we have enough info to estimate Nv30 performance.

[Nv40]

look at the chart in this link ..

http://www.nvmax.com/Articles/Previews/NVIDIA_GEFORCE_FX/index.html

there are benchmarks posted there of Nv30 vs Geforce4 ti4600 with these settings .

P4 3GHz **
1280x1024
4x antialiasing
2x anisotropic filtering

Nv30 is ...
2.5x faster in quake3 than Geforce4
3.5x faster in Doom3
3x times faster in 3dmark2001se
4x times faster in pixel shading test
(i think this is 3dmark2001 advance shading test)

** i think that it is a P4 3ghz because in all reviews this is the
processor that most benchmarks Nvidia use.

i think we can do some estimates now about the Nv30 performance in those settings against the Radeon9700pro..
so anyone with the R300 feel free to post your info using
High quality settings.. without overclocking the video card!!
only the cpu to reach something close to 3ghz !!

i estimate by looking at those numbers that Nv30 should be between 10% and 40% in these test against the
Radeon9700pro.

[-=DVS=-]

Well if you look at 3dMark 2k1 database you will find that Radeon 9700 Pro at high res with 4xaa and aniso is already 3 - 4 times faster then Geforce 4ti :rolleyes: , so NV30 doesn't look so impressive anymore :p , best we can do is wait for real product and see real performance not paper speculation ;)

[jbirney]

Also remember Doom3 has a seperate code path for NV30. Not sure which path the R300 would have to take. Also no idea what version NV used on the doom bench and you did notice that NV made the level they tested :)

[thcdru2k]

yes it is four times faster than the gf4 ti4600, but if you look at the nature scores, they are slower against the supposed nv30 nature test benches.

[Bigus Dickus]

Yes, and pixel shading is one area where the NV30 should have the undisputed advantage, simply because it's 8 shaders are running at 500MHz compared to the R300's 325MHz. If it did not have a significant lead in pixel shading, something would be terribly wrong.

I have to wonder about nV's approach to vertex shading though. While it sounds like a cool idea (being different and all), it's less powerful per clock than the R300. In fact, were the R300 just clocked 8% faster it would match the NV30's vertex shading power.

I wonder what such a design bought nVidia? Less transistors? More flexibility?

[gokickrocks]

if someone would donate a p4 3.06 with board...i'd be glad to benchmark it on the 9700 pro :D

[Nv40]

Originally posted by Bigus Dickus
Yes, and pixel shading is one area where the NV30 should have the undisputed advantage, simply because it's 8 shaders are running at 500MHz compared to the R300's 325MHz. If it did not have a significant lead in pixel shading, something would be terribly wrong.

I have to wonder about nV's approach to vertex shading though. While it sounds like a cool idea (being different and all), it's less powerful per clock than the R300. In fact, were the R300 just clocked 8% faster it would match the NV30's vertex shading power.

I wonder what such a design bought nVidia? Less transistors? More flexibility?


it is the same issue with AMD and Intel ..

clock per clock AMD AtlonsXP processor are more faster than Intel
Pentiums4 .. just raise the clock of the atlons Xp processor to
3ghz! and it will easily smoke the P4 3ghz of Intel !

does that mean that Amd its the leader in Performance ?
Absolutely not. Intel is the King .

So Faster memory and higher clocks and Wider bus is part of the strategic design of Nvidia and ATi . the ones who get the best technology and hardware combination wins .

it will be interesting the year 2003 ,with ATi and Nvidia
next products.. :)


but i agree that only real benchmarks of the Nv30 retail
can say the truth above how well it fulfill Nvidia claims
in benchmarks.

[sebazve]

who made these benchs?? if they come from Nvidia is BS!

[-=DVS=-]

Originally posted by sebazve
who made these benchs?? if they come from Nvidia is BS!

Yeah they came from Nvidia :rolleyes: review sites will get first samples in few months :( so no real bench till next year.

[Uttar]

Originally posted by Bigus Dickus
Yes, and pixel shading is one area where the NV30 should have the undisputed advantage, simply because it's 8 shaders are running at 500MHz compared to the R300's 325MHz. If it did not have a significant lead in pixel shading, something would be terribly wrong.

I have to wonder about nV's approach to vertex shading though. While it sounds like a cool idea (being different and all), it's less powerful per clock than the R300. In fact, were the R300 just clocked 8% faster it would match the NV30's vertex shading power.

I wonder what such a design bought nVidia? Less transistors? More flexibility?

I'd guess it's more efficient for Dynamic Branching ( something the R300 doesn't support )
With the huge number of calculating units used, you could possibly already prepare a few of them for the next vertex branching... Or something like that.
This is only speculation, but I find it quite likely.


Uttar

[SavagePaladin]

Originally posted by Nv40
it is the same issue with AMD and Intel ..

clock per clock AMD AtlonsXP processor are more faster than Intel
Pentiums4 .. just raise the clock of the atlons Xp processor to
3ghz! and it will easily smoke the P4 3ghz of Intel !

does that mean that Amd its the leader in Performance ?
Absolutely not. Intel is the King .

So Faster memory and higher clocks and Wider bus is part of the strategic design of Nvidia and ATi . the ones who get the best technology and hardware combination wins .

it will be interesting the year 2003 ,with ATi and Nvidia
next products.. :)


but i agree that only real benchmarks of the Nv30 retail
can say the truth above how well it fulfill Nvidia claims
in benchmarks.
Before I couldn't see the AMD vs Intel race as so clear cut...
Right now, though, I can see it as budget excellent performance vs. absolute excellent performance.

