6Gbps Performance

I installed the Intel SSD 510 in a 15-inch 2011 MacBook Pro as well as a 15-inch 2010 MacBook Pro to put together a 3Gbps vs. 6Gbps performance comparison. I turned to Xbench for some quick and dirty performance data:

SATA Performance—XBench 1.3
Intel SSD 510 250GB 3Gbps (2010 15-inch MBP) 6Gbps (2011 15-inch MBP) 6Gbps Advantage
4KB Sequential Write 157.8 MB/s 192.2 MB/s +21.8%
256KB Sequential Write 182.0 MB/s 257.1 MB/s +41.3%
4KB Sequential Read 32.5 MB/s 32.7 MB/s 0.0%
256KB Sequential Read 197.3 MB/s 315.6 MB/s +60.0%
4KB Random Write 47.8 MB/s 49.0 MB/s +2.5%
256KB Random Write 186.4 MB/s 260.9 MB/s +40.0%
4KB Random Read 14.5 MB/s 13.4 MB/s -7.6%
256KB Random Read 149.7 MB/s 207.3 MB/s +38.5%

As you'd expect, there's no real benefit to the new 6Gbps interface for random operations (particularly at low queue depths). Sequential speeds are much improved however. Xbench shows up to a 60% increase in performance in sequential operations.

You'll note that the absolute numbers are pretty low to begin with. A 128KB sequential read of the Intel SSD 510 on our desktop Sandy Bridge SSD testbed pulls nearly 400MB/s. On the new MacBook Pro we can't get more than 320MB/s.

Our sequential Iometer tests are run at a queue depth of 1 so there's no advantage there. The only explanation I can come up with (assuming Xbench's test is accurate) is that Apple may be aggressively implementing SATA controller power management under OS X. Capping the link's performance or aggressively putting it to sleep could reduce performance at the benefit of increasing battery life.

The other thing I noticed was that performance on the 13-inch MBP using Xbench was a bit lower than the 15-inch MBP. Take a look at these numbers:

SATA Performance—XBench 1.3
Intel SSD 510 250GB 13-inch 2011 MBP 15-inch 2011 MBP
4KB Sequential Write 155.3 MB/s 192.2 MB/s
256KB Sequential Write 184.8 MB/s 257.1 MB/s
4KB Sequential Read 30.4 MB/s 32.7 MB/s
256KB Sequential Read 201.8 MB/s 315.6 MB/s
4KB Random Write 49.6 MB/s 49.0 MB/s
256KB Random Write 183.9 MB/s 260.9 MB/s
4KB Random Read 13.9 MB/s 13.4 MB/s
256KB Random Read 144.9 MB/s 207.3 MB/s

I only noticed this with the Intel SSD 510, the Crucial RealSSD C300 and Vertex 3 both performed identically between the 13 and 15-inch MBPs. I'm not sure what's going on here at all, although I suspect that it's somehow related to the issues users have been having with some of these drives (more on this below).

SSD Recommendations

Where does all of this discussion about SSDs leave us? Unfortunately recommending an SSD for the new MacBook Pro today is pretty difficult but I'll try my best.

If you're the conservative type and just wants something that for sure works with little to no concern about absolute performance, the Apple SSDs are probably the safest bet. You'll get a drive that's much faster than a hard drive, fully supported by Apple and with TRIM support. Yes, that's right, OS X finally has TRIM support but Apple only enables it on it's own branded SSDs. To Apple's credit, given the number of problems I've seen with SSDs over the past couple of years it makes sense to lock down and only support drives you've validated. On the flip side however, Apple should be validating and working with controller makers to ensure all drives work under OS X. Making as much money as Apple does, I don't buy the "we didn't have the time/resources" argument.

If you are going down the Apple SSD path, at least the 128GB drive isn't super ridiculously priced, although I'm less comfortable recommending the 256GB version unless you can get it at $500.

Now if you want to get a faster SSD or actually take advantage of the 6Gbps interface, things get more complicated. I've heard reports of users having issues with the Intel SSD 510 and Crucial RealSSD C300. I've tested both drives as well as the OCZ Vertex 3 in three different MacBook Pros, and in all cases the drives worked perfectly. They were all detected as 6Gbps drives and all performed well. I should note that while I couldn't get the Vertex 3 Pro to work in the 2010 MacBook Pro, the Vertex 3 worked just fine in the 2011 MacBook Pro.

