On 2002-02-25 09:17, DemoCoder wrote:
I can't believe those games were CPU limited at 40fps. Did you try running the same benchmarks on a different card? I get better scores on a 1.2Ghz Athlon and GF2 (non-ultra)
Democoder, first of all, I hope your comments are based on using the same games (Max Payne, F1 2001 and Serious Sam Second Encounter, not the original in the last case) I used in my review as well as, and especially so, the saved games I specifically used. Unless this is the case, your comments will add doubt to the consistency and accuracy of my review.
I just stuck the GeForce3 Ti 500 back in the same machine as the GF4 MX440 review. I did a quick test using Serious Sam Second Encounter's Direct3D mode to test SW vs HW T&L performance. First of all, there was no way I could bump the GF3 Ti500's core up to the GF4 MX440's default core of 270MHz... so I settled on the maximum the GF3 Ti500 could muster, which is 255MHz. I reduced the GF3 Ti500's memory clock to match the GF4 MX440's, which is 400MHz.
The same 27.30 NVIDIA drivers were used.
All are at 640x480x32bits, no AA nor any driver-enabled aniso, etc, etc tweaks, with all Serious Sam graphics options exactly identical between the GF3 Ti500 and GF4 MX440. Serious Sam Second Encounter allows the ability to determine number of TUs to use so I thought I might as well provide the additional info regarding this.
Serious Sam Second Encounter "The Grand Cathedral Demo" (as appear in the GF4 MX440 review) :
Direct3D Software T&L
GF4 MX440 at 270/400 (core/mem) using DUAL=max texture units allowed : 20.7 fps (as per my review)
GF3 Ti 500 at 255/400 (core/mem) using DUAL texture units : 20.9 fps
GF3 Ti500 at 255/400 (core/mem) using QUAD texture units : 21.6 fps
Direct3D Hardware T&L
GF4 MX440 at 270/400 (core/mem) using DUAL=max texture units allowed : 16.7 fps (as per my review)
GF3 Ti500 at 255/400 (core/mem) using DUAL texture units : 31.7 fps
GF3 Ti500 at 255/400 (core/mem) using QUAD texture units : 32.5 fps
I'm not even going to test the GF3 Ti500 at its default core/mem clocks. The GF3 Ti500 is at a lower core clock compared to the GF4 MX400 with the same memory bandwidth in the above test. So what's wrong? Nothing... the GF4 MX440 is simply slow, like I said (more or less) in my review.
Theories (based on specs, or based on a 2+ year old game as reported by other sites) are fine... but until you test it the way I tested the GF4 MX440 you won't really know the true picture.
I stand by my ("occluded" harsh-ish) conclusion - the GF4 MX440 is slow in the games I tested it with in my review regardless of the fact that I used a "slowish" CPU or "outdated" OS. If other folks want to use Quake3, then that's their decision... I used newer games, of which I'm not going into a debate about whether this is right or wrong or, even worse, whether I should use Quake3 simply so folks can "compare" whatever Quake3 numbers I come up with with other sites'.
As for the subject of EMBM, it is *not* easy to implement but only so depending on what you're trying to achieve. *Almost all* (keywords) developers usually use what is interesting to them at a certain time - when EMBM debuted with the Matrox offering, the developers' reactions were the same as those when the original GeForce256 debuted... you use what is available while thinking of a/any fallback mode. "Matrox has EMBM... it looks really nice.. let's use it but make sure things look reasonably nice on those other vid cards without EMBM"... and "The GeForce256 has T&L... let's use what we can while making sure things look reasonably nice on those other vid cards without T&L". Oh come on guys... this is such a stupid thing to argue about. Per-pixel lighting will put EMBM in a coffin... EMBM looked nice when it became available with the Matrox part but "simple" DOT3 has more uses in general. Talking about "3D technology" per se is a world apart from talking about "installed base of games using older=when-it-became-available" which belongs in an entirely different discussion.
The GF4 MX440 will probably satisfy Quake3 gamers... but it probably won't do with some games that have appeared in the two+ (!!) years since, like those used in my review.
_________________
Reverend
Beyond3D
3DPulpit