3GB magic number?

Apparently installing 3gb of memory in the G1s provides massive performance boosts in gaming over 2gb in Vista so I was planning to install 4gb of memory in a laptop with almost identical specifications to the G1s (Evesham's Zieo) so I could get the dual channel performance boost and because memory is so cheap nowadays. But then I read somewhere ( here) that the performance boost only happens with 3gb and not 4? Is this true, and how could this be the case?

 

I would give you some rep for mentioning Evesham but apparantly I did that already on another of your posts!

Yeh, 4GB would be fine but only 3GB of that would be used because Vista only supports 3GB (I think).

Hope this helped!

 

That's not a vista limitation. Rather, that's a Vista 32 bit limitation.
But what I think he's referring to, is in reference to another thread, which found that Going from 3 to 4 GB of RAM decreases the performace of the G1s. There were quite a few things discussed there, including the possibility of a whole DIMM being ignored. nothing concrete, though.

 

Oh I see, so would it be more advisable to upgrade to 3gb instead of 4gb? If Vista-32 only sees around 3gb max, would the extra gb make absolutely no difference? I would consider this but then my memory will not work in dual channel mode (2gb+1gb) and that apparently itself leads to a big performance loss in memory intensive applications due to the FSB of the Santa Rosa CPU and the bandwidth of the memory not being equal?

 

For the record guys it's not "just a Vista 32-bit" limitation, I've been running 4GB with Vista X64 and it ran worse in a few instances than it runs with 3GB.

 

Dual Channel mode doesn't have to do with size, just speed. (As a matter of fact it will even run in Dual-Channel with different speeds, but it can only run at the slower of the two speeds, so it is sort of a waste.)

 

Hmmmm, so I can have a 2gb stick and a 1gb stick (of the same speed) and they'd run in dual channel mode? I was under the impression that the size of the memory modules had to be identical.

It seems like the modules have to be the same size for dual channel to work, which presents me with a dilemma. Either I upgrade to 4gb of dual channel memory and lose performance or upgrade to 3gb and gain that performance in games but lose it in memory intensive tasks because dual channel won't work.

 

that, my friend, is a conundrum.

 

I don't believe you will see any real benefit of going to 4GB, "dual channel" has never presented itself to me as worthwhile both in the notebook and desktop realm.

 

For the record, a 32-bit operating system IS capable of 4 GB of RAM - that is its absolute limitation. Take a look here for why you don't actually get the full 4 GB, however (if you are interested):

http://www.codinghorror.com/blog/archives/000811.html

 

I'm getting many conflicting messages here. Over here, I was advised that dual channel memory offers a 15-20% increase in memory intensive tasks, which seems like a massive improvement to me.

Are there any actual benchmarks done with 3 vs 4gb or dual channel vs not dual channel so I can clearly see what differences there are, if any? In any case, I think the notebook bandwidth will overall be faster than my disappointing PC2-5300 desktop bandwidth of 4000Mb/s (as measured in Sandra Lite).

 

The theory of dual channel is great, however i must conclude that it is not being used to full potential or it was not implemented to work as good as it could (form the results many have shared)

I'd go with 3g ram because I've heard Vista (i was under the impression it did not matter what version of it) does not recognize over 3g. So putting 4 gigs in the G1s and running Vista would have no purpose and therefore no performance boost. And if you must have dual channel, from what has been previously posted, dual channel will supposedly work even with different sized modules.

 

http://forum.notebookreview.com/showthread.php?t=131458

Evidence that 4GB isn't being used properly by Vista x64.

 

Well, that was a lot of useful information. Thank you. Following those links around seems to show that even in synthetic memory benchmarks, there is no gain in bandwidth with dual channel and 3gb really is the magic number, though I still don't understand why. I guess I'll be upgrading to 3gb instead of 4 then and getting the 7200rpm hard drive with the money I save.

 

I'm thinking of upgrading to 3GB myself, since 4GB is useless and not working right apparently.

Which 2GB sodimm are people installing? I'm looking at the OCZ Value Serie 2GB module (OCZ2MV6672G) for 169.99 CAD$.

 

I used Super Talent brand Micron d9 memory. If you are interested in purchasing one of my 2gb sticks please pm me or checkmy FS post!

 

guys, Dual Channel mode works if you have two DIMMs of DDR on a modern chipset in your computer. It doesn't matter the size, latency, make, model, speed or anything. However, Dual Channel defaults to the slowest speed and the highest latency, as well the memory controller isn't as effecient when they are of different size. There is also some lag if the DIMMs are made from different enough manufacturers. The opitmal performance for Dual Channel is matching everything, all the way down to the number of chips on each side of the module. The performance hit you take is about 5% MAX of the 20% boost that it should theoretically give you. (not counting latency and speed being tuned to the worse of the two, because that is something that should always be avoided, unlike two differently sized ones, or of a different make or whatever else.)
If you don't think Dual Channel helps then use a computer with one DIMM of 2GB then switch to one that uses 2x 1GB. Everything else equal, you should see the performance increased pretty easily.

