Are ATi worried abou the NV3x series? [Archive]

View Full Version : Are ATi worried abou the NV3x series?

Pages : [1] 2

SurfMonkey

10-18-02, 09:33 PM

ATI fights Nvidia's NV30 with new R350 chip
Charles Chou, Taipei; Christy Lee, DigiTimes.com [Monday 30 September 2002]

ATI Technologies recently revealed that it will launch a new R350 chip and later a more simplified RV350 to brace for the competition from Nvidia¡¦s NV30 in the fourth quarter.

Following their usual production policies, Nvidia will adopt the latest 0.13-micron process for its NV30 chip line, while ATI will continue using the more mature 0.15-micron process to ensure yield rates.

Besides the manufacturing process, ATI did not give any more details about product specifications for the new R350 and RV350 chips.

Though the unit cost for 0.15-micron processing is generally higher than for 0.13-micron processing, ATI noted that the better yield rate that the 0.15-micron process delivers will be able to cover the overall cost gap. Still, it is believed that the competitiveness offered by the different processes will become clearer when Nvidia¡¦s 0.13-micron NV30 chips hit the market in volume after November.

In addition to the R350 and RV350 chips, ATI disclosed that it may also introduce a RV280 chip, based on the Radeon 9000 core, by the end of this year to support the rising demand for AGP 8x technology. However, with the roll-out, ATI will no doubt experience a fast-expanding product portfolio, as the company has scheduled to launch the four-pipeline Radeon 9500 series in October as well.

To create more specific marketing segments, Wang Pao-chu, ATI¡¦s CEO for Greater China, said that the company will adjust product lines depending on future market response.

Methinks ATi may be running a little scared of what they believe the NV3x is capable of. Or maybe they have run into problems of their own with the .13u process and now realise that they may have to play catch up after all?

Megatron

10-18-02, 09:38 PM

I dont think they are "worried".

At the same time I dont think they are stupid. They know Nvidia has something coming, and they want to be ready. Its one thing to take the crown...its another entirely to keep it.

I think they just want to do the latter.

LORD-eX-Bu

10-18-02, 11:36 PM

Yeah, they are not scared, but like any smart company they are cautious and curious. They have working DDR-II cards already and nVIDIA doesn't, now when one company has surperior hardware and technology than the other, like ATI over nVIDIA, sure, they are always worried, I mean, the only way to go once you are on the top is down. And you can only hope to remain on top as long as you can struggle and compete. nVIDIA on the other hand, doesn't seem to want to disclose any information on their new part, why? maybe to build suspense, or maybe they figured that they could not beat ATI and had to go back and redesign the chip to better compete with ATI's hardware. nVIDIA are working on a untested process and are having problems on their own, ATI is sticking to proven technology while they work out the problems on their own process. Its smart competition, thats all.

Bigus Dickus

10-19-02, 02:03 AM

Originally posted by SurfMonkey
[b]Methinks ATi may be running a little scared of what they believe the NV3x is capable of. Or maybe they have run into problems of their own with the .13u process and now realise that they may have to play catch up after all?

Methinks they probably don't have much better of an idea of how fast the NV30 will be than the rest of us. They might not even have a source for "official" specs, and there probably isn't even an NV30 board running yet for performance numbers to leak out to them.

Instead, I think it's likely that they are simply excuting their scheduled product releases as planned, which has probably always included a .15u R350 with DDR-II memory. Instead of being scared, I think they're probably a bit relieved that the R300 didn't have to go head to head with the NV30 right off the bat, and that their roadmap includes something which might match up better with it. ;)

Nv40

10-19-02, 04:43 AM

Uttar

10-19-02, 06:09 AM

Originally posted by |TX|-LORD-EX-BU
They have working DDR-II cards already and nVIDIA doesn't

And you've got any proof of what you're advancing? ...

I'm sorry but i see no reason that nVidia doesn't have working DDR2 cards. And there's a HUGE difference between having a few cards working with it and actually being able to volume produce it, you know.

