In the 'Nvidia Geforce 8200 review' (of Michael Larabel) its said "The MCP785 chipset isn't too Linux friendly if using a pre-2.6.25 kernel." Were there any Nvidia/AMD collaborations that are friendly to pre-2.6.25 kernels?
They were all pretty friendly dating back to the original nforce.
Originally Posted by i8DRM
Thanks! So Geforce 8200 was the exception? Any other exceptions I need to put on my 'avoid' list?
Originally Posted by deanjo
Basically that chipset is completely useless when you use a new FMx board. AMx boards can use those chips but in theory those are more or less legacy. I don't know why AMD does not manage to shink the die to put 3rd level cache together with a gpu on a die.
No. The real solution would be to provide a FM motherboard with TWO 12V 8PIN VGA connectors directly into the mainboard. Then simply provide processor classes:
Originally Posted by Kano
-1- TDP under 140W w/cut down APU and not requiring 12V VGA connectors.
-2- TDP under 140W CPU without VGA, not requiring 12V VGA connectors, requiring external GPU (which is possible due to presence of PCIex16)
-3- TDP up to 400W with full blown APU, piledriver based, with L3 cache, with lots of steam processors. Basically the 7xxx card inside, without extra bus logic and dedicated memory.
Also, at least 4x memory controllers for third version.
This would mean the motherboard itself would require to build 4x lanes to each memory slot(4xddr3 or 8xddr3).
This will increase the MB price only slight, since the lanes are dummy and controller resides in CPU/APU itself.
One puts (2) and (1) versions in such motherboard, the only two build-in memory controllers activate and the MB switches the pairing lanes together.
One puts (3) and probably high-end (2) versions inside, all four memory controllers activate directly to memory sockets without routing.
Thats my dream
However recently I have been researching the whole FM2 socket area and it caused massive head-bang feel due to how completely limited the platform is. Its HTPC only.
They don't want own FM platform to cannibalize on discrete cards, however this is utterly stupid, because:
(1) low-end/htpc - use either (1) without GPU or discret low-power VGA for legacy AMD boards/all Intel boards
(2) mid-end workstation - use (2) + discrete GPU
(3) high-end workstation - use (3) without GPU or (2) for legacy systems + high-end GPU
(4) ultra-end workstations - use (3) with high-end GPU in HSA/CF.
That config would ROCK with possibilities. Just market the gaps cleverly. In the end if customer buys AMD or AMD - it does not matter.
The customer will decide between compatibility (mid-range/high-end discrete card, which can be moved + CPU) and energy efficiency (high-end APU(3)) or both for ultimate performance (four controllers, build in high-end VGA, crossfired with other discrete high-end graphics).
And this with opensource drivers.