NV18 and NV28 names revealed

[Pav]

I noticed @ Warp2Search that they discovered that recent 31.00 drivers appear to show what NV18/28 cards will be named and just by looking @ the names - you can't help to conclude that NV18/NV28 will be nothing more then a Geforce 4 with a agp 8x support. Now I wonder what will nVidia come up with to sell these - let me think - "Geforce4 is now twice as fast!!!" - LOL

More info @ Warp2Search (http://www.warp2search.net/article.php?sid=6443&mode=thread&order=0&thold=0)

Also I would like to make this point - It is indeed very interesting that no information has leaked about NV28 or NV18 other then the fact that it will have AGP 8x support and some number of extra transistors over NV25. At the same time so much hype and marketing BS has been released about NV30. Don't you find it a bit weird that noone is talking about NV18/NV28? No single fan nvidia website, no major hardware site - total silence. Weird huh? NV30 is still so far away yet we already heard a lot about it (nothing concrete though I guess). Makes you wonder, doesn't it............:rolleyes:

My theory - The latest news about the naming of NV18/NV28 in a way tells us that perhaps there's nothing to talk about? I mean if its only Geforce4 + AGP 8x - who cares? lol - so in a way maybe that explains lack on Marketing PR and Hype about nv18/nv28 and lack of info leaks in that area. Its NV30 that nVidia wants to hype up because with NV18/28 it might be hard to come up with something :D

[Uttar]

The NDA for the NV30 is partly lifted: the NDA for the NV28/NV18 is not lifted at all.

That has been proved by Jen Hsun Huang at the investor meeting. He could talk a little about the NV30, but not a single word about the NV28 was said ( yes, an analyst did try to get some info but failed )


nVidia strategy is simple: Hype their high-end products so that people not having the money for them say "Hey, even if it costs less, since it's made by the company making the NV30, it gotta be good too!"

*cough*Geforce4 MX*cough*

However, i *do* think the NV28/NV18 will have new features.

Remember, the Geforce4 MX, even if ti wasn't a DX8 card, it still had nView and Accuview AA. I think a similar thing is going to happen ( maybe nV is going 3DFX way for AA? Who knows... )


According to official nVidia information ( and that's the only official info on this ) , the NV28 will have more than 80 million transistors. That's as much as Matrox card! Good thing TSMC 0.15 is very mature by now...


Uttar

[Uttar]

Just looked at my 30.82 and 31.00 drivers inf. In the 30.82, i find no NV28 info.

In the 31.00, i find the NV28 and it's simply called "NV28".

Would that link at Warp2Search be saying incorrect information, eh? :)

And yes, i downloaded those drivers from Guru3D, which also seems to be that article source.

Uttar

[Pav]

the .inf file says nv18/28 because its been modded by guru3d - its the driver files that say geforce4 xxxxx with agp 8x - simple hex editor can open them and some of them say this instead of nv28/18 like in 30.82 drivers and before.

As to your thoughts about new features - think about this - if nv28 would have any nv30 features or any new features why would nvidia name it Geforce 4 Ti 4200/4600 with agp 8x? Why not call it geforce5 or even geforce 4 ti 4800?

[Uttar]

Originally posted by Pav
the .inf file says nv18/28 because its been modded by guru3d - its the driver files that say geforce4 xxxxx with agp 8x - simple hex editor can open them and some of them say this instead of nv28/18 like in 30.82 drivers and before.

As to your thoughts about new features - think about this - if nv28 would have any nv30 features or any new features why would nvidia name it Geforce 4 Ti 4200/4600 with agp 8x? Why not call it geforce5 or even geforce 4 ti 4800?

Well, to be sincere, i doubt it's REALLY going to be named that way.

I'm pretty much betting they're calling it that way because Marketing didn't choose a name. However, it is an indication nV engineers think it isn't so new.

So, yes, it won't likely be much different from a NV25. But they wouldn't add 20 million transistors that do no effect...


Uttar

[Bigus Dickus]

I still don't think the NV28 will have 86 million transistors. 66 is more likely.

[DIMA]

Originally posted by Bigus Dickus
I still don't think the NV28 will have 86 million transistors. 66 is more likely.

For crises sake, it has been CONFIRMED and broadcasted on T.V and people saw it and the numbers were clearly visible there and the CEO repeated them himself during the press conference!

