• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

R420 12 Pipelines

More pipelines doesn't necessarily mean it would be better. Look at netburst.
 
Keep in mind Rumor, not fact.

I found this interesting though
But now we can reveal that the R420 design features 12 working piplines from a possible maximum of 16.

Sounds like the kinda deal with the GFFX pipelines, being able to be 8x1 or 4x2. Maybe the pipes in the R420 are done in a new way that allows it to be 12 when needed or 16 when needed. It would be the most effecient design IMO.
 
When do you think reviewer will be able to get a sample review of the R420XT R420Pro?
 
Niether ATI Nor Nvidia have been naming their cards after the the Core name and I doubt ATI start doing it now.

As for the reliability of the source. Well I trust the inquirer as much as i would trust a wet toilet paper rope bridge.
 
Originally posted by M4d-K10wN
More pipelines doesn't necessarily mean it would be better. Look at netburst.
:p That's the exact opposite, taking what you intended to say (generously). Netburst has 1 fewer execution unit (functionally) than the P3. You're thinking pipeline stages in the CPU, but that's not the same as pixel/texture pipelines in a graphics chip. :LOL:
 
"Maybe the pipes in the R420 are done in a new way that allows it to be 12 when needed or 16 when needed. It would be the most effecient design IMO."

If this card has the ability to use 16 pipelines, when would using 12 ever be more efficient?
 
TheInq made a funny about their own Rumormongering..

Check out the Official picture of the Athlon 64 2800

http://www.theinquirer.net/?article=14696

I think they did it just to make sure people remember that they are in the business of rumors... And although they have been hitting a lot of truth lately its not always 100 percent fact that they post up.
 
Originally posted by ec_hern
"Maybe the pipes in the R420 are done in a new way that allows it to be 12 when needed or 16 when needed. It would be the most effecient design IMO."

If this card has the ability to use 16 pipelines, when would using 12 ever be more efficient?

in some operations i am sure, just don't ask me what operations :D

just thinking out loud
 
If the NV40 still uses shared resources (arrays) between the pipelines, 16 pipelines could possibly be slower than the R420's 12 pipelines except in simple fillrate.
 
Originally posted by Brent


Sounds like the kinda deal with the GFFX pipelines, being able to be 8x1 or 4x2. Maybe the pipes in the R420 are done in a new way that allows it to be 12 when needed or 16 when needed. It would be the most effecient design IMO.

I disagree. It sounds like they'll have 12 pipelines in their cards, but the design can support a "possible maximum" of 16. I have no idea why they'd do this, though. Maybe to save on cost, or heat?
 
Originally posted by Sandman
I disagree. It sounds like they'll have 12 pipelines in their cards, but the design can support a "possible maximum" of 16. I have no idea why they'd do this, though. Maybe to save on cost, or heat?

Again, this isn't official information keep in mind.
 
Maybe they designed the card to have a max of 16 but only using 12 because there not able to incorperate 16 at this time, and they are going to use the same layout for a 16 pipeline version down the road, maybe an XT version?
 
Nvidia's CEO said that with the NV40 he was hopeful the card would have 2x the performance. Given this isnt it time to stop quoting some inquirer article that it will be 16 pipelines. it would have to be a major screwup to go from 4x2 to 16x1 and only get 2x the performance.
 
Originally posted by Orbius
Nvidia's CEO said that with the NV40 he was hopeful the card would have 2x the performance. Given this isnt it time to stop quoting some inquirer article that it will be 16 pipelines. it would have to be a major screwup to go from 4x2 to 16x1 and only get 2x the performance.

I'm no video expert (far from it), but isn't 4x2 able to do 8x1? And if so, that's half of 16x1.
 
Recombining 8 pipes or 8 way data is about the most efficient data scheme without getting overly-complex according to Intel. Too complex = sometimes having to lower chipspeed to compensate... Mind you we are only talking about 500Mhz for videochips, a far cry from 3 Ghz+ CPUs.

Take for example PC3200 Dual channel DDR: With one channel of DDR you might get 1600MB/sec. With Dual channel the max is twice that, but in reality the overhead of recombining the data gives about a 50 percent speed boost or to 2400MB/sec.

