The new SSD Thread (Benchmarks, Brands, News and Advice)

Yeah, that.

 

My notebook's current spinning 250 HD has these gold pins that literally stick out sort of like a comb.
Something like this but more:

I want to upgrade to a intel SSD but I notice that their gold pin slots are different. They are flat.

How will I upgrade to the intel SSD?

 

first: that pic is funny of the hdd. i guess you don't have your pins on the side of the disk

second: your disk will be some sort of pata drive (parallel ata). all new drives (and all new notebooks and pcs) have sata today (serial ata). that's the reason for the different connector. you would need a converter (possible in pc's as you have enough space there). in a notebook, no chance really.

 

So, I want to get two of those 128 GB Samsung MLC's from Dell to run in S/W RAID-0 on my computer (remember I asked a page or two ago about Dell SSD's). I'll be wanting to run this on XP Pro x64 (currently running x86, but want to try out x64 finally). IF I want to partition the drive (windows here, storage/games there, etc.), will it work just the same as partitioning a spinner? And I assume it doesn't matter in which order I install things (as in: ideally one would install big files/drivers/games sooner than later because the inside of the platter is faster access than the outside for the read head)?

Now I just have to refind the page in this thread about diskpart instructions for the RAID-0. I assume this applies just the same to a software-based RAID-0...

 

no clue how good s/w raid0 is. it will for sure eat a bit of cpu.

my hint: don't even bother about xp 64bit. if you want to go 64bit, go vista. if you think it's slow: try it. at least with an ssd it's very very very fast.

and i wouldn't partition the disk really, it's of no use performance wise, and you should back up your data externally anyways.



edit: are those 128gb ones the ones with 200MB/s readwrite? else, you can get a 256gb one just as well and still have one disk-slot free for another 256gb later

 

The outside of a platter is faster because it is covering more area in the same amount of time to complete a 360 rotation. Think of how a record spins... Like davepermen said there is no point in partitioning ssd because there is no "faster" part of an ssd. Also listen to him and get vista. Welcome to the year 2009...

 

That is very interesting indeed to hear such a detailed review on SSDs, Nice job my friend I am very sure this techonology is very promising..

 

There are a few conventionally sized PATA solid state drives on the market... http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2013240636 1421530855&name=PATA. The problem is that most of these drives are very slow or JMicron controlled, aka junk.

If you really want to go down the PATA route, you can try looking for an early Samsung 1.8" PATA SSD or an Mtron PATA ZIF SSD. Both will need adapters as Dave mentioned, but since they are smaller than a conventional 2.5" drive, you might have room to cram the adapter into your drive bay.

 

from mtron, there are pata 2.5" disks. but they are not that cheap..

and some of the samsung pata ssds where very bad. i have one. terrible

 

Oh man, that's disappointing. Who'da thunk. What was the model? Do you have *gasp* benchmarks!? I suppose I'll be sure to stay away from those...

 

no benchmarks needed. i can just say "stutter"?

terrible.

i'll check the numbers at home again. well, the model number. it won't work anymore without adapter anyways 600$ spend for crap.. i know why i stay with mtron and slc

 

I don't know how many posts you're going to regurgitate the same information about BACKING YOUR STUFF UP, but I get it. My question had nothing to do with backing anything up. And partitioning has nothing to do with performance gains, but rather security and organization. I was wondering if partitioning is as easy as with an HDD, not whether davepermen LIKED partitioning. All I wanted to know was whether there were extra steps I should look for when partitioning an SSD - regardless of how much you don't 'like' it.

http://partition.radified.com/

Read that if you want to bash my yearning for partitions. My favorite article ever.


I mean really man, I usually respect your knowledge to a high degree, but when I ask a simple question like that you jump down my throat (as you have before almost verbatim this subject with a different question) with OH BUY VISTA IT'S 2009 and backup your stuff!!!11oneone Did you think that maybe I have XP Pro x64 and would like to try that? I've read that SSD's can run just fine on XP if the partitioning alignment is set from the start to what it should be. I have used Vista enough and I'm not interested, especially with the advent of W7 on the horizon.

 

Partitioning an SSD should be no different than partitioning an HDD. As mentioned numerous times in the past (I think?), as of now there should be basically no difference between an SSD and HDD as recognized by an OS.

