does tex fill rate mean anything anymore

[KainXS]

going back does tex fill rate mean anything anymore when buying a card since newer cards have less tmu's but more powerful shaders than older cards.

 

[cheezburger]

going back does tex fill rate mean anything anymore when buying a card since newer cards have less tmu's but more powerful shaders than older cards.

it is still mean a lot and most issue was because they want to tie up more ALU with SIMD cluster which it end up the more TMU the more shader but die space has limitation and higher TMU will cause lower yield because of even more shader. even so it's still reasonable because we had TMU number boost since g70 and even today TMU count still twice as large compare top raster pipe while back to the time rops are same equal number to TMU.

 

[cadaveca]

Yes, it's still important. Textures are used in games, and a card's ability to process them is one of the most important things of a gpu.

*start nonsensical compare:

A Shader..does what, do you think? maybe create shades of colour, or light/darkness? Textures must be applied with speed too...


*end nonsensical compare.

 

[KainXS]

seems I wasn't very clear, what I am asking is that if you compare older gpu's and newer ones, does the texture fill rate give a good basis for comparing them, like GTX480 vs GTX285

 

[hellrazor]

Well, a higher one will let the game have higher-res textures.

Of course that means it depends on the game mostly.

 

[yogurt_21]

seems I wasn't very clear, what I am asking is that if you compare older gpu's and newer ones, does the texture fill rate give a good basis for comparing them, like GTX480 vs GTX285

put simply, no



http://www.nvidia.com/object/product_geforce_gtx_285_us.html
51.8 Billion
http://www.nvidia.com/object/product_geforce_gtx_480_us.html
42 billion

yet we know the 480 is faster.

texture fillrate is heavily dependant on tmu's, memory bit rate and memory clockspeed.

note that the memory clocks are measured differently there the gtx285 should show 2484MHZ to be comperable
gtx 480 60 tmu's, 384-bit (albeit gddr5), clocked at 1848MHZ
GTX285 80 tmu's, 512-bit (albeit gddr3), clocked at 2482MHZ

so with a higher memory bit, more tmu's, and higher memory clocks speed the 285 is able to have a higher texture fillrate while haveing a lower memory bandwidth and lower performance overall.

 

[cheezburger]

put simply, no



http://www.nvidia.com/object/product_geforce_gtx_285_us.html
51.8 Billion
http://www.nvidia.com/object/product_geforce_gtx_480_us.html
42 billion

yet we know the 480 is faster.

texture fillrate is heavily dependant on tmu's, memory bit rate and memory clockspeed.

note that the memory clocks are measured differently there the gtx285 should show 2484MHZ to be comperable
gtx 480 60 tmu's, 384-bit (albeit gddr5), clocked at 1848MHZ
GTX285 80 tmu's, 512-bit (albeit gddr3), clocked at 2482MHZ

so with a higher memory bit, more tmu's, and higher memory clocks speed the 285 is able to have a higher texture fillrate while haveing a lower memory bandwidth and lower performance overall.

indeed, fermi is a junk because the texture fill rate is much less than gtx 285. while under dx9/10 it's much much slower because of it. and narrow bus are impact to fermi despite it has GDDR5 memory and higher core clock.

result: 512bit bus will make its come back soon.