3D Board Naming (from ATI's new 2004 thread)

[Quitch]

I can see it now... NVIDIA releases the NV40, which, to everyone's complete surprise, kicks ass. Marketing at ATI goes crazy and demands a huge core speed increase, which is not possible without a voltage increase, which again is not possible without a cooler overhaul. The cooler is redone, but... it's really loud. And then the R420 comes out as the Radeon 5800 Ultra.

Where's the "ballpunch" smiley when you really need it.... :?

A joke, my friend, a joke. Many people were convinced that the NV30 was going to rock just as much as the R420 is now rumored to rock, so I'm just trying to provide a bit of insight through humorous references to cards past.

In that case however, the previous generation of cards were based on an API which was now out of date. No such thing is happening now. OGL 1.5 is already with us, and DX9 isn't going anywhere. ATI would really struggle to pull a nv30 until the R5xx series of cards.

 

[KimB]

In that case however, the previous generation of cards were based on an API which was now out of date. No such thing is happening now. OGL 1.5 is already with us, and DX9 isn't going anywhere. ATI would really struggle to pull a nv30 until the R5xx series of cards.

Many of nVidia's problems with the NV30 apparently had more to do with manufacturing rather than making different decisions than Microsoft.

 

[Rolf N]

The Radeon 9000/9200 also had some advancements over the Radeon 8500, so I'm not sure that's a valid argument.

What were the advancements? I thought the 9000 had the vertex shader from the 9700, (correct me if I'm wrong here), but it was also missing hardware TRUFORM, as well.

(I only have a Radeon 9200, I'm not sure if the same applies to Radeon 9000; though I'd guess it does)

RV280 is 4x1 while R200 is 4x2. But it can do single cycle trilinear, ie when using that it's as fast as R200 clock for clock. In fact it's even faster than R200 if you use an uneven number of texture layers (R200's loopback scheme is weird ...).
Note the memory clock difference!

R200 @ 250/250 MHz
Mode                RGBA5650 Z16 S0
--Textured fillrate-----------------------------
----Bilinear filter-----------------------------
    1               983.79 MPix/s
    2               983.79 MPix/s
    3               497.50 MPix/s
    4               496.77 MPix/s
    5               331.42 MPix/s
    6               331.41 MPix/s

----Trilinear filter----------------------------
    1               502.13 MPix/s
    2               502.13 MPix/s
    3               254.50 MPix/s
    4               254.50 MPix/s
    5               170.42 MPix/s
    6               170.42 MPix/s

RV280 @ 250/200MHz
Mode                RGBA5650 Z16 S0
--Textured fillrate-----------------------------
----Bilinear filter-----------------------------
    1               956.17 MPix/s
    2               495.06 MPix/s
    3               328.71 MPix/s
    4               248.58 MPix/s
    5               198.91 MPix/s
    6               165.79 MPix/s

----Trilinear filter----------------------------
    1               843.73 MPix/s
    2               423.00 MPix/s
    3               282.19 MPix/s
    4               211.74 MPix/s
    5               169.43 MPix/s
    6               141.22 MPix/s

Texture cache sizes are equal.

Fixed function T&L is some 20% faster on RV280 clock for clock. The vertex shader may be better, too, but I haven't tested that yet.

R200 also has numeric range issues with projected textures. This has been fixed in RV280.

 

[KimB]

RV280 is 4x1 while R200 is 4x2. But it can do single cycle trilinear,

The R200 doesn't even do trilinear filtering in conjunction with anisotropic. So I'm not sure this is a real benefit.

 

[FUDie]

RV280 is 4x1 while R200 is 4x2. But it can do single cycle trilinear,

The R200 doesn't even do trilinear filtering in conjunction with anisotropic. So I'm not sure this is a real benefit.

Sure it's a benefit, when most applications don't even support anisotropic filtering.

-FUDie

 

[hkultala]

In reference to this whole Geforce 4 MX naming scheme there is also the matter that the Geforce 4 MX is touted as a DX8.1 card, when it really isn't DX8.1 at all. It is only DX7. Essentially it is based off the original Geforce 2 MX with memory controller improvements and a bumped up clock speed.

