nV40 Specs emerge @ 3Dcentre.de

It's in German but you should be able to understand the <A HREF="http://babelfish.altavista.com/babelfish/urltrurl?lp=de_en&url=http://www.3dcenter.de/artikel/2004/01-27_a.php" target="_new">trainslteatad vurshun</A>.

Basics:

8x2 Architecture
500-600mhz core
600mhz DDR2 (1200mhz effective)
Support for DDR/GDDR2/GDDR3
PS/VS 3.0
Improved FSAA methods
Better AF (up to 16X), looks like they are using rotational method.


<A HREF="http://babelfish.altavista.com/babelfish/urltrurl?lp=de_en&url=http://www.3dcenter.de/artikel/2004/01-27_a.php" target="_new">http://babelfish.altavista.com/babelfish/urltrurl?lp=de_en&url=http://www.3dcenter.de/artikel/2004/01-27_a.php</A>


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! - <font color=green>RED </font color=green> <font color=red> GREEN</font color=red> GA to SK :evil:
 

pauldh

Illustrious
Thx. Can't wait to see some benchmark results in the coming months.

By the way, please tell me you didn't read that whole article? That english was beyond my 1:00AM attention span. :wink:

ABIT IS7, P4 2.6C, 512MB Corsair TwinX PC3200LL, Radeon 9500 Pro, Santa Cruz, Antec 1000AMG, TruePower 430watt
 
Yeah I did, I tend to understand translationese, thankfully I know a bit of german and can bring up the original side by side and make out some gaps.

What impressed me the most was the mention of the NV41 close to the same time as the NV40 which will provide the PCI-EX-AGP bridge, the question for me is how that will work for them. Only 1 generation with the bridge though. Almost makes the benifits from a single core very short-lived.

BTW, it's only 11pm out here, so I guess I have that going for my reading skillz. :eek:

Love the chart on the 2nd page, I just wish it was it's own page so I could link directly to it for future questions, but this whole article should be enough.

No on to the R420/423 specs.

And yes more Benchies would be nice.


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! - <font color=green>RED </font color=green> <font color=red> GREEN</font color=red> GA to SK :evil:
 

Vimp

Distinguished
Jul 13, 2003
358
0
18,780
LOL, thats a hoot. Dunno why they even bothered trying to make it english since its almost completly unreadable in english. I had read THEINQUIRER last night already so I had already read about the specs but I didn't go to their source so I never saw the funny english. Thats funny stuff.

<font color=blue>_____________________________________</font color=blue>
Vancouver, B.C., Canada
Asus A7N8X-X, Athelon XP 2500+ Barton,
Samsung 512mb DDR400, Creative Geforce2 GTS,
SoundBlaster Live! Value.
 

ChipDeath

Splendid
May 16, 2002
4,307
0
22,790
Dunno why they even bothered trying to make it english
They didn't. It's been babelfished - babelfish will translate any foreign webpage between a variety of languages. It does produce some gibberish a lot of the time, but you can usually get the gist of it... Such are the joys of totally automatic translation software.

---
<font color=red>Those of you who think you know everything are annoying to those of us who do.</font color=red> :wink:
 

Vapor

Distinguished
Jun 11, 2001
2,206
0
19,780
Sticking to the FX name--must be crap :wink: .

It honestly doesn't look THAT great. Especially since ATi has claimed 2x the performance of the 9700Pro in the R420 and 2x 9800Pro in the R423 (don't know how reliable that info is though). I'd expect 1.5x the performance of the 5900U with this card.

Oh well, I still look forward to it...and it's AGP 8x! I can use it w/o getting a new board if it does turn out to be great. It better be great too, nVidia really needs come out with a bang (á la ATi's R300) this time around.

Oh yeah, the translation wasn't bad at all...or maybe I'm just fluent is translationese. Oh yeah#2, 1600MHz effective DDR2/GDDR2/GDDR3 sounds very promising; if nothing else comes out of this card, nVidia will still be opening bulk cases of whoopass in terms of bandwidth.

Maxtor disgraces the six letters that make Matrox.
 
Yeah like ChipDeath said, I 'translated' it through babelfish so that the majority of people would be able to get the gist of it. It's not a perfect solution by any means but it does make it better for the average person than the German original. And since it's mostly about the specs than really a deeper meaning I don't think it impacts it too much.


