Tweet
Ok sorry I just have to make clear all the arguments about RAMBUS, because even the people that think they know about it, really don't.
Just wanna say first that I don't like rambus and their underhanded tactics and overpriced parts, and never have owned any of their stuff, so I'm not a fanboy or anything.
First of all, there was a comparison: "You will see that a DDR266 based board outpreform a PC1066 based board in divx encoding."
This is *only* on a granite bay mobo, which only just came out (rdram1066 was out ages ago). If you look to the future, the new SIS chipset with quad channel rdram (1200 rdram) is about to come out will BLOW anything ddr, or ddr-2 out of the water. INCLUDING granite bay. The only thing that will come close (or surpass it) is intel's upcoming quad channel ddr chipset, but that won't be for a long time yet.
HOWEVER let's note that Intel has afaik dropped all support for the rambus platform, so don't expect nice intel chipsets out in the future - which pretty much guarantees death of the platform, which is a fair reason not to even get rambus. In conclusion though, I do not see DDR SDRAM as the clear perf winner *just yet* but possibly in the future (ignoring price).
"AMD processors are terrible if you want fast (DDR and above) memory because their fastest FSB is almost a quarter of the P4"
Ok as it currently stands p4 platforms (northwood) have a FSB that can shuffle 4.2 Gb/s of data (133 mhz quad pumped, 64 bit) to the cpu. AMD platforms (thoroughbred-b) have a FSB than can transfer 2.6 Gb/s of data (166 mhz, dual pumped, 64 bit) to the cpu. (roughly heh)
P4 is the clear winner in terms of raw bandwidth: it has nearly DOUBLE and NOT (not 4x) the bandwidth of a comparable AMD platform. However there are other issues, like whether the bus actually does transfer the maximum amount - here latencies, synchronous/asynch fsb/mem clock, sse2 etc come into play, but p4 is still the clear winner.
and then a response..... "It's true that the P4 uses a higher FSB, but Intel only achived the higher FSB by reducing the bus width."
Intel did NOT achieve a higher FSB by reducing the bus width. The p3 coppermine could transfer around 1 Gb/s also at 64 bit. What he meant was that Intel achieved higher clock speeds by *increasing* the length of the pipelines within - which of course has adverse effects on latencies and IPC as he said.
They didn't go for higher clockspeeds just to woo the customers - their engineers believe it or not think differently to their marketers - they did it to avoid the problem that AMD is currently having - where their cpu architecture is having trouble ramping up altho they have done magnificently well in trying to keep up. Their latest release though, Barton, is a bit too little, too late with an overly optimistic ratings scheme (which someone else already pointed out). The only thing going for AMD atm is that their cpus are a little cheaper, but not by a lot. Essentially its the FSB that is holding AMD back - even with the sweet new doubling of cache.
As for the 10x increase in encoding performance for I26... look at it this way: he prolly doesn't have a good measurement of frame rates... its not like he re-divx-ed a movie he did with his old machine *with the exact same software* and recorded the times.. he says 3fps to 28 fps which is say, 9x... but it could well be more like 5 fps to 25 fps which is 5x... taking into account advances in software n other changes stuff. So its a fine line, and all fairly credible IMO.
In conclusion, i dunno much about divx encoding still, i think I know more about hardware, but prolly still not that much, but I think I've been accurate so far. </rant>
Just wanna say first that I don't like rambus and their underhanded tactics and overpriced parts, and never have owned any of their stuff, so I'm not a fanboy or anything.
First of all, there was a comparison: "You will see that a DDR266 based board outpreform a PC1066 based board in divx encoding."
This is *only* on a granite bay mobo, which only just came out (rdram1066 was out ages ago). If you look to the future, the new SIS chipset with quad channel rdram (1200 rdram) is about to come out will BLOW anything ddr, or ddr-2 out of the water. INCLUDING granite bay. The only thing that will come close (or surpass it) is intel's upcoming quad channel ddr chipset, but that won't be for a long time yet.
HOWEVER let's note that Intel has afaik dropped all support for the rambus platform, so don't expect nice intel chipsets out in the future - which pretty much guarantees death of the platform, which is a fair reason not to even get rambus. In conclusion though, I do not see DDR SDRAM as the clear perf winner *just yet* but possibly in the future (ignoring price).
"AMD processors are terrible if you want fast (DDR and above) memory because their fastest FSB is almost a quarter of the P4"
Ok as it currently stands p4 platforms (northwood) have a FSB that can shuffle 4.2 Gb/s of data (133 mhz quad pumped, 64 bit) to the cpu. AMD platforms (thoroughbred-b) have a FSB than can transfer 2.6 Gb/s of data (166 mhz, dual pumped, 64 bit) to the cpu. (roughly heh)
P4 is the clear winner in terms of raw bandwidth: it has nearly DOUBLE and NOT (not 4x) the bandwidth of a comparable AMD platform. However there are other issues, like whether the bus actually does transfer the maximum amount - here latencies, synchronous/asynch fsb/mem clock, sse2 etc come into play, but p4 is still the clear winner.
and then a response..... "It's true that the P4 uses a higher FSB, but Intel only achived the higher FSB by reducing the bus width."
Intel did NOT achieve a higher FSB by reducing the bus width. The p3 coppermine could transfer around 1 Gb/s also at 64 bit. What he meant was that Intel achieved higher clock speeds by *increasing* the length of the pipelines within - which of course has adverse effects on latencies and IPC as he said.
They didn't go for higher clockspeeds just to woo the customers - their engineers believe it or not think differently to their marketers - they did it to avoid the problem that AMD is currently having - where their cpu architecture is having trouble ramping up altho they have done magnificently well in trying to keep up. Their latest release though, Barton, is a bit too little, too late with an overly optimistic ratings scheme (which someone else already pointed out). The only thing going for AMD atm is that their cpus are a little cheaper, but not by a lot. Essentially its the FSB that is holding AMD back - even with the sweet new doubling of cache.
As for the 10x increase in encoding performance for I26... look at it this way: he prolly doesn't have a good measurement of frame rates... its not like he re-divx-ed a movie he did with his old machine *with the exact same software* and recorded the times.. he says 3fps to 28 fps which is say, 9x... but it could well be more like 5 fps to 25 fps which is 5x... taking into account advances in software n other changes stuff. So its a fine line, and all fairly credible IMO.
In conclusion, i dunno much about divx encoding still, i think I know more about hardware, but prolly still not that much, but I think I've been accurate so far. </rant>
Comment