[TheOneKEA]

Doesn't mean much. PR has been wrong before. IOf you can find a link to Nvidia's website that has 86 million or 66 million, I'll believe it.

[DadGT]

So what will NV28 be? All of the sudden I am very interested.

<wild speculation on>
What if NV took a GF4 core, added a 256 bit memory bus for outrageous bandwidth, and made some other tweaks to the core...Lets see, the R300's cream the Ti4600 is all current games which are largely bandwidth dependent...so a GF4 core may be able to compete with the R300 on today's types of games. Then add in the fact that a GF4 core with "reasonably" minor changes for the wider bus and some tweaks should be relatively cheap for NV to design and validate and it's built on a money efficient 0.15 um process. That means they might be make it cheaply...Cheap enough to severely undercut a R300 board in price? Maybe. That would mean that NV would release a product with outrageous bandwidth and competes head to head with an R300 in today's games but lacks the DX9 (which isn't out yet) certifcation for significantly less money. That would be really interesting.

Of course, that is all unsubstantiated guessing with absolutely no basis in rumors or reality.
<wild speculation off>

Guess we just have to wait until tomorrow to see what NV shows.

[StealthHawk]

the answer is quite simple. nvidia isn't hyping NV28 because they know it can't stand up to the R9700Pro.

what are they supposed to do, yell "we've got the second fastest card on the planet coming out SOON." no, wait, we already have the 2nd fastest card on the market :p

[Nv40]

The Nv28 will be in performance between RAdeon9700 and Geforce4 ,
so if they add higher clocked DDR avaible memory 900Mhz ?
with a higher clocked core ,in a .15 process ,but with TWO TMUS
not ONe like the radeon 9700, that should mean
a card 15%-25% slower than the RAdeon9700 overall in AA+AF settings but for $249 .. :)

and even faster!! than Radeon9700 in many benchamrks where the RAdeon9700 only got from 10frames-20frames more than Geforce4 with AA/AF disable.. ;)

heavy multitexture games that benefit more for the EXTRA TMU
the Geforce4 already have + 900mhz+ DDR memory plus + higer clocks
should make the NV28 25% faster in multitextures games like
like Jedinight2/RTCW/QUake3/Medalofhonor/UnrealTorunament2003 ... :)

86 millions transitors for $259 for an Nvidia card is something i would like to see ,how its score against radeon9700...
i think Radeon9700 will be faster overall but not anymore for more than 20%-30% in direcx8 games ;)

Nv30 will be more a better competition for the
Radeon R350 (a .13 refresh of Radeon9700) but will
be avaible by the end of this year ;)

[Bigus Dickus]

Originally posted by Nv40
The Nv28 will be in performance between RAdeon9700 and Geforce4 ,
so if they add higher clocked DDR avaible memory 900Mhz ?
with a higher clocked core ,in a .15 process ,but with TWO TMUS
not ONe like the radeon 9700, that should mean
a card 15%-25% slower than the RAdeon9700 overall in AA+AF settings but for $249 .. :)That's got to be the funniest thing I've ever read. So they're going to bump the clock rates up, and it will go from being 100%++ slower to only 15% to 20%. Yep, I can't wait to see that GF4 clocked at 500/500! LOL
and even faster!! than Radeon9700 in many benchamrks where the RAdeon9700 only got from 10frames-20frames more than Geforce4 with AA/AF disable.. ;) Ever hear of CPU limited games/benchmarks? Guess not. Another lol for that one. :D
heavy multitexture games that benefit more for the EXTRA TMU
the Geforce4 already have + 900mhz+ DDR memory plus + higer clocks
should make the NV28 25% faster in multitextures games like
like Jedinight2/RTCW/QUake3/Medalofhonor/UnrealTorunament2003 ... :)Nope, I was mistaken previously. THAT is the funniest thing I've ever read. You do realize that the 9700 has the same number of TMU's as the GF4, don't you?

[Bigus Dickus]

psst, and there is no Radeon2, in case you hadn't actually checked in on ATI's product line in the past few years. ;)

[Matthyahuw]

BD, do you make it a habbit to just flame anyone who doesn't agree with you? People do have different opinions on things ya know...that's ok if you haven't figured it out in the RealWorld®...