And if you have a good SSD, it'll run fine with XP even without partition alignment. I have XP Pro 32bit on both of my solid state drives, no tweaks.

 

Thanks wolf. That's the response I was seeking

 

On my system I have 2 partions on the main drive and one in the Ultra Bay. I haven't done any alignment on both my SSD's and they seem to working perfectly.

 

Haven't been keeping up with the thread as much as I'd like, but I decided to install steadystate on my laptop and it works great, though the "reboot, commit changes to drive, reboot again" cycle is slightly annoying, I wish it would just commit the changes when I log out (as I type this there probably is a setting for that and I just haven't gotten around to finding it). It eliminated all of the pauses and jerkiness I had with the Super Talent MX.

 

I'm surprised that a search reveals not one mention of SanDisk's proposed industry standard drive speed metric of "vRPM", or virtual rotations per minute. When I first read this acronym back in January I immediately dismissed it as marketing fluff invented to make their upcoming SSD sound better. However, after finally reading SanDisk's white note today on how to calculate vRPM, the measure appears to be very useful. The experts on here should read and critique it.

Quick summary: vRPM is a metric of a drive's IOPS that can be used to compare HDD IO performance to SSD IO performance as well as allow comparison between different SSDs' IO performance. This metric is limited to accounting for small (4k) random read/write performance. It therefore ignores all other measures of drive performance such as sequential reads/writes or IO performance at other block sizes. vRPM is calculated by a simple formula using data obtained from a benchmarking tool such as IOmeter (queue depth = 4).

Here are my simple thoughts. vRPM looks like a very useful way of comparing HDDs to SSDs and different models of SSDs to each other in terms of small random IOPS which, as we know, have an enormous effect on drive performance. This seems significant to me because currently there is no attempt in the SSD industry to represent a drive's small random IOPs in the published specs. This makes comparing SSDs based on published specs futile. vRPM would rectify this. The lack of a figure like vRPM is the reason JMicron drives looked so great on paper but caused much confusion as an explanation of stuttering was sought. Since vRPM is solely a measure of small random IOPS it does not attempt to represent bandwidth whatsoever. That's just fine - the only spec given by SSD manufacturers right now is precisely that (max sequential bandwidth). Adding a published vRPM spec to this would allow the average joe to compare apples to apples (SSD to SSD) as well as apples to oranges (SSD to HDD).

After reading the white note, I wonder if SanDisk devised vRPM partly to highlight how much better their upcoming G3 drive is than the previous JMicron drives (the data further below highlights this well). If in fact SanDisk can market the vRPM spec then it is all but certain that the SanDisk G3, with an identical pricepoint to the JMicron drives, will be the demise of JMicron.

With that said, it appears that SanDisk made some faulty predictions on the state of the SSD market in 2009. I think they were anticipating releasing their G3 SSD with only JMicron drives to compete with at the same pricepoint. They probably didn't plan on the Intel's price dropping so much (which takes away the enormous price advantage that they would have had) nor is it likely that they anticipated the entrance of the Indilix drives appearing at a comparable pricepoint yet significantly higher level of performance. These assumptions are supported when calculating vRPM values for all of these drives. Although the G3 is clearly better than the JMicron products, it is far behind the Vertex and not even comparable to the Intel. The price differential is too small between the G3 and the Vertex to choose the G3 based on price alone - and this doesn't account for the inevitable decline in the Vertex's price before the G3 will be released in "mid 2009."

Based on my calculations (using IOmeter data from my own solid, Anand's article, and Sandisk's website, here's how things stack up:

Code:

Drive		Size		  vRPM		Price
OCZ Solid	60		    600		$160
7ms seek HDD			  7,200
Sandisk G3	60		 40,000		$149
OCZ Vertex 	60		 73,000		$210
Intel X-25M 	80		414,000		$363

What are your thoughts on using the vRPM metric as a matter of course when benchmarking SSDs?

Should we place pressure on SSD manufacturers to use and publish this spec?

 

You answered your post right there, unfortunately. Manufacturers releasing JMicron drives will fight to the death anything that exposes their drives as falling flat on their face on the one test that really matters for day to day drive operation. I'm for it, but I highly doubt it'll happen.

