4Gb > 8Gb improvements and pitfalls ?

Soldato
Joined
3 Aug 2003
Posts
15,921
Location
UK
Ok so I currently have the Corsair 4GB DDR2 XMS2 Dominator PC2-8500C5 TwinX (2x2GB) 4Gb total installed in the correct banks to make it dual channel.
The current O.S is Vista Home premium x64.

What advantages and what disadvantages are there adding 2 more sticks of the same ram (same make model and matched set) in the other two free slots to make 8Gb total. ?
 
first off, do you need 8GB of ram?? are you using the 4 you have, could the £100 odd be spent on another better upgrade.

also you overclock (assuming you have) may become a bit more unstable, as the memory controller is put under more strain as all slots are filled, extra voltage can help.
 
The only things I need are food, water and a roof over my head, all the rest are just toys to keep my idle fingers occupied. :D
 
I went 4gig to 8gig in Vista ultimate x64

Love it

Not a speed boosts per se, But when it comes to encoding or editing the extra ram helps, With more ram i never seemed to get bogged down

Would not go back to 4gig now

In fact want more ram from g-skill when/if OcUK stock it :D

For more info check the windows etc part of the forums ;)

Ciao

Def
 
I think it's the kind of situation where if you can't explain why you'll need 8GB of RAM before you buy it then you probably won't see the improvement. If you're working with huge images or models or running lots of virtual machines then it will help, but I think that if you don't have a usage case in mind then it's unlikely it will help you. Personally I find 4GB to be easily enough for everything I do in Vista.
 
I baught the 8 gig right off the bat when i built current machine just because its so cheap. 2 gig of high speed ddr1 cost me a lot more less then 2 years ago. Cant say if utilised it much, but its there if I ever do ;)
 
Biggest pitfall is the 'possible' reduction in overclocking as mentioned, and a reduction in the width of your wallet. Apart from that go for it if you think it's worth it (such as running quad CrossfireX or three way SLI :)

Matthew
 
I think its been shown in many recent good reviews its faster due to less AXX to the HDD needed.

A few peeps here claim the jump from 4GB to 8GB is better than 2GB to 4GB.

All I know is that my Vista 64Ult fully loaded with lots of stuff is sluggish with 2GB.
 
I think its been shown in many recent good reviews its faster due to less AXX to the HDD needed.

A few peeps here claim the jump from 4GB to 8GB is better than 2GB to 4GB.

All I know is that my Vista 64Ult fully loaded with lots of stuff is sluggish with 2GB.


Whats AXX?



Anyway I currently have 4Gb and this is more than enough for me I see 8Gb as way overkill :D
 
I was a firm believer in 2Gb over 4Gb as I couldn't see any great benefit but I read a very good Tom's Harware article where it explains that 4Gb on a 64-bit operating system can be slower than 2Gb on a 32-bit OS because of the way 64-bit operating systems store numbers (it takes twice as much space to store a 64-bit number as a 32-bit number). This situation is completely reversed if you go to 8Gb with a 64-bit OS.

My advice would be to do it.
 
Ive just built a new machine last night ( Still doing burn ins ). E8500 @ 4ghz on a gigabyte x48-DQ6 with 8GB of corsair dominator.

All i can say is, after running Age of Conan and Crysis with 4gb and then 8gb installed, that 8gb sure speeds things up! I had to install my vista 64 with only 2gb installed as i dont have a sliipstream of sp1 yet and it was hanging on install.

In general desktop the differance between 4 and 8 gb isnt really apparent as you can imagine. But after playing Age of Conan for two hours last night and moving between zones it was very noticeable. I imagine the game uses all the ram to store textures and stuff so loading is a lot faster. Crysis also appeared to load much faster after the initial loading, as if it also uses the free memory to store textures and such.

With the price of DDR2 being so cheap you may as well get 8gb now if you have everything else in your machine upto spec.

As a side note i tested the rig with a maxtor sata 150 drive and when i changed to a samsung sata300 drive for final install the differance in that was astounding! :p i knew it would be a bit faster but wow!
 
8gig is worth it, but you won't notice an increase in performance for most day to day stuff.

The best way to use it is to make the swap file as small as possible and do a registry tweak to page all windows files directly to RAM, reducing hard disk access as much as possible.
 
I was a firm believer in 2Gb over 4Gb as I couldn't see any great benefit but I read a very good Tom's Harware article where it explains that 4Gb on a 64-bit operating system can be slower than 2Gb on a 32-bit OS because of the way 64-bit operating systems store numbers (it takes twice as much space to store a 64-bit number as a 32-bit number). This situation is completely reversed if you go to 8Gb with a 64-bit OS.

My advice would be to do it.

Actually thats not necessarily correct, the Intel X86 processors have 8, 16, 32 and 64bit registers, and they can store their numbers in many different formats depending on the programming. Just because you run the processor in 64bit mode, using a 64bit OS, doesnt suddently mean the programs are forced to run in 64bit registers.

But yes, if a programmer needs 64bit math, and uses 64bit variables then it will take 64 bits to store the number one. A 64bit OS is still able to work with any of the previous data structures natively due to the design of the X86 processor.

BTW, another reason 64bit is theoretically slower, is because the larger storeage requirements put more strain on the available memory bandwidth. Just doubling ram up to 8GB doesnt help, as the memory interface on dual channel DDR remains 128bit wide. (Just think of the lack of performance gains on 1024meg NV 8800GT's.. the 256bit bus is just too much of a limitation for the extra ram on a graphics card)

Nehalem will help a ton, as it combines tripple channel 192bit memory, with the higher bandwidths of DDR3 (Although apparently DDR2 should work if the motherboard makers wish) Conroe's limited by its FSB speed, but if the integrated memory controller on nehalem supports say DDR2@1066, in triple channel thats a big bandwidth increase over the performance of a Conroe with a 1333FSB.

