Seagate Momentus XT Hybrid HDD w/ built-in 4GB SSD

Interesting! Tiller is saying the same thing:

Update: No wonder, 4K random read & write performance has improved by 50%!



Maybe someone can confirm with AS-SSD on a primary partition.

 

Wow. That's pretty awesome, nice benchmark too. I'll have to update mine and see for myself.

 

So, how would a pair of these perform in a raid0 ?

 

Wow, great to see benchmarks like that o_o

Are you sure it's been that drastic a change?

 

The changes were consistent at 100MB file size. At 500MB file size there was no change visible.

It would be nice if someone could rerun these benchmarks on a primary partition.

 

Poorly.

This drive isn't designed to take advantage of RAID schemes. You should avoid it.

You can achieve pretty good performances with a RAID of scorpios black for example, getting better results than the XT Momentus for cheaper.

 

Someone else I know had good luck with replacing the XTs w/ Scorps. But you know, next thing wwoods will be talking about placing the XTs in a RAID-5 volume! :wink:

 

oh god no......not a raid5.....reason I asked was for wifes system, but got my answer...goin back to my cave lurking.

Thanks

 

Actually these drives perform wonderfully in RAID 0 based on benchmarks.

http://www.overclockersclub.com/reviews/seagate_momentus_xt_500gb/4.htm

 

from what I can tell so far. SD28 seems to cache faster to nand. Previously with HD_Tune 2.55 on the first run it showed 17 average ms access and it still does. before on the second run the access was lower but still you could see it was not all in nand but on the 3rd run 98% of the access went to nand and 0.4 ms access. Well on the second run with SD28 it all went to nand and now 0.3 MS access.

I use the drive as a secondary drive on my system and can confirm somwhat the crystal disk mark results. With 50mb and 100 mb the speeds have increased but at 1,000 mb they are about the same. For average use we almost never sequentially read or write 1GB at a time so real world the lower numbers are more realisticaly used.

This is a fast impression so far but for real world it looks like the firmware improves performance....................

 

Probably a change to the algorithm used for fetching chunks of data.

 

Right now I have a raid0 of WD Scorpio Black 750's and here are the crystal disk mark benchies I get. I am wondering, and I may try later, if the XT's with SD28 will be any faster.

 

If these Scorpio Black's are 3.5" drives, I would say that the XT's will not bench any better.

However, I am willing to bet that they will perform better in actual use - even over the 3.5" or the notebook 2.5" 750GB Scorpio Blacks.

Looking forward to seeing your results.

 

They are laptop 2.5's, doin a full system image now.....will post back with the results after replacement and test.

 

Well, benchmarks are one thing, but there's also a word from Seagate (seen in this thread) that you would loose the advantage of the NAND in RAID, and could even make some operation slower if i remember.

Anyway, the XTs are mainly for system that can't get two drives. Buy two good 3.5" 7.5k rpm and you get a very good RAID, for cheap

 

Now thats interesting, I remember a while back readng and from experiance that these caused me nothing but lots of headaches in a raid5, but I dident realise that you coulent take advantage of the NAND in a raid0...(I will take your word on that seagate post, I am not gonna read all 170+ posts to find it) Also, is that word from Seagate recent ? Would that still apply with the new firmware ?

If thats the case, I will probaby stick with my scorpios

 

I believe you do in fact get the advantage of the NAND.

 

The drive's nand doesn't know if it's in RAID0 or not - it will take advantage of the nand's caching.

 

OK here are the benchmarks. I did the test after letting the comnp sit idle for 10 mins, just like with the scorpio blacks, exact same image from the scorpio blacks. I also went into device manager and made sure the performance setting was the same as on the scorpio blacks, I also set the same settings in Intel RST.

If someone could help interperate these numbers, its odd, some tests the XT's performed better, some not.

 

RAID 0 Scorpio Blacks giver better sequential reads and RAID 0 XT's give better random reads.

 

This is gonna be a tough descision on which to keep, I keep 4 VM's that I use for work on a daily basis on the raid....hmm

 

If you're using VM's your Random reads are more important. Those are the types of reads an OS makes often.

Sequentials can be important as well but for different things.

 

When you're using both as normal hard drives there won't be a worthwhile difference.

If you use the XT in a way that the cache can help (most users do) it will be much faster than the Scorpio Black. Almost like SSD vs. HDD.

 

OK, and jat way would that be....I am just trying to squeeze all the speed I can out of this system. When you say "use the XT in a way cache can help", what do ya mean ? Youu understand that my setup is thus:

C: Boot drive SSD
D: Data Drive, 2X Seagate XT's in Raid0

it used to be thus:

C: Boot Drive SSD
D: Data Drive, 2X WD Scorpio Black 750GB's

and what I am seeking to determine is which is the best (fastest) setup.

The system is really dual use, I normally run 2 - 4 VM's in vmware workstation (Windows XP a Windows7 and either 1 or 2 Linux distros) nightly for work, but I also have 3 games installed I play (everquest 2 and Company of Heroes, as well as Crysis), so I am trying to get the best setup for all the above.

