Question Does NVMe require M.2?

[Asphodelus]

If I were to purchase a datacenter-style SSD that goes into the PCIe x16 slot, would I be able to boot from it via NVMe? Or do mobos only recognize NVMe SSDs if they're plugged into the M.2 slot?

 

[Insert_Nickname]

You might not be able to see it in the UEFI, but the OS installer should see and boot from it no problem.

 

[Fernando 1]

@Asphodelus:
Provided, that your mainboard/system has an UEFI BIOS with an NVMe EFI module within it (can be inserted by the user), you will be able to boot off a PCIe connected NVMe SSD.

 

[gea]

Think of M.2, oculink or pci-e slots as different connectors to add nvme disks.

If a system can boot from either (ex M.2), it can boot from the others as well. If a mainboard has M.2 slots this is always the case.

On older systems without M.2 slots you usually cannot boot from nvme at all (beside raid hardware adapters with their own bios extensions ex broadcom 9400) but you can use NVMe ex via a pci-e card as datadisks.

 

[aigomorla]

You need UEFI as others said as your bios, and have it be able to boot from a PCI-E.
It would help to know what board you wish to install that nVME.
If you do not have the option to boot from nVME which is bios level UEFI feature, then no you will not be able to boot from it, without using some form of customized bios.

 

[BonzaiDuck]

You need UEFI as others said as your bios, and have it be able to boot from a PCI-E.
It would help to know what board you wish to install that nVME.
If you do not have the option to boot from nVME which is bios level UEFI feature, then no you will not be able to boot from it, without using some form of customized bios.

Actually, this discussion has me thinking about z68 (gen-3) and z77 motherboards running in my household. I'm PRETTY SURE they have UEFI BIOSes. I KNOW that they provide PCIE v.3.0 bandwidth, if -- if! -- the processor is an Ivy Bridge as opposed to Sandy. But I didn't want to mess around experimenting with NVME drives in those boxes. And -- why invest even the low price for an NVME drive for such an experiment, for systems with processors of those generations? Moms and Bro are perfectly happy with SATA SSD boot disks.

 

[Jimminy]

Actually, this discussion has me thinking about z68 (gen-3) and z77 motherboards running in my household. I'm PRETTY SURE they have UEFI BIOSes. I KNOW that they provide PCIE v.3.0 bandwidth, if -- if! -- the processor is an Ivy Bridge as opposed to Sandy. But I didn't want to mess around experimenting with NVME drives in those boxes. And -- why invest even the low price for an NVME drive for such an experiment, for systems with processors of those generations? Moms and Bro are perfectly happy with SATA SSD boot disks.

I have a gen 2 sandy bridge system (core i5-2400). I added a couple of nvme disks on pcie adapter boards. They work fine as data disks, but as you say, there's no way to use them as boot disks, because the bios doesn't know they exist.

I think there is software that can do it, but it's too complicated to set up (for me, at least). I haven't seen anything that would work for us dummies

I also use a sata 3 ssd for boot disk, and it works good enough, but nowhere near as fast as the nvme data disks, even though mine are only running at pcie gen 2.

 

[BonzaiDuck]

I have a gen 2 sandy bridge system (core i5-2400). I added a couple of nvme disks on pcie adapter boards. They work fine as data disks, but as you say, there's no way to use them as boot disks, because the bios doesn't know they exist.

I think there is software that can do it, but it's too complicated to set up (for me, at least). I haven't seen anything that would work for us dummies

I also use a sata 3 ssd for boot disk, and it works good enough, but nowhere near as fast as the nvme data disks, even though mine are only running at pcie gen 2.

I get flak here all the time for the "thing I've done" to all my systems. A $30 piece of software from a company in Shanghai is the center of all of it: PrimoCache by Romex Software. They have a lively and robust forum of users.

This was one reason I began increasing RAM in my systems. Just caching HDDs and SSDs to 8GB of RAM seems to transform the system, so those old Sandy Bridge systems might have more RAM than they already do, but souping them up further with PrimoCache RAM and SSD caches isn't a big priority for what they're currently used for.

I get all sorts of criticism. "It's just a benchmark score!" "You can move up to gen-10 processors and Z490 boards with Optane and PCIE v.4.0 Samsung 980 Pro speeds!"

But the licenses are "lifetime". I think I got a 3-PC license, and a fourth one is for 1 PC.

There has never been a hitch or a glitch with it. I DID discover four months ago that my "best" system wouldn't hibernate or sleep without throwing a critical stop event-ID 41 in the event-viewer. You begin to imagine all sorts of causes, like the PSU, bad memory sticks -- any number of things. But as we know, it's always best to investigate drivers and software as causes for such problems. Finally, I just ran a web-search on "failure to hibernate or sleep" with "INTERNAL_POWER_ERROR" -- the bug-check string.

