NV3x-based nForce?

[Chalnoth]

At the end of Anand's nForce2 article (http://www.anandtech.com/chipsets/showdoc.html?i=1731&p=10), I caught this paragraph:

Our only performance complaint with nForce2 is the performance of the integrated GPU. Its performance is understandable, considering the limited memory bandwidth at its disposal; with that said, in order for the IGP to be taken seriously even as a gaming platform we'll need to see some serious improvements in performance. The nForce2 IGP is already the fastest integrated graphics solution out for any platform, AMD or Intel; but NVIDIA will have to continue to raise the bar in order to undo the impact of years of poor integrated graphics solutions on the minds of consumers and enthusiasts alike. We expect the first major step to come with the NV34 based nForce solution due out next year; with a NV30 derived core at its disposal, NVIDIA should have an even more compelling chipset a year from now.

An NV3x integrated processor due next year? Next year could be an amazing year for the 3D graphics market!

With the apparent quick move to the low-end of the market of the NV3x core, though, it kind of makes you wonder why there was no such migration for the NV2x core. In fact, it doesn't look like nVidia will ever produce a low-end NV2x core.

[gemini1313]

i want one.

i think the nforce3x would have to be for the good ol' mc hamma platform.

[Chalnoth]

Now that's one thing that I'm really curious about. Since the Hammer platform uses a memory controller integrated on the CPU, unless the integrated graphics has its own memory, it will have more memory latency than it would otherwise have in one of today's chipsets. From what I've seen so far, nForce boards for the Hammer processor will not have dedicated memory for the GPU. I'd really like to know how well nVidia is going to be able to hide that latency. Logic would dictate that the Hammer versions of their integrated graphics would be noticeably worse than those for the Athlon.

[Bigus Dickus]

It would be cool if there was BGA memory on-board for the IGP, or at least an extra DIMM slot that was for dedicated video memory... but I guess that's a bit of wishful thinking. Sort of negates the whole cost-saving reason for integrated platforms I guess.

[Chalnoth]

That was apparently nVidia's argument in not having dedicated memory for the IGP.

[Matthyahuw]

put me on the waiting list for an nForce3x! :D

[heedory]

I don't expect much from nvidia
they are not what they were..
before talking about nforce3.. where is nforce2 ??

[StealthHawk]

i would like to point out that we don't know for sure whether or not motherboard manufacturers will have memory controllers or not. i've been hearing that they will, based on some rumor that a dual channel memory controller can somehow make the Hammer's integrated controller even more efficient.

dunno whether it's true or not though

[jbirney]

Whoa Hos,

read that a bit closer:

with a NV30 derived core at its disposal

Well if thats the case I am not all that Jazzed about it until we now what they mean about "derived" core. If you go by past history then its usally a lower perfroming intergrated core. Or if you look at their last GF4 derived core you then have the GF4 MX which again is not all that great.

I have two nForce mobos. They work great. I can not wait to get an nF2 one. And I am sure the nF3 will rock the house. I will just take mine with out the watered down graphics intergrated :)

[Bigus Dickus]

Except that the Nforce2 (and the GF4MX core) is not a GF4 derived core. It's a GF2 derived core, hence it's core name of NV17 which puts it firmly in the NV1x family.

[jbirney]

yea sorry I knew that but just did not word it correctly.

[borntosoul]

has anyone seen the performance of the nforce2 with integrated graphics vs video card ? it suffers a fair bit worse ,maybe they got something up their sleeve again :)

[Chalnoth]

Not sure what you mean by, "It suffers a fair bit worse."

The nForce2 is a new core, and it just doesn't have as much memory bandwidth in comparison to the current 4MX line as the original nForce did in comparison to the 2MX line when it was released.

So, how are you making this comparison?

[ben6]

keep in mind, that traditionally , codenames follow a set of features . In other words

NV10= Geforce
NV11= Geforce2 MX
NV15= Geforce2
NV16= Geforce2 Ultra
NV17= Geforce4 MX
NV18= Geforce4 MX AGP8x

notice all of the cards have the same exact featureset, whether or not they're named Geforce2 or 4 .

NV20 Geforce3
NV25_1 Geforce 4 4600
NV25_2 Geforce4 Ti4400
NV25_3 Geforce4 Ti4200
NV28 Geforce 4 AGP8x again everything in line

now, it would be logical to assume anything NV3x would be DX9 based, including their integrated and mobile cores

[Chalnoth]

Well, not exactly the same featureset, but at least the same programming interface (essentially, anyway, GF3 and 4 are slightly different).

[Chalnoth]

Originally posted by ricercar
At the rate we're going, we won't see nForce3 until end of 2004. By that time NVIDIA and ATI may be on the next generation of hardware after NV30/R300.

In the don't count your chickens category, the nForce3 IGP may have NV40 video core. I don't know where Anandtech got NV34 for nForce3. Might as well pick NV32 or NV38, huh?

This whole idea is based entirely on heresay, but it does mesh well with what we've been hearing in various rumors, from those about having an NV31 and/or NV31M out next Spring to the updated nForce.

Just don't be too expectant about the performance of an NV31 or nForce w/ NV3x. For all we know, nVidia may decide to go all the way down to two pipelines for each of these products to keep the costs in line (I hope they don't, but it's still a possibility, though fortunately it seems remote currently).

