Apple's MacBook Pro with 128GB SSD: Performance and Battery Life Investigated
by Anand Lal Shimpi on April 15, 2008 4:20 PM EST- Posted in
- Mac
Better Application Performance? Not Exactly
If battery life didn't improve, surely it must be a sacrifice made in the name of better application performance right? Not exactly.
To understand why, we must first look at the synthetic performance of the drive:
| XBench Disk Test (Sequential) | MacBook Pro (Hitachi 5400RPM) | MacBook Pro (Memoright SSD) |
| Uncached Writes (4K) | 53.5MB/s | 61.7MB/s |
| Uncached Writes (256K) | 48.2MB/s | 73.0MB/s |
| Uncached Reads (4K) | 11.2MB/s | 10.2MB/s |
| Uncached Reads (256K) | 49.8MB/s | 68.2MB/s |
In sequential access, small block reads and writes either don't improve at all or improve by an amount that's not huge (15% for uncached writes, but that's just an improvement in the performance of your disk subsystem - not the entire machine, expect real world performance improvements to be some fraction of that). Larger accesses are a bit more favorable, with reads and writes improving by 37% and 51% respectively.
Most single-application desktop usage models are actually very heavy on sequential disk access, and in these situations you won't see the biggest performance benefits from a SSD - even something as fast as the Memoright.
Looking at the random reads however tells another story:
| XBench Disk Test (Random Access) | MacBook Pro (Hitachi 5400RPM) | MacBook Pro (Memoright SSD) |
| Uncached Writes (4K) | 0.92MB/s | 1.11MB/s |
| Uncached Writes (256K) | 22.4MB/s | 27.9MB/s |
| Uncached Reads (4K) | 0.47MB/s | 9.98MB/s |
| Uncached Reads (256K) | 19.5MB/s | 79.1MB/s |
While random writes offer a ~20% performance improvement, random reads range from 3x - 20x the speed of a mechanical disk. Now since most single-application usage patterns tend to be sequential in nature, we don't see these incredible performance gains in many of our scripted tests - however, in actual usage you can easily feel a bigger difference.
This is where the whole: once you go SSD, it hurts to go back statement from the introduction of this article comes from. Within a single application, performance may not improve a ton, but your day to day usage experience will be a lot smoother.
Let's take a look at some of those application tests:
| Application Tests in Seconds (Lower is Better) | MacBook Pro (Hitachi 5400RPM) | MacBook Pro (Memoright SSD) |
| iPhoto Import | 72.1 seconds | 62.2 seconds |
| iPhoto Export to Web | 116 seconds | 119 seconds |
| Pages Export to Word | 27.4 seconds | 26.9 seconds |
| Keynote Export to PPT | 18.3 seconds | 16.0 seconds |
| Word 2008 - Compare Docs | 69.0 seconds | 62.9 seconds |
| PowerPoint 2008 + Word 2008 - Compare Docs & Print to PDF | 82.8 seconds | 85.8 seconds |
| Adobe Photoshop CS3 - Retouch Artists Speed Test | 44.6 seconds | 42.5 seconds |
The biggest performance increases here are in the iPhoto Import, Keynote Export and Word Compare Documents tests. Performance went down slightly in the iPhoto Export and PowerPoint tests, but the drops were so small that they can be considered insignificant.
The major take-home point here is that performance didn't go up all that much in most of these tests, with the biggest gain being just under 14% that's a marginal improvement given the nearly $4,000 price of admission.
But there is just one more thing...
Facebook
Twitter
RSS
39 Comments
View All Comments
Denithor - Wednesday, April 16, 2008 - link
When will these become available/affordable for desktop use? HTPC comes immediately to mind, but I would like one for my gaming rig if it yields a "snappier" system for a moderate cost.mindless1 - Friday, April 18, 2008 - link
You must be kidding. Minimal to no gain in sequential access, improvement primarily in random access, limited capacity, and extreme price per GB make this about the worst choice possible for a HTPC.Regardless, if that's what you want go ahead and do it, drive rail adapters to use 2.5" in 3.5" bays are not expensive or hard to find. You could even squeeze two or three drives into one 3.5" bay so you have a HTPC with $12,000 spent on storage instead of a $80 mechanical drive.
just4U - Wednesday, April 16, 2008 - link
Is it possible to instal Windowso n a flash drive. You know, one of those 16/32 meg jobs. This article has got me curious.. :)strikeback03 - Thursday, April 17, 2008 - link
I assume you mean 16/32GB, and yes, I believe I saw instructions for that over at mp3car.com.mindless1 - Friday, April 18, 2008 - link
Instructions? Not a complex process.1) Buy CF3 or CF4 spec'd CF card and CF-IDE adapter. CF card performance is lower than on a good SSD so staying with PATA/IDE interface is not a bottleneck.
2) Plug card into adapter, plug adapter into system.
3) You're done, there is now no difference beyond having mechanical drive instead, although if SSD is not using SLC flash chips you might want to decide how to limit # of writes to it from pagefile, temporary browser files, etc.
strikeback03 - Friday, April 18, 2008 - link
I just glanced at the instructions as I am not building a carputer yet, but IIRC a lot of it was optimizing the pagefile and other little writes.Nihility - Wednesday, April 16, 2008 - link
That needs to be fixed, no reason for flash to take more power than a hard drive, maybe they can power off some of the flash that is unused until it's needed? If it doesn't even increase battery life then what's the point? Resilience and random seek times are nice but battery life is the main concern on a mobile platform.mindless1 - Friday, April 18, 2008 - link
It's the controller, bridge and cache that use the power, these flash chips don't have to be recharged.Keep in mind that while battery life is important, and power consumption of an SSD will go down over time, they still aren't one of the larger consumers of power. Ultimiately if runtime is most important the area to focus on is designers who mistakenly assume a smaller device footprint is more important than runtime, thus squeezing in a smaller battery (capacity).
iwodo - Tuesday, April 15, 2008 - link
Intel 's SSD promise doubling the performance of current SSD Drive. I cant wait to see it.I wonder would the ARM7 chip be the limiting factor here?
skiboysteve - Tuesday, April 15, 2008 - link
"two areas of inefficiency: the drive isn't a native SATA device and it uses a FPGA instead of a custom IC for some functions."this is incorrect. using an FPGA instead of a custom IC makes no difference in performance whatsoever. the difference is in cost. there is a lot of research into cost/benefit of using an FPGA instead of a custom ic and it all boils down to volume. obviously, they dont have high enough volume to necessitate a custom IC.
but, an fpga configured to behave exactly like what your custom IC would behave like ... are the same thing. only difference again, is price.
some point might astutely point out that a custom IC can be clocked higher, but i very much doubt that advantage is applicable here.