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! - <font color=green>RED </font color=green> <font color=red> GREEN</font color=red> GA to SK :evil:
 
YEah I was most interested in the nV41, and the NV45, but for the short term the NV40 should offer some interesting improvemnts. I never trust predicted performance, but I was VERY dissapointed by the RV380's subpar peformance considering it was suppose to reach R9700/9800 levels, and here the first review shows it below the R9600XT (which itself isn't impressive), so I really am taking any predictions about the R42X with a grain of salt. Hopefully their supposed decision to rework it to make it competitive with the NV40 has made it a significantly better product than the R400 was going to be. If the RV380 figures are not driver inhibited/crippled, and the R423 comes out for under $800, then that will likely be my solution of choice (unless info comes out soon enough afterwards about the R450 [using 0.11micron process?] to make me wait, AGAIN!)

The Bandwidth should be impressive, but ATI is looking in that direction too, just not for the initial release it appears. But both cards having 50+GB/s is extremely tasty.

It appears to be <A HREF="http://www.xbitlabs.com/news/video/display/20030918134855.html" target="_new">Samsung for nV</A> and <A HREF="http://www.xbitlabs.com/news/video/display/20030925033540.html" target="_new">MICRON</A> (like the name of a Forum Member I know :cool: ) for ATI.

So I think BOTH sides will have their Bandwidth monsters. The questions is will one or the other NEED that extra bandwidth to stay competitive? Rememeber memory is very expensive, especially the faster more rarified stuff.

We shall see as always.

My question is what with the actual NAME of the new Radeon Line be? The R10,000? The R X-series? The number 10 something would imply DX10 capability. I would prefer they move to a NAME naming system like the ATI Radeon Mad Mutha XT, although the R9999FRO or the R10000MOFO have a nice ring to them. :evil:

Although the FX6666 sounds about right, especially coming from SATAN CLARA. :evil:


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! - <font color=green>RED </font color=green> <font color=red> GREEN</font color=red> GA to SK :evil:
 
Yep, but rememeber even that is just rumour.

BTW, the Q1 and Q2 are often Financial Q1 and Q2. Which depending on their timetable could be April May (if Q one starts April 1 (which is the case for many companies). I think we are currently in ATI's Q2 IIRC.

But current belief is R420 to arrive in March-April, and NV40 April-May. Considering that the R380 has been reviewed and benched publically, perhaps it's going to be the first out of the gates, even before the AGP based top tier cards. WHY/HOW I don't know considering the lack of enough equiped MOBOs.


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! - <font color=green>RED </font color=green> <font color=red> GREEN</font color=red> GA to SK :evil:
 

cleeve

Illustrious
Looks damned impressive to me.

What I'm really interested in seeing is how different the Ati and Nvidia architectures are. It seems each generation they get a little more different from each other.

The next generation of graphics card wars will be very interesting, methinks.

________________
<b>Radeon <font color=red>9500 PRO</b></font color=red> <i>(hardmodded 9500, o/c 322/322)</i>
<b>AthlonXP <font color=red>2600+</b></font color=red> <i>(o/c 2400+ w/143Mhz fsb)</i>
<b>3dMark03: <font color=red>4,055</b></font color=red>
 

Vapor

Distinguished
Jun 11, 2001
2,206
0
19,780
The preliminary performance of the RV380 has been crappy at best. I'm VERY disappointed. Pitou should compete with it :eek: (well, maybe not--we'll only have to wait to see if Matrox has any gas for gaming). Somehow I've started to not believe those Scotty and RV380 previews...at all.

Logically, it makes little to no sense for ATi or Intel to release products that are considerably worse than their respective predecessors. Also, I've never heard of those sites (not that I know all of them, but I recognize the more reputable ones), and what the hell happened to NDA?? I'm really not going to consider those scores--I have little to no reason to trust them.

50+GB/s is just amazing...I want one (screw Matrox!!--j/k!)! Point of consideration: take the well-known 9800Pro, give it 45GB/s bandwidth, changing NOTHING else--shouldn't it be at least 75% better?? From all the overclocks of the 9800Pros I've seen, performance scales almost exactly to the memory speed. Grant it 45GB/s of bandwidth would render the GPU (or VPU if you want to play by ATi's nomenclature rules) the bottleneck. So a simple upping in the core speed would solve that, no? To get those higher MHz, 130nm low-k should be ample (R42X's design, IIRC)...leading to the double performance, doing nothing more than simply upping the clock. Of course, other advancements should and likely will be present, only further adding to the performance (oh man, improvements to FSAA [or implementation of FAA in addition] and on AF would be great). Just speculating...

IMHO, ATi has the power to put nVidia out in this next-generation--they just need a card like one described above (or better) for their high-end as well as a solid mid-range card (RV380 seems to be lacking--oh wait, I'm ignoring those results for the time being...RV380 should be good) and for NV40 to be only a large evolution of the NV38 (as opposed to a revolution). nVidia really needs this next-gen to be good, "we shall see as always."

