Core i7 and Games, Memory Testing 6GB vs 3GB

Why would you use Vista x64 instead of Vista x86 for the 3Gb benchmarks?

For 3Gb using x86 would have been a better test as it's far more economical with memory thanks to all variables being 32bits long instead of 64. This alone can make applications use 20-40% more memory and negates the advantages of x64 bit while using <4Gb.
 
I'm not sure exactly what variables you are talking about. However, the 32bit version of Vista will not fully support the 6GB kit so it was not logical to use it in this test.
 
Stonedofmoo said:
Why would you use Vista x64 instead of Vista x86 for the 3Gb benchmarks?
Click to expand...

Hello Stonedofmoo, you have to take into account that even though you have 3GB of RAM installed, even if you are using a 32-bit operating system, it doesn't necessary mean that all of that 3GB is going to be addressed. In the case of the article that Yellowbeard has just linked too, because of the sort of system they have used for this test, if they used a 32-bit operating system, the amount of memory that would have been addressed would have just been under 2GB. Even if they had only used one Nvidia Geforce 280 GTX, I believe the amount of memory that would have been addressed would have been just under 3GB since the graphics card isn't the only Memory Mapped I/O Device in a system.

There is also the variable of "fair testing". Using a 32-bit operating system whilst only having 3GB of RAM installed and then a 64-bit operating system when using 6GB of RAM isn't exactly a fair test.

For 3Gb using x86 would have been a better test as it's far more economical with memory thanks to all variables being 32bits long instead of 64. This alone can make applications use 20-40% more memory and negates the advantages of x64 bit while using <4Gb.
Click to expand...

Why you would want to use a 32-bit system in this day and age is beyond me. Though, you are indeed correct that using a 64-bit system does indeed results in a large memory footprint in running applications. However, I am not quite sure it is to the extent of 20-40%. Have you got any kind of evidence to back that figure up? I would be quite interested to know if that is actually the case. :)
 
A few points.

1) If you're comparing the difference in performance between different amounts of memory, then it only makes sense to keep everything else (and that especially includes the OS!) the same. Otherwise you're comparing apples and oranges.

2) 32-bit stuff is legacy. Get over it. ;)

3) 64-bit stuff ISN'T particularly less memory efficient. Sure, pointers are 48 bits instead of 36 bits (PAE), but this won't make a huge difference. Data itself is the same size (although larger variables are supported, you have to explicitly program your apps to use larger types, and any sane programmer will make sure they use types of appropriate size for what they are doing i.e. not using a long when an int will amply suffice). Another thing that contributes to x86-64 binaries being bigger (and use a bit more memory) is that x86-64 has double the number of general purpose registers on the CPU. This makes a massive difference in performance where appliactions are bound by register contention. x86 only has 8 general purpose registers (one of those legacy crapness things). This is outrageously small compared to just about every non-embedded RISC processor in use today. x86-64 doubles this by adding another 8 general purpose registers. 16 is still relatively few compared to the competition, but it's a lot less bad.

4) Not all GPU memory is mapped into the CPU address space (this used to be known as the "aperture" in the AGP days). Personally, I think it'd be awesome if it was because I could then use the excess RAM on my GPUs via MTD drivers under Linux as very fast block devices (e.g. for swap or /tmp). But, sadly, it isn't. Only the apertured amount eats into the CPU address space. Whoever originally started to propagate the myth that 32-bit machines were unable to address all the memory in a cases where RAM + VRAM > 3GB was talking out of the wrong orifice with just enough knowledge to make them dangerous while not enough knowledge to stop them from being ignorant. (BTW, if anyone knows of a way to map 100% of the GPU RAM into the aperture, please do tell! :) )
 
Didn't you learn anything in school science?
It's all about the FAIR TEST changing only ONE VARIABLE.

Is there an fps change in gaming?
 
Yellowbeard said:
Back in April 2008 we tested 4GB vs 2GB. Now that the Core i7 memory offerings are different, we did more testing with 6GB vs 3GB. We got very similar results. Vista with 3GB on i7 is good but 6GB is much better. :cool:

APPLICATION NOTE 811
Click to expand...

It would be very interesting to see the same test performed on Windows 7
I wonder if that would perform much better on 3GB ?
 
Tuvoc said:
It would be very interesting to see the same test performed on Windows 7
I wonder if that would perform much better on 3GB ?
Click to expand...
There's no way we'll do any substantial testing on Windows 7 until there is a retail version released. It would be pointless at this time and I don't think the device driver support is very thorough yet.

And, the OS cannot improve the limitations of 3GB. The combination of a game and OS devouring more than 3GB of memory when it is available means that you simply need more than 3GB of memory.
 
Tuvoc said:
It would be very interesting to see the same test performed on Windows 7
I wonder if that would perform much better on 3GB ?
Click to expand...
There's no way we'll do any substantial testing on Windows 7 until there is a retail version released. It would be pointless at this time and I don't think the device driver support is very thorough yet.

And, the OS cannot improve the limitations of 3GB. The combination of a game and OS devouring more than 3GB of memory when it is available means that you simply need more than 3GB of memory.
 
Fledge said:
Didn't you learn anything in school science?
It's all about the FAIR TEST changing only ONE VARIABLE.

Is there an fps change in gaming?
Click to expand...

Id have thought this and similar posts are missing the point slightly. The question should be 'do I want 3gb or 6gb for games' not is vista 64bit faster at games with 3 or with 6gb

A benefit to only running 3gb is that you can run 32 bit vista, or heavens forbid, XP. I'd expect 32 bit vista to outpace 64 bit on a 3gb system (based on 32 bit linux running a lot quicker than 64 bit on slower hardware). I have some doubts about how many people with less than 4gb of RAM are actually running 64 bit os.

Obsessing over fair tests while answering the wrong question is less useful than answering the right question as well as you can, whatever school biology might teach.
Good link despite this, thanks :)
 
JonJ678 said:
Id have thought this and similar posts are missing the point slightly. The question should be 'do I want 3gb or 6gb for games' not is vista 64bit faster at games with 3 or with 6gb

A benefit to only running 3gb is that you can run 32 bit vista, or heavens forbid, XP. I'd expect 32 bit vista to outpace 64 bit on a 3gb system (based on 32 bit linux running a lot quicker than 64 bit on slower hardware). I have some doubts about how many people with less than 4gb of RAM are actually running 64 bit os.

Obsessing over fair tests while answering the wrong question is less useful than answering the right question as well as you can, whatever school biology might teach.
Good link despite this, thanks :)
Click to expand...
Excellent points and thanks. Another aspect that some users overlook is the effect of the 32bit limit on the MMIO addressing. Each of my 280GTXs have 1GB of Vram. And, X58 MOBOs are averaging about 1GB of address space just for the PCI addressing. I have not tried this system with any 32bit OS but, I would think that the OS would see quite a bit less than 2GB of memory which is worthless.
 
You must log in or register to reply here.
Back
Top Bottom