Get Two Vertex 2 or Three Older SSD's?

C

Cehr

Weaksauce
Joined
Sep 18, 2010
Messages
107
I have a 64GB G.Skill SSD that I bought at the end of 2008.

Since I'm building my photography rig, I might need the maximum bandwidth possible.

Which is why I did some research and found that the OCZ Vertex 2 is the current performance leader.

So, should I get two of these, or three of the same SSD I already have (if I can find it)?
I mean, in terms of cost, it's not a big deal.
In terms of reliability, it doesn't matter because it's only going to be used for the OS
In terms of size, I suppose 256GB would indeed be better.
In terms of I/O performance, I suppose both are SSD's and are thus similar in latency.

Thanks for your input, guys.
 
If you are storing compressed picture files or video files to the SSD, then the Sandforce drives like the Vertex 2 will NOT give you the best performance. The best performance would be with a 256GB Crucial C300.

http://www.bit-tech.net/hardware/storage/2010/08/17/ocz-vertex-2e-review-120gb/3

EDIT: I re-read your post, and I am not sure if you mean that you want to RAID 0 two or more SSDs to increase the throughput. If that is what you mean, then the best performance would be with two Crucial C300 256GB SSDs. Next best would be with two Crucial C300 128GB SSDs. That is assuming that your RAID controller is up to the task of handling the throughput of such fast SSDs. If not, your best bet is just a single 256GB Crucial C300, using a SATA 6 Gbps port (or card).
 
Last edited:
john4200 said:
If you are storing compressed picture files or video files to the SSD, then the Sandforce drives like the Vertex 2 will NOT give you the best performance. The best performance would be with a 256GB Crucial C300.

http://www.bit-tech.net/hardware/storage/2010/08/17/ocz-vertex-2e-review-120gb/3

EDIT: I re-read your post, and I am not sure if you mean that you want to RAID 0 two or more SSDs to increase the throughput. If that is what you mean, then the best performance would be with two Crucial C300 256GB SSDs. Next best would be with two Crucial C300 128GB SSDs. That is assuming that your RAID controller is up to the task of handling the throughput of such fast SSDs. If not, your best bet is just a single 256GB Crucial C300, using a SATA 6 Gbps port (or card).
Click to expand...

Let me clarify/Attempt to understand.

I'm not STORING anything on the SSD's. Only the OS and the 'temp' folder for whatever projects I'm doing, which are usually huge (10,000 images and more).
Storage, will only occur on my NAS through a Gigabit cable. The NAS's specs are in my sig. At best, it's supposed to get to the speed of the cable, approximately 120MB/s.

My original question was whether to RAID 0 two Vertex 2 SSD's or four G.SKill 64GB SSDs from 2008 (not sure which model as of right now).

Do you mind explaining what the Crucial C300 drives have over the Vertex 2 or the G.Skill?
And why you would recommend a single 256GB drive over two 128GB with undoubtably better throughput in terms of read/write bandwidth?

And lastly, unless you can recommend a not too expensive PCIe RAID card (should I buy another Perc/i?) for two drives, I'll have to use my SR-2's controller, which is 6Gb/s. But why is that bad? It's far more than the drives can pull even put together RAID 0.

EDIT: I found this review of the C300's:
http://www.maximumpc.com/article/reviews/crucial_c300_256gb_ssd_review

According to it, the only thing that the C300's have is read speed (at 50% higher than the Vertex 2). Which is great if I were going to play RTS games or only viewing my work.. instead of creating it.
But the write speeds are abysmal in comparison, and they appear to be worse on SATA 6Gb/s than SATA 3Gb/s.

So, I really don't see why they'd be better.
 
Last edited:
Cehr said:
Let me clarify/Attempt to understand.

I'm not STORING anything on the SSD's. Only the OS and the 'temp' folder for whatever projects I'm doing, which are usually huge (10,000 images and more).
Storage, will only occur on my NAS through a Gigabit cable. The NAS's specs are in my sig. At best, it's supposed to get to the speed of the cable, approximately 120MB/s.

My original question was whether to RAID 0 two Vertex 2 SSD's or four G.SKill 64GB SSDs from 2008 (not sure which model as of right now).

Do you mind explaining what the Crucial C300 drives have over the Vertex 2 or the G.Skill?
And why you would recommend a single 256GB drive over two 128GB with undoubtably better throughput in terms of read/write bandwidth?

And lastly, unless you can recommend a not too expensive PCIe RAID card (should I buy another Perc/i?) for two drives, I'll have to use my SR-2's controller, which is 6Gb/s. But why is that bad? It's far more than the drives can pull even put together RAID 0.