Below you will find benchmarks with the 2 WD Scorpio's in raid0 and the XT's in raid0. Both have the exact same software inslalled and had the same background programs running (I made a image of the WD's and just dropped that on the XT's when I benched them). I also made sure all Intel RST and Device Manager settings were the same.

 

The only way to really know is to time some of the things you do by hand.

What I mean with "use the XT in a way cache can help" is that the only benefit the XT has over the Scorpio Black is the 4GB SSD cache. You probably know that the 4GB cache is used to copy the most read blocks to, in order to speed up processes.

For example if you play the same game a few times, some files will be read from cache and those files will load a lot faster.

If money is no object and you don't really need the extra capacity of the Black I'd suggest the XT. It's likely to make at least some of your work/play faster.

 

Well, I already had the XT's so I did not need to go and buy them, and the extra capacity of the Blacks isnt used right now....I guess youre right, will just have to see over time.

 

I red this explanation from seagate when i got my XT. At the beginning of this thread i guess. Firmware shouldn't really affect that, because anyway, a disk does not know it's operating in RAID.

Can you explain why?

Sure they don't know, and that's why there are chances that it won't help. If both drives NAND had been reseted at time 0 when the RAID is mounted, i guess it could help as much as on normal operation. But if NAND aren't synchronized, the caching will differ, and while one HDD will return data fast, the striping of RAID 5 and RAID 0 will make the caching useless (the longer time to wait is the actual time you wait for the whole data).

RAID schemes are using drives in ways much different than single drive use. Striping, parallel access... kills the purpose of caching consciensously every most read block.

Benchmarks from wwoods show that you can't really say XTs are better in RAID. They perform better in random when the data is cached on both disks. Because then, you get fast access from both disks.

If you're using 4 VM's on a regular basis, be sure that the XT's NAND won't help at all, because the 4GB will just be way too small to prevent turnover of cached data.

However, XTs will perform very well in RAID as they already are good 7.2k rpm drives.

You should ask the question at Seagate's official forums for the XT.

 

Looking at benchmarks it makes sense.

And as you said, a disk doesn't know when it's in RAID. Why would it turn the cache off?

Based on?

 

The XT doesn't cache files, it caches the most read blocks.

So even if the cache is too small to hold the files, some blocks (depending on the algorithm) will be cached and that speeds up some processes.

About RAID: it's true that Seagate advises against it. Yet we have had several posters reporting that it works well.

 

It caches chunks, exactly. So it really has nothing to do with file size and that's why you can see such a huge speed increase with just 4GB.

EDIT: I don't believe "most read" has anything to do with it. It sees you access a file and then it uses algorithms to detect what you'll access next and then caches that. The "memory" you see from increased times based on how many times you do the action is just the algorithm learning what's needed to cache/ the files are already there.

 

Well we don't exactly know what the algorithm exactly is but personally I believe how often a block is read (in the last reading actions) is an important variable in the algorithm.

 

Of course it doesn't. The cache is not off, when did i say that? I only report what Seagate advises, and I try to give it a sense.

Based on the way this disk has always been reported to work (i'm surprised some people here are making up new behaviour): when a block is often read, it is "ghosted" in SSD. It acts as a proxy for this block (in fact chunks, indeed).

So, as many reviews of the disk on the web are showing, if you use equally more than 8GB of data all the time, the "proxied" chunks will often have to be overriden. You will still get some of the most used chunks cached, but the increase in speed won't be as obvious as with a small turnover.

That is, with 4 VMs, plus a host OS, you definitely get more than 4GB of very often used files. I remember a review (Tom's Hardware maybe) saying the turnover limit was reached with Windows 7 (and the usual system programs like antivirus) + Word + Photoshop + Outlook.

I guess 5 systems will definitely overcome that limit.

But (see what i wrote above), when 8GB of data is almost equally "most read", the cached chunks aren't lasting long in SSD.

Indeed, it seems to work pretty well, but, i repeat, there won't be such a big speed increase with using XTs on RAID as you notice when using an XT on a single system.

That's the first time i see this explanation of the algorithm. Honestly, i don't think it's the way it works. You're describing typical cache here, the 16/32/64MB of RAM in HDDs. Seagate announced the XT as a normal drive that holds more read files in SSD to speed up most used programs and Windows. Now, that's the commercial text, maybe the actual behaviour is as you suggest. But i doubt it.

 

Of course. And still the blocks that are cached will greatly help speeds things up. That's what I'm talking about.

Where do you base this on?

I would have to see benchmarks before I would believe such a thing.

 

I assume it works like any local cache though as Phil pointed out we really can't know.

Either way, grabbing chunks still helps. Granted you won't see a huge speed boost when running 4VM's simultaneously (I thought they just wanted one VM at a time) but really how could you with something so heavy.

 

Well, its not ALWAYS 4 at a time, thats the top end, useally its only 2.