 

Actually, that is wrong... it'll "work", but it won't be dual channel.

http://www.intel.com/support/motherboards/desktop/sb/cs-011965.htm

Rules to Enable Dual Channel Mode
To achieve Dual Channel mode, the following conditions must be met:

* Matched DIMM configuration in each channel
* Same Density (128MB, 256MB, 512MB, etc.)
* Matched in both Channel A and Channel B memory channels
* Populate symmetrical memory slots (Slot 0 or Slot 1)

Note: Configurations that do not match the above conditions will revert to Single Channel mode.

The following conditions do not need to be met:

* Same brand
* Same timing specifications
* Same DDR speed

Note: Memory channel speed is determined by the slowest DIMM module populated in the system.

 

You missed something:

Flex mode
This mode provides the most flexible performance characteristics. The bottommost DRAM memory (the memory that is lowest within the system memory map) is mapped to dual channel operation; the topmost DRAM memory (the memory that is nearest to the 8 GB address space limit), if any, is mapped to single channel operation. Flex mode results in multiple zones of dual and single channel operation across the whole of DRAM memory. To use flex mode, it is necessary to populate both channels.

Intel® Desktop Boards based on the following chipsets support Flex Mode:

* P965
* G965
* Q965
* Q963



EDIT: Secondly, I just realised, this only covers desktop mobos. Notebook mobos only have two slots, they are somewhat different. They generally will always default to Flex Mode unless you have identical sticks.

 

Flex only applies to 965-based chipsets, meaning as far as laptop goes that only applies to Santa Rosa units. Moreover, it works by making just the part that is dual channel-capable work as dual channel and let just the part that isn't work as single channel. That is still single channel for the other part of the system.

The dual channel concept is universal to both Intel and AMD-based systems. The idea is that they are *supposed* to be as similar as possible so that they can work in parallel and behave as if they were a single part that works faster. Making them different, it makes it that much more difficult on the system. Older dual channel systems used to be more picky and would actually "work" but then become unstable and crash further down the line if the parts were too different (hence, the importance of matched sets of modules). Newer systems have more tolerances for differences, but the more different it is, the more you take your chances on whether it'll actually work properly.

 

When we talk about the 3GB magic number, we are talking about Santa Rosa systems. Also, I am not saying that matching DIMMs is wrong, I am just saying that modern dual-channel (yes it has evolved), does not necessarily require complete matches. You get a performance boost through a match, but you will still get the "double-funnel effect" (to use the analogy from the kingston whitesheet from a few years back) with different DIMMs. And you can't truly compare the notebook architecture directly to desktop architecture.
Especially in this case because of TurboCache. People talk about how it never worked this well before and so-on, but they are talking about how it was used on low-end cards with insignificant amounts of actual local memory. In laptops TurboCache is used almost necessarily to reduce die-sizes, also because the parts will be very expensive anyway, to reduce costs. In addition, most desktop systems with TC are equipped with low-end RAM as well. TC benefits highly from the higher-end RAM that is found in gaming laptops, like the G1S, which have better timings and lower latency. When you add up the benefits to TC of working on a higher-end card, being coupled with better system RAM and being able to take advantage of super-clock speeds from GDDR3 on-die memory, it becomes a very, very powerful feature.
It works by defaulting to the system memory until a time of high-stress when the on-die memory kicks in and you are effectively using the amount of memory listed on the box. I believe (and this is my theory which could be quite wrong and silly sounding) that when in a system with 4GB of memory, in a 32-bit architecture (Memory controller set to x86 which can only allocate 3GB per application, maximum), that is only using 3GB (75% of resources) it never feels the need to use its own RAM. This essentially leaves your card operating entirely on system RAM and fighting (though the TCM can't see it) with the system over the RAM. This will quite obviously bottleneck performance. When the system has only 3GB (double channel or not), it can see the entire charge becoming used, and adjust the memory it uses thusly, and when that happens to a card with 256MB of extra memory to contribute (that is not hampered by the 32-bit Mem controller because it has its own) you should see a hearty performance increase in graphical applications. This is a Santa Rosa model (as I said before) so flex mode will allow some dual-channel performance.

 

Can anyone with a 3GB system check with CpuZ if they're in DDR/Flex-mode so that we settle this part once and for all.

I'm really surprised though to see that Vista is using 3GB of ram better than 4GB. Doesn't Vista have PAE enabled by default?

 

i have a hp dv9008tx laptop and i'm on 3GB and cpu-z says dual channel. but we all know that its impossible, so i take it that its in flex mode.

 

Windows has PAE on by default on NX enabled systems, very few (if any) Intel processors have used NX bit, they use XD-bit (same thing different name, but different flag, too...). However, it is also possible that XD-bit also causes it to be on by default.
Everything else equal Vista has a bunch of wonky features, I wouldn't be surprised if AWE was messed up too.

 

This should not get bumped from the first page at least until we have confirmation of its (in/)existence.