Windows uses an system called WOW which basically creates little isolated pockets for 32bit applications to safely run without the risk of a 64bit application modifying its reserved memory etc. But its still run natively on the Core 2 Processors, its not a software emulation, just a kinda protective buffer.

But even running pure 64bit applications, its still possible for data to be stored as 8/16/32 or 64 bit depending on the programs requirements. If you look at some of the early MMORPGS, they used 8 bit data structures for characters (leading to skill caps of 256), yet they were running on 32bit windows. By using smaller data structures they could run with far smaller memory requirements.

If most of your applications are 32bit, 4GB on 64bit windows will generally outperform a 2GB 32bit windows. If you run a lot of computationally heavy 64bit applications, then it could be slower, due to the amount of data being dealt with.. but then again the 32bit windows couldnt even run those applications without extremely processor intensive schemes to do 64bit math in 32bit chunks, and in that situation the 64bit OS will be a huge amount faster.
 
Last edited:
I was a firm believer in 2Gb over 4Gb as I couldn't see any great benefit but I read a very good Tom's Harware article where it explains that 4Gb on a 64-bit operating system can be slower than 2Gb on a 32-bit OS because of the way 64-bit operating systems store numbers (it takes twice as much space to store a 64-bit number as a 32-bit number). This situation is completely reversed if you go to 8Gb with a 64-bit OS.

My advice would be to do it.

There's also the pitfall of cache thrash.

Basically you have a small amount of processor cache. With 2GB the CPU will fly however because the tasks require more memory with 64bit you'll be paging more. Thus stalls the CPU.
Switch to 4GB and the majority of users will see a boost due to the lower paging requirement The CPU perhaps will go at a lower speed overall but the fact that the majority of tasks are in memory lead to a real time speed increase.
Switch to 8GB and you've doubled the memory that the same cache has to cover - putting pressure on it which results in less items being cached. Faster memory bandwidth will help but you may see a slight drop in performance because of it. However for big tasks this is still the better option. Free memory used for file caching will help increase the speed for repeated tasks such as loading libraries etc.
Switching to 16GB .. more pressure on the cache and the CPU spends it's time vainly caching data that the tasks swapping and threading will invalidate quickly. Overall expect a drop in performance for normal tasks although memory that is used to cache files will help increase repeated tasks too.

Lastly - twin quad core CPUs such as those used in the MacPro (ie PC workstation).
There is a minimum amount of ram you should have too... nowdays the cache in a CPU is so small compared to the amount of memory and the concurrent threads (cores) being processed through it that the computer's memory is more like the cache onto the real 'memory'... the hard drive.
When you get to big CPUs such as twin quad core, the minimum you should really have is 4GB each to allow distribution of tasks and allow enough data to be in memory for threaded tasks that each core has enough memory to keep working.
This is normally because of the tasks used for this type of box differ in the characteristics compared to the gamers and 95% home users.
 
Actually thats not necessarily correct, the Intel X86 processors have 8, 16, 32 and 64bit registers, and they can store their numbers in many different formats depending on the programming. Just because you run the processor in 64bit mode, using a 64bit OS, doesnt suddently mean the programs are forced to run in 64bit registers.

But yes, if a programmer needs 64bit math, and uses 64bit variables then it will take 64 bits to store the number one. A 64bit OS is still able to work with any of the previous data structures natively due to the design of the X86 processor.

BTW, another reason 64bit is theoretically slower, is because the larger storeage requirements put more strain on the available memory bandwidth. Just doubling ram up to 8GB doesnt help, as the memory interface on dual channel DDR remains 128bit wide. (Just think of the lack of performance gains on 1024meg NV 8800GT's.. the 256bit bus is just too much of a limitation for the extra ram on a graphics card)

Nehalem will help a ton, as it combines tripple channel 192bit memory, with the higher bandwidths of DDR3 (Although apparently DDR2 should work if the motherboard makers wish) Conroe's limited by its FSB speed, but if the integrated memory controller on nehalem supports say DDR2@1066, in triple channel thats a big bandwidth increase over the performance of a Conroe with a 1333FSB.

Windows uses an system called WOW which basically creates little isolated pockets for 32bit applications to safely run without the risk of a 64bit application modifying its reserved memory etc. But its still run natively on the Core 2 Processors, its not a software emulation, just a kinda protective buffer.

But even running pure 64bit applications, its still possible for data to be stored as 8/16/32 or 64 bit depending on the programs requirements. If you look at some of the early MMORPGS, they used 8 bit data structures for characters (leading to skill caps of 256), yet they were running on 32bit windows. By using smaller data structures they could run with far smaller memory requirements.

If most of your applications are 32bit, 4GB on 64bit windows will generally outperform a 2GB 32bit windows. If you run a lot of computationally heavy 64bit applications, then it could be slower, due to the amount of data being dealt with.. but then again the 32bit windows couldnt even run those applications without extremely processor intensive schemes to do 64bit math in 32bit chunks, and in that situation the 64bit OS will be a huge amount faster.


But guys do we really need anymore speed yet. Cos when I went from my amd 2.8gz to my e8400 3.6hz, I didnt notice a huge performance boost in windows. I just keep upgrading the graphics card untill the graphic cards are being slowed down by the cpu, then I either drop another cpu in or have a nice new system.:D
 
Back
Top Bottom