Look out R420, NV40 has 16 pipelines & 210 transistors!!!

[krag]

If this is true the R420 might be the newest version of the R8500 all over again! Yeah, I know its from the Inquirer but sometimes they are right. This would be a monster card!!! I might just go back to NV if this proves true! I have been a ATI fanboy since the 9700 Pro. But my money allways goes to the victor! It won't be to long now before we find out for sure!

NV 40 16 pipes...


Scope it and whine!!

 

[coolcpu2]

This card will retail at $700 though?

 

[Evnas]

same thing I posted in the other thread

All I have to say is no

It would cost WAY to much to produce the card, they would lose insane amounts of money off of such a complex, unneeded design. Also, to go with such a move would severly limit them as far as the future is concerned, and they would either have to release later in the future, a card that sticks to the standards (and thus would be severly slower), or continue to release huge, extremely inefective cores (cost wise)

To make a jump like that over the standards of the day's technology, means they would need to stay that far ahead of the technology to keep up with product cycles. Or, they would have to make a major jump back to the standards, which would look bad to the companies investors, and consumers.

Given nVidia's track record of late, it is a possibility, but it will bankrupt the company in the long run

Its very unlikely. DX 9 requires 8 pipes, DX 8 required 4...following the trend, 16 pipes would likely be the requirement of DX 10. That leaves nVidia with the options of either just ramping up the clock speed release after release, or continuing to stay that far ahead of technology

 

[Sentential]

More power to em. Its about time they pulled their heads outta their asses!

Only problem I see is with the transister count!!!

Remember Prescott has a very large number of transisters. It is entirely possible that the "dust-buster" will be reborn

 

[Krowa 02]

Dont judge it yet, wait, done assume.

 

[Th0r]

Whats the point on jumping to the 16x1 pipeline?...Direct X9 only uses 8 doesn't it?..Unless MicroShaft are planning something...:S...

 

[violineb]

What's the difference between a 8x1 pipe system and a 4x2 system? I thought the Volari duo had 16 pipes. Or maybe my memory is wrong. But either way 210 transistors is too much to be economic.

 

[JCLW]

Hello?

Doesn't everyone remember that the 5800 was supposedly 8x1?

There's a good discussion about it at Beyond3D.

 

[JCLW]

What's the difference between a 8x1 pipe system and a 4x2 system? I thought the Volari duo had 16 pipes. Or maybe my memory is wrong. But either way 210 transistors is too much to be economic.

I'm not an expert but:
8x1 can work with 8 pixels per clock, and do one thing to each pixel.
4x2 can work with only 4 pixels per clock, but has the capabitility to do two things per clock.

If you only need to do one thing per pixel on a 4x2 you're still only going to get 4 pixels per clock out of it.

So in a single texturing test the 8x1 will be twice as fast as the 4x2 (assuming same clock rates) but in a multi texturing test the 4x2 will catch up.

I believe what happens with the 5800/5900 cards is that they can only run four pipes in FP32, but they can run eight pipes at FP16. ATI runs 8 pipes at FP24.

DX9 calls for at least FP24 I think.

So when running in DX9 mode they are slow, but when they drop down to DX8 they run things at FP16 and are fast. That's part of the whole nVidia driver controvesy - Aparently the drivers are forcing the cards down to FP16 to gain speed, at the cost of image quality.

So IF the nv40 is really a "16x1", we might find it only does 16x1 in FP16 (so you can play UT2003 @ 750FPS instead of 500FPS) but when running new games (with the spec IQ) it is only 8x2.

re: 210 transistors: I could be wrong but I doubt they're going to have 210 transistors running at 500~600mhz on a 0.13 process...

... if so the Prescott will look like an ice cube in comparison.

edited for clarity

 

[micamica1217]

What's the difference between a 8x1 pipe system and a 4x2 system?

well, let's see........

