FP12 ?????

[OpenGL guy]

can accept FP32 requests but just converts them to FP24 correct?

Correct. The texture address (data input if you wish) can be FP32, but is converted to an ATI 'internal' FP24 during the pixel shader calculations. The R3x0 can expand the result from the pixel shader up to FP32 data afterwards, however.

sireric (ATI) has told us that writing out their internal format to 128b (FP32) and reading it back in as a texture 'causes no precision loss'.

Does it cause a performance hit?

Not to be rude, but this is a silly question. 128 bits per texel. That means more bandwidth is consumed reading the data into the texture cache. That also means the texture cache holds fewer texels. Of course there's a performance hit.

 

[digitalwanderer]

Not to be rude, but this is a silly question.

That's ok, I'm a bit of a silly guy...plus I really ain't anywhere near as educated about these things as most of the people here.

Asking silly questions I still find to be one of the best ways to learn silly things.

128 bits per texel. That means more bandwidth is consumed reading the data into the texture cache. That also means the texture cache holds fewer texels. Of course there's a performance hit.

Thanks, I thought as much but I really didn't know.

 

[3dcgi]

Sigh.
I never liked that format.

Using sign-exp-mantissa ordering would be more "natural" because that's the actual order in which they are stored.

I agree.

 

[Ilfirin]

Sigh.
I never liked that format.

Using sign-exp-mantissa ordering would be more "natural" because that's the actual order in which they are stored.

I agree.

Except the storage order is platform dependent. s<mantissa bits>e<exponent bits> is based around how you write the numbers in scientific notation (i.e. mathematics - a universal language). 3e6, -2.314e1, 6.626e-34, ...

 

[horvendile]

3e6, -2.314e1, 6.626e-34, ...

<OT>
OK, I can identify the last one as a physical constant, but not the two former. Are they?
</OT>

 

[KimB]

3e6, -2.314e1, 6.626e-34, ...

<OT>
OK, I can identify the last one as a physical constant, but not the two former. Are they?
</OT>

I didn't think any of them were physical constants...

Update:
Heh, n/m...that last one's "h" ... I usually use "h-bar"

 

[Myrmecophagavir]

3e6, -2.314e1, 6.626e-34, ...

<OT>
OK, I can identify the last one as a physical constant, but not the two former. Are they?
</OT>

I didn't think any of them were physical constants...

Isn't it Planck's or something?

 

[KimB]

Blah, you beat me to the reply before I could edit.

Yes, it's Planck's Constant, "h", that is frequently referred to in physics as the smallest possible unit of phase space (or something similar). Phase space is a six-dimensional space including both position and momentum. You can see this in the uncertainty principle (one manifestation of which is):
dxdp ~ h
(Where the ~ means this is an approximate equation...there are more rigorous ones...which is where, if I remember correctly, the h-bar comes from, it's h/2pi)

I guess the fact that I know all about how "h" is used (well, at least how it has been used in the physics I've taken so far), and yet can't recognize the number when I see it has something to do with the way physics is taught these days....

 

[sireric]

Couple of comments regarding my older comments:

1) R3x0 can write out its internal format as an external 128b, configured as four 32b SPFP values.

2) When the R3x0 writes out that format and reads it back for internal use, the values obtained back match the ones written out (no precision loss). This is not wrong.

3) The main issue with writing out 128b/pixel is memory bandwidth; the BW hit limits you to at most 4 pixels/clock (assuming all BW is used for the writes), but performance is, I believe, actually lower (memory efficiency cannot be 100%, due to page crossing, R/W switch, refreshes, etc...). I've never measured this performance.

 

[Ilfirin]

3e6, -2.314e1, 6.626e-34, ...

<OT>
OK, I can identify the last one as a physical constant, but not the two former. Are they?
</OT>

The last is Planck's constant, the middle is -e^pi, and the first is a typo (was supposed to be 3e8 - the speed of light).

 

[OpenGL guy]

The last is Planck's constant, the middle is -e^pi, and the first is a typo (was supposed to be 3e8 - the speed of light).

Uh, the speed of light is 1.86e5

 

[Ostsol]

The last is Planck's constant, the middle is -e^pi, and the first is a typo (was supposed to be 3e8 - the speed of light).

Uh, the speed of light is 1.86e5

Bah. . . "Miles". . . What kind of backwater still uses that archaic measurement?

 

[Ilfirin]

The English system has no place in physics

 

[OpenGL guy]

The English system has no place in physics

Well at least the commonly used value in miles has more significant digits! So many people write 300,000 km/s (or 3e8 m/s as above) when really they should write 3.00e5 or 3.00e8.

P.S. Here's the exact value: 299,792,458 m/s. Yes, it's exact because the meter is define as "the length of the path travelled by light in vacuum during a time interval of 1/299 792 458 of a second."

 

[KimB]

Then how is the second defined?

After all, the rotation of the Earth is slowing down by a very small amount over time...so the value, "1/86400th of one revolution of the Earth" is not a constant value...

Oh, and one question...what is the use of -e^pi? I've seen complex exponentials with a "pi" in them (-e^i*pi would be 1), but not real exponentials...

 

[Ilfirin]

Well at least the commonly used value in miles has more significant digits!

Bleh, significant digits are BS too

Oh, and one question...what is the use of -e^pi? I've seen complex exponentials with a "pi" in them (-e^i*pi would be 1), but not real exponentials...

There is no use that I'm aware of. I wasn't aware I had to use universal constants when giving examples of scientific notation I just mashed some buttons to get an irrational number.

 

[OpenGL guy]

Then how is the second defined?

After all, the rotation of the Earth is slowing down by a very small amount over time...so the value, "1/86400th of one revolution of the Earth" is not a constant value...

You've forgotten about atomic clocks A second is a certain number of vibrations of light emitted by radioactive Cesium. *digs around for link* Ahh here we go.

A "cesium(-beam) atomic clock" (or "cesium-beam frequency standard") is a device that uses as a reference the exact frequency of the microwave spectral line emitted by atoms of the metallic element cesium, in particular its isotope of atomic weight 133 ("Cs-133"). The integral of frequency is time, so this frequency, 9,192,631,770 hertz (Hz = cycles/second), provides the fundamental unit of time, which may thus be measured by cesium clocks

Oh, and one question...what is the use of -e^pi? I've seen complex exponentials with a "pi" in them (-e^i*pi would be 1), but not real exponentials...

Dunno. I saw that years ago (as well as pi^e) and could never see a real use for it.

 

[DemoCoder]

Actually NIST has developed an "all optical" clock that uses optical wavelengths (Terahertz) and beats the best cesium atomic clocks by a factor of 1,000.

 

[Simon F]

The English system has no place in physics

"English"? No, "Imperial", as in the Roman Empire! That is how old it is. Of course is doesn't help that the USA got many of the measurements wrong

 

[pcchen]

Dunno. I saw that years ago (as well as pi^e) and could never see a real use for it.

e^pi is transcendental, but the status of pi^e is still unknown. Therefore, it provides works for some mathematicians