there is no NV47, but G70 is on the way

[kemosabe]

Something obviously doesn't jive with that comparison because the 6800 Go performance has nearly tripled (on a slower CPU), which is far beyond what you would expect from the clock increases of the Ultra specs.

 

[Razor1]

maybe increased vertex shader performance

 

[_xxx_]

Judging on these benchies, they definitely changed something. But the first pic shows the non-ultra GF6800Go. Wasn't it a 12-pipe part?

EDIT:
Yup. Lookie.
B3d's own

 

[Razor1]

yeah it was 12 and I think the clocks were 300/300 this one has quite a bit more on both the gpu and memory 450/550.

 

[wireframe]

There was also mention of 128 and 256-bit memory interfaces depending on the model (higher end getting the wider bus, naturally).

I'm still not entirely convinced about the Pentium M argument either. It's probably because the numbers fit a NV40@450 so perfectly at times. It would be nice to know exactly what is going on.

 

[Megadrive1988]

Is a 512-bit memory bus feasible late into 2005, or say, sometime in 2006 ?


it was pretty impressive to see 3 graphics vendors have 256-Bit memory buses in 2002: Matrox`3Dlabs`ATi.

 

[Geeforcer]

There was also mention of 128 and 256-bit memory interfaces depending on the model (higher end getting the wider bus, naturally).

I'm still not entirely convinced about the Pentium M argument either. It's probably because the numbers fit a NV40@450 so perfectly at times. It would be nice to know exactly what is going on.

I am not so sure either: Pentium M can't even approach higher-end Athlons FX in any of the games Anand tested here:

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2342&p=15

 

[_xxx_]

Is a 512-bit memory bus feasible late into 2005, or say, sometime in 2006 ?


it was pretty impressive to see 3 graphics vendors have 256-Bit memory buses in 2002: Matrox`3Dlabs`ATi.

512-bit is rather problematic because of the high pincount. I guess we'll see some kind of serial interface pretty soon, or some other techniques to minimize that (think PS3 kinda bus?).

 

[KimB]

512-bit is rather problematic because of the high pincount. I guess we'll see some kind of serial interface pretty soon, or some other techniques to minimize that (think PS3 kinda bus?).

Well, DDR, and its subsequent refinements, are steps in this direction.

 

[Pete]

An MR X800 would give similar performance using the same CPU as well.

The MR X800 is 400/400, whereas this GFg 6800U has leapt to 450/550. I don't think the X800 will outperform it.

Something obviously doesn't jive with that comparison because the 6800 Go performance has nearly tripled (on a slower CPU), which is far beyond what you would expect from the clock increases of the Ultra specs.

Dothan on Sonoma should be faster than a P4, extrapolating from here. And I'm not sure if that previous 6800--a 300/300MHz part--was packing 128 or 256MB. (Both are 256-bit, according to the reviews.) Finally, the mobile 6800U is using newer drivers.

Judging on these benchies, they definitely changed something. But the first pic shows the non-ultra GF6800Go. Wasn't it a 12-pipe part?

All the GFg 6800s are 12 pipes, even the GFg 6800U. The latter is just an upclocked version of the former.

I am not so sure either: Pentium M can't even approach higher-end Athlons FX in any of the games Anand tested here:

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2342&p=15

Ah, but none of those Dothans are using a 533MHz FSB, which seemed to help the OC'ed 2.13GHz Dothan in my first GamePC link. The Dell is based on a Sonoma platform, which adds dual-channel on top of the now-native 533MHz FSB. So, lots more bandwidth for the Dell Dothan.

Anyway, there are more gaming benchmarks here, courtesy of another GamePC article that focuses specifically on Dothan vs. A64 (and P4) performance, thus offering a wider range of benchmarks. It also includes an A64 3500+ and Dothan 2.0GHz running at 400MHz FSB. You can see the Dothan outperforms the A64 across the board even with a lower FSB than my first GamePC link.

BTW, nice pics of the Dell here.

 

[ANova]

The MR X800 is 400/400, whereas this GFg 6800U has leapt to 450/550. I don't think the X800 will outperform it.

You're right, but I didn't say it would outperform it. I still think sticking an MR X800 in the same laptop will bring it closer to the Go 6800U but yes, it will be a little behind.

 

[Pete]

Right, dunno how I inferred that. But the 6800 tends to be more efficient at a iven core clock, and coupled with the GFg6800U's much higher mem clocks, the 6800U should be noticably faster. So I took issue with "similar" perf, but I have no clue how that turned into outperforming it.

Anyway, sorry about that.

 

[wireframe]

BTW, nice pics of the Dell here.

That was a wonderful link and look at the 3DMark 2005 score it posted: 5221 3DMarks. (wowza!)

Surely this 'title' cannot be benefiting from the magic Pentium M Dothan. Are we to think driver optimizations allow this or is there something more to it?

Assuming this is a 12-pipeline part with 450/1100 clocks, this would suggest a desktop 6800 Ultra score of [(16/12)*(400/450)*5221=] 6187 unless we are memory bound (very rough and nasty scaling, but still). Unless I am completely mistaken, my Ultra (desktop) scored in the high 4000s and maybe in the low 5000s with newer drivers. (I am guilty of not running this benchmark as much and often as I should).


3DMark 2003 is also spuspiciously close to what I get on my 6800 Ultra (around 12K with 'High Quality'). In fact, everything there is looking suspiciously close to what I am seeing on a 6800 Ultra (425/1200) coupled with an Athlon 64 3500+.

Someone needs to run a pure fill-rate test and put us out of our misery.

 

[ondaedg]

My guess is that they are probably using some driver optimizations similar to what ATI did with the 9600. I just don't see how a mobile part could equal or best a desktop part unless it was exactly the same adapter(i.e. 16 pipe, 256meg). Those drivers could be performing some type of enhancement as well. Another set of Nvidia miracle drivers?

 

[Jawed]

I was wondering if the drivers might be rendering everything in 16-bit. After all, who needs more than 16-bits on cruddy laptop LCDs?

Teehee.

Jawed

 

[Dave Baumann]

yeah it was 12 and I think the clocks were 300/300 this one has quite a bit more on both the gpu and memory 450/550.

450/530 according to Dell.

 

[Razor1]

oops sorry the Anandtech article had 550 on the mem, but on dell's site it is 530.

 

[Karma Police]

As a guy that's looking for a notebook in a few months (I think) & posts on notebook forums, let me tell you guys the overclocking potential of the NV42M is astounding. The .11u by itself could let it get over 500+Mhz core.

The only way I can see ATi beating nVidia in the notebook race now is if it releases the X800XL in a notebook. A 12-pipe SM3.0 card should have around the same tranny count as a 16-pipe SM2.0b card, but should allow better performance at higher resolutions.

Too bad the XPS2 can't ever run XP-Pro 64 bit version, though. At least, that's my take.

 

[Zaphod]

PCI Express? This connector looks very different from the MXM connectors we have seen in the past.

That's an AXIOM-module isn't it? Looks rather similar to the images ATI released when they introduced the Mobility X600. Also the XPS sister model in Europe, the Inspiron 9300, has an option for the mobility X300. If so: Will we get the breaking news followup article from the Inquirer?

 

[Dave Baumann]

The only way I can see ATi beating nVidia in the notebook race now is if it releases the X800XL in a notebook. A 12-pipe SM3.0 card should have around the same tranny count as a 16-pipe SM2.0b card, but should allow better performance at higher resolutions.

130nm low-k will offer better thermal properties than 110nm.