Dual Channel RAM Guide

Not sure if this has already been addressed, since I haven't read every single page of this thread:

Would I get better performance from 3GB asynchronous/asymmetric setup, or a true dual-channel 4GB setup?

On the one hand, I'll have only 2GB running at dual-channel, while the remaining 1GB runs at single-channel. On the other hand, I'll have only 3.2GB total, but I'm assuming all 3.2GB will be running at dual-channel.

Would it be correct to say that it's a matter of 2GB dual-channel (+ 1 GB single) vs. 3.2GB dual-channel?

Is it worth it, getting 3.2GB running at dual-channel, even if it means losing that 800mb of memory?

 

They run very close many say it would not matter real world. I would say very small difference, either is fine.

 

SiSoftware Sandra's memory benchmark gives a slightly lower bandwidth for 3GB compared with 2GB (results earlier in this thread).

I asked Sandra's developer how the benchmark addresses the memory space (the benchmark says it is using about 500MB) However, it seems that while the blocks tested are contiguous in "virtual mode" but not physical mode, so there's no easy way to figure out what is being tested.

I also carried out a set of tests on a Samsung R20 (ATI chipset) with different RAM configurations. Having any combination of two RAM modules on board gave significantly better bandwidth than having one module but the spread of results between symmetric and asymmetric RAM configurations was small. The PCMark05 result for 2.5GB RAM was slightly better than for 2GB.

John

 

Well John I can not say with any documents to back me up but from my readings. The VM is there so the applications including a benchmark are tricked into thinking info is stored in contiguous physical blocks (applications need). Which they in fact are not. This likely is beneficial to RAM or any storage system. It has to move less out to move new in, in theory can cut the work in half (likely more I am being simple on purpose). Now even though it is not a full physical test I think we can assume some level of randomness ( tests parts of entire array) for both real world and the SiSoft test. One thing that further indicates this is true. And supported by our data is bandwidth variation is directly proportional to the difference between DIMM sizes. I am saying it looks exactly like you would expect a random sample to demonstrate. And as I said on an earlier post, and I read in a MS document the most used migrates to the front end of the array. True dual channel. And that makes sense, 3GB's, 1 in 3 chance of ending up in the back 1/3, 2 in 3 chance of ending front loaded 2/3. No real world reason to be concerned about a contiguous physical RAM test as that is not how the OS functions so no reason and could be misleading especially only testing 500MB.

 

Slightly off-topic, but if anyone wants to see the test results for some CL=4 @ 667MHz RAM, look in this thread (thanks to hendra).

John

 

I saw John. And think intriguing. I might think about then comment. I mostly noticed the lack of difference between CL4 and CL5. Not the 20%/25% I was looking for?

 

Yes, we need to understand more about how memory access works. That tRC of 20 is very high, but there is no bank cycle time in the SPD data. Looking at my current RAM it seems that tRC is normally CAS + tRAS, so it should be down at 16. tRC of 20 is the norm for CL=5 RAM (and might be set in the BIOS as default if the SPD doesn't contain a value).

John

 

How does one upgrade RAM? Like do I just take the old one out and put the new one in?

 

Yes. The procedure is usually described in your notebook's user guide. You need to check what is compatible with your notebook. It depends on the age and chipset used.

John

 

Man, I have been reading many of these threads trying to figure out if I really need more memory. All of this has been helpful - probably a little more technical than I needed in some regards, but still insightful!

OK, I have 2gbs and with the programs/windows I use daily all open, I still hover between 55-65% usage. So really adding another GB is really not going to help me any until I start bumping....wel, in what range? 90%? 95% usage on average?

 

Not exactly. And sorry I will try and keep simple D), if you are using Vista (similar with XP but mostly in not going over 50%) it will try and keep RAM usage around the 50% mark. So what you really need to do is pay attention to "hits" to the HDD. See to maintain that approximately 50% level it might be putting things in the "Page File" which is fine most of the time as that information is the least accessed. But if you don't have the ideal amount of RAM it will put things you are needing to access in there. And that can prevent you from achieving peak performance. There is a document from Microsoft with detailed information on how to determine if you have the best amount of RAM. But even I honestly find it a little over technical. So watching the HDD light acomplishes the same thing. Unless opening an app or reading/writing data from/to the disk that HDD light should not be blinking (much). If it is a lot, more RAM could help.