BTW, what settings are you using in IOMeter to run the 4k test like in Anandtech's article? I'd like to duplicate it for my own information in IOMeter but I don't know how.

 

Since JMicron drives are only sold direct to consumers (not OEMs) I think it is safe to say that they make up a tiny fraction of the SSD market. The big players - the ones that actually produce the hardware that they market - are the ones that will set the precedent and decide if vRPM is used as a marketable drive metric or not. So basically that's just Intel and Samsung. They have nothing to lose in adopting vRPM as a way to communicate to customers since their drives clearly outperform HDDs by this measure. If they make it standard practise to report vRPM, the smaller players will be forced to follow suit. However, a company like OCZ could be pressured into using the spec since they have such a clear and direct line of communication with their customers and also have a product that shines when it comes to vRPM (vertex).

Anandtech actually doesn't follow the procedure outlined by SanDisk in order to properly calculate vRPM but it is close enough for the purposes of the comparison I posted above.

Anand specifies how he runs his IOMeter tests: 4K 100% random read or write, queue depth = 3, 8GB max disk size for 3 minutes.

SanDisk's white note specifies that to calculate vRPM (which requires only IOPS for both read and write) you should test 4K 100% random write and 8K 100% random read, queue depth = 4. They do not specify the max disk size or how long the test should be run. I'm not sure what difference the max disk size makes but if SanDisk doesn't specify then maybe it isn't important in measuring IOPS. I believe the time limit is to get an idea of what the max latency spec will be in a "realistic" useage scenario. Personally I don't think max latency is a very useful indicator of performance since by definition this is an outlier.

Using these settings my 60GB solid series gets 6 IOPS for 4K random write and 3900 IOPS for 4k random reads. This equates to a vRPM of about 600. Pretty awful. Nevertheless, overall system performance with my SSD is still far better than my old 5400rpm HDD. This demonstrates how vRPM (i.e., IOPS) is only part of the story, just like max reads/writes are only part of the story (yet have been used in marketing data to define a drive's performance by SSD manufacturers).

 

vRPM sounds silly to me. The first thing they need to do is change the name.

The next thing they need to do is come out with a benchmark that tells the true story. If a JMicron drive that gets 600 vRPM is much better than a 5400 RPM HD in typical use then something is wrong with the measurement.

 

Wow, this vRPM stuff sounds interesting. It's a bit like the 'x' factor used for CD/DVD spinning. Ultimately, everything is based on 1x which is what, 150KBps, and 1x DVD is what, 9 times that? So a 16x DVD is like a 144x CD but still based on the same metric. By the time we got around to using DVD+RWs, we lost sight of what the heck that '16x' refers to.

Anyway, it sounds like, vRPM *should* bring some sanity to the nascent SSD market.

 

I'm disappointed in the G3. I thought it would be amazing. Seems kind of strange that SanDisk would be pushing this metric when it puts their new drive miles behind the industry leader (Intel) and substantially behind the Vertex.

Then there's the fact that this metric depends on IOPS (if I interpreted the calculations properly) and thus a standardized method of measuring IOPS is needed before an apples to apples vRPM comparison can be made. But then if a standardized means of measuring IOPS is developed/defined, I feel like that figure alone is sufficient to do an apples to apples comparison between various drives (again, if I interpreted the calculations properly). Finally, since vRPM and IOPS doesn't really consider sequential speeds (yes, they factor into IO performance), I feel like they still need to be used in conjunction with rated sequential speeds to offer a complete view of drive performance.

At any rate, as long as everyone follows the same means by which to determine IOPS, I don't really think there's a need for vRPM; it's just making things more complicated.

 

the only problem comes from combining read and write onto the vRPM figure - while a HDD read and write are generally the same, things are distinctly different for SSDs. if they had a read vRPM and write vRPM then things would be much clearer.

e.g. in the ocz solid case, its read performance is probably (say) half as slow as intels, but its write performance is thousands of times slower. so then consumers, knowing both read and write performance, can make the best judgement based on the intended usage. e.g. a read-only database or similar youd be crazy to waste money on an intel compared to a jmicron cheapie, but for a 50/50 split then you definitely want to go for the drive with good write performance.

