GeForce FX 6800Ultra -previews thread

mozmo said:
Some how i have a feeling that ATI doesn't have an answer to the dual shader 16 pipe nvidia design. Remember all that 8 xtreme pipeline talk a while ago from ATI, i have a feeling they had a design that had 2 pixel shaders per pipe, but they caught wind of the nv40 16x1 design, so they must of redid the design so it's 16x1 with R300 style pipes. I have a feeling that ATI will need to hit 500mhz to match nvidias 16x 2 shader pipes with their 16x1 R300 style pipe. This is why i guess the ATI camp has been very quiet, i have a feeling they're in panic mode at the moment.
Click to expand...

Can I also have a reasonable explanation how IHVs magically pump up architectures as you seem to suggest?

ATI has been extremely quiet also before the R3xx release and I don't think it's an indication for anything. The worse case scenario for them would be IMHO to have similar or highly competitive performance; where's the reason to panic?

NVIDIA on the other hand would love to spell doom, yet so far I haven't seen anything that suggests that they're about to leave the competition clearly in the dust.
 
DemoCoder said:
Ailuros said:
DemoCoder said:
Well, technically the Radeon9800XT wasn't launched 18 months ago. :)

If ATI can manage 16pipes @ 500mhz + 1.2ghz memory and dual shaders per pipe, I think they can take the perf crown. @ 600Mhz and faster memory, it's definate.

I wonder if they've got an answer for 32x0 mode tho, which will be nice for some game engines. :)
Click to expand...

They had already 2 z/stencil units per pipe on R3xx; why would that change?
Click to expand...

But that only mattered for multisampling correct? What about non-MSAA buffers? (e.g. shadow buffer)
Click to expand...

I'm not sure where you're aiming at; who cares about non-antialiased graphics these days anyway?

If this is about stenciling, ATI has in fact an advantage with MSAA+stencil with 2x sample MSAA, since it loops with 2xAA samples at a time.
 
Evildeus said:
And if it doesn't? Aren't you a bit indifferent? That's a great improvement, whatever the R420 gives us.
Click to expand...

No doubt (though I wish they had improved the AA more), but comparing a next generation card to a last generation card doesn't really tell us what we want to know ie which is the better card this time around. I would *expect* NV40 to be much better than R350, much as I would expect an Opteron being better than an Athlon, or a Pentium 4 being better than a Pentium 3.
I need to see what the R420 brings to the table compared to NV40 before I can decide which is the better card to buy.

Crowing about how the NV40 "owns" the 9800 Pro is kind of like congratulating a grown man for beating a 12 year old at arm-wrestling.
 
Can I also have a reasonable explanation how IHVs magically pump up architectures as you seem to suggest?
Click to expand...
Well lets be honest the R420 was originally another design/project (R400) that got simplified, the R420 doesn't really introduce a new shader model and uses exsisting R300 style tech, ATI have had a lot of time to tinker around with possibilities, all i'm suggesting is probably initially they were looking to release an 8 pipe config with extreme pipes, ie 2 shader units per pipe. Since we all expected nv40 to be a 8 pipe card this would of been good for ATI, but since nvidia pulled an engineering marvel out of the hat, the other option was to use R300 quads and pack as many as they could on a single core, ie 3/4 like what we're gonna get it seems. I reckon that's what happened especially since ATI is launching later with a part with a feature set that's 18 months old.
 
The power consumption graph seem to suggest the radeon9800xt consume more power than NV40!!
Oh I can confirm the 12 pipeline, ddr1, 128mb of 6800 non ultra since it was in the launch event presentation. Democoder actually had posted it.
 
 
Lezmaka said:
Looks like Tom's is live

http://www.tomshardware.com/graphic/20040414/index.html
Click to expand...

I mean Tom is usually 'high' on Nvidia... he has some results that are 3x as good as the current competition... but somehow I'd liek to see that confirmed elswhere...

and comments like these :p

The results achieved by the 6800 Ultra in this game are simply out of this world - more than twice the performance in 1600x1200 compared to the GeForce FX 5950 and the Radeon 9800XT. Scores like these make even experienced graphics editors' eyes pop.
Click to expand...

Tom = The Sun without page 3 girl.
 
NV40 running all shaders on FP16...

farcry-scene1.jpg


As seen in that picture.
 
Ailuros said:
But that only mattered for multisampling correct? What about non-MSAA buffers? (e.g. shadow buffer)
Click to expand...

I'm not sure where you're aiming at; who cares about non-antialiased graphics these days anyway?
[/quote]

Because DX9 doesn't allow render targets to use MSAA. Ergo, shadow buffer algorithms which use render to texure can't use it.

If this is about stenciling, ATI has in fact an advantage with MSAA+stencil with 2x sample MSAA, since it loops with 2xAA samples at a time.
Click to expand...

Are we talking NV3x or NV40? I'm talking NV40. Hopefully Dave runs FableMark on NV40 with and without MSAA.
 
aarrggg

nv40 has gone angle dependant AF :!:

I'm somewhat lost for words. I was hoping that would be removed from at least r500 if not r420. There's no need for nvidia to ADD it. :?



hmm looks like they didn't show all modes, can this be disabled?
 
DaveBaumann said:
You know, sometime you feel like a right chump sitting here and waiting for the NDA times.
Click to expand...
why not just release it, others have. Im sure that means that in spirit the nda's have expired.

later,
epic
 
DaveBaumann said:
You know, sometime you feel like a right chump sitting here and waiting for the NDA times.
Click to expand...

you might come last but you will be the best ;)

so don't worry... at least this initial forum storm will pass :D
 
DemoCoder said:
Ailuros said:
But that only mattered for multisampling correct? What about non-MSAA buffers? (e.g. shadow buffer)
Click to expand...

I'm not sure where you're aiming at; who cares about non-antialiased graphics these days anyway?
Click to expand...

Because DX9 doesn't allow render targets to use MSAA. Ergo, shadow buffer algorithms which use render to texure can't use it.

If this is about stenciling, ATI has in fact an advantage with MSAA+stencil with 2x sample MSAA, since it loops with 2xAA samples at a time.
Click to expand...

Are we talking NV3x or NV40? I'm talking NV40. Hopefully Dave runs FableMark on NV40 with and without MSAA.[/quote]

I don't see anything mind-boggling here:

http://www20.tomshardware.com/graphic/20040414/feforce_6800-34.html

In fact the differences seem to be in str8 analogy to the differences in Z units between architectures. And yes Fablemark would be better suited as a pure synthetic, since it's so fill-rate limited.
 
Don't worry Dave, I'm sure yours will be the most complete. :)

If not, Tomshardware OWNS you, Muahaha, your review wears combat boots with two molex connectors!!
 
You must log in or register to reply here.

Similar threads

Rys
X800 XT/PRO Review Links
Replies
2
Views
2K
MuFu
M
Back
Top