NV35 is a 4x2 -- that can do 2 text units per pipe or 1 FP unit per pipe.

the R350 is a 8x1 -- that can do 1 text unit and 1 FP unit per pipe.

so with 8 text units and 8 FP units (PS2.0) the ATI chip can do it in one clock while it takes the nVidia chip three clocks....this is about as simple as I can make this folks.

so what's the point? well, it's more then just how many "pipes" a chip has....it's just what each pipe can do as well as the speed of the chip, plus...bla, bla, bla

now on to the new NV40....

just what can each pipe do????
does the chip have the same transistor problem that the NV3X has?
????
????
????

man, I have so many questions, and they could never be ansered untill the chip arives.

mica

 

[DDR-PIII]

Is NV known to have devloped a PCI Express GFX card yet ? If they haven't I dont think the 16 pipes will help THAT much.

-DDR

 

[JCLW]

They've said they plan on bridging it for this generation.

Game developers are saying PCI-E won't make a difference 'til H2 2005 anyways (ie: games won't be writen to take advantage of the PCI-E benifits 'til then).

 

[xtrmeocr]

Is NV known to have devloped a PCI Express GFX card yet ? If they haven't I dont think the 16 pipes will help THAT much.

-DDR

nv40 does have native pci express

 

[ashenfang]

so it will take up 3 pci slots instead of 2 now right? j/k although it wouldn't suprise me. I really take tech news with a grain of salt until it hits store shelves as a final product.

 

[Oni]

Wow. Those are some really high numbers. I mean, 16, 210. Amazingly high numbers in some cases.

I'll believe it when I see it, however.

 

[CalsonicGTR]

Posted byxtrmeocr:
nv40 does have native pci express

That's not actually true. At the moment, only ATI's R423 uses a native PCI Express design. The NV40 uses an AGP8X to PCI Express x16 bridge chip

Posted on Anandtech:
ATI went with a native PCI Express interface because it gives them a full 4GB/s upstream and downstream bandwidth at the same time. This will allow for some massive amounts of data to move between the GPU and the CPU/main memory in both directions. They also have the advantage of not needing an extra component on the graphics card itself.

NVIDIA chose to use a bridged solution (which they like to call their High Speed Interconnect or HSI solution) which gives them the ability to only produce one GPU for both AGP and PCI Express based solutions while the transition is being made to the new platform. This gives them the advantage of being more flexible to demand for AGP and PCI Express based products, and they won't have to forecast just how many of which type of GPU they will sell in any given silicon run

The full article can be found here

 

[Amino]

They should just concentrate on making faster cards. Not cards that can handle code that isn't written yet.

Pro to XT? Good step forward.

8 to 16 pipes? Why?

Despite the fact that one is just an OC'd version of the other, I still think that they were heading in the right direction. But if they made a card that was about 100Mhz faster on both core and memory, that would really be an accomplishment. I don't know if Increasing the pipeline number helps that, but I think speed is what's important until they come out with new code.

 

[Overclocker550]

16 pipes on the nv40 would efficently double the speed of the fx5900. just look, the 9800 pro is twice as fast as a 9600 pro despite the fact the 9600 pros core is clocked higher, its 128 bit ram hurts it even more. dx9 requiring 8 pipes is nonsense or the 9600 pro isnt a real dx9 card LOL! anyway take it with a grain of salt but if what nvidia claims is true and with the use of forceware, itll be twice the speed of the fx5900 and easily 50% faster than the r420. can we say 50,000 3dmarks in 2001 and 15-20 in 2003?????????? wow..........

 

[Evnas]

16 pipes on the nv40 would efficently double the speed of the fx5900. just look, the 9800 pro is twice as fast as a 9600 pro despite the fact the 9600 pros core is clocked higher, its 128 bit ram hurts it even more. dx9 requiring 8 pipes is nonsense or the 9600 pro isnt a real dx9 card LOL! anyway take it with a grain of salt but if what nvidia claims is true and with the use of forceware, itll be twice the speed of the fx5900 and easily 50% faster than the r420. can we say 50,000 3dmarks in 2001 and 15-20 in 2003?????????? wow..........

The problem with this post...is you take in VERY little of the multitude of things that go into a core. For all we know, this core may run at really low clock speeds due the the amount of transistors (and thus a large amount of heat), and thus it would NOT be twice as fast as the 5900...would it?

 

[Maxvla]

They should just concentrate on making faster cards. Not cards that can handle code that isn't written yet.

thats the best line i've seen in AGES!!!!!!!

kudos to Amino!