VPU Core speeds vs: mem speeds

[Tiamat]

In the realm of the overall computer, it is generally taken that a synchronous CPU and RAM FSB speed will grant the most performance to the user. However, I havent been able to understand why this seemingly is not the same for the Graphics cards. I have not seen many graphics cards with its memory synchronous with its VPU. Is there any reasoning for this? Does the benefits of synchronous timing not apply to the video card realm?

 

[Jeff7181]

Not really... it use to be that memory bandwidth was the main determining factor in performance... it's changing now since the pixel shaders and vertex shaders require much more processing power... we have the bandwidth we need now with 900 Mhz RAM with 128 bit wide buses... but the GPU power is still lacking for big shaders.

 

[jiffylube1024]

Originally posted by: Jeff7181
Not really... it use to be that memory bandwidth was the main determining factor in performance... it's changing now since the pixel shaders and vertex shaders require much more processing power... we have the bandwidth we need now with 900 Mhz RAM with 128 bit wide buses... but the GPU power is still lacking for big shaders.

I think you mean 256-bit buses.

 

[BoomAM]

Originally posted by: jiffylube1024

Originally posted by: Jeff7181
Not really... it use to be that memory bandwidth was the main determining factor in performance... it's changing now since the pixel shaders and vertex shaders require much more processing power... we have the bandwidth we need now with 900 Mhz RAM with 128 bit wide buses... but the GPU power is still lacking for big shaders.

I think you mean 256-bit buses.

He could mean either.
Hes right though, we`ve reached a point where Memory bandwidth is no longer at a premium.

 

[modedepe]

Originally posted by: jiffylube1024

Originally posted by: Jeff7181
Not really... it use to be that memory bandwidth was the main determining factor in performance... it's changing now since the pixel shaders and vertex shaders require much more processing power... we have the bandwidth we need now with 900 Mhz RAM with 128 bit wide buses... but the GPU power is still lacking for big shaders.

I think you mean 256-bit buses.

Even with a 128bit bus it's still 14.4 GB/s bandwidth, which is plenty for most cards. Plus z compression is being used in many cards, which is helping them to make even better use of available bandwidth.