It was the Samsung NVME driver! That's all it was!

At first, I concerned myself that hibernating 32GB of RAM would rack up TBWs and shorten the lives of my NVME drives. Always do the simple math. Even if I double the disk writes for hibernating to include some ballpark of daily disk activity, I come up with about 50 years of lifespan for the NVME drives.

The program will save RAM caches (called L1 by the program) to disk, and restore them on the next boot-up. You can cache HDDs and SATA SSDs to NVME drives, and cache all of it to RAM in a three-tiered configuration. The worst thing that has ever happened: a Windows 10 major feature update will bork the caches, and they need to be flushed.

I think there's an ongoing discussion over at the Romex forums about using the program with Optane, when you would think that Optane obviates any need to use the program. As I see it, with that little amount of trouble, a lack of any data loss or other adverse effects, I'm not giving it up as a feature of my computer systems.

Thing is -- I just thought of it so one could suspect I've lost too many brain cells already at 74 -- I could throw a Samsung 960 or 970 Pro or EVO into those Z68 and Z77 systems and make them cache drives. You'd boot from your SATA SSD, but the caches would come off the NVME drive. A 256GB NVME is enough, but I'm not sure you can buy them that small anymore! So maybe you boost your Sandy/Ivy/Z68/Z77 to 24GB or 32 GB of RAM, throw in an NVME in a PCIE x4 slot, and . . . surprise yourself . . .

I have friends who are electronics-tech PC "savvy", but they live off the old-gen dated-tech junk-pile. One of them is still using a C2D Conroe system as his main desktop. For me, I'm "just lookin'" at an i5-10600K hexa-core processor and Z490 motherboard. Just lookin'. Eventually -- maybe. But what I've got right now is fulfilling all my needs at the moment.

 

[Jimminy]

A 256GB NVME is enough, but I'm not sure you can buy them that small anymore!

Duck, look on ebay. I've had no problem finding new nvme disks which were pulled from new computers to upgrade them. I've only seen small ones, like 128 or 256 GB. The ones I bought showed only a few hours and power cycles.

Same for small 2.5" 7200 rpm hard drives. I've bought several 500 GB pulls for less than $20, which I use for usb backup drives.

So far, I've not been burned on any of these "pulls".

 

[BonzaiDuck]

Duck, look on ebay. I've had no problem finding new nvme disks which were pulled from new computers to upgrade them. I've only seen small ones, like 128 or 256 GB. The ones I bought showed only a few hours and power cycles.

Same for small 2.5" 7200 rpm hard drives. I've bought several 500 GB pulls for less than $20, which I use for usb backup drives.

So far, I've not been burned on any of these "pulls".

You know? You're absolutely positively right with this. Even so, I just took another look at the Samsung offerings. The Pro models only come with a minimum 500GB capacity, but the EVOs -- in the 970 line -- can be had as 256GB for about $60. And I think I"ll have a look at those EBay jobs you mention.

Just again, to stress my point about it. Suppose you can get a 970 EVO equivalent for $30 and you spend $30 on the caching software. That's chump change. With extra RAM on an old Z68 or Z77 board, I would be optimistic. How much would you be limited by PCIE v.2.0? You're going to cache that to RAM anyway, if you use the three-tiered approach. And there are really no drawbacks to that, either. The PCIE v.2.0 bandwidth would make the Ebay NVME's more attractive. Certainly no problem if they approached Samsung 960 performance, anyway.

Also, just to pass on the skinny -- check out the SK Hynix 1TB NVME "Gold P31" drives. You might not want to waste something like that on a Sandy generation box, but for something else -- you might have an interest. Those seem to be a bargain for the price, and people are rave about them in terms of performance and power efficiency. They supposedly run very cool. I'm going to take a look at those Ebay bargains you mentioned.

 

[BonzaiDuck]

For Jimminy -- I checked the Ebay offerings and they are numerous. Many of those are laptop pulls. It isn't the offerings per se that I have trouble with, but how they are represented (or confused) on EBay. Here and there, you could find 128 GB "NVME" drives for which the spec information would show the sequential read and write speed maximums for the drive, so you might see 2,000 MB/s, etc. Other times, the ad would say "NVME" but more thorough investigation of the information provided would show "SATA III" and spec maximums of below 600 MB/s. One has to be careful!

