The same questions are often asked on these forums: - Should I get the 2.2 GHz or the 2.4 GHz MBP? - Should I get the i5 or the i7 MBA - Is the MBA powerful enough for me or should I get a MBP? - Should I upgrade the RAM or get a SSD - Should I upgrade the CPU or RAM? - Do I need more than 4 GB RAM? - Which SSD should I get ? - Why is the MBA SSD slow …and so forth. So let me try and answer them as simply as possible: Q: Should I get the 2.2 GHz or the 2.4 GHz MBP? A: If you use the computer for productivity (Word, Excel, Powerpoint, Keynote, Numbers, Mail etc) - CPU makes very little difference. First Upgrade: SSD. Second Upgrade: RAM The 2.4 GHz 15" MBP has a faster GPU (6770m vs 6750m) so I would only recommend this upgrade if you are planning on playing games often. You can expect an increase of around 2 to 10 frames per second with this increase, depending on the game. Only you can determine if that is worth it. ----- Q: Should I get the i5 or the i7 MBA? A: Most users won't see a difference in their daily activities, unless you're rendering 3D graphics or encoding movies often ----- Q: Is the MBA powerful enough for me or should I get a MBP? A: The MBA is powerful enough for most users. Any extra CPU power is like chocolate sauce on a nice ice cream - a bonus. The MBP has firewire ports and an Gigabit Ethernet port. Do you need them? The MBP RAM and HDD can be upgraded easily. The MBA SSD can be upgraded (expensive) but the RAM cannot. If you want to play games and you can afford it, then get a MBP with a discrete GPU. 4 GB RAM and the normal HDD are good enough and upgrading won't make the game any smoother. ---- Q: Should I upgrade the RAM or get a SSD? A: Most users will see the benefit of upgrading from a HDD to SSD before the benefits of adding RAM. However, if you're running out of RAM and using a mechanical HDD, the performance penalties are severe. ---- Q: Should I upgrade the CPU or RAM? A: Most users will see the benefit of more RAM rather than a faster CPU unless you use the computer for tasks that need CPU power, for example rendering 3D models, professional audio creation and encoding movies. If you use the computer for such activities often then it's worth to upgrade. Upgrading to the 2.4 QC 2011 MBP is not worth it for most people (apart from the GPU update as above). The 2.5 GHz QC upgrade makes even less sense ---- Q: Do I need more than 4 GB RAM? A: Check your Activity Monitor and see if you have significant Page Outs versus Page Ins. 4 GB is perfectly fine for Lion, but since upgrading to 8GB with after-market brands is so cheap, I would do it anyway. ---- Q: Which SSD should I get? A: It won't make a difference to most people. Get the one that's good value for money and has a good reliability track record. Refer to the SSD Buying Guide sticky in this sub-forum ---- Q: Why is the MBA/MB SSD supplied by Apple slow in benchmarks that I've seen? A: The Apple supplied SSD is fast compared to a mechanical HDD and for most users it is "fast enough". In other words most users won't see the difference or the benefits when using a "fast" SSD. The differences between "fast" and "slow" SSDs in most normal daily tasks that involve data reads or writes are negligible. ---- If you want a more detailed explanation, then read on and check out the benchmarks. These questions are not always easy to answer without knowing all of the requirements and the particular user’s workflow. Simply looking at synthetic benchmarks does not always give one all of the answers they are looking for. If one computer scores 10,000 in Geekbench and another scores 5,000, does that mean one is twice as fast as the other? If a disk speed benchmark shows that one SSD writes and reads data at twice the speed of another, does that mean it will be twice as fast? The simple answer is no. These benchmarks show the absolute top speed a particular component of your computer can achieve. Think of it like this: if you compare a Ferrari FXX and a BMW M3 then in most conditions the Ferrari will be much faster. It will certainly be faster in a ¼ mile drag and around normal circuits. However, when you’re using your computer you’re not always doing tasks like racing around a racetrack. Sometimes you’re just going to get some groceries on roads with police and 30 mph speed limits. Just because the Ferrari is three times as fast around a