For a guy on a budget (ME) theres no way to beat AMD. Especially with my nForce shaving quite a bit off the price :D

[imtim83]

Doesn't matter if the Geforce FX won't be out till Feb. to April (maybe) because no games will take advantage of it till a few months to a year even. No one knows the real price yet of it offically from nvidia themselfs so we will have to see with that too. Remember the early benchmarks of the ATI 9700 pro on those alienware computers. It made the ATI 9700 pro seem not that fast but once the offical benchmarks came out from review sites it was a lot faster than those alienware ati 9700 pro benchmarks we saw before any of the others. Anything can happen. Time will tell.

[SocketAzzWipe]

Originally posted by Uttar
I'd guess it's more efficient for Dynamic Branching ( something the R300 doesn't support )
With the huge number of calculating units used, you could possibly already prepare a few of them for the next vertex branching... Or something like that.
This is only speculation, but I find it quite likely.


Uttar

This guy finds millions of flaws with R300, but fails to mention any with NV30...is NV30 supposed to be perfect in everyway? I guess we will witness the first fully functional 100 percent perfect piece of computer hardware in about 4 months time...

[gravioli]

Originally posted by thcdru2k
yes it is four times faster than the gf4 ti4600, but if you look at the nature scores, they are slower against the supposed nv30 nature test benches.

For a little comparison here is my Nature demo score (http://service.madonion.com/compare?2k1=5097014) run on my computer (see sig) with 4xAA and 8xAniso (quality). It's not the 40fps of the GeforceFX, but not too shabby IMO.

[StealthHawk]

Originally posted by SocketAzzWipe
This guy finds millions of flaws with R300, but fails to mention any with NV30...is NV30 supposed to be perfect in everyway? I guess we will witness the first fully functional 100 percent perfect piece of computer hardware in about 4 months time...

one reason for that is that like it or not, much of the NV30 is still shrouded in mystery.

one question here though, was the driver gfFX demoed on at Comdex the shipping driver?

[Bigus Dickus]

Originally posted by StealthHawk
one question here though, was the driver gfFX demoed on at Comdex the shipping driver?

One would certainly hope not. Though, as long as the NV30 has seemingly been delayed perhaps it's shipping drivers will be more mature than is normal for new generation products. Unless the absence of hardware is a large hinderence to bug fixing.

[thcdru2k]

looking at r300 scores right now compared to a gf fx. the improvment is less than 10%. i'm guessing, drivers could be immature. i think nv30 scores even though they are produced by nvidia will be higher than what was shown at comdex once people start putting it in their machines and overclock and tweak them.

[Nutty]

Unless the absence of hardware is a large hinderence to bug fixing.

They've had working silicon for nearly a year m8. Before that you had processor simulators. And even chips that are broken and not fully functional can still be used for testing purposes.

Believe me, nvidia doesn't have a shortage of hardware to make its drivers with.

[Bigus Dickus]

Originally posted by Nutty
They've had working silicon for nearly a year m8.

It seemed pretty clear that the NV30 did not tape out until late this year. There's no way they had working silicon for nearly a year, unless the NV30 underwent a complete design change, which seems very unlikely, and wouldn't be the same hardware as the current incarnation in any case.

[Nv40]

Originally posted by thcdru2k
looking at r300 scores right now compared to a gf fx. the improvment is less than 10%. i'm guessing, drivers could be immature. i think nv30 scores even though they are produced by nvidia will be higher than what was shown at comdex once people start putting it in their machines and overclock and tweak them.

yes but Nvidia have now finally and "adaptive" aniso .
like ATI high quality is , and a traditional one like the Geforce4 .

means that the NV30 will score much better in apples vs apples
comparisons . it is 8x "high quality" ATi vs 8x adaptive Nvidia .

if ,it is true that NAture test does not increase linear with higher cpus - its a Pixel shader benchmark)
a think it can be 20%-30% the diference in that benchmark.
with their first drivers set. something like 35 vs 50 more or less.

[StealthHawk]

Originally posted by thcdru2k
looking at r300 scores right now compared to a gf fx. the improvment is less than 10%. i'm guessing, drivers could be immature. i think nv30 scores even though they are produced by nvidia will be higher than what was shown at comdex once people start putting it in their machines and overclock and tweak them.

the answer could possibly be immautre drivers, but it could also be that the chip is severly bandwidth limited. then again, i don't see why nvidia would bother wasting time with such an extensive cooling solution as well as the power supply in order to ramp up clockspeed if it was going to end up being useless.

[StealthHawk]

Originally posted by Nv40
yes but Nvidia have now finally and "adaptive" aniso .
like ATI high quality is , and a traditional one like the Geforce4 .

means that the NV30 will score much better in apples vs apples
comparisons . it is 8x "high quality" ATi vs 8x adaptive Nvidia .

uh...so now ATI's AF is "high quality?" in the sense that you can set it higher with an ATI card than with an nvidia card (16x vs 8x) you are right.

personally i think that on equal settings nvidia's solution was superior in terms of IQ, but that is really not the point. so please, no flame wars over this.

now, the caveat is that gfFX will not support 16x, but only 8x. meaning that ATI still has the absolute image quality advantage thanks to supporting a higher degree of AF.