Now -- here's a question I pose to the forum readers. The difference between the PCIE v.2.0 and v.3.0 specs is a matter of doubling the throughput bandwidth per PCIE lane. So the bandwidth or speed of the PCIE v.2.0 lane is 500 MB/s, but PCIE v.3.0 is just below 1GB/sec. Thus, a PCIE v.2.0 x4 connection would admit 4 x 500 or 2GB/s. And 2GB/s is the upper limit of spec sequential read speed for a lot of NVME M.2 PCIE x4 drives.

Would one then infer that the an NVME PCIE x4 drive in a PCIE v.2.0 slot is only slightly limited? In other words, are you likely to see performance that is just a sliver below that of the same drive in a PCIE v.3.0 slot?

It would seem so. And if that's the case, you could do a lot of things with a 256GB NVME drive in a system providing only PCIE v.2.0 -- for instance, a Z68 or Z77 motherboard configured only with a Sandy Bridge chip as opposed to an Ivy Bridge.

This spurs me at least to investigate certain DDR3 RAM modules -- if available -- for such a system. Not saying I will buy any, but I'm curious . . .

 

[Jimminy]

I thought PCIE gen 3 was almost twice as fast as gen 2 ?

As a reference point, I have a samsung SAMSUNG MZVLB256HBHQ-000L7 256.0 GB nvme drive on an adapter card plugged into a gen 2 PCIE slot. I believe the drive is pcie gen 3, but the slot is only gen 2.

Crystal disk mark (quick 16MiB test) shows:
[Read]
Sequential 1MiB (Q= 8, T= 1): 1668.270 MB/s
Sequential 1MiB (Q= 1, T= 1): 1195.847 MB/s
Random 4KiB (Q= 32, T=16): 976.960 MB/s
Random 4KiB (Q= 1, T= 1): 36.143 MB/s

[Write]
Sequential 1MiB (Q= 8, T= 1): 1545.793 MB/s
Sequential 1MiB (Q= 1, T= 1): 1176.494 MB/s
Random 4KiB (Q= 32, T=16): 803.265 MB/s
Random 4KiB (Q= 1, T= 1): 75.833 MB/s

I have no idea if this is normal or not, but it's plenty fast for me, especially on an old sandy bridge rig.

 

[BonzaiDuck]

I thought PCIE gen 3 was almost twice as fast as gen 2 ?

As a reference point, I have a samsung SAMSUNG MZVLB256HBHQ-000L7 256.0 GB nvme drive on an adapter card plugged into a gen 2 PCIE slot. I believe the drive is pcie gen 3, but the slot is only gen 2.

Crystal disk mark (quick 16MiB test) shows:
[Read]
Sequential 1MiB (Q= 8, T= 1): 1668.270 MB/s
Sequential 1MiB (Q= 1, T= 1): 1195.847 MB/s
Random 4KiB (Q= 32, T=16): 976.960 MB/s
Random 4KiB (Q= 1, T= 1): 36.143 MB/s

[Write]
Sequential 1MiB (Q= 8, T= 1): 1545.793 MB/s
Sequential 1MiB (Q= 1, T= 1): 1176.494 MB/s
Random 4KiB (Q= 32, T=16): 803.265 MB/s
Random 4KiB (Q= 1, T= 1): 75.833 MB/s

I have no idea if this is normal or not, but it's plenty fast for me, especially on an old sandy bridge rig.

See -- you're proving my hypothesis right there. The sequential benchmarks show it.

The MZVLB256HBHQ-000L7 shows a Passmark score roughly 80% of a 970 EVO NVME. If PCIE v.2.0 has roughly half the speed of PCIE v.3.0 on each lane, of course it would have a total performance roughly half comparing v.2.0 and v.3.0 between four allocated lanes for PCIE x4. But at about 500 MB/s per lane on PCIE v.2.0, there should be adequate bandwidth for an NVME benchmarked and spec'd at about 2,000 MB/s. The Samsung drives are spec'd between 2,000 and 3,500, depending on generation and model for PCIE v.3.0 capability. So you're getting just a shade less than 2,000 MB/s for the sequential read rate.

Looking at it another way, your SATA SSDs should top out at 500+ MB/s in a sequential read-rate. So the NVME is performing more than three times faster, as I see it. Some others might scrutinize other benchmark scores, pointing to this or that, but the first thing I look at it is the sequential read rate.

Now -- given that -- my ballpark assessment -- suppose you find a way to add enough RAM at the right price to that "Old Sandy Bridge" system. I have one with 24 GB = 2x8GB + 2x4GB. I could swap out the 8GB kit for an additional 16=2x8GB (about $100), for a total of 32 GB. Then, allocate half that to caches under Primo-Cache on devices cached to the NVME drive (L2 -- L1 is Primo's label for RAM).

