PDA

View Full Version : More gossip on NV40 and R420


MUYA
03-23-04, 08:05 AM
Here (http://translate.google.com/translate?sourceid=navclient-menuext&hl=en&u=http%3A%2F%2Fwww%2Eheise%2Ede%2Fnewsticker%2Fmel dung%2F45914) Nvidias NEXT generation chip NV40 sighted
On the fair could do c't a view of still under strict secrecy standing the NV40-Chip erhaschen, with which Nvidia wants to back-conquer the performance crown of ATI. The double power connections suggest increased requirements of electric current. With the Nvidia Referenzkarte was missing still the radiator, but the view of the memory chips was free, with which it acts around Samsungs slowest GDDR3-Variante for a MEMORY clock of 500 MHz. The sample is thus no Ultra execution, but the standard model of the NV40-Karte, with which the chip clock will be presumably clearly under 500 MHz.

Nevertheless diagram chip and memory do not come without a double current infusion over two external connections to the power pack estimate the capacity out c't on approximately 100 Watts with peak values of over 100 Watts. The diagram map reaches both values naturally only with applications with complex 3D-Grafik. With the Ultra version one can proceed with a storing act from 600 to 700 MHz and a chip clock from 500 MHz. The capacity of Nvidias new over flier lies thus again a piece more highly.

After the speculations, which are considered however meanwhile as relatively safe, circulating in the net, the NV40 manufactured at IBM contains 16 pipelines, which support the Shader specification 3,0. The 200 million transistors with a structure size of 0,13, necessary for it, µm obviously lead to a new record with the capacity.

With ATIs of TSMC manufactured competition chip R420 the rumor kitchen proceeds likewise from 0,13 µm-Strukturgroesse. Since the Canadians integrate probably only 12 pipelines, which besides only minimum beyond the Shader-2.0-Spezifikation to go are, are plausible in the net circulating the data of 160 million transistors. With same number of arithmetic and logic units in the pixel Shadern of the pipelines and same effectively usable memory range ATI could make up for the smaller number of pipelines by a chip clock higher around 33 per cent again. The capacity might move with ATI then in similar regions as with Nvidia. Which of the two chips brings most pixels per second on the screen, in addition, is still completely open with publication of the clock frequencies. Only one of the numerous unknown quantities is for example, whether Nvidias is sufficient efficient memory interface, in order all 16 pipelines with maximum speed run to let simultaneous to be able

MUYA
03-23-04, 08:06 AM
From their website
http://www.nvnews.net/vbulletin/attachment.php?attachmentid=5015&stc=1

Demirug
03-23-04, 08:37 AM
Manfred Bertuch as I know him. Do not take this too seriously.

MUYA
03-23-04, 08:38 AM
just gossip ;) Nothing is to be taken too seriously till the NDA expires.

mikechai
03-23-04, 10:03 AM
The picture is real.

AthlonXP1800
03-23-04, 10:35 AM
Hmmm the pic is much clearer than the other pic, this one took closer look at the card and I am surprised somebody took the pic blanked out the card sample identify so whoever took the pic wont be caught by broke the NDA.

jbirney
03-23-04, 10:51 AM
Only one of the numerous unknown quantities is for example, whether Nvidias is sufficient efficient memory interface, in order all 16 pipelines with maximum speed run to let simultaneous to be able.

Looks like I am not the only one wondering if they have enough BW to keep 16 full pipes happy...

Brick_Top
03-23-04, 10:56 AM
The picture is real.

But is that nv40 the one we will be seeing?

Geforce4ti4200
03-23-04, 11:14 AM
is it 16 real pipes or 8x2 no one still knows for sure :(

nutball
03-23-04, 11:16 AM
Dual DVI on the reference card. Nice. About time.

mao5
03-23-04, 11:25 AM
Very early sample,GPU was made in korean,core speed is 300MHz :nono:

MUYA
03-23-04, 11:42 AM
GPU was baked in a fab in korea? or the card was manufactured in korea?

mao5
03-23-04, 11:44 AM
GPU was baked in a fab in korea? or the card was manufactured in korea?

card made in california nvidia headquarter,gpu made in korea.

Demirug
03-23-04, 11:44 AM
Looks like I am not the only one wondering if they have enough BW to keep 16 full pipes happy...

