» Login to reply to this

Kromis (675)     Dec 18, 2008 - 08:29 pm

Why stop at quad? We got 3 PCI-E slots now, man! » Login to reply to this  mr.wielder (1416)   Dec 19, 2008 - 09:38 am Heh, good point! That said, I'm not sure I expect to see that kind of Nvidia board for AM3 right away but one can always hope. [That other site I visit for Hardware reviews] has some thoughts on AA scaling limitation in SLI vs. the 4870x2 that I'm really curious about. But I DO know that Nvidia was planning AM3 chipsets for Q1 '09, so I'd also like to know what the overall performance scaling might be like on an AM3 platform with Quad-GPU.. Future dream build: Phenom 2 + SLI-GTX295 » Login to reply to this  mr.wielder (1416)    Dec 19, 2008 - 04:44 pm | Edited on Dec 26, 2008 - 04:46 pm Uhm.. no. edit: Your question is rhetorical (obviously you aren't that ignorant) so you're most likely an Intel fan, but I personally can't wait for AM3 and Phenom 2. Not like I'm exactly suffering with my current X2 setup.. » Login to reply to this  DanTheMathlete (235)    Dec 19, 2008 - 03:06 pm | Edited on Dec 20, 2008 - 09:30 am I thought amd was dead in the water? What does AMD have in the works? » Login to reply to this  DanTheMathlete (235)   Dec 19, 2008 - 07:12 am Couldn't 6Xs SLI have been run with the 9800X2? It might be a drivers issue, or heat. » Login to reply to this  Smee (20)   Dec 21, 2008 - 03:56 pm I get what your saying; however, If all cores are working on the same frame, just separate lines of the frame, then a character coming in and out of shadow (per your example) wouldn't be the same as a quad core working on a single thread. In essence each core works on a separate line of the same frame and each core has the same percentage workload of that frame (for all intensive purposes the overhead of one gpu having a very small percentage extra work is minimal... say one gpu that gets more sky to render than the others since each gpu is rendering every other/3rd/4th ect line). This in theory should cause each frame to be rendered, say nearly 4x faster, for a 4 core GPU. Anyone have Hardware GPU xp that could shed light on this? Would love to hear it. » Login to reply to this  deimos47 (445)    Dec 20, 2008 - 02:36 pm Both AMD and nVIDIA have various rendering modes - alternate lines, tiles, alternate frames, split screen etc. why stop at 4xGPU or even 10xGPU. Why not 100xGPU? Diminishing returns. Driver overhead. If each is rendering a frame, ignoring DX/OGL max render frames ahead limits, might get 100 frames of lag. Best reason is probably synchronization. I'm no expert, but guessing it works the same way as forks and threadpools in Linux programming... Same reason why your quad core CPU cant run single thread 4x faster. There are data dependencies, so one is waiting on answer from another. In GPU example, where you have 500MB+ of textures and tens of millions of polygons, when GPU1 calculates new shading for player coming out of shadow, it caches this so it doesnt have to recalculate next time. But, how does GPU2 know without recalculating. You can share memory, but then have data integrity problems of a GPU reading data another is writing to. » Login to reply to this  Smee (20)   Dec 19, 2008 - 11:16 pm GX-Brandon I remember back with the voodoo 2 3dfx's SLI had one voodoo 2 render all odd and the other all even lines and spliced them together for each frame.... why doesn't Nvidia or ATI do this??? It seems like a much more efficient way to render than alternating frames, and would not run into the same problems trying to render 3 or 4 frames in advanced does.... If each card only renders every other line (or every 3rd/4th/6th line for 3-way/quad/hex) the load of polygons will be nearly evenly split between all cards. Or is this wrong? » Login to reply to this  GX-Brandon (373)   Dec 19, 2008 - 02:04 pm Queuing ahead that many frames would be a difficult feat for driver team, I'm also not sure if DX supports it. We're already seeing diminished returns from 3 and 4-Way so I don't see 6-Way becoming reality anytime in the near future. » Login to reply to this

» Note: You need to be logged in to write a comment!