I'm going to buy a mother board, one is Biostar TA785GE 128M with ATI Radeon HD 4200 onboard - 128 sideport memory included. I don't want to share my main memory, thus I hope to have almost independent video card, but I've heard that in order to work I must give it some main RAM or computer won't even pass POST...
Is this problem solved? If not, what's the point of having sideport memory and still share main RAM? May I hope that some time later I'll be able to use sideport memory with dirver (any open source or proprietary...) using it correctly, not requiring me to share main RAM?
I just checked this with my Asrock A780GXH/128M it doesn't have a radeon hd 4200 but should be similar enough.
With shared memory disabled the X display is heavily corrupted, same when I set shared memory to 32mb. With 64mb and up it works.
AFAIK the igp can take advantage of the combined bandwidth of sideport and system ram. So if you only use sideport you lose some performance. I saw some benchmarks that showed a pretty noticable difference.
Sideport memory is primarily there for low power operation (you don't have to wake up the CPU to access CPU-attached memory). It tends to have a very narrow memory bus and so is not particularly fast. IIRC sideport memory on the system referenced in that article has a 16-bit memory bus, vs 64-128 for UMA system memory and 64-256 for a typical discrete GPU.
Just curious, what is the reason for wanting to use only sideport memory given that it is the slowest memory option ?
Thanks for your replies, guys.
I thought that then I will have better overall system (not 3D) perfomance, video chip won't be using CPU to access main RAM, and that's like having an external video card.Just curious, what is the reason for wanting to use only sideport memory given that it is the slowest memory option ?
I'm not really a gamer, so I don't really need 3D, just looking for a good motherboard, so is this a good choice?
Won't it lower the overall perfomance to use sideport+main RAM?
There are rarely simple answers when it comes to performance
I expect that using sideport + UMA (if the drivers support it, I'm not sure) would improve your graphics performance but might make background tasks run more slowly because of memory contention. I expect your overall perception of performance would be highest with sideport+UMA, followed by UMA only with sideport-only being the lowest. I don't know if the drivers even try to support sideport-only for normal use.
Once you add compositing into the mix all the performance equations change again, and there it might make sense to have the compositor target buffer(s) in sideport memory and all the normal render targets in UMA memory, but I'm not sure the compositors are sufficiently smart to deal with that kind of arrangement yet. It will make for some interesting tinkering with the open source stack over the next year or so, I think.
For the moment I would think about sideport memory as a way of boosting graphics performance when used in conjuction with UMA, at least for a desktop system.
Last edited by bridgman; 09-11-2009 at 08:49 PM.
So, should I avoid buying MB with sideport at all? As I understand it can give me (i'm not a gamer) only driver problems?I expect that using sideport + UMA (if the drivers support it, I'm not sure)
I have such suggested PCI-E card, it is not worth to buy until there is video accelleration for it. until then you could run the onboard using fglrx too. When i try to watch BD media with about 30 mbps then even a dual core E8400 is at 80% or more on each core (with mplayer mt) and tends to lose av sync. vlc would be better from sync but only runs on one core. vdpau+mplayer at least works for video but not fully prefect with a/v sync. Also mplayer currently needs -demuxer lavf for m2ts.