It depends.

1. The difference between they core and memory clock.
2. The number of clockcyles a pixelquad need to caculate in the pixelprocessor
3. The rasteroperator setup.

You will find configurations that bandwith limited but some part of the chip ist allways the limit. Sometimes it is the bandwith, sometime the fillrate, ...

MUYA
03-23-04, 11:48 AM
card made in california nvidia headquarter,gpu made in korea.
hang on? They used a foundry in Korea? Like who? news to me, I thought either IBM or TSMC would be handling this aspect of the pre-production too?

MUYA
03-23-04, 11:55 AM
Ok i get it...the G PU was packaged in Korea

MUYA
03-23-04, 01:54 PM
From guru3d (http://www.guru3d.com/article/article/125/3/)
Behind closed doors NVIDIA under the luxury of a Non-Disclosure Agreement NVIDIA was showing the next generation product to a selection of press, partners and industry analysts. Guru3D.com was one of the lucky few who where able to see it all, I've personally seen it, I've played a game, saw new technology demo's on it and I can say it rocks for sure. I can't get into details as it's all NDA stuff but I'll give you a hint: take a good look at the rumors on the web. The downside from a rumor is that often peope are starting to imagine stuff. The last couple of days days people started to shout basically anything to get a bit more attention and spread well let's say it a bit more bluntly: BS, seek in the news from last week, that's all I'm gonna say about it ::end hint:: Ah okay, one more .. also seek up a rumor on encoding certain stuff.

What can you expect from NV40, oh it's fast alright ! When can you expect the NV40 introduction ? How does three weeks from now sound ? It's close, it's in mass-production and you can likely buy it in May already despite some reports I've seen on other sites.

MUYA
03-23-04, 01:56 PM
http://www.guru3d.com/article/article/125/2/ (http://)

ATI simply is not telling anything. It was to be honest a bit of a disappointment. No confirmation on the new name, codename, clockspeed, pipelines, memory, features, technology .. nothing. A bit of a shame, but that's how these things work.

So I'll share with you the speculation/gossip on this product; we expect: 160M transistors (9800 XT has 110), 500MHz core clock, 8 pipe design, 6 vertex engines, 256MB 256-bit GDDR3 at likely ~1 GHz. The R420/423 chips will offer twice the pixel fill rate and vertex throughput of the R350 core, as well as increases in memory bandwidth. R423 is to be expected the PCI-Express version.

MUYA
03-23-04, 01:59 PM
More spam from me
From Hardware Hell (http://www.hwhell.com/articles/cebit_day_two_coverage/2.html) We sat down with Adam Foat from NVIDIA, and he was nice enough to confirm some rough dates for the next-gen chips they're working on. The debut will be in about 2 weeks, with OEM's popping cards out in about 4 weeks. We should be getting more info soon on that. It will definitely be fun, as always, to watch what ATI does in response to this release.

jbirney
03-23-04, 02:09 PM
It depends.

1. The difference between they core and memory clock.
2. The number of clockcyles a pixelquad need to caculate in the pixelprocessor
3. The rasteroperator setup.

You will find configurations that bandwith limited but some part of the chip ist allways the limit. Sometimes it is the bandwith, sometime the fillrate, ...


Granted thats true but in the past we have seen Highend parts from NV starved for BW, NV30 and its 128 bit bus for example...

zakelwe
03-24-04, 03:41 PM
From their website
http://www.nvnews.net/vbulletin/attachment.php?attachmentid=5015&stc=1

Possibly why it is red, from Xbit

"The only thing I could get about the NV40 is that it was showcased on Friday, on Saturday and will be demonstrated on Tuesday behind closed doors under NDA and only for customers of a particular company, for instance, Microstar International."

MSI cards are normally red, like Gainwards. Maybe very few samples and this is an MSI sample made for nvidia ?

Flextronics made nv30 samples in green but before that MSI made nvidia samples.

Who made the nv40 samples for nVidia ?

Regards

Andy

MUYA
03-25-04, 12:40 PM
i have heard several AIB partners have nearly finished their own boards based on NV40.... just something i heard

GlowStick
03-25-04, 05:20 PM
It dose look like NV40 will be shipping in about a month, i cant wait!