Nv30 to be 400Mhz core,128bit bus but...

[Fotis]

Yes it's from the inquirer www.theinquirer.net/?article=5794
It's not the first time I've heard of this rumor though.128bit bus but some misterius tech that will give it the power it needs.
Can anyone think of what that tech could be?

[SurfMonkey]

I would think that the most probable answer will be in their compression technology. I've been emailing a friend of mine in the states, and he says *rumour only*, that the NV30 has some pretty amazing tricks up its sleeve and some things which were going to be left out of the design have been included and work better than was projected.

Hearsay and rumour, but the word I got was that the NV30 is quite likely to make the R300 its bitch in a big way. But it is all down to yield now, and clock speed.

Hopefully that'll make ATi bring out a card that works for more than 40% of its owners. :D

[thcdru2k]

hopefully they get it out soon enough..i bet ati would press release the r350 by the time nv30 hits, and it will hurt sales.

[borntosoul]

well maybe all this talk about 128 bit mem and no ddr 2 this could mean that the nv30 would be cheaper to make than what i thought and with a nv31 (midrange card) coming out a month or so later (i dosent make sence to me for them to bring out the nv31 too close to the nv30 cause that would hurt the ti4600 sales ,looks like nv 30 ,a month later nv31 then a month or so later when they have better yields nv 30 ultra ,and then we get the 256 bit ddr 2 nv 35 ,thats how i see it , what a load of rubbish MUAHAHA ===and i love it

[Uttar]

ATI CEO confirmed to his shareholders that the R350 would be announced in H1 2003.

And in the corporate world, CEOs are often optimistic when talking to shareholders. So, really, it's unlikely it'll be launched any time before that.

Also, imagine the type of shock ATI would have if the NV30 was faster than the R350! LOL. Now that would be funny. On the other hand, that's unlikely. But we can always hope :)


Uttar

[LORD-eX-Bu]

If this no 256bit memory no DDR-2 400Mhz card is what is coming out, and if it costs more than $350 and doesn't go down quick, then I will be investing in a R350 instead.

[Bigus Dickus]

Originally posted by Uttar
the type of shock ATI would have if the NV30 was faster than the R350! LOL. Now that would be funny. On the other hand, that's unlikely. But we can always hope :)


Uttar

Imagine the type of shock everyone would have if the NV30 isn't faster than the R300!

I'm sure it will be in some cases, but being four months behind it really should be across the board. Let's hope it is.

[nutball]

Originally posted by Bigus Dickus
Imagine the type of shock everyone would have if the NV30 isn't faster than the R300!


Problem is how would we tell if it is, in real-world terms? You might well be right, in some areas maybe it is slower than R300. Are those areas important?

OK, we can benchmark UT2003, Doom ]I[, whatever else, but no games (real, or imagined) will stress half of the transistors on either R300 or NV30.

UT2003 runs at 500fps at 1600x1200 16xFSAA, 128xAF. Bore. Snooze.

zzzZZZZZZZZZ

This is just not interesting.

The real potential of both R300 and NV30 is the high-dynamic-range high-precision rendering, and that won't be exploited by any real-world games before both R300 and NV30 have been superceded.

The point I'm really orbiting around making is that both R300 and NV30 are not really about massive (irrelevant) improvements in the speed of current engines. Not even the next engines (unless they get some mid-development re-design).

They are a bit like the change that happened when floating-point units were integrated on-die onto Intel x86 CPUs as standard, rather than being a separate chip fitting in a separate socket. The performance may not change massively, but it allows software designers to think in a different way. FP is no longer an accelerated add-on for the rich boys, it comes as standard. The real significance is that you no longer have to work round short-comings of the hardware.

Or at least that's the way it will be when NV30 forms the heart of the nForce 4 platform...

[sancheuz]

I still think they will use 256 bit bus.

[koneill]

128-bit. you think wrong.

[Uttar]

Originally posted by Bigus Dickus
Imagine the type of shock everyone would have if the NV30 isn't faster than the R300!

I'm sure it will be in some cases, but being four months behind it really should be across the board. Let's hope it is.

