OK...now that Matrox has made their G450 announcement, several people have asked "what good is 64bit DDR, and what happened to 128bit DualBus?"
Well, the answer is:
The 64bitDDR interface has nothing to do with the 128bit DualBus arcitechture.
128bitDualBus relates to how the card as a whole interacts with the mobo.
64bitDDR relates to how the on-board ram interacts with the on-board processor (nothing to do with the DualBus to the mobo).
The 2 are not "directly" related. G450 will retain the same 128bit DualBus goodness as the 400, with the 64bitDDR on-board as a "non-related-to-dualbus" improvement.
Follow along...
With 32bit SGRAM (like our current G400s), an (on board) clock cycle looks like this:
Clock...Data
-------------
..0.....Idle
..1....128bit latched (32bitx4chips)
..0.....Idle
With 64bit DDR:
Clock...Data
-------------
..0.....Idle
..1.....64bit latched (64bitx1chip)
..0.....64bit latched (64bitx1chip)
Final result: 1 clock cycle = 128bits of data x-ferred, with either setup.
The real advantage of DDR comes in several forms:
-Same x-fer rate per cycle BUT:
-Clock speed should be higher, so more cycles per second
-Fewer mem chips = smaller main chip (fewer tracers for 2x 64bit chip than for 4x 32bit chips, therefore fewer pins on main processor)
While the 64bitDDR may not be a "huge" improvement over 32bit SGRAM performance-wise, it is an improvement production-wise. I would expect the G450s to be a noticably smaller chip (and card overall) than the G400s. PCBs will be simplified, tracer lengths have more room, and overall stability should improve.
So...as noted by Ant's news blurb, G450 is not going to be a huge leap in performance, and will not be marketed towards FPS counting gamers. It will be more like the G250 was to the G200...just an improved model of an existing chip, not geared as a full-blown replacement for G400, but a more cost-effective (production-wise) model with marginally better performance. Most people will not want to bother trading in their G400 for a G450, as the speed difference is not likely to be all that huge...
[This message has been edited by Kruzin (edited 26 April 2000).]
[This message has been edited by Kruzin (edited 26 April 2000).]
Well, the answer is:
The 64bitDDR interface has nothing to do with the 128bit DualBus arcitechture.
128bitDualBus relates to how the card as a whole interacts with the mobo.
64bitDDR relates to how the on-board ram interacts with the on-board processor (nothing to do with the DualBus to the mobo).
The 2 are not "directly" related. G450 will retain the same 128bit DualBus goodness as the 400, with the 64bitDDR on-board as a "non-related-to-dualbus" improvement.
Follow along...
With 32bit SGRAM (like our current G400s), an (on board) clock cycle looks like this:
Clock...Data
-------------
..0.....Idle
..1....128bit latched (32bitx4chips)
..0.....Idle
With 64bit DDR:
Clock...Data
-------------
..0.....Idle
..1.....64bit latched (64bitx1chip)
..0.....64bit latched (64bitx1chip)
Final result: 1 clock cycle = 128bits of data x-ferred, with either setup.
The real advantage of DDR comes in several forms:
-Same x-fer rate per cycle BUT:
-Clock speed should be higher, so more cycles per second
-Fewer mem chips = smaller main chip (fewer tracers for 2x 64bit chip than for 4x 32bit chips, therefore fewer pins on main processor)
While the 64bitDDR may not be a "huge" improvement over 32bit SGRAM performance-wise, it is an improvement production-wise. I would expect the G450s to be a noticably smaller chip (and card overall) than the G400s. PCBs will be simplified, tracer lengths have more room, and overall stability should improve.
So...as noted by Ant's news blurb, G450 is not going to be a huge leap in performance, and will not be marketed towards FPS counting gamers. It will be more like the G250 was to the G200...just an improved model of an existing chip, not geared as a full-blown replacement for G400, but a more cost-effective (production-wise) model with marginally better performance. Most people will not want to bother trading in their G400 for a G450, as the speed difference is not likely to be all that huge...
[This message has been edited by Kruzin (edited 26 April 2000).]
[This message has been edited by Kruzin (edited 26 April 2000).]
Comment