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

Has anybody noticed their internet browsing speed to drop with an SSD?

 

no, as internet browsing speed is 100% independent on your disk speed.

i agree with you, tidal. my priority are speed, and silence. storage size doesn't matter.

but the stuttering is terrible, i see it on the asus eee. it is NOT overblown with the original ssd's anandtech tested. the solid is like, what, third generation core drive? i HOPE they get stuttering down to nearly nothing in a third generation!!

 

I don't know about the other drives, but I'm talking about the Solid's (which I have two of in two different notebooks). Yes, hopefully the stuttering would be much better on it than previous gens! The AnandTech article was a new article and should consider the improvements that have been made since the original Core series. It doesn't sound like they have as it's doubtful they've tested or received Solid series drives.

 

In fact the latest Anandtech article tests the latest JMicron controller (JMF602B) so it is testing the best currently available (since late 2008) in the low-end SSD market. Anand doesn't specify exactly which drive utilizing the JMF602B controller he is conducting his tests with but he asserts that they are all identical. Therefore, Anand wants you to interpret his graphs as JMF602B MLC = OCZ Solid Series = any SSD mentioned in the below quote. On this page of the review he specifies that:

I think that the only difference between these drives is the firmware that they use: either JMicron's or the specific manufacturer's own in-house firmware. This difference could potentialy make a significant difference between the drives that Anand proposes are the same drive (since he simply uses the label "JMF602B" to refer to the lot of these drives in his graphs). For example, it is a known fact that the OCZ Core V2 and OCZ Solid have very different specs despite them using the exact same controller! AFAIK this can only be attributed to different firmware.

Is it possible that the firmware found on the solid series is better in small random write performance compared to the older Core V2? The solid series has slower max read/write speeds - maybe the firmware is customized for better small random write performance at the expense of top sequential speeds. This would jive with the different firmwares that have been developed for the Vertex - firmwares with lower top read/write speeds have better IOps (but I also realize that the Vertex controller is not JMicron so...maybe not).

I'd like to test my Solid Series using iometer (the benchmark that Anand uses) and see if I get the same results as he does for 4KB random writes. Here are his results:

I have not used IOMeter yet so I can't compare my solid series to his results in detail. But I *have* used CrystalDiskMark and it consistently tells me I am getting 2.0 MB/s 4KB random write - way better than the 0.02MB/s that Anand is listing. It makes perfect sense for 0.02MB/s to result in stuttering whereas 2.0MB/s should be just fine - which would be consistent with what I am experiencing. Either CrystalDiskMark is giving me bad info or the solid series should not be lumped in with his JMF602B results. Can anyone offer a suggestion on why I might be getting this result from CrystalDiskMark? i.e., is it innaccurate for SSDs? I'm going to figure out how to use IOMeter so that I can directly compare to Anand's results.

 

on the first ones, you where not able to install some os' just by pluging in the disk and start the setup, as the random write made installations fail. they where reaaaally miserable. the solids reduce the worst-cases to a minimum, but still have that worst case. i agree with the anandtech conclusion to not support drives with such worst cases, even if they are reduced to a very low amount of chance. i know that the worst case could in my case mess up a live performance. i don't want that..

 

You set up music for Swiss raves and techno clubs davepermen LOL??? Sorry that's the first image that hit my head when I remembered you do audio stuff with your machines.

 

yep i'm a dj... as written halfway before in that thread

 

LOL; quite the image.

 

Exactly. That's another reason I don't trust Anandtech. They say they're all identical because they use the same controller. Obviously that is not the case as the firmware can make a big difference (like you said) as the different results (and specs) of the different SSDs using the same controller will tell you.

 

I agree also. That sounds the same as saying that since everyone uses the same Intel chipset, that there is no performance differences between boards?

It's quite well known that firmware can be tweaked (to a certain extent) to allow for different performance. The recent Vertex firmware (1199) was released optimised more for I/O than for purely read/write performances, as that was what the user forum members wanted when polled.

 

I think ocz is working directly with indilix with the firmware upgrades, suggesting other drives using the controller could potentially use the same firmware as well.

With the jmicron drives it is possible that there are no firmware differences if jmicron didn't sort them out themselves. The jmicron B version is supposed to just be a slightly suped up version of the original not a firmware change.

 

NOT right now, OCZ gonna rewrite new firmware which can only be used for OCZ vertex.

 

I was on Dell website and I notice, you only add $160 canadian dollar for 256 SSD.

I'm a bit surprise..