Hey, i did say it was VERY unlikely! :)
And really, i wouldn't be all that surprised if the NV30 isn't faster than the R300. As i said before, the NV30 is about features. And if it IS faster than the R300, then it'll simply be because it's on 0.13 and got a higher clock.


Uttar

[tieros]

<inquirer mode on>

Since the NV30 is a 128 bit end-to-end chip, it only makes sense that the RAM would be 128 bit. Why waste all that GPU real estate on bus mux/demux circuitry when better, faster methods are at hand?

Cards based on the NV30 will be configured with 2 separate blocks of 128 bit quad pumped RAM, clocked at 200 MHz (800 MHz effective), with a modification to nVidia's existing crossbar technology that will make it operate similarly to a RAID 0/1 configuration on hard drives. It's pretty easy to do the math to see just what kind of powerhouse they've got brewing down there in Santa Clara!

On the reference card, the GPU is placed toward the back center of the PCB, and the data/address traces run out toward the top and bottom of the card like a massive silicon spider. This allows for incredible data rates since crosstalk issues are reduced. We're seeing this move to smaller data width paths in many new technologies, including serial ATA (SATA) and AMD's Hypertransport which only supports a maximum data path width of 32 bits. You can never be too rich or too thin!


The other major gain is made with the use of fractal texture compression. Fractal compression is known for its slow compression rates, but VERY rapid decompression rates. Since the card will only be decompressing the textures, it's a great technology fit.

However, the big win comes from the fact that fractal textures are resolution-independant. Say goodbye to MIPmaps! A single stored texture can be used at any distance without loss of detail or speed degredation. It also greatly reduces the need for anisotropic filtering while giving a HUGE speed increase to antialiasing. Talk about your win-win technologies!

Textures using fractal compression are also typically 87-98% smaller than their uncompressed counterparts, and 42-83% smaller than current existing texture compression schemes.

DirectX 9.1, scheduled to be released concurrently with the NV30 in early December, will have full support for fractal texture compression. Other card manufacturers will have to wait until their next iteration to add this amazing new feature.

It looks like the wait was worth it after all!

</inquirer mode off>

[Stavros]

The real potential of both R300 and NV30 is the high-dynamic-range high-precision rendering, and that won't be exploited by any real-world games before both R300 and NV30 have been superceded.

Thats the problem, potential buyers are only interested in performance they can use now (or what score they can get with 3D Mark 2002), with 8X AF and 4X FSAA turned on, and If the NV30 has no DDRII, or 256bit then its true bandwidth is going to be in the regions of a GF4Ti 4600, which we all know falls apart under these conditions (especially AF). If this happens then all the DX9 demos in the world, aren't going to save it. As far as I'm concerned, its got to perform as well, if not better than the R300 now with todays software, otherwise its going to go down faster the an Argentinian Football player in the other teams penalty box. What ever new tech nvidia have its going to have to be bloody good.

One thing though, it did come from the Inquirer (the Sunday Sport of hi-tech)

[Bigus Dickus]

Originally posted by nutball
OK, we can benchmark UT2003, Doom ]I[, whatever else, but no games (real, or imagined) will stress half of the transistors on either R300 or NV30.

UT2003 runs at 500fps at 1600x1200 16xFSAA, 128xAF. Bore. Snooze.
Um... really? Not.

The point I'm really orbiting around making is that both R300 and NV30 are not really about massive (irrelevant) improvements in the speed of current engines.

Perhaps that's not what the NV30 will be about, but it is what the R300 is about. The ability to finally run high AA and AF in high resolutions in the latest games is not an irrelevant improvement.

Games that take advantage of the DX9 attributes of either card are years away, and rather irrelevant themselves. It's the raw speed still that will decide for most gamers which is the better card for them.

[Lezmaka]

Originally posted by Stavros
If the NV30 has no DDRII, or 256bit then its true bandwidth is going to be in the regions of a GF4Ti 4600, which we all know falls apart under these conditions (especially AF).

AA is the one where bandwidth is needed most, not AF. But all the bandwidth in the world doesn't matter if you can't use it efficiently. The Parhelia had lots of bandwidth because of a 256bit bus, but didn't have anything similar to HyperZ and it doesn't perform very well. of course the low performance probably has other causes as well, but that's one of the reasons.

