UMA advantages/disadvantages?

[CTho9305]

wouldn't a major disadvantage of a unified memory architecture be that there would be obscene bandwidth requirements? Do you think that over all, the Xbox's UMA would be able to be faster than PCs?

 

[Peter]

UMA advantages: No dedicated video RAM needed (cost). All RAM is the same speed - you don't see a speed bump when exceeding the video unit's local RAM size, so you can make your scene's graphics as large as total RAM allows.

UMA disadvantage: On a standard mainboard, you can't put the RAM as close to the graphics chip as you can on a proprietary design (VGA card or console PCB), thus you can't clock the RAM as high - and you have all other RAM users (CPU and other bus master devices) eating from the same RAM's bandwidth.

As for the bandwidth, surely it has an impact ... but as far as gaming at REASONABLE resolutions and color depths (as seen on consoles) is concerned, frame rates are reasonably good with existing solutions.

Popular UMA solutions with SDRAM (like SiS 630 and 730 chipsets) are surprisingly fast when you keep the latter sentence in mind, the new DDR ones (SiS 740 and 650, as well as NVidia nForce) even more so. Sure, they're always two or three steps behind current high end graphics cards, but UMA is about providing an affordable solution to reasonable people, remember?

regards, Peter

 

[CTho9305]

ah, the resolution is probably a huge factor in cutting the bandwidth requirements.

 

[Peter]

Color depth too (16-bit vs. True Color cuts the bandwitdh by 33 or even 50 percent!), and refresh rate as well. The latter is often overlooked, yet games usually look just as good at 60 Hz as they do at 100. Windows tends to set very high refresh rates for the low resolutions by default.

regards, Peter

 

[CTho9305]

well, as I can see 72Hz flicker (slightly), I'm kinda screwed, playing games with flicker is annoying so the bandwidth is required. TV's should refresh faster too

 

[andreasl]

As I understand it, Xbox is UMA (Unified Memory Architecture) and NForce and all other integrated graphics PC chipsets are SMA (Shared Memory Architecture). While both uses the same memory chips or modules, in the case of SMA the memory space is split in half with one part dedicated to the graphics and the other part to the CPU and other devices. In UMA the memory space is shared among all devices.

Someone correct me if I'm wrong.

 

[Peter]

CTho, I doubt you'll see 60 Hz flicker with the dark and low edge contrast images in a game. On the work desktop where things
are black on a white background, things are much different.

andreasl, no difference between UMA and SMA. You still have ONE memory bus to share between graphics engine and the
rest of the system. All that NVidia did was widen that bus to 128 bit to double its bandwidth to start with. Not a new trick either,
ALi's ArtX chipset had that too, and even the oh-so-crappy SiS 630/540 chipsets could - albeit with the latter two, practically noone
offered that feature.
NVidia's pretty good at making big noise about little news, that's all I can say about nForce.

regards, Peter

 

[CTho9305]

On large flat areas the flicker is visible. yes, you are correct that generally it isn't visible.

 

[Remnant2]

UMA has one VERY large advantage for 3d gaming: There's no need to waste time uploading textures to local memory from system memory. Although with 64mb video cards the norm now, most textures do fit in vid mem, the AGP bus will become more and more of a bottleneck as textures spill over into main memory.

With UMA (particuarily if you have a very fast UMA), you can use as much textures as you have system ram, and you can modify those textures for "free" (no uploading required), allowing for nice procedural effects without the traditional upload time required.