What PCIe add in cards can boot a NVMe M.2 SSD?

[UsandThem]

Intel vROC mentioned further in the following Article:

http://www.pcworld.com/article/3199...le-crazy-raid-configurations-for-a-price.html

Add a second card (with 4 NVMe M.2 SSDs) and now the system would be able to boot eight total NVMe SSDs.

Unfortunately the article mentions multiple catches:

vROC sounds pretty cool, but once again it looks like Intel is shooting themselves in the foot by requiring any bootable drives to be Intel, and by requiring that funky key to be purchased separately to allow the person to run something other than RAID 0, and it being something that nobody knows the price or availability.

I think these drawbacks will limit the potential customers somewhat.

 

[cbn]

Two new Alphacool M.2 SSD PCIe x 4 cards:

http://www.tomshardware.com/news/alphacool-m.2-ssd-cooling-hdx,35041.html

 

[VirtualLarry]

Watercooling, for M.2 SSDs? Isn't that a bit... much? Or are they overclockable as well? (I seem to recall Intel making some overtures about that, with their 730 series drives, but I thought that they cancelled the idea of user-overclocking of SSDs at the last minute.)

 

[mikeymikec]

I'd check the PCIE link speeds and revisions these adapters use (to ensure that the storage devices aren't being unnecessarily inhibited by poor adapter design), but the ones with a fan make me feel a lot more comfortable about using M.2. It's all very well and good to say "they're fine as long as they don't get hot" and "they shouldn't get that hot unless they're getting absolutely thrashed", but when one sells a piece of equipment, one wants to be able to say with confidence that it won't overheat (whether that results in hardware damage, reduced stability, reduced performance or a combination of any of the three), IMO.

Are there external drive enclosures for M.2 drives? That's the other thing I want to be sure of, because I'm sure as hell not going to start opening my PC when I need to mount a customer's M.2 drive for whatever reason.

Admittedly that cheap-as-chips Startech one that someone provided an image for earlier should be effectively cooled if there's a chassis fan mounted in the inside front of the case and is directing air towards the adapter I would have thought.

- edit - apparently even PCIE v1 X4 can do 1GB/sec max throughput:
https://en.wikipedia.org/wiki/PCI_Express#History_and_revisions

So unless the cheap-as-chips adapters are performance inhibited in some way then they should be able to harness the full capabilities of any M.2 drive I've seen. Multi-M.2 adapter cards though are probably worth checking. I also wonder how well an adapter can handle duplex and multi-M.2 bandwidths. Definitely the topic of a tech review methinks.

 

[cbn]

Watercooling, for M.2 SSDs? Isn't that a bit... much? Or are they overclockable as well? (I seem to recall Intel making some overtures about that, with their 730 series drives, but I thought that they cancelled the idea of user-overclocking of SSDs at the last minute.)

I think watercooling is a bit much too.

However, I do wonder if PCIe 4.0 (and especially controllers for the soon to arrive PCIe 5.0) will make substantially more heat when used at their full potential?

Maybe watercooling for 4 PCIe 5.0 NVMe SSDs in RAID-0 (on a PCIe x16 add-in board) booting using Intel's vROC?

 

[Elixer]

I still can't believe they did that with the dongle.
I would assume the mobo guys will just throw that in?

 

[Billy Tallis]

but when one sells a piece of equipment, one wants to be able to say with confidence that it won't overheat (whether that results in hardware damage, reduced stability, reduced performance or a combination of any of the three), IMO.

This flies in the face of trends all over the computer industry. Most people want devices that perform well for how they are actually used, and don't want to sacrifice real-world performance or battery life for the sake of unrealistic hypotheticals. Saying that a SSD should never throttle even during a torture test is like saying that CPUs and GPUs shouldn't have Turbo Boost modes to run faster when the workload is using fewer cores and fewer SIMD FPUs. It's like saying that smartphones should include fans so that they can maintain the same clock speeds during sustained gaming and benchmarking as they do during ordinary bursts of use.

We can say with confidence that your SSD will not be damaged or less stable or perform worse due to overheating during any normal use and a lot of use cases that are pretty far out of normal. Even when torture testing your SSD, it won't be damaged by having to throttle, and it isn't any more likely to return the wrong data when its throttling.