[Chalnoth]

It really depends upon how different from the NV30 the NV31 is.

But how different they are has no bearing on release date. nVidia isn't going to release one product, and then begin developing the next. Both nVidia and ATI concurrently work on many different products.

In fact, it is very likely that both the NV31 and NV34 cores (assuming they are to be released next year, of course) are almost certainly already in the works.

Btw, the primary difference between a GeForce4 MX and GeForce4 Ti can primarily be explained by nVidia's internal model numbers: NV17 vs. NV25. Quite simply, they are entirely different architectures. The NV17 core is derived from the original NV10 core (GeForce 256), whereas the NV25 core is derived from the NV20 core (GeForce 3). The different cores are fundamentally different on the programming side just because of the support that all NV2x cores have for pixel/vertex shaders.

[StealthHawk]

Originally posted by ricercar
So when NVIDIA has a chip like the NV30, is there a lot of work to make a 31 or 32 version?

I still don't entirely understand the difference between GeForce4 MX and Ti at the same clockspeed, but I read that it's significant. Yet didn't both the GeForce4 MX and GF4 Ti series come out the same time last spring? Wouldn't the NV30 and NV31 be released simultaneously too?

And if this is the case, wouldn't nForce3 appear soon after NV30?

i would say no. rumors point to production problems with NV30. ie, all resources will be focused at getting as many NV30 chips out the door as possible. all vendors seem to use a pattern of launching high end chips followed by low end/midrange chips in order to maximize profits as well.

GF4mx is basically an enhanced gf2mx. what this means is that it doesn't have any DX8 hardware like the GF4/3 does. there are also differences in the pipelines.

[Chalnoth]

Originally posted by StealthHawk
i would say no. rumors point to production problems with NV30. ie, all resources will be focused at getting as many NV30 chips out the door as possible. all vendors seem to use a pattern of launching high end chips followed by low end/midrange chips in order to maximize profits as well.

Other than people posting on message boards about things that they're extrapolating elsewhere, I have yet to see anything remotely credible that there are production problems beyond a product delay due to the .13 micron process availability.

Regardless, we'll see, but current rumors point to the NV31 at about 3 months after the release of the NV30, which would place it right around the traditional Spring timeframe (Side note: Hopefully it's not the NV31M, which is, quite unfortunately, a strong possibility, that will be released at this time, as that would likely mean another 3-6 months before we'll see any NV31's, notebook or otherwise).

I would generally expect, mostly pulling this out of my ass, that if an NV3x-derived nForce IGP is released next year, it will be released late next year.

[StealthHawk]

didn't Anand write about the low yields of .13 at TSMC? i would classify that as a production problem.

beyond that, everything else is hearsay. things will become a lot clearer once NV30 is announced, but even then, we might not get the full picture.

[Philibob]

Originally posted by ricercar
If NVIDIA continues with a unified driver approach, won't that mean there's a lot of excess baggage? Won't the NVIDIA drivers suffer from bloated code if they have to support three architectures? I wonder if we'll have to suffer GF2 code in our NV30 drivers until the NV30 is the mainstream video card from NVIDIA.
I think they should make different sets of drivers:
1. NVx
2. NV1x
3. NV2x
4. NV3x
This would also cut down on the filesize wouldn't it?

[Chalnoth]

Originally posted by ricercar
If NVIDIA continues with a unified driver approach, won't that mean there's a lot of excess baggage? Won't the NVIDIA drivers suffer from bloated code if they have to support three architectures? I wonder if we'll have to suffer GF2 code in our NV30 drivers until the NV30 is the mainstream video card from NVIDIA.

I don't think so. This has been asked before, and nVidia still has about the best drivers you can get. There are a number of potential reasons for this, but as for filesize, it's not that bad yet.

That is, I'm reasonably certain that most of the driver code for 3D stuff lies in the two files nv4_disp.dll and nvoglnt.dll. These two files are a total of 6.18MB with the 30.82 driver version (Should be about 3MB compressed). Most of the rest of the filesize comes from driver files that would pretty much always be consistent across different architectures anyway, such as the nview files and control panel files. From this perspective, I'm really not sure that supporting only a new architecture would significantly reduce filesize (though supporting only an old architecture most definitely would....). By the way, I used the 30.82 driver file sizes because the 40.72 drivers include software emulation for NV30 features.

After all, the newer hardware still has to be backwards-compatible with older software. The differences, depending upon how nVidia has updated their hardware, between the drivers for supporting older extensions could be as small as different optimizations for different hardware.

[StealthHawk]

Originally posted by ricercar
So when NV30 arrives, NVIDIA will have three different programming/hardware architectures to support?

a) GeForce2, Geforce4 MX (¿DX7 support?)
b) GeForce3, GeForce4 Ti (partial DX8 support / NV2x pixel shaders)
c) NV30 (DX9 support)

more like

a) TNT/TNT2
b) GF/GF2
c) GF3/4
d) NV30

[Chalnoth]

What you must realize, however, is that OEMs are the prime source of revenue for nVidia's graphics cards. nVidia can make things much easier on OEMs if they can supply one driver for all of the products that they sell.

It's also a great system for businesses, as they only need the one driver for every PC in the company that uses any sort of nVidia graphics card.

And for consumers, the #1 benefit is a guarantee that the drivers for your old nVidia graphics card will never stop being updated.