ATi seems to like to do 3x2 (6,12,24) videoprocessing pipelines insted of base2 (4,8,16,32) In a lot of ways base3 makes more sense as the most critical parts of 3D are base 3. IE: XYZ coordinates, RGB colour (no alpha)
 
Originally posted by ZenOps
Recombining 8 pipes or 8 way data is about the most efficient data scheme without getting overly-complex according to Intel. Too complex = sometimes having to lower chipspeed to compensate... Mind you we are only talking about 500Mhz for videochips, a far cry from 3 Ghz+ CPUs.

Take for example PC3200 Dual channel DDR: With one channel of DDR you might get 1600MB/sec. With Dual channel the max is twice that, but in reality the overhead of recombining the data gives about a 50 percent speed boost or to 2400MB/sec.

ATi seems to like to do 3x2 (6,12,24) videoprocessing pipelines insted of base2 (4,8,16,32) In a lot of ways base3 makes more sense as the most critical parts of 3D are base 3. IE: XYZ coordinates, RGB colour (no alpha)

very astute observation, and it would also lend some credence to Brent's *shrouded* indications that 12 pipelines at base 3 could in fact be more efficient in certain situations (i.e. coords & color)
 
Originally posted by ZenOps
Recombining 8 pipes or 8 way data is about the most efficient data scheme without getting overly-complex according to Intel. Too complex = sometimes having to lower chipspeed to compensate... Mind you we are only talking about 500Mhz for videochips, a far cry from 3 Ghz+ CPUs.

Take for example PC3200 Dual channel DDR: With one channel of DDR you might get 1600MB/sec. With Dual channel the max is twice that, but in reality the overhead of recombining the data gives about a 50 percent speed boost or to 2400MB/sec.

ATi seems to like to do 3x2 (6,12,24) videoprocessing pipelines insted of base2 (4,8,16,32) In a lot of ways base3 makes more sense as the most critical parts of 3D are base 3. IE: XYZ coordinates, RGB colour (no alpha)

Yeah, but doesn't the whole 9700/9800 architecture have 8 pipelines?
 
Is it not possible that ATI is doing this for financial reasons? We all know that to REALLY enjoy the upcoming titles and some current ones you need a fast ass system. Why not take the rejects that meet a certain, predefined reject criteria...I don't think many of us know a helluva alot about chip Q&A...a market them much like the 9500s? Oh and who said that the 9500 was a money loosing venture...hell it wasn't. Granted the 9500s were grabbing 9700 sales That was the 9700s didn't make a helluva alot of money...the high end segment doesn't anyway...and mainstream makes ATI and Nvidia a crapload of dough. Hell the 4200s are still making Nvidia cash though margins have fallen drastically. The 9500 was discontinued because of the transition to the .13 process and push the 9700s that ATI had in inventory because of the 9500 out the door since they still could outperform Nvidia's top products. Yields have been plagueing Nvidia and ATI as of late. It's called ROI (Return on Investment) if not mistaken. Great business sense to improve yields and improve company profitability. Plus everyone knows how the 9500 beat the live crap out of every Nvidia mainstream card at the time. Can Nvidia stand a beating like that in the mainstream retail market again? I think not. However, this is all saying that NV40 Ultramamajammerbigasscard is on par with the R420wehavepotheadengineers. Nvidia will have nothing in its arsenal to counteract.

Now for a brief change of subject to PCI Express. From what I understand the bus is much like a switch...correct...each slot has its own path. Now could it not be possible that TWO cards could be placed in a PCI Express bus now if I'm correct (help me at the technical level here guys) ATI does some sort of 24bit percision right? 12x1 and 12x1 is (even..odd) 24 bit right in other words a card that can operate in SLI mode. 16 pipe for the single card version. I also seem to remember that someone had stated that in the drivers (4.3) there was a hint of some sort of SLI operations. Hey I'm probably talking out my ass...but while we are daydreaming a lil here why not.
 