And doesn't even support all major DX6 features (ebm) ...

 

[KimB]

Sure it's a benefit, when most applications don't even support anisotropic filtering.

-FUDie

Um. But you can force it. In nearly every application where it matters.

 

[sir doris]

Going back to the naming conventions, the current nVidia line up does not match well with the GF4. A friend (not very technically minded) had a GF4 4600Ti and it died, took it to his local PC shop and they replaced it with a new GF5600 (non ultra) told him it's the new version and would be slightly faster... Don't know if the shop was full of numpties or just con men. Whatever, haven't had the heart to tell him his new card is worse than the old one

 

[PatrickL]

Bah, it s not so bad. Go read the VN board and all theses guys asking why despite their upgrade from a gefrorce 4400 64 mo to a FX 5200 256mo they have so crappy performance

 

[KimB]

There's been a misconception for a very long time that memory size on a video card is significant. In the vast majority of cases, it isn't. It's sad, really, and the only thing you can do is inform your immediate friends that memory size doesn't mean squat. It's the other things that are a primary consideration (chipset, clock speeds).

 

[John Reynolds]

There's been a misconception for a very long time that memory size on a video card is significant.

It can be significant, but it depends on your settings. For someone like me, who runs almost every game at 12x9 with 6x AA, the extra RAM is useful. For others, probably not.

 

[PatrickL]

It is a misconception very usefull for marketing tricks

 

[KimB]

There's been a misconception for a very long time that memory size on a video card is significant.

It can be significant, but it depends on your settings. For someone like me, who runs almost every game at 12x9 with 6x AA, the extra RAM is useful. For others, probably not.

Not in the way that most users think it is. The differences you are talking about are very, very small compared to other factors (chipset and clockspeeds).

 

[John Reynolds]

Not in the way that most users think it is. The differences you are talking about are very, very small compared to other factors (chipset and clockspeeds).

Oh, I know. There's no doubt AIBs are fully aware of how to play the #s game for marketing, but I was just pointing out that there is a practical benefit from that amount of onboard RAM.

 

[ET]

It can be significant, but it depends on your settings. For someone like me, who runs almost every game at 12x9 with 6x AA, the extra RAM is useful. For others, probably not.

You're not running that res on a 5200, I'd bet.

 

[Fox5]

I can see it now... NVIDIA releases the NV40, which, to everyone's complete surprise, kicks ass. Marketing at ATI goes crazy and demands a huge core speed increase, which is not possible without a voltage increase, which again is not possible without a cooler overhaul. The cooler is redone, but... it's really loud. And then the R420 comes out as the Radeon 5800 Ultra.

Or ATi could just repeat what they did with the 8500(or be like nvidia but skip the giant fan step) and wait 6 months or so until they're capable of creating a card that can compete.

 

[Rugor]

More often than not cards are running into memory speed issues before they run into memory size issues. Also, the larger the framebuffer, the more bandwidth you need to use it effectively. The marketing driven pattern of loadign budget cards with vast amounts of slow RAM hurts performance in far more ways than it helps.

 

[PaulS]

GROUNDHOG DAY!!!!!!

Phil? Phil Connors? Phil Connors, I thought that was you! Now don't you tell me you don't remember me 'cause I sure as heckfire remember you!

 

[rwolf]

heh heh heh. Thats good Ned.

 

[T2k]

There's been a misconception for a very long time that memory size on a video card is significant.

It can be significant, but it depends on your settings. For someone like me, who runs almost every game at 12x9 with 6x AA, the extra RAM is useful. For others, probably not.

Not in the way that most users think it is. The differences you are talking about are very, very small compared to other factors (chipset and clockspeeds).

Wrong. Depends on your setup.
Let me just say one example: I'm playing everything as high as I can, as possible w/ 4-6x AA + 8x QAF - usually 1920x1200 if supported or 1600x1200.

Try to do it with half the memory - in my case 64MB...