6 Gb vs. 4 Gb

| Leefa |

n00b
Joined
Jun 5, 2008
Messages
9
I am currently running 2x 2gb Corsair XMS2 800. My mobo supports up to 8 gb, and I have 2x 1Gb of the same type of ram lying around. I was wondering if it would be worth it, due to the fact that neither set would be running at dual channel (or would they?). I am running x64 Vista Ultimate.

How much faster would it run games/vista?

Thanks for any help.
 
Put them in, they will all still be running in dual channel as long as you put them in the correct slots. 4 slots, two of them will be one color, and the other two will be a different color. If you put the 2 matching pairs on one color and the other two matching pairs on the other color then all of your ram will be running in dual channel and you will have 6 GB.
 
I have 6 GB aswell, 2x2 gb + 2x1 gb. Works like a charm.
Before that, I had 2x1 gb + 2x512 mb, worked fine aswell.
As said, as long as you put them in the right banks, they run dualchannel.
A game like Crysis actually appears to use more than 3 gb at very high detail in some areas.
 
I dont know of any games that cause the use of 4GB let alone 6GB.
 
I have been running 6GB for a while. Works great. I cant tell much of a difference at all though from 4GB to 6GB.
 
I dont know of any games that cause the use of 4GB let alone 6GB.

ya rly i have played with 4 and 8 gig in my system and i can't tell a difference. tbo it won't hurt anything but u don't have anything to loose

 
What if the sets run at different voltages?

You can't do that (at least, not on any motherboard I've ever seen). There's only one global voltage setting for all modules.
So all modules have to run at the same voltage (and clockspeed/timings/etc).
So it's about the lowest common denominator.
 
I run 8 because well... its so cheap why not! :) If something come along that uses more than 4gb, I will never know there is a problem.. get me?
 
theyll run in dual channel. you wont see much of a gain in games but im sure your experience with the OS itself will benefit. if you have the RAM lying around anyway, might as well use it.
 
Seriously guys, I run Crysis with 2gig and a 8800 gts at maximum graphics and a resolution of 1680 by 1050 with no lag at all. I think it's more a waster of money than anything.. Oh and between, I just wanted to know, I know xp is limited to around 3gig max, but what is the max with vista?
 
Its not an XP vs Vista issue, its a 32bit vs 64bit OS issue.
 
Seriously guys, I run Crysis with 2gig and a 8800 gts at maximum graphics and a resolution of 1680 by 1050 with no lag at all. I think it's more a waster of money than anything.. Oh and between, I just wanted to know, I know xp is limited to around 3gig max, but what is the max with vista?

Are you running in Vista or XP?
My brother has 2 GB, and runs Crysis in Vista x64 at Very High detail. In some areas he still has problems where the HDD goes swapping.

Aside from that, the 3 GB max is not an OS-limit but a limit of a 32-bit address space.
Vista x86 has the same limit, but both XP x64 and Vista x64 can address far more memory. Much more than any motherboard or memory module can offer today (I think the max is 16 GB on a regular desktop board).
 
As you can see from the chart below, it is an OS limitation. Driver compatibility is one of the reasons as to why Microsoft put that limit into Vista, just like they did in XP/SP2.

http://msdn.microsoft.com/en-us/library/aa366778.aspx#physical_memory_limits_windows_server_2003

Yes and no.
They use PAE to extend the memory limit on some of the server versions. But not in the same way as an x64 OS does (effectively PAE gives you a 36-bit address space, it's no longer 32-bit).
With PAE disabled (which gives better performance and better software compatibility), these OSes still have the same problem, as do all other 32-bit OSes.
That's what I meant by saying it's not an OS limit. It is physically impossible to use the full 4 GB when you use a 32-bit address space.

They didn't as much 'put in' the limit as they didn't put in the option to go beyond the limit (PAE), because it isn't useful for anything but big server/datacenter installations.
It's better for normal home/office PCs/small servers/workstations to just skip PAE and its problems altogether, and go straight from 32-bit to 64-bit.
 
In the physical address space, PAE is as good as 64bit - up to a certain point. The PAE option in the CPU is also enabled when you go into 64bit mode, and it is not that slow. Vista runs in PAE mode by default, the extra addressing bits are just ignored.
 
No, the PAE mode of addressing is not used when you go into 64-bit mode.
I think you are confusing the /PAE switch in the Windows boot.ini with what PAE is in hardware (see http://en.wikipedia.org/wiki/Physical_Address_Extension).
Windows Vista (and XP SP2) uses a feature found in modern CPUs that can make memory non-executable. In order to use this feature, you must also switch the CPU in PAE mode.
However, since you still only use a single 4 GB window, there is virtually no performance loss, there are no compatibility problems, and you are still limited to 32-bit address space.
In order to use more than 4 GB, the OS has to switch the 4 GB window, and that can cause performance loss (and compatibility problems). In 64-bit mode this is not required, you still work with a single window, just as in regular 32-bit mode, or /PAE on regular Vista/XP.