 

[VirtualLarry]

We can say with confidence that your SSD will not be damaged or less stable or perform worse due to overheating during any normal use and a lot of use cases that are pretty far out of normal. Even when torture testing your SSD, it won't be damaged by having to throttle, and it isn't any more likely to return the wrong data when its throttling.

My Intel 600p 256GB SSD is a trooper. I had it running in the "Ultra M.2" slot under the primary video card, in an ASRock Z170 Pro4S, with two video cards, both HIS 7950 3GB dual-slot, single-fan cards. In a RaidMax Cobra case, all stuffed into a desk cubby.

Well, watching the temps on the video cards, they would just keep rising... and rising... and rising! They hit like 90C+, more like 95C. And that poor SSD was shoved right under them.

The system kind of slowed down, as far as the storage went, but I think that it only crashed once. Or maybe I manually rebooted, to give it a chance to cool down. The SSD wasn't immediately detected by UEFI when I rebooted, but after it had cooled down for a bit, it seemed OK.

Not a great idea to run them that hot, I don't expect.

 

[cbn]

My Intel 600p 256GB SSD is a trooper. I had it running in the "Ultra M.2" slot under the primary video card, in an ASRock Z170 Pro4S, with two video cards, both HIS 7950 3GB dual-slot, single-fan cards. In a RaidMax Cobra case, all stuffed into a desk cubby.

Well, watching the temps on the video cards, they would just keep rising... and rising... and rising! They hit like 90C+, more like 95C. And that poor SSD was shoved right under them.

The system kind of slowed down, as far as the storage went, but I think that it only crashed once. Or maybe I manually rebooted, to give it a chance to cool down. The SSD wasn't immediately detected by UEFI when I rebooted, but after it had cooled down for a bit, it seemed OK.

Not a great idea to run them that hot, I don't expect.

I noticed you also mentioned slowdowns for SM2260 SSDs in this post. (quoted below)

I was a bit less impressed with both my 256GB Intel 600p M.2 SSD (3D TLC, I think), and my Adata XPG SX8000 SSD (3D MLC). The Intel was slower than the Samsung, by a noticeable amount, and the Adata slower by a lesser amount. (Well, maybe not slower, per se, but had some "pauses", which I attribute to the controller, an SMI something-or-other, that is used by both the Intel and Adata PCI-E M.2 SSDs.)

Was this occuring in a high temperature environment also?

P.S. One Newegg Intel 600p reviewer (Hail B on 6/21/2017, Newegg Intel 600p product link here) also mentioned slow down, but wasn't sure if it was related to temperarture.....but thought it might be.

This review is from: Intel SSD 600p Series (512GB, M.2 2280 80mm NVMe PCIe 3.0 x4, 3D1, TLC) Reseller Single Pack

Pros: It is only slight faster then my Samsung 850 EVO in some circumstances. Controller chip gets very hot quick (cant place finger on it). Also main issue with drive is random slow down. While performing Anti Virus Scan of drive, Copying data from one ssd to another etc, it will slow down for several seconds! I am using the latest firmware but I would recommend a regular Samsung EVO 850 or better yet if you full experience of X4 NVME get a Samsung 960 EVO/PRO. With far better performance.

+ Cheap for NVME x4 ( But waste of M.2 NVME slot since performance is not that great as compare to true PCIe x4 like Samsung 960)
+ 5 Year Warranty

Cons: - Poor mediocre performance as compare to regular SSD. Sometimes it performs like a regular HHD.
- Random slow down for several seconds where performance drops substantially which can be annoying.
- Controller chip gets very HOT maybe that's one reason for performance drop.
- Slowest of the nvme drives in the market!

Other Thoughts: I would not recommend this PCI X4 drive since it will just occupy your M.2 slot unless you have two slots. Drive has its issues and latest firmware has not resolved them. This points to the persistent issues that will not be resolved at all. Save your money and but a regular SSD.

 

[mikeymikec]

This flies in the face of trends all over the computer industry. Most people want devices that perform well for how they are actually used, and don't want to sacrifice real-world performance or battery life for the sake of unrealistic hypotheticals. Saying that a SSD should never throttle even during a torture test is like saying that CPUs and GPUs shouldn't have Turbo Boost modes to run faster when the workload is using fewer cores and fewer SIMD FPUs. It's like saying that smartphones should include fans so that they can maintain the same clock speeds during sustained gaming and benchmarking as they do during ordinary bursts of use.

