Sign in with
Sign up | Sign in
Your question

30 - 32GB SSD?

Last response: in Storage
Share
July 24, 2009 4:29:40 PM

Hi all ,

I am trying to decide on an SSD(s) for my operating system, games and apps. I was wondering if anyone on here had any suggestions?

I have been looking at a few but think the best is the Intel X25-E Extreme 32GB. But for £66 cheaper I could get 2 OCZ Vertex Series 30GB and put them in RAID 0. More speed more capacity. Plus an extra 20MB/s the Intel isn't that much faster than the OCZ Vertex.

What do you guys think? Am I missing any other SSD's?

Suggestion are greatly appreciated.

More about : 32gb ssd

July 24, 2009 5:22:50 PM

Just get the new G2 Intel X-25M 80GB.
July 24, 2009 6:22:58 PM

Crucial is now marketing a new line of SSDs that are supposed to be about as speedy as the first generation of the Intels, but at a much better price.

Also, if you do spot some SSD deals out there, watch out for the ones that use JMicron controllers. They've been implicated in a lot of complaints relating to SSD stuttering. Supposedly it's corrected by now, but some of the newer SSDs are promoting the fact they don't use JMicron controllers specifically because of that possible issue.
Related resources
a b å Intel
a c 415 G Storage
July 24, 2009 8:10:49 PM

Intel still seems to have the reputation for the best SSDs, and they've just announced their next-generation drives with substantially lower prices. See this article.
July 24, 2009 9:02:02 PM

Quote:
Wait for the new intel's


They're floating around out there now. Newegg got a shipment in last night, but they're sold out already.
July 24, 2009 10:50:01 PM

I am very satisfied with my OCZ 30GB Vertex and the amount of useful SSD information on their forums.
July 25, 2009 3:12:48 AM

I say new intel 80GB drive. IIRC they have the same read speed as the SLC drives, just slower writes. It's a better price and more capacity for speed that is still extremely good.
a c 127 G Storage
July 25, 2009 12:23:35 PM

Yeah the new 34nm Intel SSDs are getting quite cheap; even cheaper than the less fast but very popular OCZ Vertex.

OCZ in response is going to lower its prices too, quick comparison:
Intel X25-M 34nm 80GB = EUR 189
OCZ Vertex 60GB = EUR 190

I think the 60GB will go down to 150 or so, which still makes the Intel the better buy. The Intel controller delivers better random write and IOps than the Vertex' Indilinx controller.
July 25, 2009 6:58:54 PM

Thanks for the suggestions guys. I have looked into the G2 Intel X-25M 80GB and they seem pretty good. What is the exact release date for these drives, I am having trouble finding them on any UK sites.

Will there be a huge lose in performance by going for one Intel G2 over 2 OCZ Vertex drives in RAID 0??
a b å Intel
a c 415 G Storage
July 25, 2009 9:30:15 PM

Without the benefit of an actual performance test, and bearing in mind that some RAID controllers may not be able to keep up with the data rates from a pair of SSDs, my guess is that two OCZ Vertices in RAID 0 would handily outperform the Intel drive.
a c 127 G Storage
July 26, 2009 12:07:56 PM

The read latency is lower for the new 34nm Intel SSDs, and they are available in my area so it shouldn't take long for you; maybe a few weeks until it has propagated to distributors near your location.

So the read latency is even lower. That's good because you can't lower it with the use of RAID0, so the read latency is a bottleneck of its own, especially in single threaded apps that use Blocking I/O the result is a low queue depth and overall performance totally depends on the read latency.

The lower read latency also increases its IOps i can imagine. But since its speeds haven't been changed the controller chip is likely still the same type. Intel could easily make it write at 600MB/s but they don't want to (economical/strategical decision). But even without that, i think it can outperform two vertex in RAID0 in some benchmarks, but not all.

Nevertheless the price cut of the high end Intel X25-M will be very beneficial to SSD pricing, and is the first major drop of prices i've seen for GOOD ssds, not the junk without any controller or one of those bad JMicron ones.
July 26, 2009 3:01:53 PM

To add to that, I also think that the X25-M G2 would beat two RAID0 Vertexes in some, but not all, benchmarks. The vertexes would obviously have faster sequential reads and writes. I'm pretty sure that the X25-M G2 would maintain faster random read/writes though, and that is really what matters.

Keep in mind that OCZ has already laid out their new pricing because of the G2, so don't buy those if they're much above these prices: http://anandtech.com/storage/showdoc.aspx?i=3608
a c 117 å Intel
a b G Storage
July 26, 2009 4:41:11 PM

30GB - 32GB is kinda small for an OS drive, unless it's Win XP.

My Vista partition is 35GB and there's only 17GB free. That's with hardly any software installed other and a few video encoding / editing programs.
a c 127 G Storage
July 26, 2009 4:55:51 PM

I have a 30GB SSD with Windows 7 and World of Warcraft on it. Still has 4GB free, double if it wouldn't have that damn hybernation file, which i can't seem to shut down. :p 

But its possible, though you can't really install much else, and for writing performance at least some portions needs to be free of the disk.