racetrack it does not mean that it will be thrice as fast to get your groceries. I’ve tried to show what happens when you run the same workflows on different computers and different configurations in the real world. I’ve created a number of tests that are fairly typical tasks. One important thing to note is that these tests were run exclusively with nothing else, apart from background system tasks, running. That’s the big difference between a computer like a Ferrari and computer like a M3. One can do a lot more at the same time on a faster computer with faster storage and more random access memory. The lineup: - 2009 MBP 13” 2.53 Core 2 Duo upgraded to 8 GB 1066 MHz RAM and a 500 GB Seagate Momentus XT HDD - 2011 MBA 13” i7 1.8 GHz 256 GB SSD (Samsung) 4 GB RAM - 2011 Mini Server i7 Quad 2.0 GHz (I7-2635QM) 750 GB WD Scorpio Black 7200 RPM HDD 4 GB 1333 MHz RAM - 2011 Mini Server i7 Quad 2.0 GHz (I7-2635QM) 750 GB WD Scorpio Black 7200 RPM HDD 8 GB 1600 MHz RAM - 2011 Mini Server i7 Quad 2.0 GHz (I7-2635QM) 240 GB OCZ Vertex 3 4 GB RAM - 2011 Mini Server i7 Quad 2.0 GHz (I7-2635QM) 240 GB OCZ Vertex 3 8 GB 1600 MHz RAM The quad core CPU in the mini is the same as the baseline early 2011 15” MBP so it makes for a nice comparison. Unfortunately I do not have a late MBP to use in these tests, however the point is to look at the relative performance of these machines to draw conclusions based on that. All tests were created in Automator and run multiple times after clean reboots or caches were cleared, except for the tests with the Momentus XT since I do not know how to clear its cache and how persistent it is. In order to get consistent results, I discounted the highest and lowest benchmarks and averaged the rest of the results. I made sure that the results could be replicated a number of times to account for any spikes or anomalies during testing. Disk Copy Test – Copy 1 5.1 GB File Shorter bars are better. Scores are in seconds. This test involves copying one 5.1 GB mkv movie file from one location to another. From a hard drive point of view the data is incompressible and it is a sequential read and write. As expected the OCZ Vertex 3 pulls ahead with the MBA’s Samsung SSD coming in 2nd place 11 seconds behind. The extra RAM did not make any difference in this scenario. What’s interesting, and surprising to me, is how much slower the Seagate MomentusXT is in comparison to the Western Digital Scorpio Black. If your workflow revolves around large incompressible data then a SSD drive or multiple HDDs in a hardware RAID configuration are a must. Geekbench 64-bit test You can download and run Geekbench on your own machine or you can check out the average scores for each Mac model on the Geekbench website. I’ve run these tests to highlight a couple of things. Geekbench tests how fast your CPU and RAM are. It has nothing to with your storage speeds. Longer bars are better. The poor, old MBP struggles in last place and it’s amazing to see how CPU technology is increasing for the little MBA to get such a respectable score. You’ll note that the Mini with upgraded RAM scores more than the stock Mini. The increase in score has nothing to do with the fact that we’ve increased the amount of memory. It is because the 8 GB memory is 1600 MHz (yes, your Sandy Bridge Mac supports it) as opposed to the stock memory in the 2011 Mac lineup, which runs at 1333 MHz. Without going into the technical details of it all, memory with a higher clock rate runs faster, as long as your CPU and/or motherboard supports it. I must add that 200 points in Geekbench is not a huge difference in the real world. During these tests, results may vary by as much as 50 points above or behind the average score of your computer. However, if you look at the details of the benchmark, you’ll see that all of the points were gained in the memory tests, as expected. Cinebench R11.5 CPU This is a great benchmark that you can download and run yourself from the Maxon website. The CPU test uses all available cores. You can read more here: http://www.maxon.net/downloads/cinebench.html Longer bars are better. Scores are in points. The quad core in the Mini Server easily stretches its legs to beat the competition and the 2009 MBP comes in last. What about the Geekbench scores? Shouldn’t the faster RAM help? No, faster RAM will not make your CPU faster. It will make certain