SATA Performance—XBench 1.3
13-inch 2011 MBP Crucial C300 256GB Intel SSD 510 250GB OCZ Vertex 3 240GB
4KB Sequential Write 239.0 MB/s 155.3 MB/s 319.9 MB/s
256KB Sequential Write 217.2 MB/s 184.8 MB/s 257.8 MB/s
4KB Sequential Read 35.1 MB/s 30.4 MB/s 33.3 MB/s
256KB Sequential Read 248.3 MB/s 201.8 MB/s 311.8 MB/s
4KB Random Write 175.0 MB/s 49.6 MB/s 247.8 MB/s
256KB Random Write 226.6 MB/s 183.9 MB/s 290.0 MB/s
4KB Random Read 19.1 MB/s 13.9 MB/s 21.1 MB/s
256KB Random Read 239.0 MB/s 144.9 MB/s 304.0 MB/s

SATA Performance—XBench 1.3
15-inch 2011 MBP Crucial C300 256GB Intel SSD 510 250GB OCZ Vertex 3 240GB
4KB Sequential Write 239.3 MB/s 192.2 MB/s 316.5 MB/s
256KB Sequential Write 218.8 MB/s 257.1 MB/s 282.0 MB/s
4KB Sequential Read 34.8 MB/s 32.7 MB/s 34.2 MB/s
256KB Sequential Read 245.1 MB/s 315.6 MB/s 306.7 MB/s
4KB Random Write 160.5 MB/s 49.0 MB/s 240.5 MB/s
256KB Random Write 227.5 MB/s 260.9 MB/s 311.3 MB/s
4KB Random Read 18.7 MB/s 13.4 MB/s 20.9 MB/s
256KB Random Read 238.2 MB/s 207.3 MB/s 303 MB/s

The Vertex 3 is the fastest drive out of the aforementioned three, but its availability and firmware maturity are both unknowns at this point. If you have to buy today and are ok with the chance that the drive may not work (given other experiences online, although I haven't seen problems), Intel's SSD 510 is likely a good runner up (at least for the 15-inch, the C300 seems to perform better on the 13).

As far as the reports of incompatibilities with these drives are concerned, I'm not really sure what's going on. I've been hammering on all of the drives, putting the system to sleep/waking it up, and haven't encountered any failures or high latency IO operations (stuttering) yet. That's not to say that these problems won't appear over time (I'm currently doing long term testing to figure that out now), but just that I haven't seen them yet.

If you are having issues with the Intel SSD 510, Crucial RealSSD C300 or anything else please email me (link at the top of the page) the following information:

1) What are the full specs of your MBP? Any upgrades?

2) Tell me about your SSD. Is it new out of box? Have you done anything to the drive? What model, firmware revision, etc...

3) Describe the symptoms of the issue—beachballs, data corruption, etc...? What do you have to do create the issue?

4) Is the drive detected as a 6Gbps drive or a 3Gbps drive?

5) Take me through your drive installation procedure, did you just pop it in, partition and install OS X?

6) Any visible damage to the SATA flex cable when you installed the drive?

7) Have you tried exchanging the SSD or MBP? Any difference in behavior?

We haven't seen any issues on three different 2011 models that we've been testing here extensively with the Intel SSD 510, Crucial RealSSD C300, OCZ Vertex 3 and OCZ Vertex 2. I realize a number of you are having issues so the more details I can get the better.

Apple's SSD Strategy The GPU Comparison
POST A COMMENT

197 Comments

View All Comments

  • Anand Lal Shimpi - Friday, March 11, 2011 - link

    Thank you for reading them, comments like this really do make it all worthwhile :)

    You wouldn't believe how much time was spent making sure Apple wasn't doing something funny with the max turbo frequencies. At the end of the day it was a non-issue, but we had to be sure.

    Take care,
    Anand
    Reply
  • Ryan Smith - Friday, March 11, 2011 - link

    Just to add some technical background to this, it's actually quite complex to get a CPU speed reading on modern CPUs. Mac OS X's Sysctl reports the base speed of the processor, regardless whether Turbo Mode is active or not. So on the 15" low-end QC model you will always see 2.3GHz.

    To actually read the instantaneous speed of any given core, you need to peek at the CPU itself and count the cycles - Intel actually has a handy document detailing an algorithm to do this(1). The issue with that is that it requires peeking at the Model-Specific Registers (MSRs), which require Ring 0 access; or in other words you need a broker at the driver level to do it.

    Linux already does this (/proc/cpu/0/msr), and on Windows it's fairly trivial to load a driver alongside an Admin-level application to do this(CPU-Z, etc). Under Mac OS X this requires installing an Extension (at least as far as I know) which gets messy. If you don't go through this process you'll never be able to read the core speeds accurately, which is why there's virtually no Mac software capable of this.

    Fortunately MSR Tools exists, and it has a 32bit extension to allow it to peek at the MSRs. The right answer of course is always the last answer you try, so this was only after trying several other ways of calculating the CPU speed and a couple different OS-agnostic benchmarks to try to rule out OS differences.

