Macbook Pro RAM: 1333 vs 1600, 4vs8vs16GB

[ewestlund]

Hi folks. I have never seen a guide that compared 1333 vs 1600 RAM as well as various size configurations in a Macbook Pro. I wrote one up. It's online here, but I've pasted the gist here if you don't want to click away. Pictures and tables are included in the web link if you really wanna get into the nitty gritty.

Background

Some quick internet searching told me that later model Macbook Pros could, in fact, handle 16GB of memory (despite Apple saying 8GB is the limit) and can run DD3 Memory at 1600 MHz (despite Apple saying 1333 MHz is the limit). So, I ordered 2x8GB of Corsair Vengeance PC1600 notebook memory at $57.99 a piece. Tack on a $10 NewEgg discount, and the grand total was $104.98. Only $1.50 more per GB than I paid to upgrade my machine from 4GB to 8GB in December 2011. For whatever reason, it was cheaper to order 2 single 8GB sticks ($115.98 before discount) than the dual-pack of the same memory ($119.99). You can download the data used in this analysis.

Test Machine

Late 2011, 17″ Macbook Pro with quad-core 2.5GHz Sandy Bridge i7 CPU.

Benchmark Software

Geekbench 2.3.1 for Mac OS X x86 (64-bit)

Test Configurations


4GB (1x4GB) 1333MHz, Corsair (CMSO8GX3M2A1333C9)
4GB (2x2GB) 1333MHz, Samsung (Stock Apple memory)
8GB (1x8GB) 1600MHz, Corsair Vengeance (CMSX8GX3M1A1600C10)
8GB (2x4GB) 1333MHz, Samsung (Stock Apple memory)
16GB (2x8GB) 1600MHz, Corsair Vengeance (CMSX8GX3M1A1600C10)

Procedure

I ran Geekbench 64-bit for the Mac 5 times on each of the above configurations to account for random error. (Admittedly, on configuration 3 I forgot to record one of the results and couldn’t be bothered to re-install the memory, so n=4 for configuration 3.) While Geekbench gives about 20 different numbers, I recorded 5: the overall composite and the four summary statistics: Integer Performance, Floating Point Performance, Memory Performance, and Streaming Memory Performance.

Analysis

1. More memory improves Geekbench overall scores: about 1.5% increase from 4GB to 8GB and 1.2% increase from 8GB to 16GB
2. Dual channel makes a big difference. All dual channel configurations outperform the single channel configurations: 20-30% for Memory performance and 35% for Stream Memory performance.
3. Dual channel configurations with more memory do better than dual channel configurations with less memory on the order of 3% for Memory performance and 8% for Stream Memory performance
4. More memory improves floating point performance marginally.
5. The 16GB configurations has slightly more variance between measures on memory related tests and slightly less variance on processor related tests. Theoretically, that means more memory equals more consistent processing performance relative to lower capacity configurations and slightly less consistent memory performance relative to lower capacity configurations. This probably has absolutely no real world or practical meaning, but the claim stands up to scrutiny in statistics class.

Conclusions

1. Do not run in single channel mode. It’s inefficient.
2. One can expect a similar percentage increase in benchmarks when moving from 4GB to 8GB as from 8GB to 16GB. That is to say, there are diminishing returns as memory increases in size, but there are returns nonetheless.
3. Unsurprisingly, increasing memory improves memory benchmarks more than processor benchmarks, but there is a small improvement on the processing front.

Cautions

1. I cannot get a good handle on 1333 vs 1600 performance on synthetic benchmarks because I have to compare dual channel to single channel performance. Nevertheless, other benchmarks suggest you may notice a slight speed boost on very intensive work, but moving from 1333 to 1600 is not going to make any noticeable difference for regular human beings. Get the 1600 RAM to feel good for being on the bleeding edge, and then laugh at yourself for the absurdity of it all. The cost difference is marginal, in any event.
2. These benchmarks are for a late 2011 i7 quad core. Results may, of course, vary, but I suspect these types of changes will remain relatively consistent from Mac to Mac. I do not know how far back one can go and still run the memory at 1600 MHz, however.

