If possible, please give links to hard comparative data. An intelligent
estimate based on personal experience or competent theoretical
knowledge will also be welcome. But no wild guesses please.
--
Conor,
Same shit, different day.
A 64-bit bus is sub-par and crippling to any kind of performance. Even
the original Riva 128 (NVIDIA's first successful graphics accelerator)
had a 128-bit bus. The thing it hurts most if bandwidth and texture
uploads. Your card needs to communicate to the CPU over the AGP/PCIE
bus quite a bit, having a one way (64-bit) bus means it has to wait to
read/write or split the bus even more into two 32-bit channels for
bidirectional.
64-bit buses are generally used on very low-end or OEM system
components. These are the sub 90.00 cards or the overpriced value cards
sold at retail outlets like Best Buy.
Memory bandwidth is greatly affected by bus size/speed; in order to
compensate for a weak bus you need to use faster RAM (GDDR2/3) or pump
up some of the clock speeds, but even then you're usually dealing with a
low end card that's underclocked if anything and using sub-par RAM chips
with slow timings.
A lesser example can be seen with the 6600GT. It had a 128-bit bus
compared to the 6800's 256-bit. In order to compensate, it uses GDDR3
RAM and fast clock speeds (500/1000 compared to 400/700). It's a good
card for it's price range, but even so the larger bandwidth of a 6800/GS
will handily beat it.
How does this translate into real-life experience,
particularly gaming fps ? Can you please provide links
to comparative benchmarks ?
At the moment I'm interested in knowing the difference
between two GF6200 cards by XFX (Pine) - the 128-bit
PV-T43L-UA (256MB) and the 64-bit PV-T43L-RA (128MB).
(Both use DDR and the same clock speeds). But benchmarks
for any two otherwise comparable cards using the same
GPU will provide a good indication.
For games and settings that can be played at reasonable
frame rates with these entry-level cards, I don't think
the amount of RAM will make a major difference, right ?
I don't know of a recent low end benchmark, but you can always try
Tomshardware's VGA charts.
I can tell you from absolute experience though that yes a bus size makes
a big benchmarkable difference and RAM is generally important. RAM is
used for texel storage. Texels can be image maps, shader fragments,
screen elements, temporary ops, etc. In general the more you have, the
better, so long as a game will ALLOCATE that much video RAM to begin with.
Where RAM tends not to matter so much is the difference between 128 and
256 on 128-bit cards. The increase is marginal, but helps with higher
resolutions and may have faster texture uploads in shader intensive
games (BF2 and Oblivion for example).
Yet on a 64-bit bus card, it doesn't matter. The GPU core is hampered
by overall throughput and more RAM tends to be a marketing gimmick more
than anything. I've even seen a 512MB 6200 card! Ridiculous and useless!
Uhh.. Isn't your math off a bit?
65bits is the double that of 64bits. 1 extra bit can address twice the
values.
Lets start with smaller numbers first:
2 bits can address 4 decimal values.
3 bits can address 8 decimal values.
4 bits can address 16 decimal values.
Not when talking the width of the data path. You are correct when
referring to memory addresses, but the 128 bit vs. 64 bit in this case
is the width of the data path.
Actually he's right. Your math is right in the an extra bit allows double
the value to be held, but that is different then bandwidth.
Think of it this way. You're moving bytes. 128 bits is 16 bytes. At the
instant of a clock tick, 16 bytes are available on the bus and 'grabbed' by
the process.
If you upgrade to 256 buts, then 32 bytes are available at that clock tick
and can be 'grabbed'.
Double the bandwidth.
Tom