 

security and organisation.. hm.. want to see that espencially the security part

you have no security gain at all, as it's still just one disk.

organisation gain: what do you gain over folders? with folders, each folder can scale to the full disk size if needed. with partitions, it can't.

imagine having a d: for games, a e: for movies. what if you suddenly note you need more for the games? or more for the movies? you just lose storage space, having say only 1gb on each of 5 partitions free is not equal to having 5gb free on one partition. in one case, i could rip a dvd on it, in the other, i couldn't.

it wasn't ment as an insult, just as a why? as there is no gain having more than one partition except for one thing: performance and defragmentation gains. and that mostly only in xp as it can't handle it itself that well. and those two gains are of no use on an ssd or in vista.

i've read trough and it is completely useless afai can see. a big bla and nothing that makes sense. the analogy to the house is flawed. that's what folders are for.

the only point that fits is: partitioning for multiple os'. and there, i prefer to use virtualbox or so anyways as i never dual boot (but then, partitioning would be way to go).

I don't want to jump down my throat (and i try to visually imagine that right now ). I just want to question your plan to give you some new ideas. I can see a lot of "I've learned it that way back when i started in 1995, I still have to do this today as it's best." And this is the wrongest wrong in the computing industry, the fastest evolving and changing industry ever.

So what I really mean is, what do you gain from your partitions except that good feeling of doing the same thing you did once as you learned it should be the best thing (a.k.a. sheeping).

yep, and I saw enough that thought that way until they tried vista 64.

the problem is, xp 64 is mostly unsupported and dead before it really got alive. do you really want to mess with an os that close to no one uses? that close to no one supports? do you want to have that burden for nothing? just because you want to get rid of vista?

all i say is try vista as well. it will perform very very very very well and be much more compatible and stable and secure. and, once you get used to it's daily business workflows, you will even start to like it's enhanced usability. change is good..

w7 is vista. and it's not on the horizon. it's months to a year away. you can get w7 right now by installing vista. it performs identical on an ssd, and is mostly identical to use. if you don't like vista, why should you think you like win7? again? you want to think, or just be a sheep?

this is not an attack. this is an idea to think over what you plan to do. it may not be the best way. but we should quit this here. you can contact me personally if you like. i'd like to.

 

Well, I decided to pull the trigger on the 64GB 1.8" from newegg (took my own advice!) I will stick it in my E4200, hopefully, without any modification (I am worried about the holes lining up on the card).

The 128GB MLC will become my spare 'handy drive'.

 

The point of using vRPM as a metric is to keep it familiar. It is not a metric intended to speak to the uber nerds but rather the average Joe consumers that needa a single simple number that makes sense. RPM already makes sense to these consumers. vRPm is just an extension.

In terms of telling the "true story": as I mentioned, vRPM is one of two major measures of gauging performance. The other is the (already published) max bandwidth speeds. In the case of the solid series, a vRPM indicates of 600 indicates that its IOPS are much worse than a standard HDD. This is absolutely true and this data needs to be obvious. The vRPM score of 600 makes it obvious.

Yeah, that's kind of what I was saying. I think that the only option SanDisk has is to release the G3 substantially cheaper than the Indilix based drives. Say, $100 for 60GB (the same price that I got my JMicron drive for). Then it would definitely be a viable alternative to the Indilix drives; about half the price with performance that kills a HDD and is only slightly behind the more expensive SSD options in real world measures.

SanDisk includes a description of how to measure IOPS (i.e., a standardised method) in their white note. I mentioned the particulars in my previous post.

IOPS alone are a great measure for the uber-nerds but you can be certain that a table of IOPS data will never appear on an SSD's box or on the manufacturer's website. These tests have been done and will continue to be done only by end-users. OTOH, vRPM is a single, simple number that retains familiarity. The advantage, as I've already mentioned, is to inform the average Joes just how good an SSD is and how it will compare with a HDD. The advantage for the uber-nerds out there is that, if the metric becomes an industry standard, SSD manufacturers will be forced to produce drives that focus on having good IOPS. Consider that if vRPM was used from the beginning, the JMicron drives would never have been released to market. They would have been forced to develop further until IOPS at least equal to a normal HDD were achieved.

I'm not sure if you've understood the white note completely. Anyway, the point of combining read and write into the vRPM metric is to produce a figure that is comparable to HDD RPM. Since HDDs have synchronous read/write speeds, it is important to retain this in developing a comparable metric (vRPM). What you are proposing is to just report read IOPS and write IOPS separately. Although this would communicate important data, it fails to retain familiarity with the average Joe consumer.