So I think you sorta know a thing about PAE, but not exactly (it's not just 'extra bits', it's more like how the segmented 16-bit mode works on x86).
In short, the PAE mode is enabled, but set up in a 'bypass' mode on normal 32-bit and 64-bit versions of Windows.
 
You don't "swtich" any windows. Each process's address space just point to different locations in the physical address space. No loss here. (this is how paging works)

To enter 64bit mode, you need to flip on the PAE bit (you get a 64bit wide page table entry).
 
You don't "swtich" any windows. Each process's address space just point to different locations in the physical address space. No loss here. (this is how paging works)

You are confusing virtual address space with PAE.
The virtual address space itself is done with page translation, but this can only be done with 32-bit pointers.
So you have a window on a 32-bit virtual address space (but since it's not linear, you can map it over more than 4 GB of physical memory).
PAE also adds an extra level of hierarchy, you can have multiple page directories. But this is not used by regular 32-bit Windows. Switching between these page directories can cause extra overhead.
Also, regular 32-bit Windows doesn't map outside 4 GB of physical memory, because this causes problems with drivers that assume that physical addresses are always 32-bit.

Aside from that, each process in 32-bit PAE mode can only have 2 GB of address space. 64-bit mode does not have this restriction, so you don't have to resort to the Windows AWE API to map memory inside/outside the virtual address space if you want to access more than 2 GB of data.

So PAE really cannot be compared to 64-bit mode. I don't think you and I are talking about the same thing, or at least not from the same viewpoint.

To enter 64bit mode, you need to flip on the PAE bit (you get a 64bit wide page table entry).

Yes, but that doesn't mean you're actually making use of multiple page directories and such. 64-bit mode works the same as 32-bit mode in XP/Vista... PAE is on, but just because that's a technical requirement for other things you want to do (like no-execute memory and 64-bit mode), not because they're using the special features that allow you to address more memory. You're just enabling it and setting it up in a transparent way, then you don't touch the extra features.
 
I don't think I am confusing anything. The virtual address space is translated to physical addresses through the page table. Entering PAE mode makes each PTE 64bit wide (instead of 32), the same goes for the PTEs when you enter 64bit mode.

You don't "switch" or "slide" any "windows" by going PAE. That is a misunderstanding. The virtual address space is still 4GB, but that is not the point. Each process' user space just point to different places in the physical address space.

In non-PAE the page table is a 2-layer model. In PAE, it is 3 layer. 64 bit adds yet another layer, so you end up with four layers in total. And as I pointed out, Vista and XP/SP2 runs in PAE mode (thus 3 layers) by default.
 
I don't think I am confusing anything. The virtual address space is translated to physical addresses through the page table. Entering PAE mode makes each PTE 64bit wide (instead of 32), the same goes for the PTEs when you enter 64bit mode.

Yes, I already covered that.
So you agree with me here.

In non-PAE the page table is a 2-layer model. In PAE, it is 3 layer. 64 bit adds yet another layer, so you end up with four layers in total. And as I pointed out, Vista and XP/SP2 runs in PAE mode (thus 3 layers) by default.

I already covered that aswell.
So you agree here aswell.

You don't "switch" or "slide" any "windows" by going PAE.

I didn't say you did.
So the misunderstanding seems to be that you are thinking only about flicking the PAE bit, where I was mentioning the use of additional addressing capabilities, like the AWE API allows you to do.

Each process' user space just point to different places in the physical address space.

What is the significance of this statement?
This is true for all virtual addressing in Windows, regardless of PAE, 32-bit or 64-bit.

Oh well, nevermind. It's not leading anywhere, really.
Bottom line is that 64-bit has 64-bit virtual address space and PAE on a 32-bit system does not. And that's quite a difference for applications.
 
"Yes, I already covered that. So you agree with me here."

Where?

"I already covered that aswell. So you agree here aswell."

You said something about PAE in Vista was not really PAE, but only about a NX bit. And that there were no extra overhead.

"I didn't say you did."

Yes you did. You said: "the OS has to switch the 4 GB window". The AWE API is not PAE, and you don't have to use it, if you don't want to. Multiple applications can easily use more than 4GB together without.


"What is the significance of this statement?"

It was to your statement about the "window" the OS has to "switch".

"Bottom line is that 64-bit has 64-bit virtual address space and PAE on a 32-bit system does not. And that's quite a difference for applications."

Which is not the point at all in this discussion.
 
Memory pissing match... Yay! :rolleyes:


Back on topic, 2x2 + 2x1 works flawlessly so long as you make sure to run dual-channel and that all modules will work at "X" rated voltage.
 

