[The Godfather]

My MB has two 256MB sticks. If I buy one 1 GB stick and end up with 256MB + 1GB, will it be slower in any way than just 1GB?

Will 2GB sticks be available in the future?

[mduell]

No, it will not be slower. The 945GM chipset supports asymmetric dual-channel, so you'll get maximum performance with mismatched sticks.
2GB DDR2-667 sticks are available today for laptops; they're about $1800 each. The chipset in your MacBook supports them.

[uicandrew]

2gb sticks are available. another forum member has it prominently stated in their signature.

1.25gb is better than 1gb. faster/slower doesn't really apply to your example.

theoretically, it is only faster if the 2 ram chips are paired in size and specifications. so 2 chips of 256 is FASTER than 1 stick of 1gb and 1 stick of 256mb, but the increase in amount of ram OVERWHELMINGLY is better than the incremental speed increase of matching ram.

another analogy: 2 cars going at 65 mph moves luggage faster than a huge semi-truck going at 60mph. but the capacity of the semi renders the speed difference to be irrelevant

[harrisjamieh]

mduell, you keep saying that the Intel Macs get no speed increase at all from matched pairs. From what I have read, asymmetric dual-channel is quite a bit slower than symmetric dual channel. Also, why would Apple suggest the RAM is installed in pairs on their website if it didn't help performance one bit?

From Apple's website:

. And if both slots are loaded with an equal amount of RAM — which is strongly recommended — you can take advantage of the system's dual-channel memory architecture for an additional performance boost. With a dual-channel memory interface, both banks of SDRAM can be addressed at the same time, enabling MacBook to reach a memory throughput of up to 10.7 GBps.

[Dr.Michael]

From the barefeats website:

"April 20th, 2006 -- Matched Memory Pairs make your Intel Mac Faster. Some of you were asking whether you would gain any speed using matched memory pairs on your Macbook Pro, Intel Mac or Intel mini. In our testing, iMovie renders were 3% faster with matched pairs. Our Photoshop CS2 MP actions ran 6% faster with matched pairs."

And also this (May 18th):
"DO MATCHING MEMORY PAIRS MAKE IT GO FASTER?
We tested the MacBook Pro with both matching pairs of 1GB SDRAM and non-matching pairs (one 1GB and one 256MB module). There were gains: 51% in Quake 3, 18% in Doom 3, 2.5% in iMovie, and 3.5% in Photoshop CS."

Finally this. Interesting because it includes a statement about GPU tasks affected by matching pairs (March 7th):
"DO MATCHING MEMORY PAIRS MAKE IT GO FASTER?
We tested the MacBook Pro with both matching pairs of 1GB SDRAM and non-matching pairs (one 1GB and one 256MB module). None of the GPU test results were affected but the matching pairs did provide a small gain in CPU intensive tests (2.5% in iMovie, 3.5% in Photoshop CS)."

Could be puzzling compared with the 2nd statement.
At least the latest measurements show, that there can be a huge effect. What impact this has on the integrated graphics of the macbook is still undiscovered.

[Eug Wanker]

Originally Posted by mduell

No, it will not be slower. The 945GM chipset supports asymmetric dual-channel, so you'll get maximum performance with mismatched sticks.

That is not correct. Benchmarks have already shown in the Mac mini (which uses the 945GM chipset) that mismatched RAM sticks cause anywhere from a small to an enormous hit in 3D performance.

I'd expect the MacBook to behave the exact same way (since it uses the 945GM chipset as well).

[The Godfather]

I don't care much about gaming, since it is a Mac anyway. I do care about fixing the horrible sluginess while multitasking, and reducing beachballs.

I could even swear that my iBook G4 800MHz with 640MB RAM had less beachballs and was much smoother at multitasking than my new Macbook.

[Eug Wanker]

Originally Posted by The Godfather

I don't care much about gaming, since it is a Mac anyway. I do care about fixing the horrible sluginess while multitasking, and reducing beachballs.

I could even swear that my iBook G4 800MHz with 640MB RAM had less beachballs and was much smoother at multitasking than my new Macbook.

Yeah, the MacBook sux with 512 MB RAM.

Instead of getting 256+1024, have you considered 2x512? It's about the same price as 1x1024, or sometimes less.

What appz do you use?

[The Godfather]

Yes. I have considered two 512MB, but it would be a lost investment when I do decide to pony up for a pair of GBs.

[maximusbibicus]

Originally Posted by The Godfather

Yes. I have considered two 512MB, but it would be a lost investment when I do decide to pony up for a pair of GBs.

True. That would be a waste of cash when you upgrade. I stuck 2GB in mine off the bat and it flies. Not sure how bad it was with 512MB. I popped the sticks in before i started up the first time.

[cornwallstone]

Originally Posted by mduell

No, it will not be slower. The 945GM chipset supports asymmetric dual-channel, so you'll get maximum performance with mismatched sticks.