Maxtor disgraces the six letters that make Matrox.
 
Exactly. I can't wait to see the 12x1 or 16x1 (whatever the R420 actually ends up being) versus the 8X2. That will likely offer alot of performance differences. And the new FP32 ATI chip that's a first for them, did they do it well?

There should be alot of nice quirks I would think. And this new method of AA is it Fragment? Oohh now that would be cool, not much detail on it though. 16X AF is good, but the questions is what are these AF efficiencies? Brillinear or true Trillinear. Rotational like ATI?

I can't wait for the REAL thing from both and yeah, and that first round of benchmarking will likely show both cards in very different lights, and they will likely struggle somehwere at first until they get the drivers to fully work with the new hardware. Just hope that doesn't involve any cheats from either side.


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! - <font color=green>RED </font color=green> <font color=red> GREEN</font color=red> GA to SK :evil:
 
Unfortunately it's a little more complex than that (well actually fortunately too, it should give us fun reading for when we look back at this and think how off the mark we were on this or that item/idea). Moving to the low-k process will help, but with that many more transistors, heat will be more of any issue in general, but it should do better than the R9800 series, even if it doesn't go as high as the R9600Xts.

The increase in memory for both the R9600 and R9800 series does offer much greater gains, but I don't know if the max is reached earlier than expected.

And I agree, ATI does how within their grasp the opportunity to put nV all but out of the graphics card game with a winning part. But the money is still made in the middle class, so that will likely decide their fates, and that RV380 does suck rocks, so not even the most potent high-end knockout blow will have much effect as long as the low end is left open. If it is just lack of good drivers for it, then that would be nothing to worry about, I agree, best to take them with a grain of salt for now.
If ATI can exploit the low-k, while nV can't (remember they
said it was dangerous), then that would give them a slight edge in core speed potential.

Now here's the kicker that I eluded to earlier and Cleeve reiterated, the architecture may pose certain advantages to one or the other. The 8x2 design may allow nVidia to continue running fast partial precision with this next generation, but I'm not sure if that will be possible with the ATI 12/16x1 design. That could cause similar performance gaps to the Early R300/NV30 perfromances this time with nV benifiting from their driver optimization and 2 pipe architecture. It could offer some interesting performance issues. Rememebr there won't be a clock or memeory representative increase because ATI is also moving to 32bit precision from 24, which will likely cause some bottlenecks (hey perhaps that explains the RV380s dismal performance in those test.

This could cause problems for both. Should the RV380 suck rocks, andthe R420s single pipe design not meet expectations this could also greatly hurt ATI. So who knows what we shall see, I can't wait though.

There are som many subtle differences to make it very interesting and enlightening. We will likely learn alot more about design and performance in this round that we did even last round.


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! - <font color=green>RED </font color=green> <font color=red> GREEN</font color=red> GA to SK :evil:
<P ID="edit"><FONT SIZE=-1><EM>Edited by TheGreatGrapeApe on 01/29/04 01:08 PM.</EM></FONT></P>
 

JP5

Distinguished
Jun 26, 2003
534
0
18,980
It will be funny if ATI get the jump and releases their cards right at or before the Game Developers Forum. I have not owned an ATI crad belore, but I may be ready to make the leap.
 

Vapor

Distinguished
Jun 11, 2001
2,206
0
19,780
I hate to say this, but based off what you said...it looks like the current gen of cards may be the last generation that we can use the top PCI slot (minus nVidia, they already used the space) in a good gaming system. Glad I don't use it now and have little intention to use it in the future!

Let me qualify what I'm saying: I'm generally talking about the upper-most highend, unless otherwise specified. Additionally, if I allude to a potential (such as my last post where I talked about low-k potentially helping a lot), I generally mean in a way that the majority of people here would use it--OCed to whatever clockspeed they felt content with or could attain--not the potential performance straight out of the box. Alright, now that that has been said and done...let me continue.

Anyway, IMO, heat can be dealtwith--at a cost. That cost would be a literal cost: money. Out of the box, I don't think the R420 will hit the same clock speeds as the 9600XT at first (especially not the new AIW version!), I'd bet the Pro would be 480MHz, np at 440MHz, XT at 525MHz, but can OC like a biznatch with both stock cooling and then even moreso with better cooling (say peltier/watercooling!).