And people wonder why Rage3D has such a bad rep...:rolleyes:

You seriously need to chill...the forums will be modded more heavily because they have gotten out of hand lately, just consider this a heads-up...;)

[volt]

YA!
Prepare for Armageddon! :eek:
j/k :D

[Nv40]

Originally posted by Bigus Dickus
You do realize that the 9700 has the same number of TMU's as the GF4, don't you?


mmmm , are you sure ??


http://www17.tomshardware.com/graphic/02q3/020819/radeon9700-03.html


Radeon 9700 will be limited by its ONLY ONE Texture unit pipe aka "TMU" ,thats why the radeon2 is much faster than the radeon9000
in multitexture games like (quake3engines games,and others)
... nice try.. ;)

Nv28 is not more than a Geforce4 ultra plus with 26 millions more transistors ,with higher DDR mememory and core ,so if Nv28 is 25-35% faster than Geforce4 ,dont be surprised to see the Nv28 score
better than the Radeon9700 in many multitextures games
that are not cpu limited .(AA+AF disable)

RAdeon9700 is more cpu limited than Geforce4 ,its needs a p4 2.5
Ghz to push the real the performance of the card
thats why people with lower Cpus score even closer
to Geforce4 levels .. :(

so keep with the FUn ,
but remember RAdeon9700 has only ONE TMU and everyones knows
how important this is in performance for multitextured games ;)

lets wait and see but dont be surprised
to see a $250 card (Nv28) scoring so close to the R300 or even
better in multitextured games with AA+AF disable ..
,people will think why pay $200 more for a
card that is only alot more faster
ONLY when anisotropic+Antialiasing its enable ..

mmmmm.something to think .. ;)

[-=DVS=-]

hmm who plays without AA and AF these days lol none that i know :D your comments made little point :rolleyes: becouse noone buys a card to play with crapy graphics i am sure even you play with AA and AF i bet:p

[StealthHawk]

Originally posted by DVS
hmm who plays without AA and AF these days lol none that i know :D your comments made little point :rolleyes: becouse noone buys a card to play with crapy graphics i am sure even you play with AA and AF i bet:p

all your friends must be in the know then. i bet that 50% of the people who own cards capable of FSAA or AF don't even use either, let alone both. i'm sure the actual number is much higher than that even. it's a proven fact that some people can't even tell the difference between AF and FSAA being on and off, even when given before/after shots(as evidenced on these forums more than a few times)

[jbirney]

Originally posted by Nv40

Radeon 9700 will be limited by its [B]ONLY ONE Texture unit pipe aka "TMU" ,thats why the radeon2 is much faster than the radeon9000
in multitexture games like (quake3engines games,and others)
... nice try.. ;)


The R9000 actually does better in the games that have heavy use of multitexturing:

http://www.anandtech.com/showdoc.html?i=1655&p=8

You can see for SS a game that lays down up to 5 textures that the 9000 with only one tmu is right up there with the 8500 an her two TMUs. Then understand that in its current forum of desgin the R9700 does not have enough bandwidth to keep both tmus per pipe busy. Finally as developers start to use more and more dx8/dx9 effects then the TMUs will play a smaller roll as pixel shaders will have more of an impact.




RAdeon9700 is more cpu limited than Geforce4 ,its needs a p4 2.5
Ghz to push the real the performance of the card
thats why people with lower Cpus score even closer
to Geforce4 levels .. :(

There is no such thing as being more CPU limited as that would imply that its scores lower than another card for a given CPU. And when that happens its usally drivers that have not been optimized. Besides the NV30 will also be CPU limited in these same cases. Besides I'd rather have a card that CPU limited as I can get around that but if its MB or fill rate limited then I have to get a new card or lower the IQ.


all your friends must be in the know then. i bet that 50% of the people who own cards capable of FSAA or AF don't even use either, let alone both. i'm sure the actual number is much higher than that even. it's a proven fact that some people can't even tell the difference between AF and FSAA being on and off, even when given before/after shots(as evidenced on these forums more than a few times)

Well very true but how many of your of those people whent out and paid more than $250 for the card? Why is that important? Well be cause any one that spends that ammount of money usally knows the hardware and knows about its features. Those that get a lower priced version arent really keen on features as they want price. High end users want IQ/Features. Different market different needs/wants.

[Bigus Dickus]

Originally posted by Matthyahuw
BD, do you make it a habbit to just flame anyone who doesn't agree with you? People do have different opinions on things ya know...that's ok if you haven't figured it out in the RealWorld®...