 

SIMPLE and clear (thanks for making it easy for me!). I could probably use the extra GB - my HD is accessing/blinking just about all day.

 

So right now I'm hovering around 1 GB used out of the 3.5 GB available to me under XP 32-bit (PC2-5300 2x2GB). My paging file is configured static 512 MB, is there any benefit to having a paging file at this point or am I better off removing it.

 

Unless you are in need of that 512MB HDD space you should probably just keep it. There have been discussions about this and no complete agreement. There is no reason it should slow your system down having it, and not having could crash your system. That said with as much RAM as you have and XP I bet if you turned it off you would not have a problem. I am not encouraging you just letting you know. I would keep it.

 

I got a Clevo notebook with 1 stick of 2 Gb (667mhz) and Turion X2, do you guys think its worth trying to get +1 stick of 2 Gb to use with WinXP SP3 ?

 

Yes.....................

 

I just by the new vostro 1510 with 3GB of ram, on deII's page says that they are in duaI channeI, is this right? they use this new tecnoIogy of fIex mode?. Thanks you very much

 

Yes, this is the flex mode. If you want to be sure about the memory performance, download SiSoftware Sandra and run the memory bandwidth benchmark. You should get at more than 3000MB/s. Single channel RAM will give no more than about 2000MB/s.

John

 

Also if you download CPU-Z it will say if Dual chanel or not. But unless something broke it is dual channel.

 

Does this mean intel based laptops perform 8%-10% better with dual channel RAM? Or is this performance only for the RAM? if so, how would that translate to overall performance?

Are there any benchmarks available?

 

Phil that is a memory bandwidth only so is not going to directly affect performance to that degree with most apps. John will likely see and post some links.

John Ratsey 1x2GB Single channel
SiSoftware Sandra

Benchmark Results
Int Buff'd iSSE2 Memory Bandwidth : 3658 MB/s
Float Buff'd iSSE2 Memory Bandwidth : 3674 MB/s
Results Interpretation : Higher index values are better.
Logical/Chipset 1 Memory Banks
Bank 0 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Bank 1 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Shared Memory : 8MB
Channels : 1
Memory Bus Speed : 4x 333MHz (1332MHz data rate)
Width : 64-bit


John Ratsey 1x1GB 1X2GB asymetric Dual Channel
SiSoftware Sandra

Benchmark Results
Int Buff'd iSSE2 Memory Bandwidth : 3847 MB/s
Float Buff'd iSSE2 Memory Bandwidth : 3859 MB/s
Results Interpretation : Higher index values are better.

Logical/Chipset 1 Memory Banks
Bank 0 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Bank 1 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Bank 4 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Bank 5 : 1GB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Shared Memory : 8MB
Channels : 2
Memory Bus Speed : 4x 333MHz (1332MHz data rate)
Width : 64-bit

John Ratsey 2x1GB interleaved Dual Channel
SiSoftware Sandra

Benchmark Results
Int Buff'd iSSE2 Memory Bandwidth : 3980 MB/s
Float Buff'd iSSE2 Memory Bandwidth : 4015 MB/s
Results Interpretation : Higher index values are better.

Logical/Chipset 1 Memory Banks
Bank 0 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Bank 1 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Bank 4 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Bank 5 : 512MB DDR2-SDRAM 5.0-5-5-15 (tCL-tRCD-tRP-tRAS) CR1
Shared Memory : 8MB
Channels : 2
Memory Bus Speed : 4x 333MHz (1332MHz data rate)
Width : 64-bit

 

In the fullness of time (not sure when) I could try PCMark05 with the 2 x 1GB, 1 x 2GB and 3GB combinations on my Zepto (Intel 965 chipset). Even 2 x 2GB if you want it.

