They have been mostly displaced by enterprise SSD based drives that have high IOPs or big data solutions with transaction logs and distributed reads and writes.
Edit: the one thing I'd like from a ram disk for it to survive a reboot; poking about, most memory contents doesn't change across a reboot but AFAIK no OS has the ability to persist a RAMdisk across a restart.
That suggestion makes no sense. It's one thing to buy a SSD/PCI interface that allows you to plug in your RAM sticks. It's an entirely different thing to buy a brand new computer.
Your suggestion is like "do you need more disk space? Just buy a new computer."
..unless, of course - you are blessed with one of the devices that have RAM soldered on. I cannot express how much I dislike these devices.
2. For the cost of upgrading to larger RAM is prohibitive because you have to throw away your existing DIMMs and replace them with new-to-market DIMMs that are extortionate because they've only just appeared
No one presented such use case in this discussion.
Most have four and I'd say most people have two free.
To answer your question though, devices like that have been made from time to time, but I don't think I've seen one since DDR2. It really would be easier to increase your system ram and run a software ram drive.
If you're already maxed out on ram, you would need to get a new system with more ram slots or higher capacity per slot, etc of course. If you get some off-lease server hardware, you can get tons of slots though.
If you're wanting to do something with old ram sitting around, you should probably find a marketplace to sell it. Or plan to hoard it until it's retro.
You don't have to buy a whole new computer to add RAM to it, unless you're already hitting the limits of the motherboard, either slots or total capacity. And if you are, you should be able to get away with getting a new motherboard instead of a whole new computer.
Right.
Do you think that anyone interested in using RAM drives and willing to spend money on hardware wouldn't have already considered using freely available software RAM drive?
Maybe author is not super into "build a PC" movement, but know they need more RAM, so they ask the question that makes sense to them. I don't know, and I won't assume either, so asking basic questions first makes sense.
Maybe they weren't aware you could create arbitrary memory space on hard-drives via swap, so questions around that would be appropriate as well.
https://www.superssd.com/products/ramsan-400/ https://www.superssd.com/success/ccpgames.html
I/Os Per Second 400,000
Capacity 32-128 GB
Bandwidth 3000 MB per second
There is one "untested" on ebay at $800, 64GB model was $72K when new in 2007. Today comparable drive is something like "Verbatim 256GB Vi3000 PCIe Gen 3.0 X4 NVMe M.2 2280 Internal SSD" $38.89 with free shipping from Amazon.
> Also, I presume that a SATA interface to store data in RAM might have better performance than the average SSD.
That totally depends on the silicon capabilities and still would be limited by the SATA. Sure, you would have a very low latency on a random IO, but anything with serious throughput (especially if the block size is small) would quickly saturate the bus. You can see it comparing SATA and NVMe drives from the same vendor and model range.
And a personal anecdote - couple of years ago I did played with a software RAM drive (guess I needed something with tons of IO on a moderate sized load?) and the performance was... not good. Sure, it was a software written by some enthusiast and probably wasn't optimized, but still.
Modern SSDs are around $100-150 per TB, so you would get a better size and performance just buying an NVMe drive, with something like M.2 to PCI-E adapter[1] if needed.
PS of course you can just buy an old server mobo and fill it out with your RAM sticks, make a software RAM drive and expose it over Ethernet with iSCSI or FCoE... but again that wouldn't be cheaper than $100-150 NVMe drive.
[0] https://en.wikipedia.org/wiki/RAM_drive, under "Dedicated hardware RAM drives"
https://www.thessdreview.com/daily-news/latest-buzz/allone-c...
... not cheap ...
If there was a (fulfilled) demand than there would be more devices like this, but... by 2012 the problem of a fast storage was solved by throwing loads of SSDs in some SAN (eg 3PAR 7xxx series) and of course enterprise sector preferred volatility and scalability, of which RAM-drives has none. Same with enthusiast market, WD Raptors[4] for a "fast" low/mid-tier, SSD for the fast tier. Just to be sure I checked when OCZ-VERTEX3, which is still running in T440 as a system drive, was anounced - it was ~Feb 2011. In 120Gb and 240Gb versions.
It took some time to dig up the prices (I don't remember nor the prices nor the amount, heck, a decade ago):
> The price of high-end 2GB DDR3 (double data rate, third generation) DRAM modules declined to US$25 in the first half of November, down 46 percent from its peak of US$46.50 in the first half of this year, according to DRAMeXchange. [0] Nov 2010
> G-Skill is offering a value DDR3 kit for $36 USD and Crucial sells their own value RAM for $34.99 over at Newegg. Considering a bit more than 3 years ago (Aprill 22, 2008 via the Way Back Machine’s snapshot of Newegg), a 4GB (2x2GB) kit of G-Skill DDR3 RAM went for $279.99, or about $560 for an equivalent amount of RAM today (8 GB 2×4 GB for $36 versus two 4 GB 2×2 GB kits for $560)! [1] Nov 2011
So even if you were somehow a visionary and would see the price drop a year or two before (ie ~2009)... at 2011 a single 8Gb stick was ~$50, so with $250 you could only get 32Gb (IF the RAM drive card was $50! and it wans't!), compared to $250 for 120Gb of Vertex 3. [2]
Strange visionary who would see a DDR3 price drop, but not SSDs. I paid way less than $250 for that drive, because I bought it somewhere in 2012-2013:
> Just ordered a OCZ Vertex 3 120GB after shipping, taxes and Rebate came to $87.00, ... [4] Apr 2012
[0] https://www.computerworld.com/article/2752016/dram-prices-to...
[1] https://pcper.com/2011/11/ddr3-ram-now-at-all-time-low-price...
[2] https://www.storagereview.com/review/ocz-vertex-3-review-240...
[3] https://hardforum.com/threads/ocz-vertex-3-120-any-good-87-n...
[4] it's quite telling what the last models were announced in 2012 and then quietly disappeared a mere year later: https://forums.tomshardware.com/threads/1tb-velociraptors-di...
I'd use something like this for heavy disk I/O over an SSD any day. Plus, it lets you use memory that would otherwise likely go to waste.
However, those were expensive back then and haven't gotten any cheaper, nor are there newer devices that I know about that'd take newer DIMM types. That'd be nice, though.
Had to very sadly retire it when Windows 10 came along because their disk requirements increased.
Now that I think about it, I should dig it out and load a lightweight Linux on it.
Project for next weekend fixed!
The only potential advantage I see would be read and write speed, but aren't SSD's already fast enough to saturate the SATA bus ? Isn't that why NVME offers better performance than SATA ?
What is clear is that for equivalent capacity DRAM is much more expensive than an SSD.
My guess is that if no one is making a product like what you are asking about it's because it offers no real advantage over the much cheaper solution of adding another SSD.
Some SATA SSDs have a cached mode where data gets put into normal RAM first, and then copied onto the SSD, but it's not really useful other than getting overinflated benchmarking numbers (things get messy if your system loses power before the data can transfer from RAM to the disk). Samsung calls it RAPID, the Crucial name is Momentum Cache
https://www.scorptec.com.au/product/hard-drives-&-ssds/solid...
That's a SATA drive rated for 5,600 TBW, and it's "maximum" rated write speed is 540MB/s.
To work out how long the drive should last at that maximum speed, let's do:
540MB/s * 60 (seconds) * 60 (minutes) = 1,9444,000 MB per hour
Let's call it 2 TB per hour for simplicity.We take the rated endurance of 5,600 TBW, and divide it by the above 2 TB/h, giving us "2,800" hours worth of write endurance.
At 24 hours in a day, that's 116 2/3 days worth of endurance at the maximum write speed the whole time.
Anyway, if the project in mind is going to run for less than ~120 days, it should be ok. Any other hardware/software issues aside that is.
A niche market, but I think there would be a market for ram drives like that, especially if they supported ECC memory. Lots of old server memory kicking around but if you want to actually use it your stuck with some abysmal CPU performance and a very high CPU power draw.
However, there reasonable reasons for wanting a DRAM-Drive.
What would those be ?
> how does it fare with regards to performance?
What's your use case?
https://www.newegg.com/amp/gigabyte-gc-ramdisk-others/p/N82E...
The maximum bandwidth of a SATA interface is 6Gb/sec.
For comparison, old PC-1600 DDR RAM has a data rate of 1.6GB/sec...or >10Gb/sec because 1Gb != 1GB.
Over SATA, static RAM is good enough to saturate the interface and is non-volatile and cheaper.
Good luck.
The best thing to do would just make a RAMDrive, but x86 systems are surprisingly craptastic with real world memory bandwidth and latency and I haven't found this solution to be mind-blowing.
On ARM it might be a different story.
There exist old slow PCI (DDR1 era) cards that do exactly what you expect: https://www.youtube.com/watch?v=bYbCYgYZVT8
There are also some niche contemporary options: https://ddramdisk.store/2022/05/12/meet-the-latest-release/
And in the future there will be very fast options: https://www.anandtech.com/show/16670/using-a-pcie-slot-to-in...
I'm not trying to discourage you or anything, but I think paying $50 for an adapter to run 500gb of RAM slower than a $50 500gb NVMe drive is a bit of a goose chase. Good luck in your search regardless!
It's an interesting question, but I think the circumstances explain why nobody has really commercialized something like this. NVMe will get you 90% of the way for 30% of the price, and it's non-volatile to boot.