[StealthHawk]

Originally posted by tieros

Cards based on the NV30 will be configured with 2 separate blocks of 128 bit quad pumped RAM, clocked at 200 MHz (800 MHz effective), with a modification to nVidia's existing crossbar technology that will make it operate similarly to a RAID 0/1 configuration on hard drives.

quad pumped RAM? there hasn't been a single rumor about nvidia using any form of quad pumped RAM...ever. and the latest news in the grapevine is supposed to indicate that NV30 won't even use ddrII

[Nutty]

doesn't this at least mean, with a 128bit bus/memory, it should help keep the price down at least, right?

I would love it for nvidia's 128bit bus nv30 to thrash the 256bit bus R300. That would just be poetry..

Smart technology is always better than brute force.

[Fotis]

Originally posted by Nutty
doesn't this at least mean, with a 128bit bus/memory, it should help keep the price down at least, right?

I would love it for nvidia's 128bit bus nv30 to thrash the 256bit bus R300. That would just be poetry..

Smart technology is always better than brute force.
Unfortunatly nv30 will not be cheaper.According to creative's europe rep that was interviewed at www.guru3d.com nv30's production costs more than gf4ti4600's but it will launch at the same price.
Can someone remind me the launch price of gf ti4600?

[StealthHawk]

Originally posted by Fotis
Unfortunatly nv30 will not be cheaper.According to creative's europe rep that was interviewed at www.guru3d.com nv30's production costs more than gf4ti4600's but it will launch at the same price.
Can someone remind me the launch price of gf ti4600?

but how much of a profit margin was nvidia getting with the GF4Ti4600? depending on the performance delta between NV30 and R9700Pro, nvidia may really have no choice but to offer a lower price than $400.

everything lies on whether or not NV30 outperforms R300, and if it does, by what margin. with ATI set to release all these new cards before NV30 that take different parts of the market, nvidia may have to use some serious aggression.

[jbirney]

Originally posted by Nutty
doesn't this at least mean, with a 128bit bus/memory, it should help keep the price down at least, right?


Well if nV uses normal DDR then yes 128 mem is cheaper. However all info we had points to faster DDR2 wich is still very expensive :(


I would love it for nvidia's 128bit bus nv30 to thrash the 256bit bus R300. That would just be poetry..

Smart technology is always better than brute force.

You do realize that ATI has a number of "smart technology" helping to save bandwidth right?

[ReDeeMeR]

R300 price will fall after debut of nv30 and if what I see now that nv30 wont be significantly faster then R300 then nvidia will get her ass kicked, cause only idiot could buy 400$ card witch is slower then the 300$(current price) Radeon9700.

[PreservedSwine]

2 things the NV30 has going for it that ATI had to overcome.

Power and heat issues.


@ .13 microns, at least those (2) issues should be swept away, which seem to be the biggest hurdles for the R9700 to operate smoothly....

However, we just don't know what problems may be encountered in the move to .13.....vever having been done before,.

[Nutty]

You do realize that ATI has a number of "smart technology" helping to save bandwidth right?

sure I do, but past analysis, points to nvidia having the better technology.

And NV30 has been in production a helluva long time. When I was at the nvidia gathering event in london, months before the gf3 was released. They told us nv30 was already feature frozen back then. So like how many years is that?

I'm still expecting something very special from nvidia.

[Bigus Dickus]

Originally posted by Nutty
They told us nv30 was already feature frozen back then. So like how many years is that?

That was likely either an outright lie, or a vague statement meaning that their chalkboard that had the list of desired features was deemed "final" at the time. I'm sure lots of tweaking and adjusting has happened since then... it's even rumored that the NV30 once had a per-primitive processor that has been dropped from the design.

[StealthHawk]

Originally posted by jbirney
You do realize that ATI has a number of "smart technology" helping to save bandwidth right?

and if NV30 outperformed R300 using only a 128bit bus, then nvidia would have the "smarter technology" of the two companies, yes?

saying that doesn't mean that ATI doesn't have smart technology as well, it's just simply not as efficient