You've got your analogy the wrong way around. To have it the correct way around to make it comparable with what I wrote, CPUs would have to throttle their clock speed in order to keep going at 100% CPU usage. Most CPUs don't do that. If you look up the specs of M.2 SSDs, they don't say "can do 3500MB/sec in turbo mode", nor do they say a speed they can sustain in "non-turbo mode". Furthermore, you can switch off turbo mode for every capable CPU I'm aware of. CPUs have a proven track record of being able to clock up and down safely and without massively affecting real-world performance as well as the user being able to disable such systems and have the CPU go at 100% until the cows come home. SSDs do not. In fact, SSDs are far more temperature sensitive, and pretty much the first word I heard when hearing about M.2 SSDs is that they're even more temperature sensitive than SATA SSDs.

A smartphone is nothing like a high-end SSD. Average joe will not walk in off the street wanting to buy a 3.5GB/sec M.2 SSD so that they can play Farmville or whatever. A product like the 960 PRO is only likely to attract people who want to see that kind of performance, and the most likely scenario includes people who will be pushing that device as hard as it can go for in sustained load for significant periods of time. No-one I've ever heard of asks such things of smartphones. If that was a common scenario, then some smartphones would definitely have fans because then there would be a market for it.

 

[cbn]

Though not specifically a PCIe add-in card I do think the following M.2 heatsink looks very interesting (can be used for add-in cards without heatsink as well as directly in the motherboard M.2 slot):

http://www.anandtech.com/show/11606/ekwb-launches-aftermarket-ekm2-aluminum-heatsink-for-m2-ssds

According to the article these are pretty cheap too:

The device costs €10 – €13, depending on the finish.

 

[Billy Tallis]

You've got your analogy the wrong way around. To have it the correct way around to make it comparable with what I wrote, CPUs would have to throttle their clock speed in order to keep going at 100% CPU usage. Most CPUs don't do that.

All modern CPUs do this out of the box. Turning off Turbo Modes without also overclocking reduces peak performance and doesn't increase sustained performance. CPUs are typically advertised with a nominal base clock and a maximum single-core Turbo clock, but even the base clock isn't 100% guaranteed and comes with caveats (such as whether you're using AVX instructions). There's less reason for a consumer SSD to advertise its thermally-throttled speeds than for a consumer CPU to advertise its all-cores AVX clock. Real-world applications sometimes use AVX. Real-world client storage workloads almost never hit the queue depths necessary to trigger thermal throttling, and it's even rarer for the I/O workload to sustain that intensity long enough for the drive to get up to its temperature limits.

Consumer SSD performance spec sheets ought to be moving toward more realistic measurements by not quoting everything at QD32. Listing performance on artificial torture tests would be a step further away from reality.

 

[UsandThem]

Though not specifically a PCIe add-in card I do think the following M.2 heatsink looks very interesting (can be used for add-in cards without heatsink as well as directly in the motherboard M.2 slot):

According to the article these are pretty cheap too:

The only problem anymore with hardware review/enthusiast sites (at least the way I see it), is they all seem to make the same announcement for new products, but there is almost never a review that comes later on. Since the announcements all look the same, it's more or less an advertising "article". /climbs down off of soapbox

That being said, I wonder how it will do performance wise? XabanakFanatik played around with a similar AlphaCool M.2 cooler in another thread, and the results really weren't impressive at all......especially compared to his results with the KyroM.2:

https://forums.anandtech.com/thread...le-normal-for-nvme-ssd.2497459/#post-38704055

960 Pro bare: 26C idle, 53C peak, 27C delta
960 Pro Kryo: 23C idle, 29C peak, 6C delta

960 Evo bare: 26C idle, 54C peak, 28C delta
960 Evo Alpha: 31C idle, 48C peak, 17C delta

 

[cbn]

Intel vROC mentioned further in the following Article:

http://www.pcworld.com/article/3199...le-crazy-raid-configurations-for-a-price.html

Add a second card (with 4 NVMe M.2 SSDs) and now the system would be able to boot eight total NVMe SSDs.