EDIT: I found this review of the C300's:
http://www.maximumpc.com/article/reviews/crucial_c300_256gb_ssd_review

According to it, the only thing that the C300's have is read speed (at 50% higher than the Vertex 2). Which is great if I were going to play RTS games or only viewing my work.. instead of creating it.
But the write speeds are abysmal in comparison, and they appear to be worse on SATA 6Gb/s than SATA 3Gb/s.

So, I really don't see why they'd be better.
Click to expand...

If you RAID, stay away from the C300.

http://www.anandtech.com/show/3812/the-ssd-diaries-crucials-realssd-c300/8
 
Cehr said:
Let me clarify/Attempt to understand.

I'm not STORING anything on the SSD's.
Click to expand...

Clear as mud. If you are not storing anything on the SSD, what are you doing with the SSD, and why do you want high throughput?

I can only guess you mean that you are not storing anything LONG TERM on the SSD. But that is irrelevant. If you write any data to the SSD, then you are storing data on the SSD. Let's move beyond semantics.

What data are you writing and reading from the SSDs? Compressed picture files? If so, then you need to pay attention to the link I already gave you, which shows the performance of various SSDs on incompressible data like compressed picture files. The 256GB C300 has higher throughput for both reads and writes as compared to Sandforce SSDs.
 
john4200 said:
Clear as mud. If you are not storing anything on the SSD, what are you doing with the SSD, and why do you want high throughput?

I can only guess you mean that you are not storing anything LONG TERM on the SSD. But that is irrelevant. If you write any data to the SSD, then you are storing data on the SSD. Let's move beyond semantics.

What data are you writing and reading from the SSDs? Compressed picture files? If so, then you need to pay attention to the link I already gave you, which shows the performance of various SSDs on incompressible data like compressed picture files. The 256GB C300 has higher throughput for both reads and writes as compared to Sandforce SSDs.
Click to expand...

I'm sorry. I meant that I'm not storing anything long term.
I thought it was valid because if I am storing all the files on my NAS, then read speeds do not matter, as they are being read from the HDD's, and then written to the SSD's. Followed by transferring back to the NAS at a maximum of 120MB/s and read off of it.
So it was only meant to suggest that read speeds are of little importance to me.

And according to the link, the Vertex has far better write speed than Vertex 2.
And while C300 has great write speed, although only on that review while every other says it has bad write speeds.. wouldn't two Vertex/Vertex 2 in RAID 0 still be far superior, even in terms of read?
Although, I would like to know why the 64GB version of the C300 has much worse performance than the 256GB version. I mean if there was some sort of internal RAID configuration, it would be a detriment to realiability.
 
aznx said:
If you RAID, stay away from the C300.

http://www.anandtech.com/show/3812/the-ssd-diaries-crucials-realssd-c300/8
Click to expand...

That is not the correct conclusion to draw from Anand's tests. The true situation is more complicated.

First, Anand's test of resiliency is highly flawed. HD Tach is writing a continuous stream of what amounts to a bunch of zeros sequentially to every LBA on the drive. Obviously, this is not a realistic workload, since who writes 128GB of zeros sequentially, without pause, to an SSD?

So, there are two things unrealistic about that workload: (1) It is a 128GB sequential write and (2) it is essentially a string of zeros

Look at (1) first. Note that the C300 actually does pretty well for the first 1 to 2 GB of writes -- around 140 MB/s. And that is the 128GB C300. The 256GB C300 is faster, probably around 180-190 MB/s. So if you write no more than 2GB of data at a time, the performance will be good. Unfortunately, Anand does not quantify how long you have to wait between 2GB writes for the performance to return to peak. But I suspect it is not very long.

Now consider (2). Anand has claimed a couple times that the Sandforce drives are more resilient since their HD Tach write trace does not dip down like it does for most other SSDs (Crucial, Indilinx). Well, that is only relevant if you write a 128GB string of zeros to your Sandforce drive (the SF controller compresses/dedupes the data and physically writes very little to flash). If you write compressible data to the Sandforce drive, that picture would look very different. You can get an inkling of what might happen by looking at the link I gave in my first reply. Also, check out the READ speed in the HD Tach scan of the SF drive:

http://www.anandtech.com/show/3681/oczs-vertex-2-special-sauce-sf1200-reviewed/4

The average read dropped to 114 MB/s on the SSD. That is because the drive was filled with random data, so reading does not benefit from the SF compression/dedupe. If the read dropped to less than 1/2 of peak, I imagine the write, for incompressible data, dropped much more than that.