things faster, as we shall see later, but in this use-case it does not help. Having more RAM also does not help since the test is not constrained in this regard. Cinebench R11.5 OpenGL This benchmark tests your graphics subsystem so basically it will show you how powerful your GPU is in comparison to others. Gaming performance and how smooth 3D models are to navigate will be some of the things affected by your GPU’s speed. There is also another discussion around how some image editing suites attempt to use your GPU to speed up certain things, but that is not something I’ll be getting into in this post. Longer bars are better. Scores are in frames per second. The Nvidia 320M in the 2009 MBP loses out to the integrated Intel HD3000 in this test. The Intel graphics in the mini are a smidgeon faster in comparison to the MBA. Looking at the detailed specs for each of the CPUs, I do not see anything to explain this as the HD3000 runs at the same frequency. Perhaps someone has some views on this? The mini with 8GB of Kingston 1600 MHz RAM comes out on top by around 0.75 of a frame per second – not exactly a noticeable difference. It’s important to bear in mind that the difference is not because of the extra amount of RAM. It’s because the Kingston RAM is faster than stock and the integrated Intel GPU uses the system RAM. Encoding a 1080p 120 seconds clip This test uses the HandBrake command line tool to encode a 1080p clip that is 120 seconds long using the High Profile preset. The clip is a BluRay rip in mkv format and the output is an iTunes friendly m4v file. Encoding is all about the CPU power as we will see in the results below. Shorter bars are better. Scores are in seconds The quad core CPU wins by a quite margin as expected. It’s difficult to explain the 2 seconds difference between the configurations, but I can reproduce it constantly. I assume it has to do with the final operations at the end of the encode operation where various clip info is written into the file. Productivity – Combine Excel Files (11.6 MB each) The test uses 2 files and each of them has 9 sheets. Each sheet has about 40,000 rows of data across 3 to 5 columns. Automator is used to open the files and then they are combined and saved as a new file. This is probably about as intensive operation as an office bound excel user would undertake before considering a database-based solution. Shorter bars are better. Scores are in seconds. Unsurprisingly the SSD equipped configurations pull ahead. What’s interesting to note is the miniscule difference between the “slow” Samsung SSD in the MBA and the “fast” Vertex 3 SSD in the Mini. Extra RAM does not help in this case because we never run out of RAM. However, when using the stock 7200 RPM HDD, extra RAM did reduce the time by 3 seconds. There was no difference between the SSD equipped models and I don’t expect any caching during this test so I have to discount the difference as an anomaly, unless someone has a different opinion and would like to share it. I am going to go back to a car analogy and use it to try to explain what we’re seeing. In its simplest terms a car can be broken down into an engine & gearbox, which drives the wheel, which are shod in tyres that are in contact with the ground. A car also has suspension to keep as much of the tyre as possible in contact with the ground. So these 3 basic components (engine, tyres and suspension) are they key characteristics that determine performance. Of course there are many other variables like mass, but let’s not sidetrack ourselves and keep this simple. The engine is the most important, but a powerful engine without good tyres or on old worn-out tyres will not be very good to drive. Suspension is the other component. Upgrading the suspension will make the car better but all 3 of these have to work in symmetry for the best results. A car with a powerful engine, sticky tyres and good suspension will be faster than the same car with worse suspension. When using Excel the storage system is your engine, the CPU is your tyres and the RAM is your suspension. Excel does not require much CPU power. In fact, most of your tasks do not require much CPU power and a quad core processor is overkill for most people. In other words: not everybody needs a V8, but it’s nice to have. Productivity – Sort Excel (11.6 MB each) We’re using the same file as in the