Ask average Joe what the 100% random 4KB read IOPS are of a WD5000BEVT. How about 100% random 4KB write IOPS for a WD32000BEKT? Are you imagining a blank face? I'm guessing yes. In fact, I'll bet most people even in this thread, myself included would not be able to answer those questions even while understanding what they are asking. Reporting IOPS is useless for comparing SSDs to HDDs since essentially no consumers know IOPS figures for HDDs.

OTOH, if you ask average Joe how many RPM a good HDD is, he'll tell you 7200. OK Joe, would you like an drive with an equivalent RPM speed of 600 (solid), 5,400 (notebook HDD), 40,000 (G3), or over 400,000 (Intel)? Add max bandwidth speeds to this decision and Joe can make up his mind based on 4 figures, all of which are familiar to him, directly comparable to one another, and noticeable in real-world use: vRPM, max bandwidth, drive capacity, and price.

 

Right. I'm saying that this figure, assuming everyone measures it the same way, is sufficient to determine what vRPM is trying to measure. IOPS is just as simple and singular a number. Manufacturers should just provide Sequential speeds and IOPS, and that'll be that.

I suppose if you want to argue for familiarity, I'll have to support the vRPM spec, but how many consumers actually know the difference between a 5400RPM drive and a 7200RPM drive in the first place?

 

I think everyone should stop worrying about comparing SSD to HDD. There really is no need. SSD>HDD provided there is no stutter aka has high enough IOPS. This makes it completely unnecassary to have vRPM. Manufacturers just need to drill that fact into consumers head once and for all and be done.

Then using a standardized method of producing sequential speeds and IOPS between manufactures would be sufficient.

 

Agreed. vRPM would be a temporary fix for people who are used to HDD and need the numbers as some sort of comparison. But once HDDs start getting phased out, then we are stuck with a completely irrelevant performance metric. There are no RPMs in an SSD!

 

Dumb question since I can't seem to find much reliable info. Everyone says the Super Talent Ultradrive ME is the same as the Vertex (the vertex does have cache memory right?), does it also have the SRAM cache then? Supertalents site makes no mention of it and I can't find any reliable reviews of the thing. The price for the 64 after rebate at newegg is pretty reasonable.

 

Yes, definitely has the cache. Vertex and Ultradrive ME hardware is exactly the same, only differences will be from firmware.

 

I have really appreciated this informative thread! I am trying to choose a SSD for a Thinkpad X61s which as you probably all know is limited in transfer rate to 110 MB/s or thereabouts by motherboard design. For this reason some people recommend passing on the fastest SSDs because the Thinkpad won't benefit from their speed advantages. But to what extent are the more relevant random read/writes speeds affected by the hardware limitation?

So for example I have seen people recommend the Samsung SLC for its price advantage because its transfer speeds pretty much reach the limit of the Thinkpad, but this drive is $175 and requires an adapter while the Vertex 60GB is $210, and wouldn't the Vertex still be faster in most or all applications?

 

@jkellar - I'd go with a Corsair S128.

 

Just stumbled upon this interview with an Intel SSD engineer from January, don't remember it being posted here, so I thought I'd share in case anyone else hadn't seen it:

http://www.hardocp.com/article.html?art=MTYxMSwsLGhlbnRodXNpYXN0

 

I agree that providing separate read/write IOPS would be great for detail. But manufacturers won't do it. The proof is in the pudding: they easily could be publishing those specs right now, but they're not. They've decided instead to only publish max bandwidth. Why? The obvious answer is marketing. Marketing is about succinctness, clarity, and familiarity. IOPS has none of that, and vRPM seems to have it all.

