Ascended Saiyan
Newcomer
R300 pics?!!{56k warning}
- Thread starter Ascended Saiyan
- Start date
I think its just occured to me what the expansion strip is for.
Radeon 8500 AIW DV had issues becuase they were trying to use the Digital Tuner chip on the board but it didn't agree with the operating frequencies of the chip/mem which is why they initially released a clocked down version (and later a full speed verion using an analogue tuner, but with less functionality). I'm thinking that their thinking is that if the can isolate the digital tuner chip from the rest of the board then they will still be able to use it, hence an expansion daughter board on an AIW version. What do you reckon?
Radeon 8500 AIW DV had issues becuase they were trying to use the Digital Tuner chip on the board but it didn't agree with the operating frequencies of the chip/mem which is why they initially released a clocked down version (and later a full speed verion using an analogue tuner, but with less functionality). I'm thinking that their thinking is that if the can isolate the digital tuner chip from the rest of the board then they will still be able to use it, hence an expansion daughter board on an AIW version. What do you reckon?
Doomtrooper
Veteran
jackkoho said:
That would be correct
On first glance they look authentic to me. The card is admirably small, not anywhere near the monster a Ti4600 is. The heatsink/fan seems to be a pretty serious piece of equipment tho...
Ed: Oh, and that thing definitely looks like a 4pin power connector, which is a sensible move considering the supposed power requirements of the R300 (if we learned one thing from Nvidia's past, its never trust the motherboards to stick to the AGP specifications, hehe)
Ed: Oh, and that thing definitely looks like a 4pin power connector, which is a sensible move considering the supposed power requirements of the R300 (if we learned one thing from Nvidia's past, its never trust the motherboards to stick to the AGP specifications, hehe)
horvendile
Regular
I may have got something terribly wrong, but to me supersampling sounds like a good idea in some circumstances, namely slow games where excessive framerates are of secondary importance and you have a lot of time to look at crawling textures. One such game would be Gabriel Knight 3; there are others.
I do, however, not make any pretence at knowing exactly what I am talking about. I have been given to understand that texture crawling can be remedied with huge amounts of anisotropic filtering. But even if that is correct, wouldn't it just introduce another entity which we could do without if we had SS instead of MS? Mind you, that argument is based more on neatness than performance estimations. Is the general verdict that MS+AF gives at least equal IQ results at higher performance than a clever SS?
Of course, I recognize that in many cases MS will do by itself, I just like alternatives as much as the next man. I also recognize that my post might not seem very coherent (or, for that matter, informed), but I hope for the best
I do, however, not make any pretence at knowing exactly what I am talking about. I have been given to understand that texture crawling can be remedied with huge amounts of anisotropic filtering. But even if that is correct, wouldn't it just introduce another entity which we could do without if we had SS instead of MS? Mind you, that argument is based more on neatness than performance estimations. Is the general verdict that MS+AF gives at least equal IQ results at higher performance than a clever SS?
Of course, I recognize that in many cases MS will do by itself, I just like alternatives as much as the next man. I also recognize that my post might not seem very coherent (or, for that matter, informed), but I hope for the best
Hi,
Yes.. those pics are most definitely genuine. The R300 board is revision "00B"; the sucessor to the Computex board (PN 109 94200-00A). And A12 silicon makes sense too; they should ramp using this second-pass silicon if I'm not mistaken.
The DC connector is the long strip to the left of the GPU, yes. DCs are configurable by the manufacture only (NOT by the end user - although they can be swopped fairly easily) and require a BIOS flash and some component changes on the mainboard. The other options are set by the DC using straps. DCs are used mainly from an economical point of view; as far as I know there is no DC with a tuner at this time. The main cost cutting comes from the integration of the secondary TV DAC onto the die, which keeps PCB/component costs very low; no need for an RT chip on the standard boards.
Some of the DC configurations available are MEAN! I am aware of one that uses a Silicon Image TDMS resulting in an R300 that can provide clean 1920x1xxx DVI-I into two displays.
