How to calculate memory bandwidth

[StickHead]

Just wonder what the formula is for this. Anyone know?

 

[StickHead]

Nevermind, guess the people here on the fourms don't know much about video cards.

 

[hominid skull]

I'll give the general principals here from what i can remember off hand
say a geforce has a 64bit data bus and 250MHz DDr ram (500MHz effective)

Total bits transfered in one second:

= databus width x bus frequency
= 64 x 500000000
= 32000000000 bit per second

take that number and change it into something we recognize, how many bits in one gigabyte

= bits in a byte x bytes in one k x k in one Mb x Mb in Gb
= 8 x 1024 x 1024 x 1024
= 8589934592 bits in a gigabyte

so a geforece has 32000000000/8589934592 Gb/s memory bandwidth.

= 3.725 GB per second memory bandwidth..(using my numbers)

Since you asked so nicely 😉

 

[hominid skull]

DP...

 

[miniMUNCH]

Yep!

I concur wit da man above...

Most all hardware terms like this are actually simple. But this calc. explains why high-end workstations have 256 bit wide (or wider)data paths. If you need more bandwidth, either increase the access speed of memory (which is material limited and thus tough), make a wider data path (expensive motherboards, mem. modules, drivers), or double/quad (DDR/QDR) pump the bus to transfer twice per clock cycle on the rising and falling edge of the cycle. I don't really know how the double/quad pumping is accomplished in hardware; maybe having mulitple sets of column and row strobes for mem scans. Anyone?

The fastest SDRAM modules available have a 4ns access time (somebody correct me on this if I'm wrong) which gives them a effective clock cycle of 250 MHz. These modules are used on the GeForce2 ultra's right now.

 

[StickHead]

Thanks guys. That is what I was thinking. I was wrong, somepeople do know stuff about videocards!