 

you can still trust anandtech. no matter how much you can tweak firmware, a bad base chip is a bad base chip. you can hide it more, make the problem happen to be your problem less often. but still, the max write latency is in the order of a second, and you can't fix that. you can try to make it happen as less as possible, that's what the firmware is for. but still, you can't "fix" it. sadly.

but of course, if you limit the chance of the worst case to happen, and get the users to just not do what triggers a worst case, they won't notice.

that's like you could say a cpu is never long on 100% cpu usage, so we can passive cool the system, cooling will fit. now imagine one having such a system running 100% of the time at 100% cpu. it would overheat.

that's the problem i see with such systems. they may not show the problem at all for most users. doesn't make it non-existant.

 

Bottom line here:

Some people have JMicron controller drives that do not stutter. But enough do, OCZ Solid, etc, that they should be avoided at all cost. I bought an EEE 900 with SSD and installed XP on it. My system would literally lock up for 5-20 seconds when I could right click to rename a folder. I couldn't browse the internet without stutters and lock-ups, it was unusable. I returned the system the next day.

The point is, Jmicron's solutions and all these OS tweaks are band-aids on what is still a profusely bleeding bad controller. Somebody said this earlier:

Random reads and random writes are the vast majority of your operating system's disk usage, and random writes are incredibly day to day. You change a setting, you hit apply, your browser writes a cache, and so much more behind the scenes. Sequential reads and writes, something JMicron drives excel at like any SSD, are not common at all.

 

OK, so this is interesting. Is Anand trying to make the current JMicron drives look worse than they actually are? From the looks of things, YES!

Anand changed the way he benches small random writes with IOMeter in his March 2009 review compared to how he tested in his fall 2008 review - but makes no specific mention of this change. In fact, he leads the reader to believe that the test is the exact same by stating:

NO! He changed the way he set up the IOMeter test!
In the first review he tests 4KB random writes with a queue depth of 1 ( LINK).
In the most recent review he uses a queue depth of 3 ( LINK).
This would have a minimal impact on the high-end drives but quite a substantial impact on the low-end JMicron drives. The only reason I can think of for changing the tests is to make even the revised JMicron controller (602B) look worse than it is in order to prove a point. The 602B controller has definitely improved compared to the original controller (602A) yet his test results actually show it getting worse and he offers no explanation for this at all

Furthermore, it appears to me that there definitely is a difference from one JMicron JMF602B drive to the next - firmware matters and Anand is probably not justified in lumping all the drives that use a JMicron controller into the same category. My results with an OCZ solid series, although still really bad compared to the Intel, indicate why I have not had a bad user experience (i.e., stuttering): my IOs are 120% higher than what Anand reports in his first review (JMF602A) and 66% higher than what he reports in his latest review for the JMF602B. Furthermore, the average and max latencies of the OCZ solid series are about half of what Anand reports in both his fall 2008 and most recent tests - this is a pretty big difference!

Here is a comparison of the Solid Series (in a "used" state) compared to what Anand says I *should* be getting:

Code:

Queue depth = 1		OCZ Solid Series	Anand's JMF602A results (fall 2008)
IO per second:		8.90			4.06
MB/s:			0.03			0.016
Average Latency:	112ms			244ms
Maximum Latency:	463ms			991ms



Queue depth = 3		OCZ Solid Series	Anand's JMF602B results (March 2009)
IO per second:		9.33			5.61
MB/s:			0.04			0.02
Average Latency:	320ms			532ms
Maximum Latency:	1145ms			2042

In summary, I think Anand has purposely written a harsh review of the current SSDs that use a JMicron 602B controller. On one hand, this is actually a good thing - it will force both SSD manufacturers and controller manufacturers to abandon any new products that approach this behaviour since the issue has been so loudly covered by Anand. On the other hand, this is poor methodology and misleading. He should just state facts without a personal agenda and let the end users make up their mind, IMO.

Incidentally, can I make a call out to INTELUSER? I remember you mentioning that you figured there is a threshold of IO performance past which a user won't notice a difference, i.e., the drive will not be able to stutter no matter what you throw at it. Can you comment on an IO/s performance of ~9 ? This is very low, but I'm not seeing any "stuttering" - only poor performance doing a few very specific tasks which are out of the ordinary (uncompressed audio edits).

 

You've misunderstood my quote. I've bolded the essential word. Yes, you are correct that small random writes happen all the time - but they are not usually occurring in a constant, consistent, dedicated fashion - so it is not a 100% guarantee (as many people posit) that a user will overload the capability of the JMicron controller and experience slowdown/stuttering. In fact, it seems that with the revised 602B controller that many (most?) users do not experience this behaviour in day-to-day usage. It seems that the 602B controller can handle the small random writes of the OS (Vista and 7 anyway), internet browsing, messaging, etc without a problem.