Also, keep in mind nVidia barely says anything about the NV30. Now that ATI got the crown, they want to make people think NV30 will be slow so that their cards sell fast.

And one thing is certain: they're doing their best for making DDR2 work on their card if it doesn't already. See: http://www.extremetech.com/print_article/0,3998,a=29479,00.asp , New nVidia Iron Coming

Sure, it means they're delayed according to what's said there ( october in volume ) , but really, it does proof that at the time they wanted it to launch in october, they wanted DDR2.

Uttar

ReDeeMeR

10-19-02, 09:12 AM

Nice, if there were no r300 I think Nv30 would of been 128 bit with simple DDR mem for same 400bucks, talking about sweet competition heh

gstanford

10-19-02, 10:38 AM

me thinks that if Nv30 comes with 8 pipesx2tmu's , 256bit bus and
with 1Ghz ddr2 memory at 400mhz core
(like reactor critical and the inquirer claim..)
the Ati Radeon9700pro or even the R350 at .15 will have
some problems to stay competitive in the top of the line
performance
That's almost correct. But the memory will be a 128 bit implementation, not 256 bit.

These figures line up very nicely with historical rumors on the NV30.

Another rumored number is the memory bandwidth - 48 gig/s. FWIW the figure of 48.0 gig/s has been mentioned many times in lots of different rumors covering a very large time span. To get memory bandwith simply multiply memory speed by core speed.

Here is a possible table of outcomes:
mem 1000, core 480 = 48.0 gig/s
mem 1000, core 400 = 40.0 gig/s
mem 1000, core 350 = 35.0 gig/s
mem 900, core 400 = 36.0 gig/s
mem 900, core 350 = 31.5 gig/s
mem 700, core 400 = 28.0 gig/s
mem 700, core 350 = 24.5 gig/s

for reference here are the bandwidth figures for some of today's cards:
Radeon 8500 = 8.8 gig/s
GeForce4 Ti4600 = 10.4 gig/s
Radeon 9700 Pro = 19.8 gig/s
Matrox Parhelia = 20.0 gig/s

While on the subject of old rumors, I reckon NV30 will be known as Eclipse. This name has been around since discussion of NV30 began (back when Geforce3 and the XBOX graphics chip were brand new). I wonder if nVidia has chosen their launch venue and date so that an eclipse actually occurs during launch? (eclipses are actually quite common events)

Greg

Uttar

10-19-02, 01:18 PM

Originally posted by gstanford
That's almost correct. But the memory will be a 128 bit implementation, not 256 bit.

These figures line up very nicely with historical rumors on the NV30.

Another rumored number is the memory bandwidth - 48 gig/s. FWIW the figure of 48.0 gig/s has been mentioned many times in lots of different rumors covering a very large time span. To get memory bandwith simply multiply memory speed by core speed.

Here is a possible table of outcomes:
mem 1000, core 480 = 48.0 gig/s
mem 1000, core 400 = 40.0 gig/s
mem 1000, core 350 = 35.0 gig/s
mem 900, core 400 = 36.0 gig/s
mem 900, core 350 = 31.5 gig/s
mem 700, core 400 = 28.0 gig/s
mem 700, core 350 = 24.5 gig/s

for reference here are the bandwidth figures for some of today's cards:
Radeon 8500 = 8.8 gig/s
GeForce4 Ti4600 = 10.4 gig/s
Radeon 9700 Pro = 19.8 gig/s
Matrox Parhelia = 20.0 gig/s

While on the subject of old rumors, I reckon NV30 will be known as Eclipse. This name has been around since discussion of NV30 began (back when Geforce3 and the XBOX graphics chip were brand new). I wonder if nVidia has chosen their launch venue and date so that an eclipse actually occurs during launch? (eclipses are actually quite common events)

Greg

First of all, where the heck did you get that core clock x memory clock = memory bandwidth idea?

