6970 vs 6990 in Eyefinity - Page 6

Originally Posted by XXXfire

Beg your pardon? I find your analysis to be moreso idiotic than your reasoning for why "we" are idiots. I, in fact, upgraded from three 6970 separate PCBs to dual 6990 precisely because the plx 6847 bridge chip acts as full bandwidth x16 crossfire gpu to gpu bridge.

Secondarily? SLi does NOT exclusively use the SLi port & bridge for gpu communication and finalized rendered frame passthrough. A GREAT deal of communications occur via the pci-e bus

Originally Posted by trippinonprozac

unless you are going for 3x30" displays for eyefintity you will never have to look past tri-fire . 6990 is a great place to start and then when the time comes, simply add a 6970 to the setup and you will be screaming!

Originally Posted by XXXfire

You're impressive.

My belief is substantiated by facts, sharp guy. I've clearly articulated just why I consider there to be an intrinsic advantage (cleared up PCI-E bandwidth, easier frame synchronicity between the gpu devices). Especially on bandwidth limited chipsets such as 1155/1156, where maintaining a "clear of traffic" PCI-E bus is especially important to ensure optimal hardware performance. Thus the benefit of PCI-E bridges, such as the nf200 chip which pioneered several functions that exist in all dual processor graphic cards. PW Short provides a point to point data path between a "master" device and 1-3 GPUs. When people criticize the nf200 because it doesn't actually add PCI-E lanes, the assertion is only half true. While the CPU is limited to 16 lanes and always will be, increasing the bandwidth between PCI-E slots & providing a route of communication works (quite well) reducing bus traffic. The benefit derives from an increase in PCI-E bandwidth via addition of lanes and, secondarily from the decrease in pci-e bandwidth congestion between the CPU & nf200 bridge.

Besides PW Short and more impressive as a performance optimization feature is the function Broadcast; the nf200 or PEX 6847 directs the CPU to, instead of 2-4 sent packets targeting every device in the array, submit a single copy of the transmission. The nf200 or PEX 6847 receives, duplicates, and delivers to all respective intended targets. Depending on how many graphic processors are in operation, this feature alone can reduce bandwidth 2-4 times in circumstances of redundant data transfer (something very common in AFR rendering).

You've referenced "history" as your paramount argument for dual GPU inferiority? Is that a valid argument in any substantive way? Why on earth would you reference a different architecture, different components in a different configuration to rebuke Antilles? Your basis of argument is as fundamentally flawed as previous iterations of dual-gpu cards. Speaking toward the latter, is a flaw that has not been reproduced in the 6990.

The benefits, beside the updated PEX bridge chip, are related to the re-architecture of Crossfire that was initiated in the 10.5 Catalyst driver & hit stride subsequent the release of Barts (6800) series AMD GPUs. The difference between Cypress & Cayman is night & day, really.

Powertune has provided a great safety mechanism for users keeping their devices on air. The voltage regulators, RAM, PCB design, et al. are absolutely stellar & have a strong, specific capacity for aftermarket cooling & overclocking potential.

So please, tell me how history has any relevant basis for reference in attempted defamation of the 6990? Or is it more likely you've just dug deep within your intellectual prowess and dusted off a lame, hackneyed cliche (history repeats itself) and built a fallacious & flimsy argument upon it. Don't use a nonsensical quip as the premise of your argument. As a foundation for subsequent points, it'll collapse the second scrutiny & reason is applied, even in the most modest of degrees.