Unfortunately the article mentions multiple catches:

AMD now has NVMe RAID on CPU available for Socket TR4 (Threadripper):

https://www.anandtech.com/show/11892/amd-launches-nvme-raid-support-for-ryzen-threadripper-platform

P.S. Here is that ASUS M.2 x 16 card (from the quote above)-->
https://www.bhphotovideo.com/c/product/1358493-REG/asus_hyper_m_2_x16_card_hyper_m_2_x16_pcie.html ($55)

 

[cbn]

Here is a video (originally posted by SamMaster here) showing Threadripper working with 8 NVMe SSDs (in CPU RAID) using two of those ASUS Hyper M.2 x16 cards (from the above post);

So the ASUS card does work.

 

[mnanandtech]

That is not the question. An add-in adaptor card has zero to do with being able to boot from a device. That is entirely down to UEFI* support, and any OROM present on the device you're trying to boot.

An adaptor will not magically allow you to boot an NVMe SSD on a system without NVMe support in UEFI, and any PCIe-to-M.2 adaptor will allow you to boot on a system with NVMe support. It is simply a passive electrical adaptor.

*No legacy BIOS to my knowledge is able to boot from an NVMe device.

Wrong on two points. Or I should say, mostly right but there are two important exceptions.

I am writing this note on a Win10 installation installed on an M.2 NVMe drive plugged in to a legacy BIOS motherboard I bought in 2010. The trick is to buy a Samsung 950 Pro. Samsung went through extra effort to make this -- and only this -- generation of SSDs bootable on legacy BIOS.

I have both the preceding Samsung SM951 and following Samsung 960 Pro, installed on the same system as Storage Spaces RAID0. Neither one can boot on this motherboard. And the former as AHCI, so NVMe has nothing to do with it, it's all about the option ROM apparently. And Windows 10 is running Samsung's NVMe driver after boot, so there is also no problem with legacy BIOS taking full advantage of the NVMe protocol, since the BIOS is not involved in drive communication post-boot. The benchmarks I've run show the drives all preforming very well up to the limit of the motherboard's PCIe gen 2.0 speed ceiling.

A 950 Pro isn't quite the fastest drive in the world, but it is extremely fast even by today's standards, and if you are upgrading a legacy machine with one, it is probably the fastest single + boot drive performance possible for you and you should be happy to have it.

Further, there is actually one possible trick you can pull to boot *any* M.2 drive on a legacy mb -- by using a Clover UEFI USB boot drive. I wasn't able to get it to work on my mb, but it may work on yours:
http://www.insanelymac.com/forum/to...ash-with-macos-distr-under-windows/?p=2522798

Secondly, yes no basic M.2 adapter card will change a drive's bootability; but that doesn't mean cards "have nothing to do with it". RAID add-in-cards provide bootability on-board, independent of what drives are plugged in to them. I haven't looked at availability of such cards recently, but as of a year ago Highpoint was promising a family of such cards; if they are not available by now I would think it is just a matter of (short) time. Plugging a single M.2 into such a board would indeed convert a recent drive into being bootable on legacy BIOS, provided the RAID card itself isn't UEFI-only.

 

[cbn]

I am writing this note on a Win10 installation installed on an M.2 NVMe drive plugged in to a legacy BIOS motherboard I bought in 2010. The trick is to buy a Samsung 950 Pro. Samsung went through extra effort to make this -- and only this -- generation of SSDs bootable on legacy BIOS.

Yep, I found that out also (Very interesting):

https://forums.anandtech.com/threads/nvme-drive-booting-in-ahci-mode-solved.2500796/#post-38776555

I have both the preceding Samsung SM951 and following Samsung 960 Pro, installed on the same system as Storage Spaces RAID0. Neither one can boot on this motherboard. And the former as AHCI, so NVMe has nothing to do with it, it's all about the option ROM apparently. And Windows 10 is running Samsung's NVMe driver after boot, so there is also no problem with legacy BIOS taking full advantage of the NVMe protocol, since the BIOS is not involved in drive communication post-boot.

And you answered the question I had about the 960 Pro:

https://forums.anandtech.com/threads/nvme-drive-booting-in-ahci-mode-solved.2500796/#post-38779779

Thank you!