I've got a stack of results for the ATI chipset in the Samsung R20 which can be summarised as:

2 x 512MB PC4200 : 3498 PCMarks
1 x 1GB PC4200 : 2964 PCMarks
1.5GB PC4200 : 3479 PCMarks
2 x 1GB PC4200 : 3562 PCMarks

Also:

2 x 512MB PC5300 : 3450 PCMarks
1 x 1GB PC5300 : 3342 PCMarks
1 x 2GB PC5300 : 3375 PCMarks
2.5GB PC5300 : 3572 PCMarks

I can't explain why 1 x 1GB of PC4200 performed much worse than 1 x 1GB of PC5300.

John

 

John what I posted was from this:
Model : Inventec Corp Mobile PM965/GM965/GL960 Express Processor to DRAM Controller

Is that your Zepto?

 

Yes.

BTW, Zepto are dumping Inventec. Their BIOS writer seemed to be someone moonlighting from another job.

John

 

***, why did powerpack get banned?

 

would the AMD Puma laptops run in 3GB dual channel like the intel sta.rosa does..? TIA

 

This topic is great. Shame I did not find it earlier .

 

So currently this is what i have using the cpu-z. I am running a travelmate 8103 which came a long with 512 ram which i upgraded to another 512. I was planning to up grade it to 2x 1gig and looking at the rams from newegg. I cant decide what latency to buy.

should i go with :

Kingston 2GB (2 x 1GB) 200-Pin DDR2 SO-DIMM DDR2 667 (PC2 5300) Dual Channel Kit Laptop Memory Model KVR667D2K2SO/2GR - Retail

cas latency 5
Voltage: 1.8v

or

Kingston HyperX 2GB (2 x 1GB) 200-Pin DDR2 SO-DIMM DDR2 667 (PC2 5300) Dual Channel Kit Laptop Memory Model KHX5300S2LLK2/2G - Retail

Latency 4
timing 4-4-4-12

and what brand should I go with.

 

The HyperX looks to have lower latencies and should give slightly better performance, but you will probably need to use a stop-watch to see the difference.

Yet another option would be to use PC6400 (800MHz) RAM. Most, but not all, these modules, have good timings at 667 and 533 MHz. However, some PC6400 modules don't support 667MHz. See the attached.

John

 

You guys really are experts when it comes to computer setups and components.. I never realized that RAMs could be customized in so many ways.

 

do you mean cas latency?

here:
http://en.wikipedia.org/wiki/CAS_latency

anyway cas latency does not effect performence very much.
you're better off with cas6 and 6gb then cas3 and 4gb.

It's more important to have a higher frequency and the bigger capacity then low timmings.
cas3 2gb ddr2 667mhz is probably possible. you can "flash" the SPD of your ram to tell it to run with cas3. use SPDtool (just google spdtool) for this. Although this will make your ram less stable and possibly kill it.

I would not worry about timmings. In fact most people would increase the cas latency to allow ram to be overclocked higher.

 

That's fine. 800MHz vs. 667MHz makes no difference in performance either...

 

Higher RAM frequency does help performance of integrated graphics, but there is little performance difference between CL5 800MHz and CL4 667MHz as the latency, measured in microseconds, is identical. I don't think I have seen any notebook RAM with lower latency.

John

 

you can overclock your ram by overclocking the FSB.
Your CPU would also be overclocked.

I used setfsb to overclock my fsb.
You may have to use some other software if you have an nvidia chipset.
I'm assume you don't have overclocking options in the bios?

You could also try a PLL pin mod to overclock the FSB.
check the thread in my sig for some ideas. (the thread about setfsb and clockgen)

 

if you can find the PLL (clock genorator) on your motherboard then you can use setfsb to overclock the FSB. when the FSB is overclocked your ram is also overclocked.

Also setfsb is very easy to troubleshoot and it's constantly being updated.
Also you can email abo (the creator of setfsb) and he will provide support and he can add your PLL if it isn't already supported.

I can also help you with using setfsb to overclock.