geForce 5900 wallops a 6600

D

damian24

Senior Member
Joined
Apr 5, 2005
Messages
862
Reaction score
92
Location
Middlesbrough, UK
Can anyone explain this one...

An 3 year old AGP Geforce 5900/128mb memory in Vista get a 3.7 rating...

Now for the wonky one, brand new GeForce 6600/256mb gets a paltry 3.4.

I tried installing the latest Nvidia driver, but no go the 6600 goes to 3.5, how one earth is a three year old card reading better than a new (supposedly better performing) card?

D
 
probably a lady who invented the rating system in vista

??
 
i thought as much this 'rating system' of vista would have its problems.........
 
To quote myself:

Next up is the Gigabyte 5900XT offering. It was at the bottom rung of the-top-of-the-range cards of the FX series, but quite a step up from the 5700 series. It comes with a blue PCB just like all Gigabyte motherboards. The ones tested here is already a bit old and thus the fan can be quite noisy when the PC is first switched on. Its also a quite heavy card, due to the big heatsink.

The 256bit bus provides it with quite a push but due to nVidias poor implementation of DirectX 9 it suffers alot in new shader intensive titles, only just being able to compete against the 6600 series, but falling behind a lot in the newest of games.

Gigabyte nVidia GeForce FX 5900XT 128MB 256bit (Core: 300 Memory: 700)

3DMark2001SE: 15011

3DMark2003: 5316

3DMark2005: 987

Counter Strike Source Video Stress Test: 92.79fps average

---------------------------------------------------------

The next card was the previous generations mid range card and is now moving into the value price range. Its comes with a blue PCB, and a smallish silver heatsink that features a small (but pretty quiet) fan. I am pretty sure this card could have been passively cooled with a bigger heatsink, since the temperatures were quite low, this would prevent the change of the small fan becoming noisy over time.

Performance wise is has the same performance as the 5900XT in older games, but even with the same core clock and a slower memory clock it will beat the 5900XT in newer titles. This card wont let you run the newest games in all their glory, but you should get reasonable framerates with everything on medium.

Asus nVidia GeForce 6600 128MB 128bit (Core: 300 Memory: 500)

3DMark2001SE: 13384

3DMark2003: 5075

3DMark2005: 2033

Counter Strike Source Video Stress Test: 80.12fps average
Click to expand...

Source
 
Well, the 5900XT is faster in CSS cause it runs in DX8 mode. But in reality the 6600 should be capable of better using new features. Guess its a bit complicated...

Is the 5900 a DDR3 card?
 
Well a 5900 is a pretty good card actually. I'm pretyt sureit is better that a 6600.

And a 5900 vanilla is better than a 5900XT, thats how Nvidia used to work.

Or thats what i always thought, im pretty sure about that though.
 
256-bit Vs 128-bit? :o
You can't and shouldn't compare bit-width differences between two generations of GPU's. In most if not all cases, the comparison will be meaningless...

NV43 was a ~143million gate part
NV35 was a 135million gate part. (extra 5 million over NV30 were also to do with the move from an external 128-bit bus to a 256-bit one, so in pure processing logic the difference between NV43 and NV35 {FX5900} is in the order of 13 million transistors)

6600 is faster than 5900 when doing the same type of processing.
FarCry benchmarks on the two graphics cards would show the 6600 superior in image quality and speed. :)
 
Is this perhaps a vista design bug, clearly 3dmark 2005, rates the newer card as better. Vista rates the 5900 as better for gaming and windows graphics performance, my guess vista not actually testing the full range of feature of the cards properly and thus I once again (like Microsoft time) can't believe what MS tells me.

My understanding is that the vista performance indicator should remain static for a particular config, this would suggest that the vista performance guidlines will need to be upgraded from time to time meaning the figure would need to be qualified by a performance test engine version?

D
 
There is a BIG difference between the budget 6600 and a 6600GT

The 5900XT has a higher pixel fill rate (memory clock and wide bus) and will beat the 6600 even if it lacks Direct X 9 features.
 
How you figure that tibby.dude?

6600 like all 6600 is NV43. That is 4 raster outputs and 8 fragment shaders/8 texture samplers.

NV35, FX5900 has also 4 raster units each having an exclusive 2 texture memory units. The 5900 is 400MHz while the 6600 is 300MHz for the budget.

However, for the 5900 most of the time is spent in 4x1 config mode meaning is only puts out 1.6GTixels/sec while the NV43 always puts out 2.6GTexels/sec. Sure the FX5900 can put upto 3.2Gtexels/sec in dual texture mode, but those situations are few these days. And the way the NV35 operated is that each pipeline is coupled with a single pixel shader that is not capable of applying an effect on both textures simultaneously, so for an effect that requires the effect to be done on 8 texels, the NV35 has to pass that twice into the pipeline, hence the advantage of the second texture unit is lost.

On the NV43, the texture sampler sits inside the fragment pipeline, so it always has 8 texture units to use and can always* apply a shader to 8 units.

The architectural differences between the two are hundreds of pages, but this is just a basic fill-rate difference.

memory bandwidth is important, but that assumes an efficient engine where there are minimal stalls. being able to fetch 30GB/sec on the NV35 is no where as useful as being able to fetch 8 samples, process them and output them in a timely manner....

would explain more but this would be a very long post :P
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X