Clearly the situation is complicated, and Anand's test is simplistic and unrealistic. The only conclusions you can really draw from that test is that if you are doing sequential writes of tens of GB of highly compressible data without pause, then a SF drive would be a better choice than a C300 or Indilinx drive, and if you are writing data in chunks of 2GB or less (with an unspecified pause between chunks), the C300 has good performance.

Also, from the bit-tech review I posted, if you are writing a lot of mostly incompressible data to your SSD and write speed is important to you, then you should probably avoid SF drives, even if you have TRIM.
 
So you are saying the only thing that matters in your application is the speed that the data is written to the SSD? I still do not understand, since if the data is not read back (locally) from the SSD after it is written, then why write it to the SSD at all? Leave it on the NAS. If you have a reason for copying the data from the NAS to the SSD, then it seems that you will be reading the data locally. In which case SSD read speed is also important.

As for write speeds, I think you are missing my point: the Sandforce drives have a controller that compresses/dedupes that data before physically writing to the flash memory. No other SSDs have that. It works great if you write a string of zeros to a SF drive, since it compresses it and physically writes virtually nothing. But if you are talking about JPEGs or other already compressed data, the SF controller cannot compress it further, and it actually hurts performance compared to a good SSD controller with no compression.

The maximumpc review that you linked to use HD Tune and (I assume) the old version of Iometer, both of which write essentially streams of zeros.

http://www.anandtech.com/show/3667/oczs-agility-2-reviewed-the-first-sf1200-with-mp-firmware/6

Note that the 2MB sequential write for a SF drive drops from 252 MB/s to 144 MB/s when you switch from writing highly compressible to incompressible data.

As for your question about smaller capacity SSDs having slower writes, that is because the smaller capacity SSDs have fewer channels writing in parallel. Of course, if you are writing a string of zeros to a SF drive, the number of channels hardly matters since it will physically write very little to the flash. But if you write incompressible data to a SF drive, then you will see that the smaller capacity SF drives have lower write performance as well.

As for whether a RAID 0 of two SSDs would be faster than a single SSD, the answer is complicated. It is likely that the RAID would be faster, but if the alternative is a single, double-capacity SSD with TRIM in use, it is possible that the RAID would not be faster (the SSDs in RAID will not be able to be TRIMed). But if you are writing incompressible data to the RAID, it is likely that two C300s would be faster than two SF SSDs. Personally, given a choice between RAID 0 of two 128GB SSDs, and a single 256GB SSD with TRIM, I'd go for the single SSD.

As for the 256GB Crucial C300 having poor sequential writes -- that is completely untrue. Check out this comparison:

http://www.anandtech.com/bench/SSD/81

The C300 is only beaten by the Sandforce SSDs, and as I already mentioned above, the SF drives drop to less than 150 MB/s on sequential writes when you write incompressible data. So for incompressible data, the 256GB C300 has the fastest sequential writes of any MLC SSD (and even a few SLC SSDs like the X25-E).

The same holds true for 4KB-aligned 4KB random write:

http://www.anandtech.com/bench/SSD/82

Again, the C300 only loses to the SF SSDs, and only with highly compressible data. With incompressible data, the C300 and SF drives are about equal. Also, if you are using Windows Vista or Windows 7, your data will be 4KB-aligned.
 
Last edited:
@john4200

Thanks for the long and very informative post. There are only few who have these knowledge and willing to share for others to learn.

Cheers!
 
I don't have time to read through the details of the thread, but I did something similar in testing for a new server. I use software that processes data in large temp sequential files. I used 4 Western Digital black drives and partitioned only about 50GB on each drive, left the rest unused. Using the onboard Intel ICH10r matrix RAID, I created a single 200GB RAID0 for processing scratch space, files are auto-deleted at the end of processing. The OS was on another drive, the RAID was only used for processing space. Final output files were written out to another drive (I used 6 drives total, one for the OS, 4 for scratch processing, one for storage but it could've been mapped to a NAS or anything).

I still have some ATTO benchmarks and 4 drives in RAID0. They gave write speed of 421MB/s Writes and 493MB/s Reads (last line in ATTO benchmark). I think the i7 920 was running at about 25% when processing off this RAID.