Thanks to the TRIM function the SSD can be made aware of "stale" data sectors too. Modern SSDs use free space to accelerate random write I/O, so it doesn't have to go through a erase-reprogram cycle of the flash block (usually 640KiB or 1MiB). This speeds up random write I/O considerably, and the Intel X25-M does a very good job of maintaining random write performance on MLC flash memory.
July 26, 2009 5:42:43 PM

Once again thank you.

So RAID0 will speed up my sequential read/write times but the random read/write times will remain pretty much the same right?

When are sequential reads and writes performed anyway? What makes it sequential? I read an article that states that most modern OS's use random writes most of the time.

Sorry for any noob questions, just trying to get the full picture and I always find it easier to understand by asking people directly.
a b å Intel
a c 415 G Storage
July 26, 2009 6:11:15 PM

Peaks said:
When are sequential reads and writes performed anyway? What makes it sequential? I read an article that states that most modern OS's use random writes most of the time.
When the OS loads it needs to read a ton of files. Each file has to be looked up in the directory first. All those lookups and moving back and forth to pick up all those files means a lot of random I/Os - and for that kind of workload an SSD beats the pants off any mechanical disk drive.

Sequential reads occur when you're dealing with relatively large files - for example MP3 files, video files, or large pictures (RAW images from a DSLR camera are a prime example). In that case you spend a relatively small amount of time finding the file and a larger amount of time just reading chunk after chunk of it.
a c 127 G Storage
July 26, 2009 6:13:59 PM

RAID0 can increase both sequential throughput as well as random-like I/O IOps. What it can't do, is reduce read latency. That ultimately is bottlenecked by the seek-rate of the HDD. The mechanical part that is the cause for virtually all 'slowness' on modern computers.

When benchmarking, you get figures like:

(source)

This is a rather extreme scenario: reading 4KiB from random locations. Not the most extreme - 512 bytes random read across the whole drive capacity, is the ultimate test. As you can see the only HDD in the list, the fastest desktop-class also, the Velociraptor appears to have erection problems of some kind.

Seriously, this is the weakness and huge bottleneck of HDDs. HDDs can read and write fast, but they have to SEEK. You asked "What makes it sequential". Its not really being a question of sequential or not, but to what degree is it sequential. It could be 50% or 20%. Huh, you don't understand? You will :) 

100% sequential:
Read sector 1, Read sector 2, Read sector 3, Read sector 4, etc...

100% random:
Read sector 1, Seek to sector 47839275, read sector 47839275, seek to sector 27395, read sector 27395, etc..

In reality, when writing one big file (copying) or other single stream of I/O, it will be highly sequential. It may have a few seeks, but otherwise its pretty much read and write for the HDD head which it can do very fast. But if it has to seek alot, which happens countless times on a desktop, this can put an extremely fast system down to the point of applications not responding and user input being ignored, until apparantly it finished what it's doing and it still reacted to whatever keys you pressed. The supermodern quadcore cpu can't do anything, because its waiting on the HDD to seek. Even 1 seek will mean the CPU has wasted 30.000 cycles or Hz it could have spent doing work, but couldn't because it had to wait for the HDD.

With many seeks, the CPU can only do 1-4% work and has to wait the other 96%+. SSDs can finally solve this bottleneck, by removing the last mechanical component of modern storage.
July 26, 2009 6:15:15 PM

Ah well explained. So for an SSD being used for OS/applications it is clearly the random writes/reads and latencies that are the important things to worry about? Latency being a thing that RAID0 cannot improve? However, you say RAID0 can improve sequential throughput and random access so that's surely an improvement overall?

@sub mesa : lol think we replied at the same time.
a c 127 G Storage
July 26, 2009 6:31:50 PM

(yeah i also responded before i saw sminlal's nice summary, anyway the chart is cool :)  )

Random read/write IOps is very important yes, the latency is also important, but both latency, IOps and MB/s (throughput) are connected in a way that can get a little complicated to understand without first-hand experience with low-level benchmarking.

So what's more important, latency or IOps when using RAID? Well it depends, on the application. If the application uses multiple I/O streams (non-blocking I/O) the I/O queue can go up and disks can work in parallel. This works like this: disk1 handles request A, disk2 handles request B, disk3 handles request C. They all have to seek, but now you have three I/O requests being handled at the same time instead of one. All three I/O requests still have a latency the same of a single I/O request without RAID, but the IOps is three times higher in this example.

So if your application uses alot of I/O streams, RAID0 can make a real difference, if properly setup. But it can't make things like booting and application-launch much faster, as much of this I/O is random blocking I/O in nature, so only 1 I/O stream, so your RAID disks can't work in parallel. In this case you have close to single-drive performance. In this case, it ultimately depends on the latency. So modern SSDs would totally kill any HDD RAID config in this area.
a b å Intel
a c 415 G Storage
July 26, 2009 8:55:15 PM

Peaks said:
However, you say RAID0 can improve sequential throughput and random access so that's surely an improvement overall?
It's kind of like having a multicore CPU - how much performance improvement you can expect to see depends an awful lot on what you're doing.