 

Some info regarding my research on the XTs - http://forum.notebookreview.com/har.../564036-momentus-xt-raid-5-a.html#post7285659 and more here - http://forum.notebookreview.com/har...64036-momentus-xt-raid-5-a-2.html#post7285704 Now, this data is gathered from research of third party tech articles regarding the XTs, but only reading XT drive's micro-code would I know this is *exactly* how it caches SSD data.

In regards to wwoods VMs - remember the VM files will be split over X drives. Depending on how much of the VMs files end up in the XT's SSD portion would be based on what host OS is inside the VM, how much of those files are read over and over again upon boot.

Since wwoods is on a x7200 (like myself), he might easily be running 2-4 VMs at any given time.

 

I am wondering, if this whole thing is really just acedemic, I mean really, between a raid0 of WD Scorpio Black 750's and a raid0 of those XT's. I cant imagine the speed difference will really be THAT much, I mean, I am starting to wonderif if any speed increase will REALY be noticible outside of benchmarks.

I have tweeked vmware workstation to run the VM's in memory space, so they really hit the disks verry little once started, and therefore, since they run in memory space, once they are started they are running verry fast (for a VM)

 

Well if the cache works in RAID 0 the XTs will definitely be faster. But does the cache still work in RAID 0... I don't know.

Should be quite simple to establish.

 

Phil, unless this Seagate employee is flat out lying, it works -

"To the system the Momentus XT just looks like a hard drive. The performance benefit from a read perspective is 100% handled by the drive’s Adaptive Memory Technology. The drive will determine what files / bits of data will need to be placed in flash to increase read performance without dependence on a RAID controller or software."

Then again, in this post, he did not believe RAID-5 would be a problem either. We now know this to be incorrect. Some have suggested it has to do with a return value the drive makes to the RAID controller for volume participation when the drive has powered down, but the SSD portion is still active.

As you said, it is easily verified with some testing.

In any case, here is the entire article for interested parties - the comments are also interesting as well - Momentus XT – what the experts are saying – The Storage Effect

 

OK....how would I go about establishing this. I am willing to make the tests, short of trashing my system.

 

Looks like it's already confirmed by the link above

Momentus XT – what the experts are saying – The Storage Effect

 

One way to do it would be after a fresh install, using boottimer.exe and then repeatedly booting. If the cache is working there should be a big improvement in boottime.

In case you don't want to do a fresh install, the same procedure could be done with one large application or several smaller. If there are improvements over the third or fourth it should mean the cache is helping. (Rebooting in between is necessary otherwise Windows processes can influence results)

Here's what application launching looks like after several runs with one XT:
Seagate Momentus XT Application Launching in Windows - YouTube

 

Well, the fresh install and booting wont work, as I dont boot off them, my setup is like this:

c: OS - Boot Drive - 256GB SSD
d: Data - Data/files/etc/ - XT's raid0

And I have noticed, what I percieve to be improvements in opening VM's, which I do a lot....

 

Would be nice if you could time something, then we know for sure.

I'm pretty sure it does work but would be nice to see some 'hard' evidence.

Putting a large (1 - 2GB) image on your D: would also work, and then opening it in Photoshop.

 

Let me se what I can do....I am sure I can get a demo copy of photoshop, open it 2 or 3 times with normal scorpio blacks in raid0 then do same with the XT's. I will do my best to time it with a stopwatch here I have on my blackberry. Pretty sure I can get something set up.....

May be a day or 2 before I find some time to pull it off though.

 

OK that would be interesting.

The run with WD7500 would be interesting too, although not mandatory to check if the cache is working.

PS. Photoshop was just an example. A large PDF file in Adobe Reader would work just as well.

 

wwods,

Remember, you'll have normal drive caching issues as well. Perhaps you can open the large file enough times where you feel it has made it to the SSD portion. Then try to open the file AFTER a cold-boot the x7200. This way you ensure the 32MB cache on the drive itself is not in play.

 

Yes you have to reboot basically after every run, otherwise Windows ready fetch (or what ever it's called) will also come into play.

 

Superfetch.

 

The cache does work in raid0 but not as efficient as in a stand alone. The drives each have to cache data blocks of 4KB. If the stripe leaves a drive without data from the file size then nul data is cached. This is a waste of the cache. Now access time is not improved but where both drives store data, just as in a HDD with raid0, the data stream is 2x the speed of that of a single drive. Because of the null data the cache doesn't act as a true 8GB of nand.

With VM's it is an unusual situation. The system itself caches alot of files to ram. Frequently used files will be accessed by the ram cache where lower accessed files will tend to migrate to the NAND. If you only have 4GB ram and 16GB data of course the XT's NAND will be of very limited use. In those cases a true SSD is a much better option. I also guess this is the reason the XT's are not an enterprise drive too.

People seem to try and get more from this drive than what it is meant for. The XT's are a consumer device meant for low to mid level personal laptop usage. A true enthusiast drive would have been 1TB with 64GB NAND and 256MB ram cache, but at what cost?

I am not saying there could be limited bennefits to the XT's in lower level professional envoirments but you do have to realize it's limits. Once you do and use it accordingly the drive is great. At some point the returns deminish from only 4GB NAND and the HDD section, while fairly fast, is not the fastest out there................