    1) http://download.intel.com/design/processor/applnot...
    Reply
  • tno - Friday, March 11, 2011 - link

    +1

    I've been planning to plunge into Mac ownership for sometime, especially with grad school looming I really want something that's more comfortable to work on than my netbook but still fairly portable. This review really helped me gauge whether it was worth putting in the extra cost for a 2011 13" MBP or settle for a discounted 2010.

    So am I all set? Hardly! Now I need to see what the 2011 13" MBA has to offer! I'm praying that cost stays roughly the same and a move to a ULV SNB leads to 12+ hour battery life and a similarly huge leap in performance as the move lead to in the MBP. I am a sucker for lightweight form factors.

    This article is also the first one to make me ever consider the 15" MBP. I have been fairly opposed to the bulk but the performance is quite something. If I went that route then I would probably have a C2Q, water-cooled, ATI and SSD driven rig to put up on AT forums. Taking offers!
    Reply
  • tno - Wednesday, May 04, 2011 - link

    Rezzing a dead thread! I bought the 13" MBP! $999 at MicroCenter, too good to pass up! So . . . who wants my rig? Reply
  • JasperJanssen - Saturday, August 06, 2011 - link

    I, on the other hand, have gone the other way. My MBA13 is being put together in China now. Reply
  • ltcommanderdata - Thursday, March 10, 2011 - link

    A great review. I do have some additional questions though. First, given Apple was the instigator of OpenCL, it'd be great if you could run some OpenCL benchmarks. Are the Sandy Bridge MacBook Pro's disproportionately faster than the Arrandale MacBook Pro to indicate that OS X has CPU OpenCL drivers that can take advantage of AVX? Probably not, and this will hopefully come with Lion. Given nVidia's GPGPU push can the HD 6490 still keep up with the 330M GT in OpenCL? How does the HD6750 do?

    http://www.bit-tech.net/hardware/graphics/2011/01/...

    "'[Intel] will be releasing OpenCL graphics drivers to developers during the course of 2011. [Intel] continue to evaluate when and where OpenCL will intercept various products"

    And is there secret Sandy Bridge IGP OpenCL support? Bit-tech got a quote from Intel that Sandy Bridge IGP OpenCL support was inbound sometime this year and if anyone would be motivated to get it done it'd be Apple.

    And finally, does Apple now support hardware H.264 decoding on ATI or Intel GPUs? Previously, only a few nVidia GPUs were supported in Snow Leopard, such that the Arrandale MacBook Pro actually had to power up the 330M GT to decode H.264 wasting power compared to the perfectly fine Arrandale IGP if Apple just wrote the drivers. Do the new Sandy Bridge have the ATI GPUs doing H.264 decoding now, is the Intel IGP supported, or in the worst case is no H.264 hardware acceleration available now that nVidia GPUs are gone? Perhaps lack of hardware H.264 decoding is what makes the FaceTime HD CPU usage so high? QuickSync is only accelerating the encoding phase?
    Reply
  • Anand Lal Shimpi - Friday, March 11, 2011 - link

    Some answers:

    1a) I don't know of any good GPU based OpenCL tests under OS X at this point. I'm not even sure if Apple's Intel HD 3000 driver supports OpenCL.

    1b) Intel mentioned SNB's GPU technically supports OpenCL however there are no plans to release a public driver at this point.

    2) Hardware H.264 decoding is enabled on the 2011s and it is used while FaceTiming, at least according to Apple.

    Take care,
    Anand
    Reply
  • ltcommanderdata - Friday, March 11, 2011 - link

    Thanks for the reply.

    http://www.macupdate.com/app/mac/33632/smallluxgpu

    In regards to OpenCL testing, most people in OS X seem to use SmallLuxGPU which is an OpenCL raytracing benchmark. I don't have much experience with it, but it might be worth a try.

    In regards to hardware H.264 decode, do you know if the IGP is doing it or does the discrete GPU still have to be powered up as in the 2010 Arrandale MacBook Pros?

    Thanks
    Reply
  • Anand Lal Shimpi - Friday, March 11, 2011 - link

    It's my understanding that the IGP can do the decoding, although note that while FaceTime is running the dGPU is enabled by default.

    Good call on SLG, I had forgotten about that :)

    Take care,
    Anand
    Reply
  • secretmanofagent - Thursday, March 10, 2011 - link

    Hello authors,
    On one of the pages, you mentioned this:
    "This isn't Mac specific advice, but if you've got a modern Mac notebook I'd highly recommend upgrading to an SSD before you even consider the new MacBook Pro. I've said this countless times in the past but an SSD is the single best upgrade you can do to your computer."

    Is there an article where you recommend the best update for my model? Should I even bother with the drive? I realize the X3100 is going to still hamper any sort of graphical performance, but wondering if it's worth the effort.

    Out of curiosity as well, would a Time Machine restore be possible if you update the drive?
    Reply

Log in

Don't have an account? Sign up now