I'd honestly like to see a modded 9800Pro 256 (I'm not sure if what I'm about to say is even possible, well, maybe if manufactured this way) that has 1.2nsish DDR2 (that would get 1600MHz effective if I did the math correctly) so that we could find at which point the 9800's core simply doesn't need any extra bandwidth.

As for ATi putting out nVidia, I did mean in the upper-echelon of performance, which wasn't clear at all. In the midrange it looks like the RV380 (I personally take those benchies with a whole f-ing saltshaker full of grains of salt--maybe I'm in denial or some sh!t like that) is really going to suck. nVidia will almost always seem to have a grasp of where the money is made (low-middle end for Dell and the like).

May I ask why either of the companies intends to run partial precision? AFAIK, there is no need for 32-bit precision right now. I'd personally rather have them both run at DX9(.1) standard (32-bit spec isn't until DX10, right?--if it's in DX9.1, ignore me) and give performance NOW rather than later. Frankly, that is what pissed me off most about the FX series thusfar--nVidia planned for the future and it hasn't come yet.

Also, maybe you, Cleeve, or someone else could clear this up for me: how does a 8x2 pipeline design differ performance wise from a 16x1 pipeline? I'm clueless...afterall, I'm still pretty much a n00b when it comes to how graphics cards work--just a year ago I thought the only basis of performance was bandwidth, vertices/sec, and fillrate.

Anyway, I wish I could look into the future and just see what the designs and performance to put an end to all my speculation...I'll probably look back at this post in a few months and just say: "wow, how could I have been so wrong?"

Maxtor disgraces the six letters that make Matrox.
 

Vimp

Distinguished
Jul 13, 2003
358
0
18,780
I woulda thought a 16x1 design would be better then an 8x2 design because a 16x1 design allows twice the pixel fill rates then an 8x2 design at the same core speeds. However seeing as both equal 16 the texel fill rates would be identical using the same core speeds.

<font color=blue>_____________________________________</font color=blue>
Vancouver, B.C., Canada
Asus A7N8X-X, Athelon XP 2500+ Barton,
Samsung 512mb DDR400, Creative Geforce2 GTS,
SoundBlaster Live! Value.
 
Well if you see the recent talk about DX9.1 etc. IT's not really DX9.1 etc. It's a question of PS3.0 and VS3.0 it's commonly being called(misnamed) DX9.1, but really it's a subset issue. PS3.0 was included in the DX9 spec. So it causes issues. PS/VS 4.0 will arive in DX10 (DXnext as some call it [ sounds like XboxII ])

However PS/VS3.0 does call for 32 bit precision throughout IIRC. Remember that the R3XX series is 32 bit in all but Pixel Shading.

A nice general overview by B3D (old but good);
<A HREF="http://www.beyond3d.com/articles/nv30r300/index.php?p=12" target="_new">http://www.beyond3d.com/articles/nv30r300/index.php?p=12</A>

In any case the R4XX is heading in that direction.

I think if it's done well and only an 'option' that can be enable and disabled, then partial precision performance improveents would be a good thing. Surr you may on get 99.44% of the IQ, but at twice the frame rate, why not. The nV run-time compiler shows that it can be done effectively, but newer games show that it needs to be prepared because it doesn't work on something it wasn't preped for. So having control would be nice if it decides in later releases to apply partial precision to a 2003 game when really it is only necessary for a mid-late 2004 game. The option to switch back and forth would be a must IMO.

As for OCing the R420 vs the R9600XT, we will have to see. Rememeber that the R9600XT can often reach above 600mhz on that tiny cooler that comes on most of them. Even a volt moded air cooled R9600PRO can breach the 600mhz level. But increasing the transistor count will leaad to much more heat. But I don't see why 500mhz wouldn't be an easy stock target, just that it would be close to the higher end under 'normal' HSF solutions if you asked me.

As for Xx2 and Xx1 comparison with the NV3X, supposedly it can calculate two simultaneous FP16 calculations and just one FP32. But the R3XX has to calculate one at a time regardless of precision. This may or may not be the case for the next gen R4XX, but likely is. This is supposedly a great help to partial precision. Lemme come backto this tomorrow and I'll do a better job when I'm a little more lucid in my own head. Good Long Night. :wink:

I think this should offer alot of interesting comparisons because BOTH will be able to go with FP32 and thus it will be a truer comparison of the benifits of both.

And yes I can't wait. I just wish they would paper launch information about the technology even if they don't do it qbout the specs. Yeah, I can't believe I'm asking for a paper launch either, but it really should offer some interesting insights on ATI's approach which will be relatively new.


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! - <font color=green>RED </font color=green> <font color=red> GREEN</font color=red> GA to SK :evil: