SSD vs RAMDrive

[MrJinje]

Wanted to see which is faster a RAMDrive or an SSD. Since I don't own a SSD yet, (leaning towards OCZ Vertex) I will post the results of Qsoft Ramdrive, one of an older SSD, and another for a 2 disk RAID Stripe. If anyone has a newer SSD, preferably with firmware updates and W7, they should post the speeds of each brand/model.

Decided to install W7 into a RAM based VM and it turns out it is more responsive than the Host OS.

Had to use allocate all disk space now setting or VMware will error claiming not enough room, even when there is.

To show over-head associated with VMware's Hyper-Visor, here is the direct access benchmark from Windows 7 x64 on the same RAMDrive.

Qsoft RAMDrive - Direct Access Benchmarks. 8GB DDR2 RAM

Sequential Read : 2005.759 MB/s

Sequential Write : 2898.645 MB/s

Random Read 512KB : 1793.318 MB/s

Random Write 512KB : 2817.644 MB/s

Random Read 4KB : 771.885 MB/s

Random Write 4KB : 539.490 MB/s

Gigabyte I-RAM 4GB DDR SSD - For comparison.

Sequential Read : 136.622 MB/s

Sequential Write : 110.078 MB/s

Random Read 512KB : 135.524 MB/s

Random Write 512KB : 109.589 MB/s

Random Read 4KB : 48.650 MB/s

Random Write 4KB : 41.430 MB/s

2 Disk RAID Stripe Sata - While Running Windows 7

Sequential Read : 74.621 MB/s

Sequential Write : 66.418 MB/s

Random Read 512KB : 24.852 MB/s

Random Write 512KB : 44.955 MB/s

Random Read 4KB : 0.349 MB/s

Random Write 4KB : 4.022 MB/s

Use Crystal Disk Mark with default settings.

[jaclaz]

Just for the record, latest version of Firadisk has also bettered noticeably it's speed:

http://www.boot-land.net/forums/index.php?...8804&st=199

though cannot say if HDTACH results can be compared to Crystal Disk Mark ones

jaclaz

[dencorso]

Here's a true SSD:

RunCore SSD 128 GB (mini-PCIe SATA) on Asus EeePC 900 (Win XP SP3) FAT-32

--------------------------------------------------

CrystalDiskMark 2.2 © 2007-2008 hiyohiyo

Crystal Dew World : http://crystalmark.info/

--------------------------------------------------

Sequential Read * : 126.5 MB/s

Sequential Write * : 81.3 MB/s

Random Read 512KB : 120.7 MB/s

Random Write 512KB : 63.2 MB/s

Random Read 4KB : 16.3 MB/s

Random Write 4KB : 2.0 MB/s

*Test Size : 100 MB

And, below, I quote some other older data I obtained with Gavotte Rramdrive and the Gigabyte i-Ram, on my main desktop (overclocked Athlon XP-M 2800+ @2415 MHz on an ASUS A7V600-X with 3 GiB of DDR RAM @256 MHz):

Gigabyte i-RAM hardware ramdisk vs. Gavotte's software-only NT-only RRAMDISK

GB i-RAM 1.5GiB (Win XP SP3) FAT-32

------------------------------------------------------------------

CrystalDiskMark 2.2 © 2007-2008 hiyohiyo

Crystal Dew World : http://crystalmark.info/

------------------------------------------------------------------

Sequential Read * : 132.9 MB/s

Sequential Write * : 126.2 MB/s

Random Read 512KB : 132.9 MB/s

Random Write 512KB : 125.8 MB/s

Random Read 4KB : 58.6 MB/s

Random Write 4KB : 51.7 MB/s

*Test Size : 100 MB

Obs: The GB i-RAM is a SATA-I device, so its theoretic

maximum allowable data transfer is 150.0 MB/s...

=====================================

Gavotte's RRAMDISK 1.5GiB (Win XP SP3) FAT-32

------------------------------------------------------------------

CrystalDiskMark 2.2 © 2007-2008 hiyohiyo

Crystal Dew World : http://crystalmark.info/

------------------------------------------------------------------

Sequential Read * : 308.4 MB/s

Sequential Write * : 315.6 MB/s

Random Read 512KB : 270.7 MB/s

Random Write 512KB : 276.1 MB/s

Random Read 4KB : 36.1 MB/s

Random Write 4KB : 34.9 MB/s

*Test Size : 100 MB

Obs: 1)Gavotte's RRAMDISK.SYS is a WDM device that

only exists after Win XP has fully initialized.

2) RRAMDISK.SYS v. 1.0.4096.5_200811130 was used for this test.

As you can see, both our results on Gigabyte i-RAM compare well.

The direct comparison between ramdrives is not fair, however, because mine is on underclocked DDR RAM, while yours is on full-tilt DDR II RAM and presumably on a much faster bus...

But, anyway, on every hardware, the ramdrive wins, as expected.