 

Samsung 32GB SLC for $150 is back in stock at Geeks.com
http://www.geeks.com/details.asp?invtid=MCBQE32G5MPP-0VA

And 250GB Vertex is $670 right now
http://www.amazon.com/gp/product/B0...=1789&creative=390957&creativeASIN=B001NPCTBY

 

Anecdotally, reading from users and reviewers over on Xtremesystems, even the dual JMicron drives are stuttering for many of them in daily use. So I think the problem is still there.

 

The bolded word may be the key here

 

No, he's right. My G.Skill Titan 256GB still freezes up from time to time. Even with 4GB of RAM in my laptop and as much as possible available for FS cache, if I do a large write that causes a cache flush, reads will completely stop until the write completes.

But this is a pretty rare occurrence. The most common case for a desktop user would be extracting a lot of files from a ZIP archive. In my case, it happens when I update a git source tree that had a lot of queued patches, extract large archives with lots of files inside, or install system updates that rewrite lots of existing files. If the cache is large enough a lot of the small file creation can be performed entirely in RAM, and then the cache flush is totally sequential. But if the stream of updates is too large, or it's mostly updates to existing files, then you just can't avoid the cost of the random writes...

By the way - some of this focus on "random writes" is still missing the point. Even if you're doing a streaming sequential write, the flash arrays will still be 100% busy, and thus unable to service reads (that would otherwise have been very fast). E.g. on Linux, try using dd to write a series of 2MB chunks to the disk. Given your usual sequential write speed, (say, 150MB/sec) write enough blocks to take a measurable amount of time (e.g., 20 seconds). Leave that in the background, and then immediately try to launch new programs or open documents in currently open apps. You'll see massive stuttering then... Try it on an HDD and you'll have no problem. The difference is, once a flash memory starts to erase a block of memory, it's stuck on that task until it completes, and each block takes a long time. The controller can't decide to suspend what it's doing to handle some reads in the meantime. An HDD can break up any I/O request on a sector-by-sector basis if the controller knows that the sectors for a particular request are going to fly by the R/W head soon.

 

I am sorry for spoiling the JMicron talk here (for me is a "dead" product) but have you checked lately the OCZ forum regarding the Vertex SSDs and the firmware 1199 ? They pulled it of as many users were complaining about lost data and partitions. Back to 0112 (first release firmware) and numbers.

Wrong turn OCZ !!!

 

As a previous core owner and current samsung owner, I say stay away from OCZ products as long as they are not rebranded sammy's.

 

As I posted earlier... I believe the upcoming OCZ Summit will be a Samsung rebranded. We'll need to see about the sale price..

 

I still believe the JMicron stuttering problem is overblown with the Solid series and other "good" implementations (if you want to say that). The point is not that the problem is eliminated completely, but that it is reduced so that many people won't notice it. I still think power users should still stay away but for someone on a budget who isn't a "power user", I would recommend the drive.

Regarding the Vertex, I think it will be great when the firmware gets stabilized and some time has passed to prove its stability and reliability. Otherwise I'd stay away. It just seems to "BETA" for me with all these firmware releases. I would not recommend the Vertex at this point though, until it's stability and reliability is more proven.

 

Anyone know what is the difference between firmware 8610 and 8790 of X25-M/E?
Is that anyway can get the 8790 firmware for X25-M?
I can't really fiond it on Intel's website

 

If you want an excellent well rounded SSD just do yourself a favor a buy a Samsung.

 

Anyone who is not a "power user" will not know how to tweak it and it will stutter even more so those things are not good for anyone.

 

i had some jmicrons and i couldn't intall an os on it just the ordinary way. those drives all lie in a corner. i call them dead as they are of no use for me. tweaking or not. a drive should by default be able to handle what i feed to it.

and the biggest jmicron problem is multitasking. do something in the background and the chip can't handle something else. like installing windows updates => you can't really start apps fast anymore.

TidalWaveOne, you should accept that it's not overblown. it's like going back to single-core systems. they where perfectly well, but you can't compare them to the snappiness of a dual or quadcore system any day.

 

I am curious...how are all the folks here doing their RAID0 with SSDs? Are you all using software-based RAID in the latest Intel chipsets or what? Is there inexpensive but reliable RAID available for the masses? The last controller I worked with was the 3Ware in one of my personal servers a few years back. It beat my Adaptec SATA 6-Channel controller but it was expensive as hell. What's the landscape now for the consumer looking for affordable RAID0 and 1 (not anything else)?

 

i use the onboard raid of my mainboard. on the pc, that is. on the notebook, i can only have one drive anyways.