And people wonder why Rage3D has such a bad rep...:rolleyes:

You seriously need to chill...the forums will be modded more heavily because they have gotten out of hand lately, just consider this a heads-up...;) Is that a threat? :rolleyes:

I make it a point to flame (if you insist on calling it that) anyone who posts such obvious nonsense. I get the distinct impression that... nah, I'll leave it alone.

Opinions are opinions, and everyone is entitled to their own. I'm also free to point out when someone else's is based on nothing but wishful thinking.

[Bigus Dickus]

Originally posted by Bigus Dickus

You do realize that the 9700 has the same number of TMU's as the GF4, don't you?
Originally posted by Nv40

mmmm , are you sure ??


http://www17.tomshardware.com/graphic/02q3/020819/radeon9700-03.html


Radeon 9700 will be limited by its ONLY ONE Texture unit pipe aka "TMU" ,thats why the radeon2 is much faster than the radeon9000
in multitexture games like (quake3engines games,and others)
... nice try.. ;)mmmm, quite sure. I think I learned multiplication in the third grade or so. GF4 = 4 pixel pipes X 2 TMU's per pipe = 8 TMU's total. Radeon 9700 = 8 pixel pipes X 1 TMU per pipe = 8 TMU's total. Therefore (in case this slipped by) the 9700 has the same number of TMU's as the GF4. In dual-textured situations the GF4 Ti4600 and Radeon 9700 has nearly the same pixel fillrate, but the 9700 has a distinct advantage of more efficient core architecture and memory bandwidth. In games with 3 textures, the 9700 actually has a 33% fillrate advantage. They're even again at 4 textures, and after that the loopback ability of the 9700 keeps it well ahead of the GF4.

Why do I get the feeling that this explanation is completely wasted...
Nv28 is not more than a Geforce4 ultra plus with 26 millions more transistors ,with higher DDR mememory and core ,so if Nv28 is 25-35% faster than Geforce4 ,dont be surprised to see the Nv28 score
better than the Radeon9700 in many multitextures games
that are not cpu limited .(AA+AF disable) Despite the fact that we've dismissed the "multitextured advantage" theory, one simple observation remains: the 9700 is nearly always CPU limited without AA and/or AF.

[DadGT]

Originally posted by Bigus Dickus

mmmm, quite sure. I think I learned multiplication in the third grade or so. GF4 = 4 pixel pipes X 2 TMU's per pipe = 8 TMU's total. Radeon 9700 = 8 pixel pipes X 1 TMU per pipe = 8 TMU's total. Therefore (in case this slipped by) the 9700 has the same number of TMU's as the GF4. In dual-textured situations the GF4 Ti4600 and Radeon 9700 has nearly the same pixel fillrate, but the 9700 has a distinct advantage of more efficient core architecture and memory bandwidth. In games with 3 textures, the 9700 actually has a 33% fillrate advantage. They're even again at 4 textures, and after that the loopback ability of the 9700 keeps it well ahead of the GF4.


You make one assumption that is not a given here: that the texture units of the R300 can be combined at any arbitrary (or specifically dual texture) combination with at as much efficiency as a dedicated dual-texture (or whatever) unit. I would imagine some overhead and speed loss should be involved.

However, it is interesting to note that the graphs in Tom's article later for multitexturing performance show that the r300 is faster than the GF4 by about 8%. That 8% matches the MHz increase from the GF4 to the R300. Right on tap with an efficient combination mechanism. Good for Ati.

More interesting to note is the single texturing behavior. The R300 creams the GF4 by an increase of 76%. But with twice as many units and an 8% speed increase, I would expect the R300 to beat the GF4 in single texture by somewhere around 116%. But the real value is not near as much. So under single texturing behavior, the R300 is way away from it's theortetical behavior. Why?

[Bigus Dickus]

Originally posted by DadGT


You make one assumption that is not a given here: that the texture units of the R300 can be combined at any arbitrary (or specifically dual texture) combination with at as much efficiency as a dedicated dual-texture (or whatever) unit. I would imagine some overhead and speed loss should be involved.Actually, I think the single TMU architecture is more flexible, and performance is more consistent among a variety of texturing demands.

