During the NCIX Boxing Day sale, I picked up a Patriot Warp 32GB SSD for $50 after MIR. While I had read all about the stuttering issues associated with the JMicron controller-equipped MLC SSDs, I reasoned that for $50, the potential benefits outweighed the risk of a crappy user experience.
Boy, am I ever glad I went ahead with that purchase.
It’s been about 3 months with the SSD installed in my Dell XPS M1330, and for over two of those, I’ve been running Windows 7 Beta. I’ll save the Windows 7 discussion for a later time, but I’ll say for now, if this is the direction of the computing experience for the next couple years, I welcome it with open arms.
Many new solid state drives tout extremely high transfer rates, sequential reads over 200MB/s and writes well over 100MB/s. While these speeds are a couple times faster than a good 7200RPM drive, it’s the random access times that really put the SSD on top. With sub 0.5ms seek times, it’s easily an order of magnitude less than hard drives. In typical productivity work, linear transfers of large files back and forth isn’t the focus. Instead most of the interactions with applications and files involve many small transactions, which at first glance, seemed perfectly aligned with the benefits of solid state drives.
JMicron – Performance Killer
Unfortunately, it wasn’t that simple. The JMicron controller used by the vast majority of MLC solid state drives was utterly poor at random reads and writes. Furthermore, the built-in request cache is quite miserly, somewhere around 16KB, compared to 256KB for the Intel controller, used in the much more expensive Intel MLC drives. With poor random read/write performance practically no buffer to mitigate the issue, most of the ‘budget’ MLC drives became an order of magnitude worse than spinning hard drives when it came to some of the things that are most prevalent in standard computing use: web browsing, instant messaging, email. When the request buffer became full, nothing could happen and performance would drop off a cliff. Windows would freeze for seconds at a time.
So, with all the issues I just mentioned, you’re probably wondering why I went and bought one of these performance crippling drives, right?
Because everyone was in agreement on one point, when they work, they work hellishly well.
Solid State Drive Tweaking
I wasn’t on the bleeding edge of SSD adoption, so there were plenty of resources aimed at mitigating the detrimental effects of these drives. The OCZ forums were an especially active area of experimentation. Their first MLC drive, the Core was received with anticipation due to its very low cost, relative to SLC drives. However, the poor performance resulted in a significant outcry by the community. As a result of the commotion, there are posts like Vista 32/64 SSD Windows Registry tweaks. The main objective of these tweaks was to reduce the number and frequency of random reads and writes and deeply queued requests that caused the stuttering.
The work flows I use my laptop for aren’t highly multitasked. While I may have many applications open at once, I rarely perform multiple strenuous things at once. For example, I don’t compile code while extracting gigabytes of archives and encoding videos. Most of the time, my actions are focused. Consequently, the causes of stuttering wasn’t very prevalent in my use case. Still, I didn’t bother trying the system out without the tweaks. My first order of business after installing the OS was to disable indexing, SuperFetch, and apply the registry tweaks mentioned. I could have gone further, such as working with Windows SteadyState, but didn’t want the tweaks to get in the way of how I used my computer.
We’re finally at the section that really means the most of all, overall performance. With the tweaks applied, it’s good, it’s very good. The Dell XPS M1330 the Patriot Warp currently resides in used to have a 120GB Hitachi 5400RPM drive. Here’s an ATTO bench of a 2+ month old system with the SSD.
Compared to the Hitachi 5K160 drive it replaced, it’s slightly slower at block sizes below 8KB; however, whereas the Hitachi topped out at around 35MB/s for sequential reads and writes, the Patriot continues on through to 130MB/s read and around 80MB/s writes. What does this mean in terms of real world usage?
In terms of launching applications, my laptop (1.8GHz Core 2 Duo, 3GB RAM) is now significantly faster than my desktop (2.4GHz Core 2 Duo, 4GB RAM, WD 640GB 7200RPM), with both running Windows 7. Here are some example application load times, timed from the launch until the application reaches a usable state.
Adobe Photoshop CS3
Desktop – 6s
Laptop w/ SSD – 3.5s
Microsoft Office PowerPoint
Desktop – 2.6s
Laptop w/ SSD – 2s
Microsoft Office Outlook
Desktop – 4.5s
Laptop w/ SSD – 2.5s
Desktop – 2.5s
Laptop w/ SSD – 1.5s
The difference can be significant. Furthermore, typical usage is more fluid, and applications are a bit more responsive. A friend, upon using my laptop for a few minutes got the feeling that the applications he was opening for the first time were already open somehow. It was as though he were simply switching applications and not launching them at all. There’s a sense of instantaneity.
While I applied many tweaks to improve performance right after installing the operating system and didn’t have a baseline to compare to, I was still able to witness first-hand the impact they had on the system. A couple weeks ago, I noticed extremely frequent stuttering in Firefox. At first, I couldn’t understand it. I thought it might have been caused by the performance degradation claims made about solid state drives. However as it turns out, SuperFetch and drive indexing had both been re-enabled. After disabling the two features, performance was back to normal again and stutters became so infrequent they no longer affect everyday usage.
Moral of the Story
What should you take away from this user experience? Solid state drives, even budget MLC ones, have the potential to vastly improve system performance. With the tweaks applied, stuttering is no longer an issue on my laptop. If there’s one thing I regret slightly is the anemic size of the SSD I purchased, 32GB, but for the price, $50 after MIR, I can’t complain about that much. The experiment was a huge success and I’m looking forward to upgrading my laptop to a larger SSD and my desktop to one.