 

I am using onboard RAID 0 as well, on my Gigabye P35 ICH9R DS4 motherboard. Seems to work well but I don't really have a choice because I don't have any available PCI-e slots for a RAID controller since I use two video cards. Is this considered hardware or software RAID?

Curious too though, are there any good & well-priced RAID controllers out there that would significantly outperform motherboard RAID 0?

 

not as long as you get x times the performance with x drives.

and as far as i can see, you get that => no need for anything else.

 

I don't know what the threshold is, it would depend on usage. I'd actually put it at 50 as the minimum but I guess there are people out there that's ok with 9-10.

Crystaldiskmark isn't wrong, but until we know more details about the way it benchmarks the drives, IOMeter tells us more about it. Plus, IOMeter doesn't show the HIGHEST of the results as the final result unlike CDM(with CDM if you want to see the average results you better sit through the benchmarking process and see all the 5 results and calculate it manually).

Again the problem is this. SSDs seemed to be marketed one step higher in market segment than they are supposed to be.

The JMicron MLC drives are really only suited for mom and pa systems. Those who knows enough about computers to know what the heck SSDs are is probably needing more space and is better off with platter HDDs.

 

I replaced the DVD drive with 2nd SSD and using Mac OS software RAID0 in 15" MacBook Pro

Sadly doing it this way I can't install windows anymore since it doesn't support Mac software RAID

 

you could use fusion or parallels

 

Yup, I have both VMware and Parallels, but sometime I need to be in native windows also..

 

I tested the solid series with no tweaks at all. No stuttering.
I'm also currently using the solid series with only one tweak (indexing off). No stuttering in 99% of tasks. The few times I've experienced slowdown was for a decidedly "power user" tasks (uncompressed audio editing and a massive RSS sync).

In short, you are just plain wrong on this assessment. From what I've found, those who are not "power users" will actually benefit the most from the cheap JMicron SSDs - they can pop it in their system without tweaks, perform their non-"power user" tasks, and see enormous speed gains compared to their old (probably slow) non-"power user" HDD.

No use to you and no use to anyone are two different things.
The drives you had are obviously first-gen JMicron. They have both a different controller and different firmware. Claiming that the first-gen controller is the same as the second-gen is comparing apples to oranges. Although the current JMicron controller is still awful for small random writes, it is improved enough from the original so that stuttering is not an issue for many users (including myself). Some of the newest JMicron based SSDs (at least the solid series) do not require tweaking, will not stutter, and will drastically outperform notebook HDDs for many users. Therefore they are NOT dead, they are just NOT for everyone. I refuse to spend more than $100 on a drive at this time, so they are the only option for people like me.

50 IOps is pretty low - it seems feasible that OCZ could release a firmware that could allow the Core V2 and current Solid Series to increase their IOps to this number. Then again, I don't know exactly how or why decresing top speeds translates into higher IOps as seen with Vertex firmwares and if the same would occur with the JMicron controller.

Also, I don't know about "mom and pa" systems only...many geeks don't require or value large drive space for what they do. Lots of hardcore computer geeks also have Lenovo T6x/X6x/R6x/Z6x or other notebooks which are limited to SATA-I speeds and won't get any value from an SSD with transfer rates higher than ~120MB/s. Also consider that there are lots of users out there that have very demanding needs but don't have high budgets and are limited to considering the lowest cost SSD drives. For this rather large audience of tech-savvy nerds, a low-cost SSD is in high demand. I don't think we should write them off.

What if OCZ could release a firmware for the Solid Series (and Core V2) that would increase small random write performance so that stuttering was essentially impossible (50 IOps by your assessment) at the expense of decreasing the top read speed from 155MB/s to, say, 100MB/s and the top write speed from 90MB/s to 60MB/s? It would still retain several of the huge advantages of SSDs (silent, low power, miniscule access time, comparitively high transfer rates to notebook HDDs). At around $100 for 60GB (and inevitably going down), this would be the perfect drive for a lot of users (especially those limited to SATA-I speeds anyway). Anyone agree?

 

I think we're on the way there Jackboot, Jmicron is still in the market, and is probably doing everything they can to make a cheap controller that works great just for that. And with more competitors entering the arena, prices will continue to fall. The X25-M is approaching half it's release cost, thanks to competition from dual Jmicron drives, the new Samsung, and now Indilinx drives, and it wasn't even released that long ago.

 

Too bad the price is always trailing though. For example, someone (perhaps like me) says they'll buy the Intel for $X, then a new SSD comes out and gives Intel some competition... the person then says they'll buy the Intel for $Y (which is less than $X). Intel then drops the price to $X but the drop is too late since the new target for the person is now the lower $Y price. Intel could make a killing if they would lower the price faster but I guess that is not their strategy.