Memory bandwidth *isn't* dependant of core clock! It's depandant of memory clock and memory bus width!

Seen the 48 gig figure from the original newsposts saying that. That wasn't raw bandwidth, but effective bandwidth using several Z, Color & AA optimizations.
The rumor actually was that even with 128 bit and DDR1, they figured great optimizations so that at a high AA level, it's just like if it was 48 gig bandwidth.

Anyway, Eclipse is sadly impossible :( ATI seem to have taken the name recently.

This might actually mean nVidia was too slow to trademark the name. Let's hope they're faster to issue patents! :)

Uttar

-=DVS=-

10-19-02, 01:37 PM

Originally posted by gstanford
That's almost correct. But the memory will be a 128 bit implementation, not 256 bit.

These figures line up very nicely with historical rumors on the NV30.

Another rumored number is the memory bandwidth - 48 gig/s. FWIW the figure of 48.0 gig/s has been mentioned many times in lots of different rumors covering a very large time span. To get memory bandwith simply multiply memory speed by core speed.

Here is a possible table of outcomes:
mem 1000, core 480 = 48.0 gig/s
mem 1000, core 400 = 40.0 gig/s
mem 1000, core 350 = 35.0 gig/s
mem 900, core 400 = 36.0 gig/s
mem 900, core 350 = 31.5 gig/s
mem 700, core 400 = 28.0 gig/s
mem 700, core 350 = 24.5 gig/s

for reference here are the bandwidth figures for some of today's cards:
Radeon 8500 = 8.8 gig/s
GeForce4 Ti4600 = 10.4 gig/s
Radeon 9700 Pro = 19.8 gig/s
Matrox Parhelia = 20.0 gig/s

While on the subject of old rumors, I reckon NV30 will be known as Eclipse. This name has been around since discussion of NV30 began (back when Geforce3 and the XBOX graphics chip were brand new). I wonder if nVidia has chosen their launch venue and date so that an eclipse actually occurs during launch? (eclipses are actually quite common events)

Greg

lol by your calculations Matrox should be faster then Radeon 9700 Pro :D :D :D :D :D

We can safly say bandwidth doesn't always mean's more speed :p

StealthHawk

10-19-02, 09:50 PM

Originally posted by ReDeeMeR
Nice, if there were no r300 I think Nv30 would of been 128 bit with simple DDR mem for same 400bucks, talking about sweet competition heh

it's not an easy fix to redesign the memory controller. moving from 128 to 256 would also mean optimizing for 256 instead of 128. i'm sure NV30 will ship with whatever bus it was intended to have. either that, or we may see a lower performing NV30 that should hypothetically perform much better. or perhaps a very buggy NV30.

MuFu

10-19-02, 09:55 PM

Bizarre calculations, lol.

Originally posted by Nv40
the radeon9700pro already gets really hot at 325/600mhz
and goes close to AGP3.0 power limits.

Close?! It breaches them... hence the need for the external power feed.

Originally posted by Uttar
And you've got any proof of what you're advancing? ...

I'm sorry but i see no reason that nVidia doesn't have working DDR2 cards.

Maybe now that they actually have silicon, a few ASICs have been mounted and they have begun debugging/verifying, but I'd be reluctant to say that they have fully-functioning boards equipped with DDR-II for sure (or any kind of memory for that matter). If they have, then it will have been for a matter of days only.

And yes, they have been delayed - originally, R&D was due to be completed by September with intent to ramp for an October launch. Puts them a couple of months behind schedule at this stage; JC was right when he said idly (I think it was around the time of E3) that they were about half a product cycle behind ATi, assuming he meant 3 months.

MuFu.

Mono

10-21-02, 05:57 PM

NV40: Have you even looked at some of the speeds people are overclocking thier 9700's with the stock cooler? Sure it gets warm but it's not like it dosn't have room to grow in that respect.

