Ati Roadmap HOLLY CRAP!!!!

[Daemon_UK]

I don't know if this has already been posted!!

Road Map

Rage 3d Original Post

:Q

 

[Killrose]

Dual Radeon R200 chips on one card. If they can pull it off it ought to be good. The R300 looks interesting.

 

[Sunner]

Now lets see if they can keep that map, and most of all, avoid another utter fiasco like the preview, and launch of the R8500 was.

Well Im keeping my fingers crossed

 

[Daemon_UK]

Well, I saw somewhere (someone find the link for me) that the Geforce 4 Ti-500 + will have 6 pipe-lines, whereas the ATi R300 will have 8.

Im sure Feb-Mar is going to be an interesting time. Hopefully, a test demo of Doom3 will be out by then!!



<< Well Im keeping my fingers crossed >>



Yeah me too, this KYRO II is doing nicely, but I want to try out all these new features like smooth vision, and 16 tap anistropy!

 

[BFG10K]

Dual Radeon R200 chips on one card. If they can pull it off it ought to be good.

Ugh, I'm getting flashbacks to the appalling MAXX technology. Let's hope they go SLI this time like 3dfx did.

 

[DClark]

The problem with the Maxx wasn't AFR, it was the lack of an AGP bridge to allow both cores to run as AGP parts (the secondary core on the Maxx ran as a PCI part).

That roadmap is old though - I think they've skipped the "R200 Maxx" and settled on the R200 to be followed in spring 2002 by the R300. My guess is that ATi was anticipating that nVidia would release something other than just a tweaked GeForce3 (the Ti series), so they had to put something impressive in their roadmap to get attention until the R300 is made. That was exactly why the Rage Fury Maxx was made in the first place - to buy time until the R100 (Radeon) core was completed.

 

[AA0]

right, MAXX wasn't the bad part, actually, it made 3dfx's dual chip technology look like crap. MAXX renders alternate frames, where as SLI renders alternative lines, MAXX is far more efficient. Apparently ATI has been working on drivers, and the actual efficiency of it the whole time, if they can get it working right, it will just kill the competition in terms of speed. Unfortunately, MAXX only works in full screen games, some games which are windowed won't work, unless they fix that too.

 

[js1973]

I don't see why people get excited over design stage specs.

 

[Innoka]

AAO can you explain please why MAXX "made 3dfx's dual chip technology look like crap". Are you party to some highly technical knowledge you can point to.

 

[DClark]

I wouldn't say that AFR made SLI look like crap, but AFR is more efficient than SLI.

In SLI, both cores have to process all the triangles of every frame, then each core does half the filling of the triangles. With AFR, you don't have two cores rendering the triangles, because each core has its own frame to render all by itself. Here's ATi's AFR FAQ regarding SLI and Wicked 3D's PGC:

There are 2 other techniques for multiple chip parallel processing:


3Dfx's SLI - Scan Line Interleave, and

Metabyte/Wicked 3D's Parallel Graphics Configuration.

3Dfx's SLI's technique:
One chip draws out even lines of an image and the other chip draws out the odd lines of the image. This technique is inefficient because both chips process triangle set up on one frame.

Unlike SLI, AFR processes separate frames.

Metabyte/Wicked 3D's PGC technique:
Both chips work on different section of a frame. That is, one chip renders the top half of the screen, while the other chip renders on the bottom half of the screen. This technique has potential for the render loads to be unbalanced between the two chips. One half of the screen can have fewer polygons to render than the other. For instance, a scene with a sky can pose obvious load unbalancing because the sky requires less rendering than what the bottom half would potentially render. Further image tearing, difficulties in DAC calibration also poses inherent problems with this technique.

In ATI's AFR technology, load unbalancing does not occur since each chip processes separate and complete frames. The render load from frame to frame is reasonably balanced unlike the wide variance found within the complexities of a frame.

Here is ATi's developer info on "what is AFR?".

 

[anime]

well road map always look good on paper--the thing is can they do the roadmap in time? as in the past there always delay and more delay.

 

[BFG10K]

right, MAXX wasn't the bad part, actually, it made 3dfx's dual chip technology look like crap.

Err, no. MAXX produced widly fluctuating framerates plus it produced horrible tearing because there was no way to synchronise one GPU with the other. If the next GPU's frame was ready the existing one would be ripped off the screen regardless of whether or not it had been fully drawn.

OTOH 3dfx's SLI technology was much better because the next frame wouldn't appear until existing frame was constructed (as both GPUs were tied to drawing the same frame) so this produced smoother and more consistent framerates.

 

[Jen]

i like to see it past paper and out to the public. proove to me what you say on paper.



Jen

 

[DClark]

AnandTech covered the Maxx's lag rumour in their Rage Fury Max review (here is the page concerning lag). AnandTech ends up describing how the Maxx uses AFR as kind of like an complex version of triple-buffering. As for tearing, so long as V-sync was enabled, I never noticed any tearing at all. Since I get the same performance with my single cored Radeon LE (some tearing when V-sync disabled, none when enabled), tearing was not a problem caused by AFR.

I own a Maxx (though it's been sitting on my shelf for most of 2001 after I bought my LE), and I was very impressed by the card when I used it. Image quality was excellent for the day (and is almost as good as the Radeon's image quality), and the card played every game I had (and was fast, especially in games like Unreal Tournament and it was excellent for EverQuest when Kunark was released). While playing EverQuest I'd hear constant complaining about the GeForce cards choking on the massive, detailed (non-hardware T&Led) zones in Kunark, yet I was fine with my clipping plane out past the maximum allowable for most zones. My highest 3DMark99 score (the last 3DMark not to use T&L) at the default setting (800x600x16) was 9033marks with a 1Ghz processor - an impressive score from any graphics card.