This statement contradicts Intel's documentation:

Mobile Intel 945 Express Chipset Family Datasheet

http://download.intel.com/design/mob...s/30921902.pdf

Page 353

The entire IGD [Integrated Graphics Controller] is fed with data from its memory controller. The GMCH's graphics performance is directly related to the amount of bandwidth available. If the engines are not receiving data fast enough from the memory controller (e.g., single-channel DDR2 553), the rest of the IGD will also be affected.

Page 331

Dual-channel Symmetric Mode

This mode provides maximum performance on real applications. Addresses are between the channels, and the switch happens after each cache line (64-byte boundary). The channel selection address bit is controlled by DCC[10:9]. If a second request sits behind the first, and that request is to an address on the second channel, that request can be sent before data from the first request has returned. Due to this feature, some progress is made even during page conflict scenarios. If two consecutive cache lines are requested, both may be retrieved simultaneously, since they are guaranteed to be on opposite channels. The drawback of Symmetric mode is that the system designer must populate both channels of memory so that they have equal capacity, but the technology and device width may vary from one channel to the other.

Dual-channel Asymmetric

This mode trades performance for system design flexibility. Unlike the previous mode, addresses start in channel A and stay there until the end of the highest rank in channel A, then addresses continue from the bottom of channel B to the top. Real world applications are unlikely to make requests that alternate between addresses that sit on opposite channels with this memory organization, so in most cases, bandwidth will be limited to that of a single channel. The system designer is free to populate or not to populate any rank on either channel, including either degenerate single-channel case.

[The Godfather]

Originally Posted by Eug Wanker

Yeah, the MacBook sux with 512 MB RAM.

Instead of getting 256+1024, have you considered 2x512? It's about the same price as 1x1024, or sometimes less.

What appz do you use?

My usual applications:

Parallels (Win2k with 128MB)
Safari
Office v.X
Toast Titanium
iChat, MSN Messenger

Sadly, I still haven't banned PPC applications in my system.

[Eug Wanker]

Originally Posted by The Godfather

My usual applications:

Parallels (Win2k with 128MB)
Safari
Office v.X
Toast Titanium
iChat, MSN Messenger

Hmmm... I'm guessing shouldn't make much difference for Safari, Office, Toast, iChat, or MSN Messenger. Dunno about Parallels, but I suspect having the extra 256 MB is better than having the matched RAM here too.

Your upgrade path makes sense, especially if you plan later to go to 2 GB. However, I did what some of the other people did, and jumped to 2 GB right away, since the memory was cheaper (3rd party) than i thought it was going to be. You can get 2 GB for about US$180. (I paid $219, but that was because it was Samsung RAM, and many of the stores selling the cheaper RAM don't ship up to Canada.)

[Dr.Michael]

Originally Posted by miniMoe

This statement contradicts Intel's documentation:

Mobile Intel 945 Express Chipset Family Datasheet

http://download.intel.com/design/mob...s/30921902.pdf

Page 353
etc.

Thanks, very interesting.

This clearly shows that the MacBook comes with hidden costs. The two 256 MB chips are worth nothing (meaning you cannot sell them) and two 512MB chips are hardly sufficient for that machine.

[Eug Wanker]

Originally Posted by Dr.Michael

Thanks, very interesting.

This clearly shows that the MacBook comes with hidden costs. The two 256 MB chips are worth nothing (meaning you cannot sell them) and two 512MB chips are hardly sufficient for that machine.

What's wrong with 2x512? That leaves you with 944 MB of usable RAM.

[gametime10]

I'm using 2 512MB chips and it is a world of difference b/w the 2 256MB it came with. I'm even using WIndows XP on Parallels w/ 256MB dedicated to it, and both Windows and OS X are running very smoothly.

[mduell]

Originally Posted by harrisjamieh

mduell, you keep saying that the Intel Macs get no speed increase at all from matched pairs. From what I have read, asymmetric dual-channel is quite a bit slower than symmetric dual channel. Also, why would Apple suggest the RAM is installed in pairs on their website if it didn't help performance one bit?

From Apple's website:

. And if both slots are loaded with an equal amount of RAM — which is strongly recommended — you can take advantage of the system's dual-channel memory architecture for an additional performance boost. With a dual-channel memory interface, both banks of SDRAM can be addressed at the same time, enabling MacBook to reach a memory throughput of up to 10.7 GBps.

You only get the dual channel performance when you're using less than twice the capacity of the smaller module. Once the smaller module fills up, you're back to single channel performance. 1GB+256MB is still faster than 256MB+256MB.

Dr.Michael: None of those benchmarks appear to directly address the issue.

Originally Posted by Eug Wanker

That is not correct. Benchmarks have already shown in the Mac mini (which uses the 945GM chipset) that mismatched RAM sticks cause anywhere from a small to an enormous hit in 3D performance.

That may be an OS-specific issue resulting from OSX's caching strategy. See above.

miniMoe: With an intelligent OS memory allocation scheme, dual channel performance can be achieved in asymmetric mode with the limitation mentioned above. I'm not sure if OSX implements such a scheme.

[Dr.Michael]

Originally Posted by Eug Wanker

What's wrong with 2x512? That leaves you with 944 MB of usable RAM.

