G.Skill Titan 128GB SSD Reviewed

http://www.xcpus.com/GetDoc.aspx?doc=119&page=1

We took this drive and ran it through the ringer, comparing its performance to the Intel X25-M.

Some interesting results, to be sure.


We also confirmed our write results on the Titan with other companies looking at the drive, to make sure that our data wasn't out of line with what other people are seeing out of this drive.

 

nice and useful read. thanks!

 

Thanks.

Honestly, I was really hoping for more out of this drive. While it didn't perform badly, it most tests it was beaten by the X25-M.

Guess we will have to wait and see what the OCZ Vertex does, or for 3rd generation MLC drives in the fall.

 

Very interesting. Thanks.

 

The Intel drive is also a lot more expensive... $400 for 80GB ($5/GB) vs. $500 for 256GB (<$2/GB).

 

The G.Skill Titan HD Tune Write graph looks really scary. I'm kinda surprise that with the new dual JMF602B controllers, there is still huge jumps in write speed.

Nice review but I am still glad I got the Corsair over the G.Skill.

 

Your Everest Linear Write results are pretty much as I'd expect for a drive that had already been heavily used; the NAND chips appear to be using erase block sizes on the order of 1MB. For writes smaller than that, you're experiencing the write amplification delays because the controller has to read the current contents of the block before writing the new data. For writes of the block size or larger, it can just overwrite a block without reading it first. I suspect if you continued to 2MB and 4MB you'd still see high throughput.

 

The JMicron controllers still suck, basically.

 

So does it stutter?

 

Yes. Perhaps less, but bottom line is that it still stutters.

 

Yes. The perceptible stuttering is greatly decreased, but the drive still leaves you at a few times twiddling your thumbs.

 

You'd think an SSD controller designer would be smart enough to include at least as much cache in their controller as the full size of an erase block... Having spare pages is one thing; being able to coalesce many small writes into a single block would have made a lot more sense.

 

I find it interesting that the review comments on "Mediocre write performance with small file sizes" while the benchmarks indicate that both SSDs outperform the reference disk at all block sizes above 4KB.

How many writes are below 4K in size for a typical user?

 

Then you have to be very careful to make sure that you maintain cache coherency in the event of a power failure or that the same block is re-requested.

It can be done, but it's a giant pain!

 

Thanks, that was a really nice review. I'm not sure we should call what the Titan does stuttering though. In my experience, while the X25-M was certainly consistently snappier than the Titan, both drives in terms of responsiveness were heads above the single JMicron SSDs and Velociraptors I've used.

I agree with your recommendation in that the Titan should still be on every enthusiast's list of possible upgrades.

 

so i won't be disappointed if i cave and buy a 60gb apex (don't need all the space/price of the 128gb titan)? right now i'm running 200gb 7200rpm hdd. did you guys have to do any tweaks at all to the titan for an OS drive?

 

In theory yes, but bit-tech notes that the Apex's performance isn't consistent and that the Titan is a much more well rounded SSD:

http://www.bit-tech.net/hardware/storage/2009/02/09/ocz-apex-120gb-ssd-review/7

 

Lots, actually. This is why firefox bogs down on the drive. The FF cache reads/writes tons of small images and index files as you browse.


FF usage was my single biggest beef w/ the Titan. Literally annoyed to the point of wanting to scream.

 

No tweaks what-so-ever. We cloned the OS over from a drive, and that worked perfectly well also.


For the money, the Titan is a GOOD buy, but if you have a little extra $$$ to spare, I'm 100% convinced that users will be happier w/ the X25-M.

 

HD Tach is absolutely horrible for SSD testing. I strongly recommend you disregard any benchmark of SSDs that uses HDTach. For the reasons pointed out in my intro, it's simply a BAD TEST of SSDs.

 

Bit-tech makes note of that in their review and compensates via FC-Test as can be read here:

 

Gotcha, gotta admit I didn't read all of their review.

 

So would it be safe to assume that the 60gb apex would make internet browsing with internet explorer just as anoying as the titan with firefox?

 

That would be my assumption, yes.

 

by definition, it' can't be done, as the cache is not part of the storage pool itself. it would cache non-written data: a.k.a. it would cache unsaved data. if anything happens (not simply power failure, but not finding a free cell to write, what ever, data corruption on the ssd, etc. simply any sort of failing disk write) then the disk lied to you and the data will be lost.

data has to be written directly, and only when done, the ssd is allowed to report "done". else, it could result in silent corruption and you could sue g.skill to death

 

but y does it have to report "done" when it is written to cache? could it be programed to only report done when it is actually written to the cell but still use the cache for it's buffering benefits?

 

there are no buffering benefits, then. if the os still has to wait for the actual write. the write can't get any faster than it is without cache. so the cache has no use, then.

it can be used for a lot of different things, still. but simply caching the write-data is of no use, then. it would, at worst, make writes even slower, as it would write to cache, wait a second (why ever ), and then write to disk, and report "done".

the only thing i can see is this:
write (tiny data) to cache and to disk on a new, free sector. report okay when on disk.

collect such small writes in the cache (while writing each to disk on a free sektor, again).

once you have a full big chunk in the cache, write that on a new cell on disk, and free all the small-filled writes from before away from the disk.

clear cache.

and this is more or less, what the intel drive does.. (+funky magic tech, as always).

this is sort of the best one can do. write chaotic into free space, and then recollect to write cleaned up again. and for this recollecting, one can use a cache.