In response to your question about consumers knowing the difference between different RPM HDDs and also to demonstrate the utility of a vRPM spec, consider the following 3 hypothetical notebook configuration pages that a consumer could encounter at dell.com:

Code:

Currently:

[o] 80 GB 5,400rpm Hard Drive 	(included)
[ ] 80 GB 7,200rpm Hard Drive 	(add $30)
[ ] 80 GB Solid State Drive 	(add $300)


Proposed vRPM scenario:

[o] 80 GB   5,400 rpm Hard Drive 	  (included)
[ ] 80 GB   7,200 rpm Hard Drive 	  (add $30)
[ ] 80 GB 414,000 vrpm Solid State Drive  (add $300)


Proposed IOPS scenario:

[o] 80 GB 5,400rpm Hard Drive 	(included)
[ ] 80 GB 7,200rpm Hard Drive 	(add $30)
[ ] 80 GB Solid State Drive
	 5,923 4KB Random write IOPS, queue depth = 3
	13,883 4KB Random read IOPS, queue depth = 3
				(add $300)

It seems obvious to me that the vRPM scenario is the only one that will massively drive SSD sales. This will result in lower prices, faster drives, and faster development. The current scenario lacks any ability to understand why one should pay 10x more for the same capacity SSD. The IOPS scenario doesn't mean a thing to the average consumer and even 90% of the readers of this thread would have to go look at scattered end-user benchmarks of varying methodology to be able to make a comparison (remember, MFRs don't and almost certainly won't publish these specs).

Check out this screenshot that I took at Dell just now to demonstrate my point of what consumers are currently confronted with:

There are hundreds of thousands of MFRs trying to "drill facts into consumer's heads" every day. What makes you think that consumers care enough about SSDs to spend the requisite hours studying bandwidth, latency, IOPS, and the like so that MFRs can just publish the nerd specs (which they won't anyway)? Computer consumers currently understand RPM, why move away from that unecessarily?

Are you suggesting that optical drive performance metrics are irrevelant? LOL.

Original optical drives were rated at 1X. Eventually we got 2X, 4X, 16X, and so on. Even though mainstream consumers jumping on the optical drive bandwagon later on didn't understand what "X" was (150KB/s), it didn't matter. They could see that one drive was 16X and another was 48X. Thus one could compare prices and buy the faster one if it made sense.

The same metric was then applied to memory cards (initially compact flash, then SD et al) despite the fact that they don't spin like an optical disk. Again, this made sense to do even though by this time very few consumers actually understood what "X" was. The important thing was that they knew how fast a 48X optical drive felt to use; it read a whole CD in about a minute. By extension, this means that Joe consumer understands that if he buys an SD card rated at 60X that he can read an equivalent amount of data from it in about a minute. It also allows Joe consumer to quickly compare the performance of a 60X SD card to a 233X SD card very easily.

Using vRPM as a performance metric for SSDs will not suddenly become "irrevelant" once HDDs are replaced. IOPS will always be important. All vRPM does is takes IOPS and turns them into a standard measurement.

The more I think about it, the more I think that vRPM is a necessary metric to implement as an industry standard.

 

Whether an optical drive is 48X or an SD card is 233X, they are both measuring transfer rate in KB/s. 7200RPM is measuring Revolutions Per Minute, which does not apply for SSD drives.

 

Weird, but A-Data looks like they want to be the #1 SSD company on the market...at least in terms of having the best SSDs with the A-Data name on them.

They are releasing A-Data branded versions of both the Intel X-25M and the Indilix-based SSD (Vertex and Ultradrive ME).

THIS one looks like the Indilix-based drive.

THIS one is clearly the Intel X-25M.

This should at least offer competition and lower prices for the Indilix-based drives. I wonder if it will bring down the price of the Intel drives. The Kingston version is actually more expensive than the Intel branded version.

 

But IOPS scales with RPM. A % increase in RPM results in a similar % increase in IOPS. In other words, they are essentially synonymous. The difference is that consumers understand RPM and not IOPS.

Edit: I'd like to add that from a personal perspective, I had never even heard of IOPS until I looked at SSD tech. I'm guessing that most other enthusiasts were in the same boat. I think Anandtech's first SSD article which identified the big problem of JMicron SSDs being IOPS was what brought IOPS into the lexicon of drive performance for many users. Before the advent of SSDs, RPM was used as a more simple proxy measure to communicate IOPS.

 

problem is. a drive with 7200rpm can have much, or not much iops, depending on data density. so rpm does not equal faster drive per se. from the same production line, a 7200rpm drive is faster than the 5400 line. but an old 60gb 7200rpm drive can not beat a new 500gb 5400rpm disk, f.e. and people can get confused by this.

i like the iops. but i don't think it's the final answer for a simple number that shows how "fast" a drive is..