Yes, thats a lot and I don't have more in my tibook. Absolutely sufficient - except if you plan to use Parallels. Since the MacBook allows that and for me this is the best option, much better than dual boot (with linux), 1GB is tight.

But its right, for normal use 1GB is a good choice. The only issue is that apple puts in memory that is useless. You have to throw it away. On one hand they keep the price of the MacBook low, but on the other hand (hehe) because of the cheap technology they have chosen, you need to invest more into upgrades than what they have saved initially. At least if you want to keep the machine powerful.

[Dr.Michael]

Originally Posted by mduell

Dr.Michael: None of those benchmarks appear to directly address the issue.

I really don't understand how your brain works. Maybe it would support your personality if you would learn to say: "Sorry, I was wrong"!

[cornwallstone]

Originally Posted by mduell

You only get the dual channel performance when you're using less than twice the capacity of the smaller module. Once the smaller module fills up, you're back to single channel performance. 1GB+256MB is still faster than 256MB+256MB.

Not according to Intel. You don't get dual channel performance PERIOD. Read how the memory is handled. The two banks are used serially.

Originally Posted by mduell

miniMoe: With an intelligent OS memory allocation scheme, dual channel performance can be achieved in asymmetric mode with the limitation mentioned above. I'm not sure if OSX implements such a scheme.

No, dual-channel performance can't be had. Even if the application or OS wrote 64 bytes of memory at X address, then 64 at X + end of bank 1, then X + 64, the calculations necessary to do this would result in MUCH more of a performance loss than simply incrementing addresses and letting the hardware Northbridge chip handle the distribution across the two banks of memory. That's why Intel said, "Real world applications are unlikely to make requests that alternate between addresses that sit on opposite channels with this memory organization..."

Originally Posted by mduell

I'm not sure if OSX implements such a scheme.

Why don't you take a hint from Apple's recommendations to use matched memory? Wouldn't that give you a clue they don't?

Originally Posted by Dr.Michael

Maybe it would support your personality if you would learn to say: "Sorry, I was wrong"!

Well said, Doc. According to Intel, mduell has been pedaling bad advice regarding this and those who search the Internet for one voice saying what they want to hear, in spite of that of corporate engineers, like those at Apple, have swallowed it. Now he's backpedaling saying it's possible in the OS rather than apologizing and admitting he's wrong.

[mduell]

Originally Posted by Dr.Michael

I really don't understand how your brain works. Maybe it would support your personality if you would learn to say: "Sorry, I was wrong"!

Your first benchmark shows almost no difference in performance; the 3-6% difference could be attributed to the increased memory.
Your second and third benchmarks again fail to vary the number of channels ceteris paribus.

Given the constraint on the number of slots, it appears that keeping the overall quantity of memory the same is impossible. I'd be convinced by results from a comparison of 2x256, 2x512, and 1GB+256 rather than 1+.25 vs 1+1.

Originally Posted by miniMoe

Not according to Intel. You don't get dual channel performance PERIOD. Read how the memory is handled. The two banks are used serially.

<snip>

Now he's backpedaling saying it's possible in the OS rather than apologizing and admitting he's wrong.

<rearrange>

Why don't you take a hint from Apple's recommendations to use matched memory? Wouldn't that give you a clue they don't?

Originally Posted by Your previous post

Real world applications are unlikely to make requests that alternate between addresses that sit on opposite channels with this memory organization, so in most cases, bandwidth will be limited to that of a single channel.

To me "unlikely" means that current OSs/apps don't do it, not that they couldn't. If your reads do alternate between banks, I still think that you get the performance of two channels based on the above statement from Intel.

Apple's recommendation is consistent with my previous post. You only get the dual channel performance until you fill the smaller DIMM, so matched DIMMs yield higher performance for longer.

There's no need to make personal attacks.

Have a nice evening,

mduell

[jaybert]

i went ahead and just ordered 1x1gb stick from newegg for the macbook, so I will remove one and add in the 1gb, for 1.25gb total.

You all do see in that test that someone mentioned above, they are comparing 1.25 asymmetrial (1gb + 256mb) and 2gb symmetrical (2x1gb) right? So it seems that the tests they are doing really have no validility since of course high memory usage apps will show a 50% increase when you increase the memory by almost 50% as well.

I do not plan on doing anything that memory intensive, so I think this will suffice for now. The reason I chose to upgrade to 1x1gb is so that if in the future I do decide that I would like 2gb, the upgrade path is much cheaper than if I had originally just purchased 2x512mb.

[The Godfather]

^^^ And I will do the same

[davidsi]

Originally Posted by The Godfather

^^^ And I will do the same

can anyone who did this tell me how the performance is?

[jeff25624]

While I have not statistical evidence to part, my my 1X1GB stick of Corsair RAM came this week, so Installed and have had a few days to play around with it. Peformance has definitely improved, the occurence of any spinning beach balls has decreased dramatically, applications run faster as well as open faster. These sucks need more than 512, that much I know now.