AM3 Reference Diagram Disclosed 65
psyph3r writes "Chilehardware has released what appears to be a confidential image showing the future customer desktop AM3 reference boards for AMD and ATI. Here is an English site talking about this reference design image and the features it enables. 'The biggest improvement for this generation of chipsets is the audio and video capabilities integrated into the motherboard. The new features packed into these chipsets are beginning to look like standalone platforms. The RS780 supports DirectX 10 and has a UVD, which is similar to most High-end cards of today.'"
Re:Why usb 1.1 and 2.0? and why not use HT for the (Score:3, Informative)
As for PCIe vs. HT, they're probably so similar in latency and throughput at that level that its just a difference in transistor count or something similarly insignificant.
Re:What I want from a motherboard... (Score:5, Informative)
hint: their internal arch. resamples ALL data to 48k. even 48k gets resampled (man!, that's dumb) to 48k. hopes of having literal bit-perfect 44.1 is hopeless with creative brand.
envy24 - full-on pro chipset. I've used that one in my studio for years.
before that was the cmedia 8738 (still a gem if you can find it). also bit-perfect and has some great free drivers (sourceforge) that allow kernel streaming (win-xp) and bit perfect mixer-goof-proof output.
almost all else is drek. ie, junk.
Re:Call me ignorant, but... (Score:5, Informative)
In the case of the discrete cards (PCI-E, AGP...) they have a pool of memory that's accessible via the bus and that's directly accessible by the GPU's own memory bus (That memory size when you see 128, 256, 512Mb, etc.)- which is faster than just about anything out there and has no contention spots for the GPU to have to wait any longer than the access latency to the memory from the second access port. The peak speed of the GPUs when compared to an IGP solution comes from the contention-less, very, very fast access to the card's memory pool so that you don't stall the graphics pipeline. A stall of a microsecond can cost you FPS (Duh...) and larger stalls can drag framerates to the slide show domain- it's part of why the older ATI fglrx drivers were roughly 50% slower under Linux when compared to Windows. They had a stall in there somewhere that was introduced by their way of getting their then Windows-ish codebase to work under Linux.
Now, having said this, Hypertransport's suspiciously close to the same performance level of most of the local GPU buses and you only need to deal with bus contention issues for the only real performance snag. IGPs start making sense at that point for many applications because the memory's now close to the same speed as the add-in card's memory with similar latencies. The only real slowdown would be that you don't have dual pathways now.
Re:Looks like a fake (Score:3, Informative)