To say that AFR was rough around the edges is kind of like saying a car engine sounds rough after removing the exhaust pipe. With V-sync enabled, AFR was a dream and a half - it took two scrawny little Rage128Pro cores and helped them compete with the brand new (at the time) GeForce cards. I've said it before and I'll say it again: The Rage Fury Maxx was the Rodney Dangerfield of graphics cards. It had hardware DVD playback, excellent 32-bit performance, line and edge Anti-Aliasing, Bump-mapping, and the first and as yet only useage of AFR. It was made at a time when ATi knew that they had a winner in the yet-to-be-released R100 (Radeon) core, and the Maxx bought ATi enough time to finish the R100 properly and not rush it out in fear of not having a card that could compete with the original GeForce.

If the R100 would have been rushed out the door, ATi would have most likely had to fix any problems that may arise when a product is released prematurely <cough> P3 1.13Ghz <cough>, and not been able to concentrate on making the R200 the powerhouse core that it is today. That's why I say that I don't think ATi will use AFR for the R200 - not because it doesn't do its' job, but because it did its job back in 1999 when it had to, and allowed ATi to make cores that were powerful enough without double-teaming the competition.

 

[Sunner]

About AFR, I dont know very much about the inner working of the technology, bu in the end, SLI worked great, and increased performance greatly without any sideeffects, never heard anyone complain about problems that were directly related to SLI.

AFR on the other hand, they could never make it work in Win2K, lots of people reported problems like the one BFG described with fluctating frames, Sharky or Firingsquad(dont remember which) even made a whole article about this, and concluded that there was indeed a problem.

 

[AA0]

I never had tearing on my MAXX. The problem with SLI is its just not efficient enough, putting two gpus on a card now is expensive, and getting moderate performance from the two isn't what you want.

I never had any luck overclocking the MAXX, don't think any tools were really compatitible. As for frame rates, I really didn't notice slow downs, or major drops in performance, but I don't monitor my fps constantly, I actually play the game.

 

[DClark]

I think that this is the graph from Sharky Extreme you're referring to in regards to the difference in FPS with V-sync disabled. It must be remembered though that the graph represents less than 7 seconds worth of images at an average of 30fps. To see the 200 changes from high to low and back to high in the span of 7 seconds is beyond the ability of the human eye and brain (though you would still be able to see tearing, as the V-sync was disabled in the bench).

But if you think that you could see the differences, the fix is simple (and shown later in the same article here): enable triple buffering. The Maxx actually gains performance with triple buffering enabled, and the frame per frame performance fluctuation is reduced significantly.

As for overclocking utilities, Powerstrip worked great for me. In regards to Win2000, you have to remember at the time the Maxx debuted, Win2000 wasn't even released. I think it was a bad decision on ATi's part to say that they'll supply Win2000 drivers later and then not do so, but at the time the Maxx was created, Win98 was the PC gaming platform.

 

[vedin]

I can't see putting two GPUs on a Radeon board anyway. In order for it to actually BE much faster would be to increase the RAM speed and amount by a long shot. Which, of course, drives up prices insanely.

 

[ZenOps]

Err, the reason that there are fluctuation framerates is because in a triple buffered game, only one frame can be displayed when the other is being rendered. The frame is never "ripped" away from the screen, it uses a very simple and very old programming algorithm its called first in, first out. If it is visually "ripping" you have set V-synch wrong (in either the main drivers, or game settings)

If the frames aren't rendered at the same speed (and they rarely are, because of scene changes) there can be up to a 50 percent delay in rendering the second frame, while the FIFO is waiting for the first, now last frame to be drawn. So instead of the supposed 200 percent speed increase two AFR chips might have, it may drop down to being only 150 percent better than a single chip (theoretically of course, in actuality the numbers are even lower than that)

In a double buffered game (maybe a two+ year old Directx6 game) the performance drop can go all the way to 100 percent or equivalent to a single chip...

SLI is not all that great either, although there is much less of an initial delay, the overhead in calculating every other line starts to eat away the performance at high resolutions. Its OK for games with 640 or 800 lines, but when it has to calculate 1280 or 1600 lines, there is a lot of overhead of recombining an image.

 

[Innoka]

The Gf2 32 pro will show as 30% faster than a V5 at low resolution but increase that to 75% at 1600x1200 in QIII. Not bad going against two chips.

 

[DClark]

The Voodoo5 was never meant to compete with the GeForce2 core, it was meant to compete with the GeForce core. The reason 3dfx went bankrupt was largely due to the fact that the V5 was supposed to be released around December 1999, but didn't make it to store shelves until around May 2000.

That would be like me bragging that I can kick the crap out of two midgets - I'm a 6 foot 1, 250 pound jail guard; I would be embarassed if I couldn't.

The GeForce SDR, GeForceDDR, Voodoo5, Rage Fury Maxx, and Savage2000 are all in the same performance and age category.

The GeForce2 series have the Radeon series and KyroI and II series as their competition.

The GeForce3 series has the Radeon 8500 series and to an extent the Radeon 7500 as competition.

 

[Innoka]

The Voodoo 5 was released after the Geforce 2 at a similar price... I don't chasten reviewers for putting the V5 in benchmarks of reviews for "later" designs, and I don't know who else does.
I think the dual R200 is just "big talk" anyway. How are the bitboys doing?