 

[Insert_Nickname]

I am writing this note on a Win10 installation installed on an M.2 NVMe drive plugged in to a legacy BIOS motherboard I bought in 2010. The trick is to buy a Samsung 950 Pro. Samsung went through extra effort to make this -- and only this -- generation of SSDs bootable on legacy BIOS.

This is because Samsung actually bothered to put an OROM on the 950 Pro. I thought everyone knew that. The OROM is what makes the drive bootable in a BIOS based system, not the BIOS. The BIOS does not include the necessary drivers/firmware to recognise the drive, much less boot from from it. Trouble is, as you write, its limited to the 950 Pro.

The Kingston HyperX Predator and Plextor M6e also provides their own OROM, but they're SATA drives.

Further, there is actually one possible trick you can pull to boot *any* M.2 drive on a legacy mb -- by using a Clover UEFI USB boot drive. I wasn't able to get it to work on my mb, but it may work on yours:
http://www.insanelymac.com/forum/to...ash-with-macos-distr-under-windows/?p=2522798

Seems like a lot of work for such old systems. I'd rather buy either a high performance, or low cost SATA drive. Besides most such older (Intel, in particular) systems are limited to SATA2, or best case a 3rd party SATA3 controller. Neither is ideal.

Another thing is that such systems are also limited to PCIe 2.0, so even the 950 is limited to about 1600MB/s sequential. PCIe 3.0 wasn't introduced until Ivy Bridge/7-series chipset.

BTW, its not really a problem for me since my Crosshair/Ryzen system has native support.

Secondly, yes no basic M.2 adapter card will change a drive's bootability; but that doesn't mean cards "have nothing to do with it". RAID add-in-cards provide bootability on-board, independent of what drives are plugged in to them. I haven't looked at availability of such cards recently, but as of a year ago Highpoint was promising a family of such cards; if they are not available by now I would think it is just a matter of (short) time. Plugging a single M.2 into such a board would indeed convert a recent drive into being bootable on legacy BIOS, provided the RAID card itself isn't UEFI-only.

Again, this is because the cards themselves have an OROM. OROMs don't care about what adaptor you use.

I should mention OROMs (Option ROMs) have been here since the first IBM PCs. They're nothing more then extensions (or you could think of them as "BIOS drivers") to the BIOS. Since they precisely provide the capability for a BIOS to recognise hardware it doesn't know natively.

 

[Smoblikat]

That is not the question. An add-in adaptor card has zero to do with being able to boot from a device. That is entirely down to UEFI* support, and any OROM present on the device you're trying to boot.

An adaptor will not magically allow you to boot an NVMe SSD on a system without NVMe support in UEFI, and any PCIe-to-M.2 adaptor will allow you to boot on a system with NVMe support. It is simply a passive electrical adaptor.

*No legacy BIOS to my knowledge is able to boot from an NVMe device.

Do you have any links to this info? Or any info at all on how PCIE -> M.2 adapters work internally? Because I ran into a situation over the past couple of days thats making me question this. I purchased this card:
https://www.ebay.com/itm/PCI-Expres...e=STRK:MEBIDX:IT&_trksid=p2060353.m2749.l2649
along with a 960 EVO for my buddies new computer. His motherboard is an ASRock Z68 Extreme3Gen3, and I have successfully added NVMe support to these boards before, so I know it *can* work. With that card plugged in, along with the 960EVO in that card, the BIOS refuses to detect it as a bootable option. It does appear as an option in the windows installer once I load the correct drivers (windows 7), but cannot be installed to due to the board not "supporting" it. After a lot of trial and error, I finally took my adapter card out of my Z77 computer:
https://www.ebay.com/itm/2-Port-NGF...e=STRK:MEBIDX:IT&_trksid=p2060353.m2749.l2649
and with my M.2 drive still plugged into it (Samsung PM851NVMe) the BIOS very happily detected "Microsoft Windows" as a boot option. I did not try using the PM851 in the black adapter though, so im wondering if the magic came from the OROM on the PM851 or if there is somthing that makes certain PCIe adapters bootable and others not. I have used the red adapter on multiple boards before, but ive only ever used the Samsung PM851 drive with them, this i smy first time using somthing other than that with hardware that doesnt natively support NVMe.