MuFu.
Yes.. those pics are most definitely genuine. The R300 board is revision "00B"; the sucessor to the Computex board (PN 109 94200-00A). And A12 silicon makes sense too; they should ramp using this second-pass silicon if I'm not mistaken.
The DC connector is the long strip to the left of the GPU, yes. DCs are configurable by the manufacture only (NOT by the end user - although they can be swopped fairly easily) and require a BIOS flash and some component changes on the mainboard. The other options are set by the DC using straps. DCs are used mainly from an economical point of view; as far as I know there is no DC with a tuner at this time. The main cost cutting comes from the integration of the secondary TV DAC onto the die, which keeps PCB/component costs very low; no need for an RT chip on the standard boards.
Some of the DC configurations available are MEAN! I am aware of one that uses a Silicon Image TDMS resulting in an R300 that can provide clean 1920x1xxx DVI-I into two displays.
MuFu.
horvendile said:I may have got something terribly wrong, but to me supersampling sounds like a good idea in some circumstances, namely slow games where excessive framerates are of secondary importance and you have a lot of time to look at crawling textures. One such game would be Gabriel Knight 3; there are others.
I do, however, not make any pretence at knowing exactly what I am talking about. I have been given to understand that texture crawling can be remedied with huge amounts of anisotropic filtering. But even if that is correct, wouldn't it just introduce another entity which we could do without if we had SS instead of MS? Mind you, that argument is based more on neatness than performance estimations. Is the general verdict that MS+AF gives at least equal IQ results at higher performance than a clever SS?
Of course, I recognize that in many cases MS will do by itself, I just like alternatives as much as the next man. I also recognize that my post might not seem very coherent (or, for that matter, informed), but I hope for the best
There is no right and wrong in this discussion, but SS can not "replace" anistropic filtering if that's what you were wondering about, as they're doing two fundamentally different hings, that happen to have a similar side effect. While SS reduces texture aliasing as well as redicing polygon "jaggies", anistropic filtering does not only reduce texture aliasing, but also achieves better IQ by by giving you better depth detail and more correct filtering of textures. The only scenario where anistropic does not Increase overall IQ are polygons directly facing the screen, every other polygon benefits from anistropic by having increased detail and sharpness, additionally to reduced aliasing, depending on the number of texel samples taken from the texture.
So the question shouldn't be wether SSAA or MSAA+Anistropic are preferable, but rather wether SSAA + Anistropic or MSAA + higher degree Anistropic (a higher degree is needed to better reduce texture aliasing and achieve comparable overall IQ) should be the way to go. IMHO, if done intelligently, it would be easier to achieve good FSAA by going the MSAA + Anistropic route in the next few generations of hardware, as the performance hit of SSAA is still too dramatic.
Maybe in a couple of years, when 3D accelerators have truly overtaken the resolutions our display devices can offer us and SSAA is possible 16x @ 1600x1200 @ 120fps, then I wouldn't see the point of MSAA anymore, until then...
horvendile
Regular
Thanks for the clarification Gollum.
It seems to me like we have some kind of ATI AF-scenario regarding MSAA+AF vs SSAA; vastly superior performance at the expense of IQ in some very special situations (polygons facing the screen at certain angles).
Edit: It also seems I'm very, very off topic.
It seems to me like we have some kind of ATI AF-scenario regarding MSAA+AF vs SSAA; vastly superior performance at the expense of IQ in some very special situations (polygons facing the screen at certain angles).
Edit: It also seems I'm very, very off topic.
Firstly, the card looks genuine to me.
Secondly, I would argue that you are incorrect in saying that pure supersampling can not replace Anisotropic filtering.
Take this screenshot from Quake 3 taken by me as an example: http://www.users.on.net/triforce/6000/glide0009.jpg
That's 8x Supersampled Jittered Grid Anti Aliasing from a V5 6000. The filtering method is just plain old bilinear (with subsample mipmap dithering aka performance trilinear). Notice that the textures are rather sharp and detailed, yet there is a distict lack of texture aliasing.