[MrJinje]

According to this, your SSD is actually making the announced SPEC (125MB/90MB) Which is rare from what I understand.

DATA TRANSFER RATES

Sustain Read Speed up to 125MB/s

Sustain Write Speed up to 95MB/s

Here's a true SSD:

RunCore SSD 128 GB (mini-PCIe SATA) on Asus EeePC 900 (Win XP SP3) FAT-32

Sequential Read * : 126.5 MB/s

Sequential Write * : 81.3 MB/s

Random Read 512KB : 120.7 MB/s

Random Write 512KB : 63.2 MB/s

Random Read 4KB : 16.3 MB/s

Random Write 4KB : 2.0 MB/s

Mainly I wanted to find out if OCZ Vertex or any other SSD that claims of 200MB read speed can actually back it up.

Rumored spec on OCZ Vertex from an online retailer. But the other rumor is that you will never achieve this in real world uses. Can anyone prove or disclaim this ?

Read Speed Up to 200MBps

Write Speed Up to 160MBps

[dencorso]

According to this, your SSD is actually making the announced SPEC (125MB/90MB) Which is rare from what I understand.

Well, it's a RunCore...

Consider it's SATA I (the Eee PC 900 southbridge is ICH6-M), so it performs very near the max. 150 MB/s allowed by the interface (and performs almost on a par with the Gigabyte i-RAM, which also is SATA I, but uses DDR SDRAM, and hence is a good indicator of what the realistic maximum for SATA I truly is)!

[CoffeeFiend]

the ramdrive wins, as expected.

No kidding. Decent DDR2 @ 800MHz has a memory bandwidth of 6400 MB/s (DDR3 @ 2200MHz would be an impressive 17600MB/s). Even with bottlenecks elsewhere, no SSD on earth can touch this (even in RAID0 on a unlimited budget you'd have a hard time even approaching that).

Which is rare from what I understand

It very much depends. As they get more usage, SSDs do slow down (as pages fill and it has to erase them before writing again basically). There was a good explanation here.

This RunCore SSD doesn't seem to support TRIM (although the Indilinx controller could probably do it, with the right firmware and all) so it will likely slow down over time (then again, some firmware versions seem to do some maintenance -- free space consolidation and erase in its idle time, so it may not be too bad still). Then again, XP doesn't support it either.

[MrJinje]

I got a dumb question, can anyone tell me if 3GB SATA is limited to 300MB on the entire BUS, or is it 300MB per channel. If it is the entire bus, then it would be a waste of time to try and RAID multiple SSD's.

Otherwise, if it is 300MB for each port, couldn't a four disk Vertex stripe reach 800MB, at least in theory.

@Coffee, was planning on SSD/Windows 7 specifically for the TRIM command. Does anyone know of a retail SSD that is confirmed pre-loaded with firmware updates. Really hate having spending money on crap that doesn't work fresh out the box.

What's your take on Multi-Level vs. single level. Is that just fanboy talk, most the reports seem contradictory to each other and I cannot tell if it is another bluray vs HDDVD mudslinging or not.

Edited December 15, 2009 by MrJinje

[CoffeeFiend]

I got a dumb question, can anyone tell me if 3GB SATA is limited to 300MB on the entire BUS, or is it 300MB per channel

SATA channels aren't shared like IDE. And yes, you can RAID them for extreme performance. Your limitations will be that of your controller itself (e.g. ICH9R) and various interconnects on its path (ICH to MCH, FSB, PCI-e lanes, etc) and so on. I very much doubt you'll manage 800MB/s, likely more like half that (guesstimating, not knowing anything about your hardware). If you want extreme speeds, plan on having to spend serious coin on a high end RAID card (and of course have the x8 slot for it on your mobo too)

SLC vs MLC... That's just one factor. Even the absolute best SLC NAND flash when paired with a poor controller and firmware will perform poorly. Either ways, even MLC is still WAY too expensive for me (and I don't want to settle for a tiny drive either). And SLC pricing on the better SSDs like the Intel X25-E Extreme is just unreal (close to $15/GB, when plain old magnetic storage is about $0.08/GB). I don't foresee buying a SSD in 2010 (I need space, NOT speed -- I'd be more interested in even cheaper storage)

@dencorso: I'm not totally sure these benchmarks are perfectly valid. There's the whole dirty blocks thing again... Benchmarking writes to NAND flash is tricky, unless all the blocks are erased first.

[dencorso]

Here's a real-word comparison of SLC vs. MLC using pendrives, and with the same (or almost) controller to make things easier. Bear in mind that the attached table was created with the objective of comparing pendrives, not of comparing SLC vs. MLC, so the Kingston pendrive is there just to show your average el-cheapo performance, as a reference. The Corsair Flash Voyager GT 8 GB and the OCZ ATV Turbo 8 GB are SLC, while the the non-GT Corsair Flash Voyager 8 GB and the Kingston are MLC. As a side note, only the Corsair Flash Voyager GTs of 8 GB or less, all dicontinued by the manufacturer at present, are SLC, while the 16 - 128 GB Corsair Flash Voyager GTs are MLCs, albeit very fast ones at that. So, the OCZ ATV Turbo 8 GB is the only commercially available high-performance SLC pendrive at the moment, AFAIK. Then again, good MLCs perform very near the performance of SLCs, or, at least, that's how I interpret the results of my test in the attached .pdf, so this issue far from being such a big deal as it's advertised.

