load but under max load in turbo boost mode.
it should be fairly straightforward to permamently disable turbo boost by swapping the high/low pin state for a certain pin on the processor but that is a last resort. if i determine the issue is not thermals i would much more likely start cutting traces from the cpu to the chipset to disable the mac pro's internal temperature/power management circuitry before i would permanently disable turbo boost. if i have to feign a certain level going to those pins after cutting the traces i am prepared to do that as well. it will either work fine or i'll burn out the cpu. i have a kill switch and a fire extinguisher handy.
https://bjango.com/mac/istatmenus/
i don’t think i can run the thermal interface test during a normal macos boot. hopefully i can come up with a mojave repro for the emergency shutdown.
it is my hypothesis that the reason the mac pro (late 2013) only supports processor tdp up to ~130 is not only related to heat but also related to the cpu riser card's ability to properly deliver and dissipate sustained sufficient power for a processor with a higher tdp (like ~150) not under typical load but under max load in turbo boost mode.
it should be fairly straightforward to permamently disable turbo boost by swapping the high/low pin state for a certain pin on the processor but that is a last resort. if i determine the issue is not thermals i would much more likely start cutting traces from the cpu to the chipset to disable the mac pro's internal temperature/power management circuitry before i would permanently disable turbo boost. if i have to feign a certain level going to those pins after cutting the traces i am prepared to do that as well. it will either work fine or i'll burn out the cpu. i have a kill switch and a fire extinguisher handy.
Last edited:
@CodeJingle Is your current CPU the E5-2667 v2? If so, an outside option would be to find the E5-2673 v2 CPU which has the same specs but the TDP is 110 W instead of 130 W. I think it's an OEM CPU so it's probably rare but maybe an eBay saved search can crawl for it.
i think you are out of the loop. did you see my inside-out mac pro?! i am not intetested in avoiding the sustained limit of 130 tdp for the mac pro. i'm using a 150 tdp processor right now (Xeon E5-2687W V2 SR19V) and it fails the asd os thermal interface test (i'm passing all other asd efi and asd os tests). i am not swapping as my processor is epoxied to the cpu riser card [using thermally conductive epoxy]! are you gonna help me mod my mac pro to bypass the thermal and power management systems to sustain turbo boost on a 150 tdp processor without triggering a hardware-level emergency shutdown? at least from a purely thermal capacity (not electrically) the tdp of my modified mac pro should already be 600+.
Last edited:
noting the visually-synonymous nature of thunderbolt 3 and usb c. even though the external enclosure for nvme is internally usb 3.1 10 gigabit, externally it is a usb c connection, which some might mistake for a native thunderbolt 3 device. though on some level it might be legit equivalent. i wonder if instead of a 'usb c' to 'usb a' cable/adapter it would work better to use the apple 'thunderbolt 3 (usb-c) to thunderbolt 2' adapter.
Last edited:
Apparently the apple adapter only works with thunderbolt 3, not with just usb c.
I'm looking for a solution that would allow a fast connection to a cMP and the nMp now for a while.
The caldigit TS3 plus offers a native usb c gen. 2 port and works with the apple adapter quite good.
another solution would be a a pice box like the sonnet echo express sel + the sonnet allegro usb c card..
I'm looking for a solution that would allow a fast connection to a cMP and the nMp now for a while.
The caldigit TS3 plus offers a native usb c gen. 2 port and works with the apple adapter quite good.
another solution would be a a pice box like the sonnet echo express sel + the sonnet allegro usb c card..