I can post my Anvil Benchmark scores, but our friends here will just say "those are just benchmarks! Bench-marks!" It doesn't matter, though. It's the reverse of the HAL 9000 computer in the Kubrick movie "2001". "Dave? My mind is going . . . . my mind is going . . . . I CAN FEEL IT . . . I CAN FEEL IT."

As opposed to "mind going", for the improvement -- I can feel it.

I was going to post this PDF "white paper" I found yesterday on Romex Software's web-site. I'm not paid or inclined to "promote" their software, and I mentioned the mild criticisms. You would think that Romex wouldn't have anything to add with Intel's recent-vintage Optane memory hardware. Reading the PDF, it may only have something to do with the inconvenience of tweaking the BIOS as opposed to configuring Primo-Cache. But it appears they have something to say -- even about using PrimoCache with Optane.

The Acceleration Experience of the Cooperation with Intel Optane Memory and PrimoCache

For the lifetime license price, at a very, very bare minimum -- "It does no harm". That being said, I've been happier than a pig-in-s*** to use it for the last (. . . mmmm . . . ) seven years.

 

[Jimminy]

For the lifetime license price, at a very, very bare minimum -- "It does no harm". That being said, I've been happier than a pig-in-s*** to use it for the last (. . . mmmm . . . ) seven years.

Sounds wonderful.
I wonder why there is no open source software that is similar. It must be very difficult to write this type software.

 

[BonzaiDuck]

Sounds wonderful.
I wonder why there is no open source software that is similar. It must be very difficult to write this type software.

That's a good question, and the elements of it were always in the back of my mind, given the online research I did to find PrimoCache.

I should have taken notes about the search. Somehow I became aware of caching as a possibility, or the existence of some software that would do it. Looking for options, I found maybe two programs that had worked with Windows 7, but were no longer being updated. I can't even remember if you could download them for free, or if they had a price-tag. They may have had American sources. I don't even remember their names anymore, but I wasn't going to adopt the use of "stale" software with no new revisions.

I discovered another software product called SuperCache developed by a company named SuperSpeed LLC -- based in Boston, Mass. When I took a closer look some four or five years ago, they had their own web-page, touting various certifications like "Microsoft Partner". There was plenty of explanatory material about what SuperCache would do and how it did it. It was enough information to prove that it fell considerably short of Romex's PrimoCache features.

Looking for it now, I find a web-page "under construction" and downloads for SuperSpeed available on SoftPedia: SuperCache Download and Information

On SoftPedia, it apparently even has a license price of close to $80, more than twice the lifetime license expense for PrimoCache's "single PC" offering. But based on what I'd learned about it when I first examined the available information years ago, I wouldn't bother with it when PrimoCache more than fills the bill. PrimoCache had many more features that just left SuperCache in the dust.

Some of our forum members have argued that Windows itself does some caching, thus insinuating that Primo is redundant. Today's argument seems to be the availability of NVME's that provide 7,000 MB/s sequential read benches, provided you have a motherboard with PCIE v.4.0 capability and the processor to match. Of course, you could still cache a Sammy Pro 980 to RAM, and then double your benchmarks again. Would a human being notice it? I'm not so sure of that.

The worst thing that ever happened due to Windows feature updates and active caches under Primo appeared at boot time when the cache definitions for drives would disappear, and you had to recreate them. Primo gives an option to re-configure "from previous", so all of that is very easy. Further, this problem with feature updates seems to have gone away, disappeared or diminished, since the only two episodes I remember were feature updates of 2017 and 2018. But there has never been a case when there was corruption on the source disks that were cached, and Windows always boots without event when those episodes occurred. You just had to flush or re-create the caches.

The SOLUTION?! Simple! Windows always announces feature updates before it installs them, unless you have that feature turned off. So you can turn it back on. Simply pause or delete the caches, make the upgrade, then reactivate the caching.

As I'd mentioned before, Romex has a very active forum. In addition to PrimoCache, they'd offered their RAMDisk software, possibly other items. Primo had originated with a product called "FancyCache", and they still have a forum and info about that item, as I recall. I'm occasionally poking my nose into their forums under the same handle that I use here. Once, maybe a couple years ago, I started a thread about configuration strategies for Primo, and the posts seem to have piled up on it since then without much intervention or visitation from me.

FOOTNOTE AND UPDATE: Now that I think of it and looking at an "Alternatives To" web-page, I recall this program: MaxVeloSSD

I'm wondering if this product doesn't originate in the EU. I think I also found shortcomings with it, as compared to Primo, because it was among the options I looked at sometime between 2014 and 2017.

 

[BonzaiDuck]

Just continuing, rather than append to my previous post, here is the user-guide for MaxVeloSSD.