I'm going to go out on a limb and say the approach you're considering is practically infeasible OR (best case) would need an oddball/bespoke device to pull this off. I'm not aware of such a device.
Closest you could get is using tmpfs or similar on RAM in RAM slots but it kind of sounds like your existing RAM is electrically/physically incompatible with the system you're attempting this approach with.
It's perfectly fine for some uses, mine included.
> I'm going to go out on a limb and say the approach you're considering is practically infeasible.
To each his own, but I know for a fact that hardware RAM drives were indeed sold in the past, and after posting this Ask HN I've stumbled upon companies marketing them.
https://ddramdisk.store/2023/01/19/the-ddr4-pcie-x8-lady-has...
Yes! I saw that device from elsewhere in the thread. Wondering when it will be available and what use will look like. Also curious about the battery backup and what maintenance looks like on that.
RAM storage has been tried before but was done as PCIe expansion card. It didn’t catch on cause of special drivers, but it only made sense back before NVMe and SSDs were expensive.
SATA is the computer bus interface, while SSD is the generic name for a storage device that more often than not comes with SATA interfaces.
SATA was not a hard requirement, just the one that I felt was the most popular.
> Also, SSDs are much larger and cheaper than RAM.
SSDs are not cheaper per GB if you already have RAM sticks lying around. Nevertheless this is besides the point, if the goal is to have low latency temp storage when you already maxed out on RAM.
The PCIe RAM drive cards were enterprise and super expensive. You could buy a whole computer for one. It is probably cheaper to sell the RAM sticks and buy SSD. As I mentioned, they are super cheap now and cache would only need small one.
Also, setting up RAM drive is a pain. The old cards required special drivers; new ones could probably emulate NVMe. But setting up cache drive and getting programs to use it for temporary storage is annoying. All SSD is easier. There is a lot of experience with using SSD as durable write cache and not as much using disposable temp space.
For those who need speed, buying more RAM is a better way of doing it. If you really need speed, then new computer with more RAM and NVMe is better approach.
In any case, PCIe RAM drives aren’t made anymore. They were a niche at the time and now there is no market at all.
1. Someone already has maxed out their motherboard and has some unused sticks of ram and would like to repurpose them instead of selling them or worse recycling them (remember reuse is before recycle). Reuse as battery powered storage might be an option. (This is the use case of the OP)
2. Someone wants a faux-Optane cache, maybe for the paging file/swap. Maybe to avoid wear on the SSD. Theoretically it should be possible to get higher performance than a SSD (same era RAM is faster than Optane which is faster than same era SSDs), in practice IDK. Maybe Optane is no longer available, or again maybe they already have the RAM
2. Someone wants a Volatile drive, maybe to be used in conjunction with a security oriented Linux distro like QubesOS or Tails. If a switch is tripped, everything on the computer is truly gone.
Elsewhere in the thread is a mention of gigabyte ramdisk, but what would be really interesting is a second bank you could have on a PCI card that could hold 8 sticks of RAM, and that RAM is considered lower-priority by the OS (use only if the main RAM is getting a bit constrained, or say "use this with the web browser".
But I doubt such features exist even in the Linux kernel.
Eventually at least. Ram loses its contents slowly, and temperature influences the rate it happens at, hence cold boot attacks where you can super-chill ram with some compressed air, move it out of a secured machine and into one you control, and extract in memory secrets.
2. Optane is available though, for less than a dollar per gigabyte. And if you have RAM already, you can sell it as above.
3. Encrypt it with a random in-memory key.
2. Intel abandoned Optane, so availability will be spotty
3. Encryption is not always plausibly deniable. And when it isn't, rubber hose cryptanalysis can be applied even if the key is unrecoverable. Data on a RAM drive, if it was turned off and the attacker didn't immediately freeze it (harder when they don't even know you use a RAM drive, and the case may be hard to open) should be obvious it is gone for any attacker who knows how RAM works. The point is less to protect the data and more to protect the owner from coercion through violence.
And a rubber hose can be applied to "give me the backups" no matter what you do, so I don't know how convincing that is. And if making a new key is automatic on boot I don't know if the additional risk of them not believing you is that high.
I appreciate the idea as an interesting hacker experiment but it has no practical value.
I have something myself that is in the ballpark. It's e-ink Thinkpad that has 4GB of soldered memory. It would be plenty fast if only it had more memory. They don't make a newer e-ink laptop like this. It's not SATA but if there was a way to stick a 32GB DDR4 memory stick onto an NVME drive, use it in the second slot as a pure swapfile, that would be a cool little trick to extend the lifespan of this thing.
I might try this with Optane if the prices come down more.