Consider: in a (hypothetical) single texturing benchmark/game, the GF4 can use only half it's theoretical texturing capability, while the 9700 can use 100%. In a (hypothetical...) dual textured benchmark/game, both can use 100%. In a triple textured game, the GF4 is limited to 75%, the 9700 is at 100%. Four textues, both at 100%...

i.e., in odd numbers of textures the GF4 has an idle texture unit on some clock cycles. This makes you wonder what the Parhelia designers were thinking, since it can have two or three TMU's idle per clock. That card as an 8 X 2 would have considerably better performance.

This whole argument can be related to the GF2 vs. Radeon256 designs. Since one TMU of the Radeon was almost neve used, it had a serious handicap by design. With only 2 pixel pipes, it only kept up with (nearly) the GF2 due to the bandwidth limitations suffered by both cards (more so by the GF2, since it had much more pixel fill rate capability).

A single TMU design is flexible in that it maximizes performance for an average of texturing situations. Further, the emphasis with the arrival of DX8 games will shift from look-up texturing to computational ability (pixel/vertex shading). The 9700 is fairly strong in this area, and we can expect this to be the category which most future cards concentrate on.

As to the single-texturing efficiency of the 9700, I have wondered about that as well. You noted that the 9700 is roughly equal to the GF4 per clock in dual texturing. It could be the case that the 9700 is very efficient at multitexturing - relative to its single texturing - due to the (previously discussed) efficiency of the 8 X 1 architecture, where the GF4 isn't as efficient at dual texturing - relative to its single texturing - because its difficult to keep both TMU's working constantly. Perhaps the GF4 suffers from some passes which use only one TMU for some reason. Just a hypothesis.

Also, it's important to note that the loopback ability of DX9 hardware keeps heavily multitextured demands from seriously crippling a 1 TMU/pipe architecture, since another cycle to apply another texture is minor compared to another rendering pass.

For my final digression... I'm thinking about the NV30 here. Let's suppose that it's an 8 X 2 TMU card as some have predicted. I think the most important performance consideration will be whether it is limited to 3 math ops per pixel pipeline, as the R300 is, or whether it will be able to process 3 math ops per TMU. I'm almost sure the former is the case (I don't think the GF4 or 8500 can process math ops per TMU either... the TMU is just one part of the pixel pipe, just as the caculational units are - adding TMU's doesn't by defacto increase computational ability). For DX8+ situations, the extra TMU won't help that much. Only in DX7 games, where performance is already approaching insane levels, will the advantage be significant (given proper bandwidth).

It would seem that an 8 X 1 architecture was a smart design choice, yielding maximum performance in a wide variety of situations given a certain transistor count limit.

[Bigus Dickus]

I think I neglected to mention something at the very beginning: the R300 doesn't need to combine TMU's arbitrarily to render multitextured applications (2, 3, whatever). Instead of "combining texture blocks" to become a 4 X 2 card, or a 2 X 4 card, it simply uses loobacks to apply multiple textures per pass.

An 8 X 1 arrangement using loopback texturing for a dual textured game would actually be significantly more efficient than a 4 X 2 "TMU combined" arrangement in the same game. The 8 X 1 arrangement would render 8 dual textured pixels per pass, with each pass requiring a small increase in clock cycles to apply the second texture. The 4 X 2 arrangement would render 4 dual textured pixels per pass without the extra clock cycles, but would require two passes (thus sending geometry information twice) to render 8 pixels. One pass + extra texturing clock cycles takes significantly fewer clock cycles than two complete passes.

In fact, an 8 X 1 card with loopbacks is, mathematically, superior in any given multitexturing application to any multiple TMU architecture, given that geometry and texture information can be sent fast enough to saturate all possible designs (certainly not a given... this is one of the primary concerns of a well-balanced design).

[-=DVS=-]

Originally posted by StealthHawk


all your friends must be in the know then. i bet that 50% of the people who own cards capable of FSAA or AF don't even use either, let alone both. i'm sure the actual number is much higher than that even. it's a proven fact that some people can't even tell the difference between AF and FSAA being on and off, even when given before/after shots(as evidenced on these forums more than a few times)


lol i feel sorry for people who play without AF and AA :p

realy how can anyone not see a difference lookeing at AA picture and without :rolleyes: maybe they are in denial :eek: or have bad eyes/medical condition so then its offcourse not good to laugh ;)