And here is the user-guide for PrimoCache.

If one user-guide appears more thorough than the other, or the formatting and link-structure of the manual outstrips the competitor's, it might not be a significant factor in choosing which software is better.

To make that determination, one would have to install both products on different computers -- more troublesome to do it successively on one machine -- then test and evaluate them. At this point, I haven't the time for that. I'm satisfied with PrimoCache. It appears that MaxVeloSSD also provides two-tiered SSD and RAM caching. But my parting remark about PrimoCAche: It works for me.

ADDENDUM -- I see part of this conversation about PrimoCache includes posts I made on a thread about "Is PrimoCache Worth It?" -- also very current. So for both threads, here is one more advisory about PrimoCache.

If you are using a program like Macrium Reflect to backup all of your persistent storage -- NVME, SATA SSD and HDD -- a program like Macrium will throw up a message that it is about to start a scheduled backup. I have mine scheduled for every morning at 9AM, excluding Saturday and Sunday. You should have time to bring up PrimoCache and "Pause" all the caches, one after another. In fact, it doesn't seem to matter if you pause the caches as the backup begins to run. Basically, I have only two cache tasks, and this is pretty easy.

The reason you might wish to do this is straightforward. If you are going to back up storage media that is cached, and especially for a full backup or differential backup, you're going to read a lot of data from the source disks, and it's going to fill your caches, pushing out the data you want in there for programs and files you access frequently. So you pause the caches to keep that from happening.

 

[Jimminy]

If you are using a program like Macrium Reflect to backup all of your persistent storage -- NVME, SATA SSD and HDD -- a program like Macrium will throw up a message that it is about to start a scheduled backup. I have mine scheduled for every morning at 9AM, excluding Saturday and Sunday.

If primo has a command line interface, you might be able to schedule a command to pause it. Maybe at 8:55, or something like that. And another later to re-enable it.

Just a thought.

 

[BonzaiDuck]

If primo has a command line interface, you might be able to schedule a command to pause it. Maybe at 8:55, or something like that. And another later to re-enable it.

Just a thought.

I was thinking the same thing . . . I'll have to poke around and see what can be done.

 

[BonzaiDuck]

If primo has a command line interface, you might be able to schedule a command to pause it. Maybe at 8:55, or something like that. And another later to re-enable it.

Just a thought.

Here's the ROMEX PrimoCAche explanation of their CLI.

PrimoCache Command Line Interface

You would have two executions of their program mentioned in the explanation: one with parameters that would pause each cache, and a second with commands that would resume each cache task.

 

[Jimminy]

Here's the ROMEX PrimoCAche explanation of their CLI.

PrimoCache Command Line Interface

You would have two executions of their program mentioned in the explanation: one with parameters that would pause each cache, and a second with commands that would resume each cache task.

Exactly. Schedule pause command right before your 9:00 backup, and schedule the resume command after an amount of time that you think the backup runs. You might look at the macrium logs to see if you can determine how long the backup takes to run. Good luck!

 

[BonzaiDuck]

Exactly. Schedule pause command right before your 9:00 backup, and schedule the resume command after an amount of time that you think the backup runs. You might look at the macrium logs to see if you can determine how long the backup takes to run. Good luck!

Yes -- that's true. I think Macrium offers you a set of "canned strategies" but otherwise let's you choose your preferences as a custom configuration. I set mine up to do a full backup once a month, a differential backup on Mondays, and incremental backups for the remainder of the business week. So the incremental backups always take mere minutes, while the Monday backups can take as long as an hour, and those full backups take more time than that.

 

[BonzaiDuck]

ON THE PRIMOCACHE CLI SOLUTION FOR BACKUPS

I think this may be the plan for doing it. You'd simply schedule the command-line script to pause the caches, because the backups all start at the same time of day. Then, create a link to the script that resumes them. Two mouse clicks on a shortcut? No problem, and you can choose when to execute it based on the status of Macrium in the system tray.

Even so. I remember reading somewhere in the voluminous Primo instruction guide that it has a built-in feature that limits the type of throughput occurring during a large disk backup. That is -- I remember "something like that". At my age, there's a limit to remembrance quantity and quality.

 

[HappyCracker]

I think there is software that can do it, but it's too complicated to set up (for me, at least). I haven't seen anything that would work for us dummies

I've done this. The Clover bootloader on an SD card is what my R720 boots to. It loads an NVMe module, which can then find the Optane drive in a PCIe slot (through an M.2 to PCIe adapter), and finally boot from the SSD. It wasn't a horrible setup; it took a little persistence to get the config file nailed down, but I haven't had to mess with it in a long time.