 

I'm not too familiar with IOPS. I just agreed with the point that vRPM is not the way to go. Really, it doesn't matter what is used, companies are probably going to use whatever metric helps them move product. As consumers, we either need to bite the bullet and do our own benchmarks to get a truly useful understanding, or read in-depth analysis like this:

http://www.anandtech.com/storage/showdoc.aspx?i=3531

Thank you anandtech!

 

I think PNY is selling a rebranded Samsung MLC: http://www.newegg.com/Product/Product.aspx?Item=N82E16820178228

 

Although it is true that an old 7200rpm HDD may have fewer IOPS than a new 5400rpm HDD, this does not make a difference for the matter at hand. The proposed metric of vRPM is not for comparing one HDD to another like you are doing here; it is for indicating how a SSD compares to a HDD in terms of IOPS.

Also, vRPM is not supposed to be a single number to show how fast a drive is - it is an indicator of IOPS performance only. Such a measure is currently not available from SSD manufacturers. vRPM must be used in conjunction with at least bandwidth to give a good overall picture of drive performance.

If you are not familiar with IOPS then how can you concurrently assert that vRPM is not the way to go?

Also, you state that it "doesn't matter what is used" when it clearly does matter! To date manfuacturers have only used max bandwidth to "help them move product" and this has been to the detriment of the industry as a whole as well as consumers who were duped into thinking that a JMicron drive would perform as well as any other SSD with the same max bandwidth rating. Like I said before, if a measure like vRPM existed previously, JMicron drives would never have been released to market with the terrible IO performance that they did. They would have been required to keep them in the oven until they were at least as good as HDDs. If the enthusiasts (us) demand that manufacturers adopt a metric like vRPM then they just might do it.

I'm not sure who you hang out with, but my understanding of "consumers" or people in general is that they will absolutely not "bite the bullet" and spend a couple of hours reading an Anandtech article which requires a huge amount of tech background to understand in the first place!

Here's my point: vRPM is not for the enthusiasts. It is for ordinary consumers. However, the implications of making it an industry standard would have benefits for everyone, including the enthusiasts.

Look at the screen capture I posted from Dell; I think if vRPM was included in that screen that consumers would adopt SSD tech faster. The result for the enthusiast is lower prices and faster performance sooner than later.

To everyone:

Personally I'd like to have a discussion on the methodological soundness of vRPM as a measure of IOPS, not whether or not you personally find it useful. There is clearly a need to market SSDs with some indication of how its IOPS performance is. Can vRPM do this? If not, why? What might be a better metric to use that can be understood by regular consumers?

 

I've only read about 20% of this thread so far so forgive me in advance...

If I understand correctly most ssd's have (SLC & MLC) have similar read times and the write time is the critical difference with the intel x-25m being at the top of the almost affordable ssds.

Pragmatically, when used as the boot drive how relevant is the write access, does it matter whether it's a vertex/titan (mlc) or an x-25m (slc)... oops just realized the x-25 is mlc ... samsung is slc.

 

You almost got it.

All sequential reads and writes and all random reads and writes (With no outstanding IO requests) down to about 64K will be almost indistinguishable (In real-world situations) on most SSDs (At this point). It's when you start adding in queued IO requests and very small reads and writes that the Intel, Samsung, and Indilinx based drives start to show their dominance, and lower quality drives start to choke and die.

 

Nice find! Cheaper than Corsair too. I only wish we could get the 64GB model at $150 in the channel...

EDIT: http://www.newegg.com/Product/Product.aspx?Item=N82E16820178227!!!

 

Exactly the point I was trying to get at. It is very doubtful that manufacturers will spend the time to do in-depth testing like what was done at anandtech and then publish the results. From the perspective of people who would be buying (consumers) you either take the numbers given by manufacturers at face value, or you dig a little deeper by reading an in-depth review or running some benchmarks if you really want to know how the drive will perform. Unless the manufacturers change, this is how it is and whatever numbers they publish are not that useful.

vRPM might help average users understand better how an SSD stacks up to an HDD overall. But still, there are so many variables and the consumer might be just as easily duped. vRPM could be mentioned, but it shouldn't be used as the industry standard. Measuring performance of an SSD drive based on Revolutions Per Minute is flat out wrong from a technical standpoint. It would be much better if the industry developed and agreed on a true standard that is more accurate and could be used accross a wide range of media, whether it is spinning or not.

