Advanced multiplyin'???

[nutball]

'Scuse my ignorance here but...

Mr Anand from Anandtech he say this:

The lossless compression algorithm can obtain up to a 4:1 compression ratio which explains where NVIDIA gets their 48GB/s of memory bandwidth from. Remember that the 500MHz DDR2 memory on the GeForce FX provides 16GB/s of bandwidth on the 128-bit memory bus, but multiply that by 4 and you'll get the 48GB/s of memory bandwidth NVIDIA is claiming.

Now back when I went to school, 4 times 16 was 64. Now it's 48? I want my money back!

[Uttar]

Originally posted by nutball
'Scuse my ignorance here but...

Mr Anand from Anandtech he say this:



Now back when I went to school, 4 times 16 was 64. Now it's 48? I want my money back!

Hehe. Well, I got a math test tommorow, so if my scores aren't that great, I'll say it's Anand's fault!

More seriously, the point is that nVidia multiplies only by 3 because there is more to bandwidth than color. There also is Z Buffer access and more. And putting that in the calculation ain't fair since the GF4 got that too.


Uttar

[Stealthy]

Originally posted by nutball
'Scuse my ignorance here but...

Mr Anand from Anandtech he say this:

The lossless compression algorithm can obtain up to a 4:1 compression ratio which explains where NVIDIA gets their 48GB/s of memory bandwidth from. Remember that the 500MHz DDR2 memory on the GeForce FX provides 16GB/s of bandwidth on the 128-bit memory bus, but multiply that by 4 (don't forget to subtract out the original 16GB/s of data) and you'll get the 48GB/s of memory bandwidth NVIDIA is claiming.

Now back when I went to school, 4 times 16 was 64. Now it's 48? I want my money back!

Dont you read between brackets! :p

[tieros]

That makes absolutely zero sense to me!

Why would you subtract out the uncompressed bandwidth from the compressed bandwidth to give a compressed bandwidth figure???

Please help me to understand :)

[nutball]

Yes, precisely.

[legion88]

Originally posted by Uttar
Hehe. Well, I got a math test tommorow, so if my scores aren't that great, I'll say it's Anand's fault!

More seriously, the point is that nVidia multiplies only by 3 because there is more to bandwidth than color. There also is Z Buffer access and more. And putting that in the calculation ain't fair since the GF4 got that too.


Uttar

It is either still a bogus math or an assumption being made that is not made public. Hypothetical example:
color read: 10 bytes
color write: 10 bytes
z access: 10 bytes
Total 30 bytes

4:1 color compression:
color read: 2.5 bytes
color write: 2.5 bytes
z access: 10 bytes
Total 15 bytes

So the 4:1 compression saved 1/2 of the bandwidth. So where did the three come from?

And what is not made obvious is that compressed data is worthless until it is uncompressed. And it takes time to compress data as well as uncompress it. So again, where did this three come from?

Out of someone's ass, that's where. It is just some funky math someone at NVIDIA came up with so they can post some large number on some press release, interview, box cover, whatever.

If one were to calculate a more realisticl "effective" theoretical bandwidth, then they will have to take into consideration time spent on compressing and uncompressing data.

Hypothetical example.
16GB/sec bandwidth.
64 GB of data, compressed 4:1 ratio --> 16GB.

Time spent on transfering 64GB uncompressed data would be 64/16 = 4 secs.

Time spent on transfering 16GB of compressed data would be 16/16 = 1 sec.

Add in time to compress/uncompress 64GB of data --> 0.5 seconds (assumption here).

Total time: 1 + 0.5 = 1.5 seconds.

So ratio of savings is: 4/1.5 = 2.6667

So perhaps the three is a rounded number where it is assumed that XYZ amount of data is compressed/uncompressed in ABC seconds. (Or they assumed that the time spent was 0.3 second.)

Subtracting 16 from 64 looks like nonsense.

[Bigus Dickus]

Anand is still an idiot. A ratio is defined between the beginning and ending values. To "subtract out" the original value is ludicrous, and wrong. He should know that.

Besides, his tech "preview" clearly stated that "multiply 16 by 4 and you get 48." Now he's just trying to back his way out of looking like a moron. :D

[tieros]

Originally posted by Bigus Dickus
Besides, his tech "preview" clearly stated that "multiply 16 by 4 and you get 48." Now he's just trying to back his way out of looking like a moron. :D

Yup, that little blurb in parens was added in after the article went live.