Valve Rolls Out A New Steam Storefront

[Luke_Wolf]

Although this isn't actually accurate. If a user is running a 32-bit app on a 64-bit OS, they have access to the full 4GB address space. It's only the 32-bit OS's that have to reserve memory the games can't access.

However, real games actually are bumping up against that 4GB virtual memory limit, and have been doing so for quite some time.

well under windows you have to edit the binary's flags to allow it to access 4gb as opposed to 2gb but yeah

[dragorth]

This is really miguided... Video cards currently utilize GDDR 5 which is much faster than DDR 3 that the desktop uses. This trend of video cards using different RAM tech will stay that way and 64-bit will not bridge that gap...

Don't conflate the video card having its own memory with its own bandwidth as being the same as regular memory. First, GDDR has to be faster than regular memory thanks to its dual purpose. It has to act as a buffer for the texture/model data that the system gives it, and has to act as a buffer for the actual calculations being performed by the GPU.

Regular memory acts as a buffer for the HD, and as a buffer for the cpu calculations. Since both of those are slower than the GPU/Memory combination, GDDR5 and whatever new that will be coming out has to do a much faster job.

Also, 64bit is going to bridge that gap in one very important way. Mainly this Heterogeneous Computing that Intel and AMD are pushing toward leads to having one heterogeneous memory space, which reduces the copy to GDDR5 memory operation. This will allow them to use system ram as GPU memory in their integrated graphics. But, Nvidia is in on this little game as well, since this will increase the ease of coding CUDA and OpenCL applications, as well as simplify their driver.

(The reality is you are offloading it from the cpu onto the GPU by simply requesting the relevant data, and the GPU doesn't care where it is. This will make compilers much more important for performance, as they will need to guess what needs to be closest to the GPU at any one time, and arrange for that state, or games will have to do that step themselves.)

[mwpow3ll]

Don't conflate the video card having its own memory with its own bandwidth as being the same as regular memory. First, GDDR has to be faster than regular memory thanks to its dual purpose. It has to act as a buffer for the texture/model data that the system gives it, and has to act as a buffer for the actual calculations being performed by the GPU.

Regular memory acts as a buffer for the HD, and as a buffer for the cpu calculations. Since both of those are slower than the GPU/Memory combination, GDDR5 and whatever new that will be coming out has to do a much faster job.

Also, 64bit is going to bridge that gap in one very important way. Mainly this Heterogeneous Computing that Intel and AMD are pushing toward leads to having one heterogeneous memory space, which reduces the copy to GDDR5 memory operation. This will allow them to use system ram as GPU memory in their integrated graphics. But, Nvidia is in on this little game as well, since this will increase the ease of coding CUDA and OpenCL applications, as well as simplify their driver.

(The reality is you are offloading it from the cpu onto the GPU by simply requesting the relevant data, and the GPU doesn't care where it is. This will make compilers much more important for performance, as they will need to guess what needs to be closest to the GPU at any one time, and arrange for that state, or games will have to do that step themselves.)

I am well aware of GDDR vs DDR, and again... 64-bit architecture with DDR3 / 4 isn't going to make a dang bit of difference against GDDR5, but don't take my word for it. If you look at the difference between PS4 and XBox One, both are running AMD x86-64 architectures but PS4 has straight GDDR5 and XBox One has DDR3 with ESRAM. The only reason XBox One can compete is because of the ESRAM which completely invalidates your idea that system RAM is enough... it's not.

[smitty3268]

well under windows you have to edit the binary's flags to allow it to access 4gb as opposed to 2gb but yeah

It's just a compiler flag in MSVC, and most apps which can take advantage of that much memory go ahead and set it.