Long story short I ordered another red adapter anyway, but im wondering if I should get my hopes up about it, or if theres somthing special about the older PM851 that makes it work better with legacy hardware.

 

[Insert_Nickname]

Do you have any links to this info? Or any info at all on how PCIE -> M.2 adapters work internally? Because I ran into a situation over the past couple of days thats making me question this. I purchased this card:
https://www.ebay.com/itm/PCI-Express-4x-to-M-Key-M-2-SSD-Slot-Adapter-Converter-Card-For-2242-2260-2280/292096193411?ssPageName=STRK:MEBIDX:IT&_trksid=p2060353.m2749.l2649
along with a 960 EVO for my buddies new computer. His motherboard is an ASRock Z68 Extreme3Gen3, and I have successfully added NVMe support to these boards before, so I know it *can* work. With that card plugged in, along with the 960EVO in that card, the BIOS refuses to detect it as a bootable option. It does appear as an option in the windows installer once I load the correct drivers (windows 7), but cannot be installed to due to the board not "supporting" it. After a lot of trial and error, I finally took my adapter card out of my Z77 computer:
https://www.ebay.com/itm/2-Port-NGFF-M-2-B-M-Key-SSD-to-PCI-E-PCI-Express-4X-4Lane-Adapter-Converter-Card/112418754728?ssPageName=STRK:MEBIDX:IT&_trksid=p2060353.m2749.l2649
and with my M.2 drive still plugged into it (Samsung PM851NVMe) the BIOS very happily detected "Microsoft Windows" as a boot option. I did not try using the PM851 in the black adapter though, so im wondering if the magic came from the OROM on the PM851 or if there is somthing that makes certain PCIe adapters bootable and others not. I have used the red adapter on multiple boards before, but ive only ever used the Samsung PM851 drive with them, this i smy first time using somthing other than that with hardware that doesnt natively support NVMe.

Long story short I ordered another red adapter anyway, but im wondering if I should get my hopes up about it, or if theres somthing special about the older PM851 that makes it work better with legacy hardware.

Did you by any chance try downgrading the link to PCIe 2.0? PCIe 3.0 can be quite finicky with adaptors.

 

[Smoblikat]

Did you by any chance try downgrading the link to PCIe 2.0? PCIe 3.0 can be quite finicky with adaptors.

Its actually a sandy bridge chip in there, so it has no PCIE3 support anyway. Definitely in 2.0 mode

I made a post in this thread:
https://www.win-raid.com/t871f16-Gu...or-all-Systems-with-an-AMI-UEFI-BIOS-163.html

It seems to be the most active hub for this sort of thing.

 

[cbn]

ASRock's new Ultra Quad M.2 card:
https://www.tweaktown.com/news/60511/asrocks-new-ultra-quad-2-card-4-nvme-ssds-pcie-x16/index.html
https://www.tweaktown.com/news/60511/asrocks-new-ultra-quad-2-card-4-nvme-ssds-pcie-x16/index.html

Motherboard, of course, does need PCIe bifurcation support:

You can't just throw the ASRock Ultra Quad M.2 Card into any system, as it has no PCIe switch on-board, so you'll need to own a motherboard with PCIe bifurcation support. High-end chipsets have this, with ASRock pointing in the direction of Intel's latest X299 chipset, as well as AMD's new X399 chipset.

 

[aigomorla]

This is the one i got for free with my Aorus G9. It came as an extra...

It boots fine...

 

[Topweasel]

It is actually a PCIe x16 device that can raid all four drives.....but I found out it cannot boot with all the drives in RAID:

http://www.tomsitpro.com/articles/hp-reveals-turboz-quad-pro,1-3022.html

To use as a boot device...... one of the drives must not be in RAID, so a person (as an example) could have 1 M.2 NVMe as boot and 3 M.2 NVMe in RAID 0.

I wonder with something like that or maybe some of the others, if they are just parsing out PCIe to the drives, if you could get away with a 4 way RAID 0 bootable on X299 or X399?

 

[paperfist]

Here is one that is pretty amazing:

http://www.storagereview.com/hp_reveals_new_z_turbo_drive_quad_pro

http://www8.hp.com/us/en/workstations/z-turbo-drive-g3.html

Neat I didn't know you could use multiple m.2 drives. I have an Asus PCIe adapter card for my m.2 drive.