@CoffeeFiend: all four pendrives were brand-new, except that I repartitioned them to have a single active primary 4th partition (Zip100 standard) and reformatted them with FAT-32, using 4 kiB clusters (half the MS recommended value), so they have about 2 million clusters, each.

Pen_Drive_Performance_Tests.pdf

[MrJinje]

Settled for the OCZ Vertex 30GB ($109). First thing I did was run benchmarks on the fresh drive. Here are the results. No where near the stated but good enough to boot from.

--------------------------------------------------

Sequential Read : 159.698 MB/s

Sequential Write : 93.776 MB/s

Random Read 512KB : 128.192 MB/s

Random Write 512KB : 93.281 MB/s

Random Read 4KB : 24.811 MB/s

Random Write 4KB : 11.684 MB/s

Can anyone tell me how to determine if my SSD already supports the TRIM command or if I need a firmware update.

Edited December 19, 2009 by MrJinje

[CoffeeFiend]

That's pretty good performance. I wouldn't know what to do with a tiny 30GB though Windows alone (including pagefile.sys, hiberfil.sys and temp files) will eat right through that, all by itself (not even counting large user profiles, the windows installer cache, the client side cache, system restore points, the recycle bin or any of that stuff). A pair of the 120GB (or a single 250GB) OCZ Vertex would be usable (including some apps), but that's like $900 right there Considering twice that much storage on plain old hard drives is worth $50... For $900, I'd be getting five 2TB drives instead (10TB total, or about 40x more space). Yep, looks like I'll be waiting a while longer.

Can anyone tell me how to determine if my SSD already supports the TRIM command or if I need a firmware update.

The Vertex has full TRIM support as of firmware v1.4, which was released a while ago. It should already have it but you can always check.

[MrJinje]

LOL, my vLited install only takes up 3.66 GB, considering I run Windows 7 from a 4GB I-RAM this 30GB will be a major improvement.

Have fun waiting for your HDD's to spin up, I don't really have .09 thousandths of a second to wait each time I click on my start menu. Some of us just prefer to have a computer doesn't act all slow and old.

I do have about 5TB of HDD space laying around, so this is more about responsiveness of the OS.

Edited December 19, 2009 by MrJinje

[CoffeeFiend]

Have fun waiting for your HDD's to spin up, I don't really have

Drives spinning up is the least of my worries

.09 thousandths of a second to wait each time I click on my start menu

So the menu is still loaded before I've released the start key or mouse button. Hardly a problem (and it's not like it doesn't do any caching either). Nevermind 99% of my time wasted there is looking for the program I want and not for the start menu to respond. And Vista/Win7 fixed that with the search.

Some of us just prefer to have a computer doesn't act all slow and old.

It hardly feels slow, and if $900 is what it takes to make it faster, then it'll wait Nevermind, I'd be due for more storage, a faster CPU (which also basically means a new motherboard and RAM) and a bunch of other things first, so not happening anytime soon

[Octopuss]

I will add my question here rather than starting a new thread.

How bad is the wear-over-time "feature" of SSDs? I was looking at Intel (various articles state that it's the only brand worth buying), specifically "Intel SSD High Performance X25-E SATA II (SLC)", but it's quite expensive (for me), so I need to be sure about it.

Does anyone have any experience with these? Especially the speed in the problematic situations where SSDs generally are slow etc.

[MrJinje]

Wear leveling has improved the situation and modern specs claim it will take me at least 30 years of above average use before my crappy little 30GB burns out. (copy/delete 90GB per day (3 full write cycles) * 10,000 days). No chance of that happening.

TRIM command helps with the speed issues (or lack thereof) when writing to previously deleted sections. If the OS is aware, it can tell a newer SSD to reclaim space from deleted data, improving the speed next time it writes to that section.

As for the articles you read, take it with a grain of salt, SSD's manufacturers seem to be in the dis-information stage. Remember when HDDVD was so much better than BluRAY. That all stopped once Bluray's price point (free with PS3) become a contributing factor. After my studies I decided on the Indilinx controller, which is used in some of the newer SSD's (namely Runcore IV, OCZ Vertex, and Patriot Torqx).

@Coffee - Considering I paid $200 for 8GB of DDR a few years back, getting another 30GB of RAM for $109 is a steal. Not sure where the $900 figure is coming from, but for that kind of money I could get 4 SSD's and an SAS/SATA PCIe card.

Edited December 22, 2009 by MrJinje