Final Thought

Benchmarks aside, 16GB feels faster than 8GB and 8GB feels faster than 4GB. This is especially noticeable when many applications are running. The upgrade is worth it for the cost.

[iamthedudeman]

So essentially the late 2011 models can take the same memory as the 2012 models?

[tninety]

Originally Posted by iamthedudeman

So essentially the late 2011 models can take the same memory as the 2012 models?

Yes, they are architecturally very similar.

[ewestlund]

Originally Posted by iamthedudeman

So essentially the late 2011 models can take the same memory as the 2012 models?

Well, the Retina Displays have the memory soldered on, so technically -- no. The non-retinas are the same form factor... so, yes, you could pop in and out the same memory and it will run and register at the correct speed.

See below:

[ivoruest]

Originally Posted by ewestlund

Well, the Retina Displays have the memory soldered on, so technically -- no. The non-retinas are the same form factor... so, yes, you could pop in and out the same memory and it will run and register at the correct speed.

See below:

Image

Maybe the MBP 2012 Non-Retina can handle 1866Mhz memory?? Ivy Bridge is supposed to support it right?

[ewestlund]

Originally Posted by ivoruest

Maybe the MBP 2012 Non-Retina can handle 1866Mhz memory?? Ivy Bridge is supposed to support it right?

My understanding is that it will work with the late 2011s (and maybe before) too. But, two things:

1. No 1866 8gb DIMMs yet, so you're stuck at 8GB, no 16GB.
2. Everything I've read on desktop machines, at least, finds almost no difference at all between 1866 and 1600. The best articles have concluded to go with 1600, because it offers the best balance between speed and latency and cost. (Speed and latency tend to be inversely related.)

I wonder if any wild ones have tried 1866, I'd be curious to know if it works and/or is stable.

[ivoruest]

Originally Posted by ewestlund

My understanding is that it will work with the late 2011s (and maybe before) too. But, two things:

1. No 1866 8gb DIMMs yet, so you're stuck at 8GB, no 16GB.
2. Everything I've read on desktop machines, at least, finds almost no difference at all between 1866 and 1600. The best articles have concluded to go with 1600, because it offers the best balance between speed and latency and cost. (Speed and latency tend to be inversely related.)

I wonder if any wild ones have tried 1866, I'd be curious to know if it works and/or is stable.

I'm really wondering if there is any "human perceivable" difference between the 1333mhz and the 1866mhz... I have my MBP with the 1600mhz just for the sake of knowing it has the "fastest". Never tried 1866mhz though.

[ParsiKade]

Thanks a lot, this article helped me.
I didn't know to go for the 1333 or 1600 for my new 16GB memory, and after reading your post I think I'll go with 1333.

[Rhinoevans]

Originally Posted by ewestlund

Hi folks. I have never seen a guide that compared 1333 vs 1600 RAM as well as various size configurations in a Macbook Pro. I wrote one up. It's online here, but I've pasted the gist here if you don't want to click away. Pictures and tables are included in the web link if you really wanna get into the nitty gritty.

Background

Some quick internet searching told me that later model Macbook Pros could, in fact, handle 16GB of memory (despite Apple saying 8GB is the limit) and can run DD3 Memory at 1600 MHz (despite Apple saying 1333 MHz is the limit). So, I ordered 2x8GB of Corsair Vengeance PC1600 notebook memory at $57.99 a piece. Tack on a $10 NewEgg discount, and the grand total was $104.98. Only $1.50 more per GB than I paid to upgrade my machine from 4GB to 8GB in December 2011. For whatever reason, it was cheaper to order 2 single 8GB sticks ($115.98 before discount) than the dual-pack of the same memory ($119.99). You can download the data used in this analysis.

Test Machine

Late 2011, 17″ Macbook Pro with quad-core 2.5GHz Sandy Bridge i7 CPU.