If you have two or more programs trying to access the RAID set at once (or one program that's smart enough to ask for two different things from the disk at once), then both disks can often respond at the same time. Each response may take just as long as if you had a single disk (ie, the same latency), but both of them will occur simultaneously instead of one after another. This means more I/O's per second.

But if you've just got one program and it's asking for file A, waiting for it to arrive, and then asking for file B - well then having a two-disk RAID set isn't going to help that much (although it can still help for sequential reads because of read-ahead and caching).
July 26, 2009 9:15:29 PM

Another question that could swing this for me is whether the second generation Intel drives are going to support TRIM? (I am aware that I will also need Windows 7 for this).

I feel that TRIM could be a major contributor to my decision as it seriously prolongs the performance of an SSD over time (that's not to say that the performance wont get worse, it will).
July 26, 2009 10:11:55 PM

Intel is releasing a firmware update supporting trim for the G2s when Win7 is released.
July 27, 2009 1:33:38 PM

sub mesa said:
I have a 30GB SSD with Windows 7 and World of Warcraft on it. Still has 4GB free, double if it wouldn't have that damn hybernation file, which i can't seem to shut down. :p 

But its possible, though you can't really install much else, and for writing performance at least some portions needs to be free of the disk.

Thanks to the TRIM function the SSD can be made aware of "stale" data sectors too. Modern SSDs use free space to accelerate random write I/O, so it doesn't have to go through a erase-reprogram cycle of the flash block (usually 640KiB or 1MiB). This speeds up random write I/O considerably, and the Intel X25-M does a very good job of maintaining random write performance on MLC flash memory.



regarding the space that hibernation eats up in Vista, if you disable hibernation from advanced power settings, it should deactivate hibernation, but won't free up the space. you have to delete "hiberfil.sys" in the Windows directory somewhere, think it's in the win_32 folder, but not sure. That will get you back about 3 GB of Vista-wasted space.
July 29, 2009 9:31:51 AM

SO I have just seen that the second generation Intel drives, that after this thread I was pretty much set to get have a maximum sequential write speed of 70MB/s. Its slower than the Vertex and the previous generation Intel drives. Does this not seem pretty slow to you guys (for and SSD that is)?
a c 127 G Storage
July 29, 2009 9:46:18 AM

Who cares about sequential performance? Its not like its a download disk or something and even then 75MB/s will be enough.

True performance is IOps, where HDDs can't even get to 1MB/s; they can't. Its too much for them. And this is where you will notice a difference. It will also remove "slowness over time" due to file fragmentation. SSDs don't care about files being fragmented, some even do this on purpose, so that its easier to fill all 8 channels with data, to be processed in parallel. This is something a HDD can't do with a single head.
July 29, 2009 9:55:24 AM

yea I know that the sequential reads/writes aren't that important (think you told me that actually - lol) Just thought it was pretty slow. I am a little confused though, don't IOps include sequential reads/writes? Aren't they just a common benchmark that includes total IOps, random reads/writes and sequential reads/writes?
a c 127 G Storage
July 31, 2009 12:15:16 AM

New versions of HDTune do have most low-level tests: raw read/write, buffered read/write (the Files benchmark), random read/write IOps and latencies. So its pretty good actually, because not many easy windows I/O benchmarking apps exist that give meaningful numbers.

When reading or writing sequentially, these are also just I/O operations so 10MB/s throughput would compare to 80 IOps of (each) 128KiB. So do the math: 80 * 128KiB = 10MB/s. So yes MB/s and IOps are related. IOps is used often in random I/O benchmarks, MB/s is used in sequential throughput benchmarks.

Look at gstat output, as Windows doesn't give you detailed information about I/O - its nice to look under the hood once:



Just some random screenshot grabbed by google, you can see the "da0" disk is being read from.

The first column is the queue size; howmany I/O requests are waiting to be processed. This is especially important in RAID. In this case its a blocking I/O app so the queue is 1.

Next two columns we see the IOps, first the total IOps then the IOps for read requests. As there is only read activity, we see 114 twice here. Next you see the throughput, 14493KB or 14MB/s. Then you see the latency (3.9 and 7.9ms) and lastly a rough percentage of utilization at the end.

You'll notice more lines appearing with roughly the same numbers. This is because this is an encrypted volume. You can see the "da0" (unencrypted) having a lower latency than the "da0.eli" (encrypted) volume. So the disk has a latency of 3.9ms, but because each I/O request has to be encrypted/decrypted, the latency increases to 7.9ms. By this you can also tell the CPU is not very modern because modern CPUs can encrypt much faster than this.

But its nice you can actually SEE this on your system - unless you're using Windows. :( 
But i'm sure some commercial third party tools exist that do roughly the same.
a c 127 G Storage
July 31, 2009 12:20:03 AM

Oh and when calculating: 114 IOps * 128 KiB (request size) = 14592KiB. Since the actual read is 14519, it appears at least one I/O request was smaller than 128KiB.

So IOps, throughput and latency are all connected in a way, just like Volts, Amps and Resistance are related in electronics.
!