Well, I referred to it here, although not specifically stating 64-bit:
"So you have a window on a 32-bit virtual address space (but since it's not linear, you can map it over more than 4 GB of physical memory)."
...
"Also, regular 32-bit Windows doesn't map outside 4 GB of physical memory, because this causes problems with drivers that assume that physical addresses are always 32-bit."

You said something about PAE in Vista was not really PAE, but only about a NX bit. And that there were no extra overhead.

I said a lot more than that.
For example: "PAE also adds an extra level of hierarchy, you can have multiple page directories."
My point was that PAE is used because the NX bit is only available in the extended page structure of the PAE mode. Aside from using this NX bit, the addressing isn't changed, and still only uses 32-bit physical address space.

Yes you did. You said: "the OS has to switch the 4 GB window". The AWE API is not PAE, and you don't have to use it, if you don't want to.

Yes, but then you're limited to the 2 GB address space limit, which you're not in 64-bit.

Multiple applications can easily use more than 4GB together without.

In theory yes, but in practice most 32-bit versions of Windows ignore the fact that they now have 64-bit tables, and only map the first 32-bits of physical memory. So they use PAE, but they don't use PAE. That was the point, wasn't it?

It was to your statement about the "window" the OS has to "switch".

If you want to address more than 2 GB in an application, yes.

Which is not the point at all in this discussion.

Well I think it was.
Namely, the discussion started when I said:
"Aside from that, the 3 GB max is not an OS-limit but a limit of a 32-bit address space."
Then you came up with PAE (which isn't a 32-bit address space in the first place), to which I said that it's not the same as having an x64 OS, where you get a 64-bit address space for all processes (as opposed to 32-bit virtual windows on a 64-bit physical address space).
What do you think the point was?
 
I could surely say a lot more, but I don't think I get anywhere. So lets stop here.

Edit: just a quick note about your window. When you say "the 4GB window", how can I possible read that otherwise than you saying: "the OS has to swtich a window to address more than 4GB". How can a 32bit application possible allocate a 4GB AWE window?
 
Edit: just a quick note about your window. When you say "the 4GB window", how can I possible read that otherwise than you saying: "the OS has to swtich a window to address more than 4GB". How can a 32bit application possible allocate a 4GB AWE window?

I can't be arsed to look at all the 2 GB and 4 GB numbers and OS and applications thrown about really. Forget it.
 
Man people getting lost in here, well have run 4 and 6 in vista 64 and can't tell the difference, have a good one
 
You can't do that (at least, not on any motherboard I've ever seen). There's only one global voltage setting for all modules.
So all modules have to run at the same voltage (and clockspeed/timings/etc).
So it's about the lowest common denominator.


Rather, the highest common denominator... you'd have to run the highest minimum voltage necessary for all modules.

Take for instance 2x2GB modules requiring 1.8V and add 2x1GB modules requiring 2.0V... you would have to run 2.0V to all of them.
 
Rather, the highest common denominator... you'd have to run the highest minimum voltage necessary for all modules.

Take for instance 2x2GB modules requiring 1.8V and add 2x1GB modules requiring 2.0V... you would have to run 2.0V to all of them.

Depends on how you look at it.
I was thinking of running at the lowest maximum voltage (which should have better stability/overclockability than the minimum voltage).
Aside from that, I wasn't thinking about voltage alone, but also frequency and latency timings.
Ofcourse you can be creative, because memory can often be run well outside spec. When I had 3 GB, I used Kingston valueram of 667 MHz, 1.8v and 5-5-5-15 timings.
I however ran all 4 modules (2x1024 and 2x512) at 833 MHz, 2.05v and 4-4-4-12 timings. It was perfectly stable for well over a year, until I went to 6 GB. Ironically enough I bought actual 800 MHz memory for the 2x2048 models that replaced the 2x512 ones, but I can't run them at 833 now, not even 800... I have to settle for 667, probably because my motherboard doesn't handle 2 GB modules very well.

Also the 'lowest common denominator' doesn't literally mean the lowest number, but the weakest of them all (which could in some cases be the highest number, depending on what that number means). So it's still not 'highest common denominator'.
 
i just bought some ocz reaper 2x2 8500 sticks to replace my crucial ballistix 2x1 8500's. when i received my package, just for shits i simply added the reaper sticks and booted up, its been running smoothly for a few days now, prime stable 24 hours. i am, however using lower speed and tighter timings then both sets of ram come stock with. im running them 800mhz 4-4-4-12, with 1.9V
(i have my q6600 clocked at 3.2ghz (400x8) w/ a 1600mhz fsb. the ballistix have been great for about 10 months now, its only with my recent switch to vista 64-bit that i felt i needed >4 gigs.
im sure the diff between 6g and 4g is not noticable.
 
Back
Top