If you get 4 cheap 640GB blue drives, I'll bet you are still close to this throughput. This would be slower, but still decent performance, and cost a fraction of the price for a pair of big expensive SSDs.
 
najames said:
I don't have time to read through the details of the thread, but I did something similar in testing for a new server. I use software that processes data in large temp sequential files. I used 4 Western Digital black drives and partitioned only about 50GB on each drive, left the rest unused. Using the onboard Intel ICH10r matrix RAID, I created a single 200GB RAID0 for processing scratch space, files are auto-deleted at the end of processing. The OS was on another drive, the RAID was only used for processing space. Final output files were written out to another drive (I used 6 drives total, one for the OS, 4 for scratch processing, one for storage but it could've been mapped to a NAS or anything).

I still have some ATTO benchmarks and 4 drives in RAID0. They gave write speed of 421MB/s Writes and 493MB/s Reads (last line in ATTO benchmark). I think the i7 920 was running at about 25% when processing off this RAID.

If you get 4 cheap 640GB blue drives, I'll bet you are still close to this throughput. This would be slower, but still decent performance, and cost a fraction of the price for a pair of big expensive SSDs.
Click to expand...

But if it's a bunch of small files, the 4 HDDs will be slow..
 
aznx said:
But if it's a bunch of small files, the 4 HDDs will be slow..
Click to expand...

I can only tell you that our old production Solaris server with 4 single core CPUs running off our crappy SAN took 7.5 hours to process our monthly data, but little $1450 PC did it in only 2.5 hours. I had some big files, but certainly a lot of smaller ones too. YMMV.
 
I'm sorry for the late reply. I've read everything and I guess I'll be going with the C300.

Just two things I'm still wondering about:

1. So it's true that all SSD's have an internal RAID0-like configuration where more chips = more bandwidth? That means that two 64GB SSD's in RAID 0 would have the same bandwidth of a single 128GB SSD. Is that really true?
I've heard of people getting 500MB/s and up with two SSD's in RAID 0. No C300 can do that, and that's what I want.
I mean I want the best - that's why I'm here.

2. What are examples of compressible and incompressible data? OS files.. JPEGs.. RAW files.. etc.

Also.. this review..
http://www.anandtech.com/bench/SSD/81

Really makes it seem like the SF drives are far superior. Are you SURE that writing anything above 2GB would result in superior bandwidth?
 
Cehr said:
I'm not STORING anything on the SSD's. Only the OS and the 'temp' folder for whatever projects I'm doing, which are usually huge (10,000 images and more).
Click to expand...

It would, perhaps, help to explain what you're doing to the images. Depending on what you're trying to do there are lots of different configuration choices. Hell, you can even consider using a RAM disk for certain tasks (some programs work better with temp in a RAM drive than when trying to use actual RAM in the program itself).
 
Cehr said:
Are you SURE that writing anything above 2GB would result in superior bandwidth?
Click to expand...

I have no idea what you are asking with that question.

For your question about RAID 0 of two 64GB C300s, no, it is not that simple. The read speed of flash memory is faster than the write, particularly if you are dealing with a nearly full SSD where the blocks have to be erased before they are written.

http://www.bit-tech.net/hardware/storage/2010/07/21/crucial-realssd-c300-64gb-ssd-review/3

Notice that the 64GB C300 gets about 350 MB/s sequential read, but only about 75 MB/s sequential write. So a pair in RAID would have higher read speed than a 128GB C300, but probably about the same sequential write speed. Although your application might benefit from TRIM, since it sounds like you will be writing a lot of files and then deleting them, and repeat. But you cannot TRIM the SSDs in RAID, so it is possible that you might have better write performance with a single SSD with TRIM than with two half-sizes in RAID.

Your question about incompressible data is difficult to answer, since it depends on the compression technique that Sandforce is using, and Sandforce is keeping mum on that. I suspect it is a simple technique, since compressing data at 150 MB/s with a several Watt processor is not going to be very sophisticated. If that is correct, then most real data will be closer to the incompressible side for a SF drive.

AS-SSD does some copy tests on real data. The copy test includes a read from and a write to the SSD, for three types of data folders: ISO (large files), game (mix of large and small files), and program (mostly small files). The 240GB Sandforce SSDs get around 90-110 MB/s on those tests. The 256GB C300 is in the 125 - 200 MB/s range. Note that is a copy, and read speeds are faster than writes, so the read speeds are more than double the reported copy speed, and the write speeds significantly less than double the copy speeds. But the actual computation is more complicated, since, especially for the large files, the reads and writes can occur in parallel to some extent.

http://www.tweaktown.com/reviews/3388/ocz_technology_agility_2_240gb_solid_state_disk/index10.html
 
Last edited:
You must log in or register to reply here.
Back
Top