Benchmark Software

Geekbench 2.3.1 for Mac OS X x86 (64-bit)

Test Configurations


4GB (1x4GB) 1333MHz, Corsair (CMSO8GX3M2A1333C9)
4GB (2x2GB) 1333MHz, Samsung (Stock Apple memory)
8GB (1x8GB) 1600MHz, Corsair Vengeance (CMSX8GX3M1A1600C10)
8GB (2x4GB) 1333MHz, Samsung (Stock Apple memory)
16GB (2x8GB) 1600MHz, Corsair Vengeance (CMSX8GX3M1A1600C10)

Procedure

I ran Geekbench 64-bit for the Mac 5 times on each of the above configurations to account for random error. (Admittedly, on configuration 3 I forgot to record one of the results and couldn’t be bothered to re-install the memory, so n=4 for configuration 3.) While Geekbench gives about 20 different numbers, I recorded 5: the overall composite and the four summary statistics: Integer Performance, Floating Point Performance, Memory Performance, and Streaming Memory Performance.

Analysis

1. More memory improves Geekbench overall scores: about 1.5% increase from 4GB to 8GB and 1.2% increase from 8GB to 16GB
2. Dual channel makes a big difference. All dual channel configurations outperform the single channel configurations: 20-30% for Memory performance and 35% for Stream Memory performance.
3. Dual channel configurations with more memory do better than dual channel configurations with less memory on the order of 3% for Memory performance and 8% for Stream Memory performance
4. More memory improves floating point performance marginally.
5. The 16GB configurations has slightly more variance between measures on memory related tests and slightly less variance on processor related tests. Theoretically, that means more memory equals more consistent processing performance relative to lower capacity configurations and slightly less consistent memory performance relative to lower capacity configurations. This probably has absolutely no real world or practical meaning, but the claim stands up to scrutiny in statistics class.

Conclusions

1. Do not run in single channel mode. It’s inefficient.
2. One can expect a similar percentage increase in benchmarks when moving from 4GB to 8GB as from 8GB to 16GB. That is to say, there are diminishing returns as memory increases in size, but there are returns nonetheless.
3. Unsurprisingly, increasing memory improves memory benchmarks more than processor benchmarks, but there is a small improvement on the processing front.

Cautions

1. I cannot get a good handle on 1333 vs 1600 performance on synthetic benchmarks because I have to compare dual channel to single channel performance. Nevertheless, other benchmarks suggest you may notice a slight speed boost on very intensive work, but moving from 1333 to 1600 is not going to make any noticeable difference for regular human beings. Get the 1600 RAM to feel good for being on the bleeding edge, and then laugh at yourself for the absurdity of it all. The cost difference is marginal, in any event.
2. These benchmarks are for a late 2011 i7 quad core. Results may, of course, vary, but I suspect these types of changes will remain relatively consistent from Mac to Mac. I do not know how far back one can go and still run the memory at 1600 MHz, however.

Final Thought

Benchmarks aside, 16GB feels faster than 8GB and 8GB feels faster than 4GB. This is especially noticeable when many applications are running. The upgrade is worth it for the cost.

Link bad

[el-John-o]

Just FYI, it may be cheaper to buy the ram in separate sticks (though I purchased 16GB DDR-1600 a couple weeks ago for $70), however many times the 'two packs' are matched 'kits' designed to work well with one another. Sometimes it can be hit-or-miss to get two sticks not packaged together to actually function in dual channel mode (though obviously it worked out for you!)

Of course, it'll still work otherwise, but performance is greater if the RAM actually operates in dual channel mode, as you pointed out.

I used to think that was a load of crock when I first read it and it made no sense to me, but when I did my last PC build I had a heck of a time getting dual channel to work, until I sent the RAM back and bought a 'dual channel kit' which has worked like a charm for me ever since.

[daneoni]

Also only the high-end quad core Core i7 2011 MBPs support 1600MHz.

[spoonie1972]

since we're sort of future-looking with the faster ram chips (haswell) - is the form factor going to be the same on intel's next low(er) power update? 204 pin?

will we see 16gb x 2 sodimms clocking-down and working for our 2011-2012 macbook pros?

[circa7]

Originally Posted by el-John-o

Just FYI, it may be cheaper to buy the ram in separate sticks (though I purchased 16GB DDR-1600 a couple weeks ago for $70), however many times the 'two packs' are matched 'kits' designed to work well with one another. Sometimes it can be hit-or-miss to get two sticks not packaged together to actually function in dual channel mode (though obviously it worked out for you!)

