![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png)
Apparently AMD has announced plans to put a GPU on the same silicon as a CPU.
That's the next step in the classic cycle of reincarnation.
I was thinking about this a few months ago as SLi became popular. The recent progression has gone: specialized proprietary graphics bus ("local bus") => standardized graphics bus ("VESA local bus") => high-speed general bus ("PCI") => standardized graphics bus ("AGP") (3 upgrades) => specialized variant of general bus ("PCI Express") and now it looks like even that isn't fast enough, and the graphics system will head straight back to the CPU.
Oy.
In the meantime, non-gamers have it made: even a $35 card has enough power to render fantastic 2D graphics and movies. Even some integrated-on-the-motherboard chipsets, traditionally the dumping grounds for barely-competent leftovers, are doing pretty well (especially in Intel land, where the 945G and successors have carved out very nice niches for themselves).
That's the next step in the classic cycle of reincarnation.
I was thinking about this a few months ago as SLi became popular. The recent progression has gone: specialized proprietary graphics bus ("local bus") => standardized graphics bus ("VESA local bus") => high-speed general bus ("PCI") => standardized graphics bus ("AGP") (3 upgrades) => specialized variant of general bus ("PCI Express") and now it looks like even that isn't fast enough, and the graphics system will head straight back to the CPU.
Oy.
In the meantime, non-gamers have it made: even a $35 card has enough power to render fantastic 2D graphics and movies. Even some integrated-on-the-motherboard chipsets, traditionally the dumping grounds for barely-competent leftovers, are doing pretty well (especially in Intel land, where the 945G and successors have carved out very nice niches for themselves).
(no subject)
Date: 2006-11-10 05:28 am (UTC)AMD also has their new effort (I forget the name) where they're allowing direct hooks into HyperTransport, and licensing their sockets for non-CPU chips. So we could see GPUs that use an Athlon socket for really high speed interconnects.
In other news, nVidia just announced their new G80 GPU architecture, which is an all new design with support for the coming DirectX 10. AnandTech tested it - and *one* G80 GPU outperforms a Quad-SLI setup (4 of today's best GPUs). AnandTech's review, which was quite long, could be summed up as "Holy Shit! This thing is a monster!" The G80 was also designed for the growing trend of GPGPU - General Purpose GPU - using the GPU for physics calculations, etc.
(no subject)
Date: 2006-11-10 02:48 pm (UTC)Shades of the visual cortex, really, which seems to do a lot of thinking by itself.