Originally posted by ZenOps

ATi seems to like to do 3x2 (6,12,24) videoprocessing pipelines insted of base2 (4,8,16,32) In a lot of ways base3 makes more sense as the most critical parts of 3D are base 3. IE: XYZ coordinates, RGB colour (no alpha)

What the hell are you talking about? The R300 was 8x1(9700)/or 4x1(9500). Where is the base 3 part?
 
I love how Brent acts like he doesn't know anything ;) ....


but then again...maybe he doesn't :D
 
Argh! When is this damn thing coming out so we can put an end to the rumors! I WANT MY R420!
 
There are such crazy rumors going around now. NV40 steppings A0/A1/A2 are supposedly fully functional, but NV's 210 mil trans chip isnt working yet. What does that tell you? I think nV is scared shitless and tried clamping extra pipes into an already bloated core. :confused:
 
^ hey, if it works, I won't complain. I couldn't care less if the core architecture is based on the tnt2 if it was faster and had higher IQ than the competators.
 
I'm doubtful that something like that would happen in the future, Merlin. The man is keeping us down.
 
I love the TNT2. I had one of those up until September of '03, when I finally realized it was outdated while playing CoD (none of the DX9 textures were showing).
 
I think the idle speculation at B3D is more accurate.

The R300 is 'designed' as a 12-pipe core, but they build the chip with 16 pipes. Since only 12 are 'needed', they just test the chips to see which 3 quads are the fastest, most reliable, most stable, whatever...and disable the other quad.

This let's ATI do a 'super-high end' core at the same time that is more than merely a bump in clock speed (a la the 9800XT) - the cores that test all 4 quads to work perfectly will be clocked a little higher and sold as their 'top end' product.

In any case, only needing 3 of the 4 of the quads you put on the chip to work for it to meet your 'ship it' criteria means you get VERY, VERY good yields.

Think about it. If nVidia really has a 16-pipe design, and put 16-pipes on the chip....each chip has to work 100% flawlessly to be shippable. ATI's plan (if, indeed, that's what this is) allows a 25% failure rate of the pipes to still yield a shippable product.

That should allow them to undercut nVidia significantly on price (as they have been this entire past generation, in fact).
 
Even TnT 1 was good. I still have it on my wall :) Should still be in working order.
 
Nvidia claims that with new tech in the nv40 (this actually said by the CEO) they are hopping for close to 100% yeilds on the nv40 and its derivatives.
 
Given this isnt it time to stop quoting some inquirer article that it will be 16 pipelines. it would have to be a major screwup to go from 4x2 to 16x1 and only get 2x the performance.

It's probably not going to be 16x1 any more than the NV30/35 was 8x1 in any meaningful sense (i.e. pixel pipelines vs. Z/Stencil ops). It will almost certainly be 8x2 in the same way the NV35 is 4x2.
The inquirer seems to believe multiplying the number of pixel pipelines by the texture sampling units gives you more pixel pipelines somehow.
 
IIRC the inq said that it would be true 16x1 not 8x2, I think that they grasp the difference.
 
"R420" / "Pipeline"

There is a joke in here somewhere, I can feel it.

"420" / "Pipe"

It's right there, I can almost see it.


This has to be the greatest video card code name in history.
 
Originally posted by Ibanez
There are such crazy rumors going around now. NV40 steppings A0/A1/A2 are supposedly fully functional, but NV's 210 mil trans chip isnt working yet. What does that tell you? I think nV is scared shitless and tried clamping extra pipes into an already bloated core. :confused:

this sounds a lot like nv30 redux...think about the implications of this statement.
 
The R420 won't use pipelines: it will be the first videocard to use a combination system of GPS, lasers, and WiFi.

I see you found the joke, Brent :D
 
IIRC the inq said that it would be true 16x1 not 8x2, I think that they grasp the difference.

It's clear they don't. You're right in that their newest story contradicts their previous efforts - Check their latest NV40 piece:
NV40 is a 16 pipelines part/ AFTER DAYS, nay weeks of inquiring we can now confirm that NV40 is not 8x2 marchitecture part as we previously suggested.
 
no, they are saying that they said before that it wasn't an 8x2 part, they aren't saying they were proved wrong, they are saying they were proved correct.
 
Back
Top