 

I heartily disagree with you here. You can be certain that the manufacturers do much more in-depth testing than Anandtech does. Even the "marketers" like OCZ et al that don't actually produce their own hardware are certain to do robust testing. Their respective OEMs (JMicron included!) do much more. It would be impossible to design and develop hardware without doing so. Testing isn't the issue. Publishing test results is one major issue, with publishing comprehensible results being the issue at hand in the discussion of vRPM.

Also, the numbers that MFRs publish at present (max bandwidth) are both useful and accurate. They just don't tell the whole story.

I'm really not trying to come off as demeaning when I say this (I'm certainly no expert), but I'm really not sure that you understand how vRPM is derived or what it represents. First of all, what are these "other variables" that there are so many of that, if ommitted from published drive specs, would dupe consumers? Secondly, vRPM is not being proposed as a single, all-encompasing industry standard to represent overall SSD performance. It is limited to representing IOPS. That's it, that's all. It must be used in conjunction with other measures, like bandwidth. Furthermore, vRPM is not RPM so it is not "flat out wrong from a technical standpoint." I think you are claiming it is wrong because technically an SSD doesn't spin. vRPM is not trying to say that an SSD spins or even that it "virtually" spins. The inverse is what it is trying to get at, i.e., how fast would a HDD need to spin to get the same level of IOPS as a certain SSD.

I'm also not sure why you think vRPM can't be used for other types of solid state media. It could just as well be used on any type of memory card (SD, compact flash, etc).

----------

Back to general discussion aimed at everyone:
Overall, It seems to me that all we need to get a simple and succinct performance snapshot of a drive is two specs: max bandwidth and random 4KB IOPS. Am I missing anything here?

Consider this: we could a table of all SSDs on the market today from best to worst performance using published max bandwidth specs. If we then did comprehensive benchmarks of all these drives and created another table of best to worst performance, the two tables would be completely different.

Now, if all drives on the market today had a listed vRPM spec as well as a max bandwidth spec, we could create another table listing them from best to worst performance. If we then did the same comprehensive benchmarks and created another table those drives, the chart should remain in exactly the same order. Is this assessment correct?

 

All this talk about IOPS sparked me to grab this from wikipedia:

Do you think this is an accurate way to present drives to the general public?

 

Personally I think no.

Problems that I see:
1. Within that list, a 7200rpm drive is probably the only one with which the "general public" is familiar. Outside of that list, I'd say that mainstream consumers are only familiar with 5400rpm drives . They aren't even aware that there are two types of SSD (SLC and MLC).
2. Related to the first point, mainstream (and even enthusiast) users are not aware of IOPS figures for different types of drives. i.e., you can't ask a consumer "what are the typical IOPS for a 7200rpm drive?" As a result, no direct comparison can be made without doing more (much more!) research.
3. IOPS is not a standardised measurement. IOPS are dependent on a number of factors: block size, queue depth, randomness, length of test, and (probably only a small extent) the benchmarking program used. Therefore, reporting IOPS without specifying these other variables is useless - as bad as simply claiming "max bandwidth". This leaves open the possibility for MFRs to fudge their IOPS numbers to make a drive look better than it is.

I see that vRPM rectifies all of these problems.

I suppose that a "standardised" IOPS measure could be devised which would fix point #3. But this would not fix the first two points.

 

I think the real crux of the situation is that SSD is a new technology that doesn't have a proven track record and they are obscenely expensive compared to HDD alternatives. The whole JMicron issue has probably been a big turn off as well. This is what is preventing consumers from buying SSD drives and I doubt that advertising vRPM would do much to sell more of them, especially considering the economy.

IMO, measuring performance based on RPM is not very useful. Some 5400RPM drives perform just as well as 7200RPM drives in certain situations. So then, adding vRPM just makes things worse.

Now would be a good time to figure out what is really useful and create industry standards around this that are more informative to consumers. I'm still learning about all these benchmarks. What two parameters are most important? Random read/write and Sequential read/write? Perhaps standardized testing could be adopted around whatever is most important for different aspects of performance and then we would have something useful.