 

Yeah, Intel's pricing strategy is really weird. But as long as it works in tandem (whether they like it or not) with the competition to keep driving prices down that's okay with me =)

 

Question:
how do you guys tweak Windows 7(7057) for SSD, I am using Samsung 256G on DELL e6400, and the score is 150MB/s read and write. I have installed the latest intel matrix storage and done few tweaks by Vista way.

 

The OCZ Vertex is apparently recommended by the Anand article as an 'inexpensive alternative'. The problem is that at $245 without rebate it is not cheap at all. The x25 has 25% more space, so to do a fair price comparison, we need to make it 25% cheaper (80GB vs. 60GB). That places it at $272 which is not very far off from the Vertex. If you include the $20 rebate, there is a $47 differential. I don't know about you but I am unwilling to make the compromise for proven speed and performance and stability for a mere $47 after rebate. At $27 it is a no-brainer (how many of us have been burned by OCZ rebates?)

Now if the Vertex came at around $175 without any rebates it would be a different story. As it stands, IMHO, it ain't a contenda. Until then, Intel FTW.

(Edit: All prices above from newegg/US only)

 

I've never purchased on OCZ - their rebates are iffy?

 

I think they are better now though I have had a bad experience with some of their RAM rebates (as have many others). The web is full of the horror stories. Admittedly, the fault may lie with the handling company doing the rebates for OCZ but whom else should we hold accountable?

I don't think it's software RAID but then again, you don't have a dedicated card so it's not hardware either. It's just onboard RAID which relies on some hardware and software to get the job done. Better than software (Windows) RAID.

 

they have processd my reabte, and i will get it in 8-10 weeks. Gee, its hella long.!!!!

 

wow. twin SSD mbp. I might just do that to my laptop after its warranty is over. Do you know if it is possible to do the same for a 12" powerbook?

 

People overestimate how much IOPS they really need. Not many people will notice "stuttering" if the drive can main constant 50 IOPS. As long as its constant its pretty interesting how low the bar for acceptance is. You know, in movies 30 fps is the measure where we don't notice flickering, yet in games people do. Only difference is that in games, that 30 fps is not constant.

Anyway I point 50 because 100-200 with the old SLC SSDs are where people don't notice. 50 is probably the minimum for most. Might feel a bit sluggish with lots of writes but nothing people would consider a problem.

And for the last sentence on the quote, let me explain why.

SSDs do not have direct logical to physical address table relation. The controller keeps a log table and using that it ensures that every part of the drive is evenly written, so no one part of drive fails suddenly(ideally anyways).

After some time the drive will look like the data is all over the place. Sequential access speeds will go down because of that. The amount of sacrifice is probably dependent on how the manufacturer wants to optimize for.

And while firmware can mitigate it somewhat, it will NEVER substitute what a well built controller will do.

I've read that Intel's flash memory is a strategic venture rather than a financial one. What that really means I don't know, but certainly they are taking it differently.

 

You're comparing the 60GB OCZ Vertex with the 80GB Intel. Perhaps a better comparision will be the 120GB OCZ vs Intel 80GB since IMHO most ppl will need at least that much space -

OCZ Vertex (120GB) - $309 after rebate
Intel - $363 for 80GB (after rebate), which works out to $545 for 120GB

To me, that $236 is a pretty major difference.

Maybe the best option is the Corsair S128 at $327, since it seems tried and tested.

 

main difference in games is, frames are just one sample in that 1/30rd second. in a movie, it's a collection of all "frames" in that 1/30rd second. a.k.a. perfect motion blur. and games have more issues due to the latency that increases with slow fps.

i have 50iops on the mtron in the worst case and it's still very snappy and all.

main problem is, if one io blocks all others. it wouldn't matter if some save-operation would take say 1/10th of a second. but if during that 1/10th of a second YOU CAN'T READ FROM THE DISK, you're 100% blocked. that's the main difference from jmicron based ones to others. there you can still access data, on jmicrons, the system gets into complete freeze state as you can do nothing else. that's why hdd's, while having sometimes much worse iops, don't feel _that_slow_ (but my pc at work took 15minutes to react this morning because of some disk trashing.. well, it's monday ).

and that's why anandtech changed the iops test to do several work in parallel right now. they wanted to test if those disks fit dual and quadcores that may read and write at the same time. jmicrons don't stutter anymore if you do only one thing at a time. they start stuttering if you do more than one thing. they can't handle more than one thing. that's why they created that crazy raid0-jmicron thing.