Macbook Pro can run games that require 872 MB Video RAM?

[TheStranger55]

I got the Santa Rosa Macbook Pro with 128 mb of VRAM but when I run DiaxDialog in Windows Vista Bootcamp it says I have 872 MB of Video ram. I'm guessing the macbook pro is currently taking main memory and using it as video memory as well?

Does this mean if a game requires more than 128 mb of vram to run, my Macbook meets the system requirements? I know it won't be as fast as having the actual videocard ram on the card but some games won't even open unless you meet that minimum amount of ram, which is what I"m concerned about rather than it stuttering a little.

 

[MRU]

^


Yeah my ATI X1900XT reports the similar in Vista. It says 1.25gb in dx-diag. I'm guessing it can access memory if needs be.

 

[TheStranger55]

well I guess if I ever do buy a game that requires 256 meg of VRAM and it doesn't work I'll just send the company my dxdiag which shows I have more VRAM than their game requires and they gotta support me until it works.

Considering most PC games don't say you need a "IBM PC compatible" on their requirements box, they just list processor, ram and videocard then I guess the MAC does technically meet those requirements.

 

[Freyqq]

I got the Santa Rosa Macbook Pro with 128 mb of VRAM but when I run DiaxDialog in Windows Vista Bootcamp it says I have 872 MB of Video ram. I'm guessing the macbook pro is currently taking main memory and using it as video memory as well?

Does this mean if a game requires more than 128 mb of vram to run, my Macbook meets the system requirements? I know it won't be as fast as having the actual videocard ram on the card but some games won't even open unless you meet that minimum amount of ram, which is what I"m concerned about rather than it stuttering a little.

I assume it would b\c if you think about it..all the integrated cards are only shared ram and they're considered to have video ram

 

[TheStranger55]

Well I found the answer to my own question.

http://www.whoismadhur.com/2006/04/08/directx-10/

The biggest change for game developers will be virtualized memory for the GPU. The video card will now have its own space in system RAM to store information that can’t fit on local video card memory. High-end video cards ship with 256MB or 512MB of memory, but games can still use the extra space in system memory to store large chunks of information, like textures.

So with a DX 10 gpu Geforce 8600 GT and Vista there is a feature called "Virtualized memory", where if more VRAM is needed it will use System memory when previously in DX 9 cards it was only limited to Video Memory.

Unfortunetly this gives me another reason to get the PC version of games for Vista rather than the Mac Version as Mac OS X doesn't have DX 10 and that's why it only reports 128 MB of VRAM for my card while Vista reports 872 MB.

I guess it also gives me a reason to add more ram too.

I wonder though will a game company offer tech support for a game that say requires 512 MB of VRAM and your card physically has 128 MB of VRAM but with the virtualized memory it has 872 MB or will they only take into account the physical VRAM and not the virutalized memory?

 

[Krevnik]

All video cards from the AGP-era and newer can share system memory if it runs out of RAM on the card. DX10 just has some fancy bits so that you can see both chunks of memory as a single hunk of space, rather than having to manually manage both slices of memory manually.

So really, this isn't new, and DX10 doesn't do anything new either, just makes it so that existing functionality is easier to manage.

 

[jdechko]

Maybe this partially explains why Apple decided to limit the (dedicated) VRAM to (up to) 256 in the newer machines-minus the Mac Pro, of course. With the limit of actual RAM bumped to 4GB there's probably gonna be a little extra left over in most cases, so Apple saves costs.

 

[Krevnik]

Maybe this partially explains why Apple decided to limit the (dedicated) VRAM to (up to) 256 in the newer machines-minus the Mac Pro, of course. With the limit of actual RAM bumped to 4GB there's probably gonna be a little extra left over in most cases, so Apple saves costs.

VRAM still has MUCH better throughput than system memory to the GPU. There is a reason why integrated graphics never keep up with a video card with its own VRAM. It is just that a GPU being able to access system memory means that performance will degrade instead of the game not running at all on systems without enough VRAM for all the textures a game uses.

 

[Erasmus]

VRAM still has MUCH better throughput than system memory to the GPU. There is a reason why integrated graphics never keep up with a video card with its own VRAM. It is just that a GPU being able to access system memory means that performance will degrade instead of the game not running at all on systems without enough VRAM for all the textures a game uses.

Now, I know that it's not nearly this simple, but the system RAM runs at 667MHz. The GPU ram runs at like 700MHz. It might run faster being GDDR, and it will certainly have far, far lower latencies, but the point is, the system RAM should be perfectly fast enough to act not as a source of texture information for the GPU core, but as a massive pool of available storage so that the GPU RAM needs hold only what the core needs right now, and is fast enough to keep the GPU RAM up to date with what the core requires. If the core needs more textures than the GPU RAM can store, you might get some slowdowns, but it should still run well because system RAM is fast.

And of course it is always better to have extra storage space. Can you imagine what it would be like if the GPU used Virtual Memory (Off the Hard Disk) to store the information it needed???