previous test, but in this time the file is open and then we sort by column A, then B, then C in the first sheet. The file is not saved and then we quit Excel. This is another fairly typical task in Excel. Shorter bars are better. Scores are in seconds. Again the Vertex 3 equipped Mini shoots ahead with the MBA in second place. It’s worth noting that I’ve rounded the times up or down for readability, but the real differences were around 0.7 of a second. The extra RAM in the mini makes no difference and the results were nearly identical. This is also one of the few tests where the 2009 MBP managed to keep up with the Mini in its default (7200 RPM HDD) configuration. The majority of the time in this test is taken by opening the file in the first place. The actual sorting operation uses the CPU, but even the Core 2 Duo chip in the MBP is good enough for this since it is hardly a taxing task. When allowing the test to run cached it drops down to 7.5 seconds for the Mini SSD and MBA systems. The faster SSD makes no noticeable difference in these types of daily tasks when compared to a slower SSD. Resize 140 images and create thumbnails This test resizes 140 images into 1280 x ??? (proportional scale) and then creates 128 pixel size thumbnails. Each image is around 3.8 MB large and 3800 x 2533 resolution. A fairly typical task if you’re doing some amateur photography. Shorter bars are better. Scores are in seconds. As usual, the Mini with a Vertex 3 SSD takes the lead. Looking at the Mini with a mechanical hard drive the extra RAM does speed up the operation by 4 seconds so there is some caching going on. The 2009 MBP is not far behind the Mini server, despite its age and slow CPU. The upgraded RAM and Momentus XT HDD are helping to keep it in the running. Again the difference between the dual core + slow SSD MBA and the quad core + fast SSD Mini is minimal at 2 seconds. The CPU takes a backstage in this test and the key factor is storage system speed. If the workload increases (number of images / resolution of images) then a faster SSD and more RAM will help. As long as your computer is not running out of RAM, having more generally does not make it faster. In this case the storage is the engine, the RAM is the tyres and the CPU is the suspension. Resize 140 images and add watermark This test is similar to the previous one, except that instead of creating thumbnails, it adds a watermark to the top left corner. The original images used for input are the same as in the previous benchmark. Shorter bars are better. Scores are in seconds. This is actually the most intensive benchmark that I ran in terms of the storage and RAM. CPU plays an insignificant role. The test uses Pixelmator actions in Automator to add the watermark and this uses a lot of memory resulting in large amounts of page outs and much swapping to the HDD or SSD. When we run out of memory we end up in a situation of disk thrashing because the virtual memory subsystem is paging in and paging out large amounts of data at the same time as the disk is trying to read and write the data required for this user operation. The Mini with 4 GB of RAM and a 7200 RPM HDD suffers badly during this test and is more than twice as slow as the 2009 MBP, which has been upgraded to 8 GB of RAM. If your typical workflow involves using large amounts of memory (restart, check your page outs versus page ins after your typical work day – search for more info on this forum if you’re not following) , then upgrading the RAM is the best upgrade you can do. Take a look at the the performance of the Mini with 8 GB of RAM and 7200 HDD versus the Mini with 4 GB of RAM and a Vertex 3 SSD - the Mini with more RAM is faster, despite its slower, mechanical HDD. Even with the excessive disk usage of the virtual memory system during this test, the slow Samsung SSD in the MBA again manages to keep up with the much faster Vertex 3 and is only 3 seconds slower. The clear winner is the Mini with 8 GB of RAM and SSD. What would be interesting to see is how a Mini with 16 GB of RAM and a mechanical HDD would perform. Considering the fact that we end up with around 11 GB of page outs after this test, even when using 8 GB of RAM, I believe that it would be faster than the 8GB SSD equipped mini. As I’ve said before on this forum, 1333 MHz RAM is about 20 times faster than the fastest consumer SSDs.