"the radeon9700pro already gets really hot at 325/600mhz"

And it's 620 on the memory dude, don't understate it :p

That said, I hope the NV30/R350 are both miracle cards. The better a product is the more we benefit and I could really care less who makes it :) Though it's going to be awhile before I upgrade again (NV40ish time prolly)

StealthHawk

10-21-02, 08:30 PM

Originally posted by Mono
"the radeon9700pro already gets really hot at 325/600mhz"

And it's 620 on the memory dude, don't understate it :p

a lot of people seem to think that the R9700Pro is more core limited than memory limited though.

Mono

10-21-02, 09:09 PM

stealth: why would they think that? I mean, I know it gets warm, but it's pretty common to see people hitting 375+ with the stock cooler. My core will do 380 stable, but I run it @ 365 just to be safe.

thcdru2k

10-21-02, 09:13 PM

i don't think they are worried at all, they're really just speculative and wonder what nvidia has up their sleeve. i think nvidia is the one worried..

StealthHawk

10-22-02, 05:49 AM

Originally posted by Mono
stealth: why would they think that? I mean, I know it gets warm, but it's pretty common to see people hitting 375+ with the stock cooler. My core will do 380 stable, but I run it @ 365 just to be safe.

huh? what i said has nothing to do with overclockability.

it has to do with the performance on the card. ie, overclocking just the core will give more performance(exept for maybe in FSAA) than overclocking just the memory.

jbirney

10-22-02, 10:05 AM

I cant resist:

I'm sorry but i see no reason that nVidia doesn't have working DDR2 cards.

Simple. ATI has publicly showed an ATI R9700 running on DDR2. While nV has not shown the nV30 at all :) :) :)

And there's a HUGE difference between having a few cards working with it and actually being able to volume produce it, you know.

And there is a SUPER DUPER HUGE difference between having a few cards that you can show off vers not having the part back from fab yet :) :) :)

Just kidding so dont take it the wrong way.

DDR2 was on ATIs road map for a long time. The fact that the R9700 memory controllers had support for DDR2 shows that this was planned from the begining.

ReDeeMeR

10-22-02, 02:14 PM

Um Ati already has a 9x part out and selling good and showing off DDRII and Nvidia is *just about* to launch their's witch only god knows the preformance... So logicaly whos worried?

Juntari

10-22-02, 03:52 PM

I do not know if 256 bit memory controller will be available with the first variation of NV30, but rest assured that NVidia will release many derivation of NV30 like they did with previous versions of the cards. That's my opinion on the matter. ATi is not worried since they have the product that is sold right now, and has the capacity to design and improve upon what they have. It is not as if they have no chance of surpassing NV30 with their upcoming products.

StealthHawk

10-22-02, 06:20 PM

Originally posted by ReDeeMeR
Um Ati already has a 9x part out and selling good and showing off DDRII and Nvidia is *just about* to launch their's witch only god knows the preformance... So logicaly whos worried?

except that everyone seems to think NV30 will outperform R9700. if you believe the Inquirer, even ATI is worried about that, and hence why they are changing production schedules and releasing an R350 soon after NV30.

LORD-eX-Bu

10-22-02, 06:37 PM

PreservedSwine

10-22-02, 07:46 PM

Bigus Dickus

10-22-02, 10:54 PM

Originally posted by StealthHawk
except that everyone seems to think NV30 will outperform R9700. if you believe the Inquirer, even ATI is worried about that, and hence why they are changing production schedules and releasing an R350 soon after NV30.

I think even ATi assumes the NV30 will be faster simply because later technology tends to be faster. Of course, with the NV30 being more delayed than being a later designed part, it may be NVIDIA who is worrying that the jump in speed may not be what some people are expecting.

StealthHawk

10-23-02, 12:34 AM

Originally posted by PreservedSwine
Is that being scared, or being smart? (rhetorical question)

the answer depends. is moving from a 1 year cycle to a 6 month cycle smarter at this point?