 

[Krevnik]

Now, I know that it's not nearly this simple, but the system RAM runs at 667MHz. The GPU ram runs at like 700MHz. It might run faster being GDDR, and it will certainly have far, far lower latencies, but the point is, the system RAM should be perfectly fast enough to act not as a source of texture information for the GPU core, but as a massive pool of available storage so that the GPU RAM needs hold only what the core needs right now, and is fast enough to keep the GPU RAM up to date with what the core requires. If the core needs more textures than the GPU RAM can store, you might get some slowdowns, but it should still run well because system RAM is fast.

You are right, it isn't that simple.

The bridge between the graphics card and system memory is slow, and is a bottleneck. You also have more bottlenecks when you have the CPU /and/ the GPU trying to use the system RAM at the same time. This means you have to share the RAM bandwidth, which makes the performance of both suffer. Also, VRAM tends to have wider buses (128-bit or 256-bit) optimized for pixel-sized blocks of memory that you don't have on standard memory, doubling, or even quadrupling your bandwidth compared to system memory (assuming 64-bit system memory) at the same clock-speed.

Now, what you described is pretty standard for how games are programmed for texture memory. The catch is that this management of sending textures to the GPU is automatic. If you don't have enough VRAM to render a single complete frame, you get texture swapping during a frame render, which is bad for performance, because of how draw order is usually tweaked to address a limited fill rate, rather than limited texture memory.

 

[TheStranger55]

Well there is no onboard gpu that is DX 10 capable so would it be fair to compare DX 10 memory sharing with non-DX 10 onboard memory sharing? I also believe DX 10 is supposed to be a new architecture that requires less calculations for graphics task which means a less performance hit.

Maybe with DX 10 in Vista the system ram use for the graphics won't be as bad as the non-DX 10 onboard system memory sharing performance-wise?

Well I don't know much about this area, I hope DX 10 will make it possible for me to play high intensive games on my low end Geforce 8600m GT gpu.

 

[Krevnik]

Well there is no onboard gpu that is DX 10 capable so would it be fair to compare DX 10 memory sharing with non-DX 10 onboard memory sharing? I also believe DX 10 is supposed to be a new architecture that requires less calculations for graphics task which means a less performance hit.

All GPUs have memory sharing, integrated or discrete. It isn't fair to isolate DX10 as if it is some sort of miracle, because it isn't. While a lot of work has gone into DX10 to make it better for programmers, there isn't really all that much difference between a DX9-class and a DX10-class GPU with the exception of the shader model /must/ be shader model 4.0 compatible. The rest of DX10's 'features' (which benefit the programmer directly, making it easier to do what was already possible, rather than trying to make more possible)... don't need hardware support.

Maybe with DX 10 in Vista the system ram use for the graphics won't be as bad as the non-DX 10 onboard system memory sharing performance-wise?

Not really. What I described was a set of hardware limitations. Software cannot overcome these limitations, although it can find ways to be more efficient about it. But system memory sharing performance for video cards will always suck, and DX10 can't fix that, since the reason for it revolves around PCIe, and other pieces of hardware.

 

[Evangelion]

All video cards from the AGP-era and newer can share system memory if it runs out of RAM on the card.

Well, not really. AGP-vidcards could use system-RAM for texture-storage, and that's it. PCI-E-cards could transparently use both system-RAM and VRAM together, and not just for textures. They could use it for framebuffer as well. This is possible because PCI-E has a lot more bandwidth in both directions than AGP did. Whereas AGP had reasonable amount of bandwidth from RAM to the vidcard, PCI-E has more bandwidth and that bandwidth runs in both directions.

DX10 just has some fancy bits so that you can see both chunks of memory as a single hunk of space, rather than having to manually manage both slices of memory manually.

That functionality was implemented in some GPU's long before DX10. 3DLabs had it in their last vidcards, and those were aimed squarely at OpenGL.

 

[gnasher729]

Well I found the answer to my own question.

http://www.whoismadhur.com/2006/04/08/directx-10/



So with a DX 10 gpu Geforce 8600 GT and Vista there is a feature called "Virtualized memory", where if more VRAM is needed it will use System memory when previously in DX 9 cards it was only limited to Video Memory.

Unfortunetly this gives me another reason to get the PC version of games for Vista rather than the Mac Version as Mac OS X doesn't have DX 10 and that's why it only reports 128 MB of VRAM for my card while Vista reports 872 MB.

I guess it also gives me a reason to add more ram too.

I wonder though will a game company offer tech support for a game that say requires 512 MB of VRAM and your card physically has 128 MB of VRAM but with the virtualized memory it has 872 MB or will they only take into account the physical VRAM and not the virutalized memory?

It is very simple: VRAM is fast, and virtual VRAM is sloooooow. To be precise, it is exactly as fast as on one of the integrated graphics cards like the MacBook, that some people are so horrified of.

If a game says that it "requires" 512 MB of VRAM, it most likely doesn't mean that it needs 512 MB to work, but 512 MB to run at an acceptable speed. So with 128MB VRAM + 872 MB virtual memory you can expect this game to run dog slow.

On top of that, some games will adapt to the amount of VRAM; for example, if you have only 128MB, they can reduce texture quality and run fast that way. With virtual VRAM, the game might not realise that there isn't enough real VRAM and use higher texture quality and be unnecessarily slow because of that.