Of course, 8x Supersampling comes at a mighty huge fillrate hit, which I do agree, is a little too high at the moment.
-Colourless
p.s. There is actually a tiny bit of texture aliasing present on the left hand side. There is a very faint moire pattern on the ground. This is a side effect of subsample mipmap dithering with an agressive LOD bias. Remember that subsample mipmap dithering is an approximation, a hack if you will, to emulate the effects of trilinear filtering. One side effect is it begins to break the texture antialasing qualities of supersampling. So, it was probably not the best choice of screenshots to show.
Secondly, I would argue that you are incorrect in saying that pure supersampling can not replace Anisotropic filtering.
Take this screenshot from Quake 3 taken by me as an example: http://www.users.on.net/triforce/6000/glide0009.jpg
That's 8x Supersampled Jittered Grid Anti Aliasing from a V5 6000. The filtering method is just plain old bilinear (with subsample mipmap dithering aka performance trilinear). Notice that the textures are rather sharp and detailed, yet there is a distict lack of texture aliasing.
Of course, 8x Supersampling comes at a mighty huge fillrate hit, which I do agree, is a little too high at the moment.
-Colourless
p.s. There is actually a tiny bit of texture aliasing present on the left hand side. There is a very faint moire pattern on the ground. This is a side effect of subsample mipmap dithering with an agressive LOD bias. Remember that subsample mipmap dithering is an approximation, a hack if you will, to emulate the effects of trilinear filtering. One side effect is it begins to break the texture antialasing qualities of supersampling. So, it was probably not the best choice of screenshots to show.
Gollum said:
Heh.... supersampling can replace anisotropic filtering, if you're willing to take enough samples.
Don't forget that bilinear isotropic samples are the basis of most hardware anisotropic filtering implementations that we see today. You can get to the same place by either taking more isotropic bilinear samples at the texturing stage, or more bilinear isotropically sub-sampled pixels at the rasterization stage.
Granted, super-sampling is a rather inefficient way to get anisotropically filtered textures... with super-sampling, you're looking at a number of bilinear samples equivalent to the degree of anisotropy squared, whereas in the texture stage you can accomplish the same thing with a number of bilinear samples equivalent to the degree of anisotropy times four.
That is to say, the four bilinear-textured sub-samples (16 texel samples total) in 4x supersampling gives a functional degree 2 anisotropy, whereas 16 texel samples could deliver degree 4 anisotropy if applied directly in the texture sampling stage....
Ok, I managed to get some information regarding the board in the pic...
The reason that the first boards (e.g. the one at Computex) didn't have the external power connector was because the AGP spec violation was not anticipated; therefore no need for the connector when designing the "00A" PCB. The boards did run ok, but only on high quality mobos such as those made by Intel (and perhaps VIA), which could handle the extra current demands.
The revised boards ("00B"), such as the one shown, most likely have the connector just so that they can be fitted with the older stepping for testing purposes. It is likely that final boards ("00" - no suffix) will not have this feature, with the possible exception of the AIW. They will also probably be fitted with dipswitches so they can switch between NTSC/PAL like the RV250s.
The R300 card shown does use an 8-layer board. According to the PCB file anyway... *commences violent winking*
MuFu.
The reason that the first boards (e.g. the one at Computex) didn't have the external power connector was because the AGP spec violation was not anticipated; therefore no need for the connector when designing the "00A" PCB. The boards did run ok, but only on high quality mobos such as those made by Intel (and perhaps VIA), which could handle the extra current demands.
The revised boards ("00B"), such as the one shown, most likely have the connector just so that they can be fitted with the older stepping for testing purposes. It is likely that final boards ("00" - no suffix) will not have this feature, with the possible exception of the AIW. They will also probably be fitted with dipswitches so they can switch between NTSC/PAL like the RV250s.
The R300 card shown does use an 8-layer board. According to the PCB file anyway...
*commences violent winking* MuFu.
Similar threads
- Locked