RobsTV,
?
A GeForce 2 "MX" beats a Kyro2 in some instances?
Such as what? Low resolution 16bit rendering on T&L titles? Show us some links?
?
Don't forget that the Kyro 2 "requires" an ultra fast CPU before it can compete with any other cards. In a normal or weaker system, (under 800MHz), it fails to compete, and other cards don't require this extra CPU power to run fast.?
I asked you to explain this previously ? how or why does KYRO ?require? more CPU performance than any other non T&L card?
?
The only edge the 64megs gives the Kyro 2 is the ability to use 4x FSAA, as 32 meg cards are limited to 2X FSAA.?
KYRO?s FSAA works differently from other cards FSAA. KYRO performs its FSAA at the tile level, and the downsampling is done as the tile is completed ? i.e.
before it goes to the external frame buffer. This means that where a card such as GTS needs a 1600x1200 worth of frame buffer space to perform 4XFSAA @ 800x600 KYRO still only needs an 800x600 space in the frame buffer.
There is no difference between the available FSAA modes between the 32MB and 64MB variants of KYRO?s.
In the case of KYRO 64MB is
purely there for extra texture space, as well as a nice marketing figure for Hercules.
?
Note that just because the card has the memory to run 4X FSAA doesn't mean you can do it. In most games the performance hit at 4X FSAA would be far to severe to actually use it. For some it may be fine if you have a fast enough processor.?
There you go again with CPU misconceptions!
FSAA is a fillrate intensive task, not a CPU/Geometry intensive task ? CPU variances are likely to make little or no impact when running large FSAA depths, unless you are running in seriously low res. In all probability running at most games 1024x768 with 4XFSAA will perform exactly the same on a Celery 500 as it will a PIII 1GHz because that massively intensive operation is purely dependant on fillrate.
Coincidently the reason I mentioned above, with KYRO performing the downsampling before going to the external frame buffer, is the exact reason why KYRO?s FSAA is far more efficient than any others ? it doesn?t have the 4X bandwidth penalty that all other cards suffer (when running 4X FSAA).
Leon,
?
And one more thing. According to latest Anandtech review, KyroII trails regular GTS in all three games tested, including Serious Sam.?
Well, the SS test used by Anand here was ?Karnak Peaceful Night Coop? which is a farily open and flat level (not KYRO?s best case) ? had this been the ?Memphis? demo (which much of SS is similar to) these probably would have been fairly different because of the overdraw factors involved.
Also Quake III Arena V 1.27g was used, which AFAIK doesn?t have texture compression enabled by default. KYROII suffers slightly on texture bandwidth, and had TC been enabled KYRO?s scores would have looked better ? better than GTS? Dunno.
?
Gap widens in T&L enabled games.?
That particular benchmark may not be due to ?T&L? per se, but because of a peculiarity of DX8 runtime. Giants uses ?Render To textures? a lot during rendering, and currently there is a bug in DX that prevents KYRO from utilising this effectively. Also,
according to Hercules, Giants has been set to use the external Z buffer (to circumvent the render to texture issue? I don?t know?) which results in slowdown in operation of KYRO ? KYRO normally uses the much, much faster internal tile Z buffer, and forcing it to use an external one results in considerable slowdowns as it is using the external bandwidth, something it doesn?t have to much of (as normally it doesn?t need it).