Asus GF4

[Ovi]

Originally posted by Ant
He's been around the forums for a long time. The forums first opened in 1969.

Heh,heh
Yeah sure, Ant!

[Jammrock]

YOU DO NOT BELIEVE IN THE GREAT ANT!!! YOU FOOL!!! May Phil, the ruler of all heck, cast you into the pits of insufficient light and beat you with his spoon for all eternity!

Ant, nice new avatar!

Jammrock

[Technoid]

He seems to have spawned at 2nd March 2001 18:11 Matrox Hardware forum where he also has been seen on variouse occasions....

He's bad omen....

[Wombat]

63 million transistors of <I>what</I> though? Graphics core math is not CPU work. CPUs do decodes, and branches, and lots other work. The video card is going to be largely parallelized work. And as the P4 showed, if you want more Hz, you just throw transistors at it until you've created twice as many stops along the way.

[superfly]

There is one other little aspect that nobody mentioned yet...the power limits that agp 4x slots are limited to...


In the case of standard agp4x,total power that the card is allowed to use can't exceed much past 25w(that includes the memory as well),while current cpu's eat up to 50~60W of power all by themselves....


That's one of the reasons why clock speeds in video cards are still relatively slow compared to recent cpu's and why some take a chance by volt modding their video card which allows it to be clocked quite a bit higher that it normally would be possible,even with the best cooling in the world.


Some of the highest scores in 3d mark 2001,which the highest score now stands at 11 700 points at the default bench settings were in part acheived by volt modding the video card,in this case a Radion 8500 in which the core was working at 380 mhz....


Pretty impressive,no?...

[superfly]

I'm sorry...it now stands at 15 791 points at the default test.

Here's the linkgamershq.madonion.com/compare2k1.shtml?2609584.

[Novdid]

That is if you don't take the AGP pro spec into consideration. And the heat would be difficult to handle.

But non the less, that score is a mighty one!!!

[Indiana]

Originally posted by superfly
I'm sorry...it now stands at 15 791 points at the default test.

Here's the linkgamershq.madonion.com/compare2k1.shtml?2609584.

Have you actually looked at that score? That's no way the default benchmark, but the fastest possible setting (640x480, everything at 16Bit).

[superfly]

My bad,i just focused so much on the score that i didn't even notice the resolution he was running it at....


Here's his updated score for the default bench thought..

gamershq.madonion.com/compare2k1.shtml?2524118..


A still very impressive 12.000+score regardless.

[superfly]

Yeah but nova,the vast majority of consumer cards out there aren't built with the agp 50 standard,so the point still stands that both graphic and memory chips have very stringent power limits to respect...

Cranking up the mhz with recent chips already pushing 60+ million transistors and to make matters worse, adding more and faster memory as well while keeping everything operating at or under the standard agp power limit is no mean feat....

[Wombat]

Actually, having more transistors is much more likely to <B>lower</B> power dissipation, as well as make the circuit design easier.

When you're really trying to tweak the speed, you go to Pseudo and dynamic circuitry, or even faster, nastier, power-hungrier, royal-bitch-to-engineer methodologies. Timing, precharging, and drive fights, yay! These tend to save you both fets and time, but cost you in safety and power (big time).

If you've got the time, and the fets, use static logic. The same circuit takes about 2x the fets, but it's much easier to engineer, and consumes a LOT less power.

How about process? Smaller processes mean a lot more leakage current.

Another thing to consider is fet size. Let's take that Athlon as an example, and look at the cache. A Tbird has 512kB of cache on-die. Those cache array banks are driven by a network of <B>huge</B> fets, probably twenty times or more the average size. All those cache lines drive fight, too, I bet.



Besides superfly, don't you think that if nVidia came out with a top-end model, and said it required an AGP Pro slot, people would buy the right boards? Don't you think nVidia could just whisper to the MB mfrs that they really need to have AGP Pro slots within 3 months or their MB will be undesirable?
People that <I>really</I> wanted those first P4s went out and got new MBs, new power supplies, and new cases even! This is not nearly so drastic.

[superfly]

How many boards are out there that support the agp pro spec...i'm guessing not that many since even intels own p4 850/845 boards don't have them and a lot of oem's use those to build their systems..


Not to mention the existing user base issues,since there are a lot of users out there that are already using ghz+ cpu's in wich their mainboards don't have an agp pro slot and somehow i don't think making those users buying new boards so that they can use a new video card will go down very well,no matter how good the card ends up being....


Nvidia still doesn't have the clout that either intel or amd does and when anyone buys a new system they usually keep it for at least a couple of years before doing another major upgrade,something that doesn't usually apply to video cards,if the current trend continues....

[Wombat]

I'm not sure what you're replying to, but it doesn't appear to be my post.

From reading your post, I don't think you read mine at all. You're repeating arguments I already addressed, and ignoring most of my post entirely.

[superfly]

Errrrr.......The last part of your post...The one about Nvidia having enough clout to convince motherboard makers and users alike to ditch their current board should their upcomming GF4 require an agp pro slot...

[Jon P. Inghram]

I'm suprised nobody's mentioned 3Dfx's solution to the power issue... an external power supply for the video card.