Of course, it'll still work otherwise, but performance is greater if the RAM actually operates in dual channel mode, as you pointed out.

I used to think that was a load of crock when I first read it and it made no sense to me, but when I did my last PC build I had a heck of a time getting dual channel to work, until I sent the RAM back and bought a 'dual channel kit' which has worked like a charm for me ever since.

Where did you buy that ram for that price?

[el-John-o]

Originally Posted by circa7

Where did you buy that ram for that price?

http://www.newegg.com/Product/Produc...82E16820148609

That's what I bought. I got it around Christmas for 20% off, though even the regular price is cheaper than what you have listed above.

There are also some cheaper 1600MHz kits than that. It doesn't have to be 'Apple specific', though I have heard of people having issues with certain brands of RAM and being marketed for Apple products gives you piece of mind that it won't be one of those finicky pieces.

My geekbench memory score, with that RAM, is 6956. Overall system score is 7557. That's running a 13" 2012 MBP, dual core i5 @ 2GHz

[ParsiKade]

Originally Posted by el-John-o

http://www.newegg.com/Product/Produc...82E16820148609

That's what I bought. I got it around Christmas for 20% off, though even the regular price is cheaper than what you have listed above.

There are also some cheaper 1600MHz kits than that. It doesn't have to be 'Apple specific', though I have heard of people having issues with certain brands of RAM and being marketed for Apple products gives you piece of mind that it won't be one of those finicky pieces.

My geekbench memory score, with that RAM, is 6956. Overall system score is 7557. That's running a 13" 2012 MBP, dual core i5 @ 2GHz

What's the difference between this RAM and the one on crucial's website? The one on newegg.com is noticeably cheaper.

[el-John-o]

Originally Posted by ParsiKade

What's the difference between this RAM and the one on crucial's website? The one on newegg.com is noticeably cheaper.

I just went on Crucials website and found it for $85.99. Are you looking at something else? Perhaps desktop RAM?

[alphaod]

Most of the time, faster [clocked] memory will also have higher latency [timing].

Since we can't tweak the timings on a laptop, sometimes there is no advantage to having faster clocked memory at the cost of higher latency.

[throAU]

Another point I'd make for potentially against going for 1600 vs. 1333 is that it is easier to make lower clocked chips reliable and heat up less.

If there's no discernable difference between 1333 and 1600, then i'd go for the 1333, personally (it is in spec for apple, the latency is lower, it is cheaper and easier to make).

I doubt you'd see any heat difference, but my tolerance for hardware failure is exceedingly minimal and anything i can do to minimize that is a good thing in my book, especially if it doesn't hurt performance.

[ParsiKade]

Originally Posted by el-John-o

I just went on Crucials website and found it for $85.99. Are you looking at something else? Perhaps desktop RAM?

The one on crucial website is 1333 but I believe the one on newegg.com is 1600.

http://www.crucial.com/store/mpartsp...A3E8DAA5CA7304

Please correct me if I'm making a mistake here.

[el-John-o]

Originally Posted by ParsiKade

The one on crucial website is 1333 but I believe the one on newegg.com is 1600.

http://www.crucial.com/store/mpartsp...A3E8DAA5CA7304

Please correct me if I'm making a mistake here.

I'm not finding the higher priced one at all.

http://www.crucial.com/store/mpartsp...32ED07A5CA7304

That's not the same model number as on NewEgg, but spec for spec it's the same (I couldn't find the NewEgg model number, but I suspect it's the same). Same price on Crucials website AND on NewEgg.

That's the DDR3-1600 module. I don't know if it lists compatibility for your model or not, but, as you know, Apple doesn't list compatibility for 16 gigs in a 2012 13" yet I'm running it just fine! So you'll have to google and see what is compatible.. but if it can handle 16GB @ 1600MHz, then that will work.

[circa7]

I just bought a 2012 Macbook Pro that obviously takes 1600mhz RAM. However, I have two Crucial 8gb 1333mhz sticks laying around. Will these work in my new Macbook or do I need 1600mhz?

Exact specs: Crucial 8GB DDR3 1333 MT/s (PC3-10600) CL9 SODIMM 204-Pin