Some new benchmarks for NV30

[jAkUp]

uttar replied this info in another topic, but i think its worthy of a whole new topic :D

anyways there are some more benchies here:
they show nv30 vs. ati 9700 in unreal tourny 2003
http://www.forum-3dcenter.org/vbulletin/showthread.php?threadid=51204&perpage=20&pagenumber=1

its in german, but here is what it says:

XP 2600+, 512 MB RAM, Nforce2 chipset
UT 2003 Version 2166 Flyby
9700 Pro Ti-4600 FX Ultra
201 170 196 @ 1.024x768, 32 Bit
110 65 143 @ 2X FSAA + 4:1 AF


UT 2003 Version 2166 Botmach

9700 Pro Ti-4600 FX Ultra
71 71 68 @ 1.024x768, 32 Bit
65 65 65 @ 2X FSAA + 4:1 AF



looks like its on par with the 9700 in low resolutions (1024x768) im curious as to how it does in resolutions such as 1600x1200... one more bit, the guy also said the fan is still really noisy

[sebazve]

IMHO that performance kinda sucks...

[undercover]

Originally posted by sebazve
IMHO that performance kinda sucks...


Why ? It's 30% faster in the only benchmark shown that at all matters.

[noko]

Why not 4x AA? Hopefully this week we will see a better comprehensive suite of comparisons. I am interested in the 4x AA and 8x AF benchmarks, the higher IQ or better pixel setting tests. Oh, throw some 16bit ones in there as well :).

[netape]

Originally posted by noko
Oh, throw some 16bit ones in there as well :).

May I ask why are you so obsessed with 16 bit? :D

[noko]

LOL,, for my GF3 it worked well if you wanted both Max AF and AA on at the same time. IQ wise I concluded it was a good trade as well if using 32bit textures, anyways old new. Still a game that uses 16bit only, there are a few nowdays would be a good kicker for ATI because ATI official drivers fails to AA in 16 bit :D.

[The Baron]

Don't forget, botmatch is completely CPU-limited.

[undercover]

Originally posted by The Baron
Don't forget, botmatch is completely CPU-limited.

Exacty :)

So, given the benchmarks available so far, GFFX is looking very good.

*ducks and covers*

[UDawg]

You know what. Either way this works out it is still good new for us comsumers of video cards.

If the FX kicks butt, we get a awesome card.
If it doesn't, we get to buy a card that the price just droped on (9700) because the FX is $400.

All is good in videoland.

[Uttar]

xxx ( who got a 2100+ IIRC ) just did a similar UT2003 test with a Ti4600.
He got 40FPS in the botmatch & 60FPS in the flyby when using 2X AA & 4:1 Aniso.

So, as you can see, the botmatch score for the Ti4600 is exagerated. It's actually the same as for flyby.
I'd assume, thus, that the poster of those scores did an error and copied the flyby score of the Ti4600 on botmatch too. The real score is probably more around 45.

However, I'd suppose the GFFX & Radeon 9700 Pro are really on par there. I'd be surprised if it was all a big mistake and all of those botmatch scores were incorrect.


Uttar

[randsom]

first of all, the # for the ti4600 are way to high.Look at the numbers the ti4600 is actually faster at 1.024x768, 32 Bit .

the ansio numbers show a very cpu bound benchmark, but even though with cpu bound scores, they wouldnt be identical, different cards render diferently with the exact same cpu.

so something fishy with these #s.

let the flaming begin as to my next comment

if this is the actual performance of the fx its a big disappointment.
all the waiting and the delays, for a ccard thats just on par with a card thats been around for months.

before you do start flaming, i want to say that i was(am)going to buy geforce fx

[The Baron]

Originally posted by Uttar
However, I'd suppose the GFFX & Radeon 9700 Pro are really on par there. I'd be surprised if it was all a big mistake and all of those botmatch scores were incorrect.

What's to prevent the scores from being faked like 95% of other GFFX benchies?

[PreservedSwine]

I'd trust this source more than Anand and Tom combined. These are legit.

So far so good, I think Aniso on the NV30 is going to kick a** (finally!), but I'm still concerned about higher levels of FSAA and higher resolution combo's....

It will have a nice sweet spot, but I fear it may be a fairly unbalanced solution...

[noko]

Well considering these are on initial drivers and Nvidia has an excellent record of performance increases after a new GPU is released can only make the GF FX look better over time. My Tyan Tachyon 9700 Pro is working fine for me but I don't expect the drivers to improve all that much more. Supposenly ATI is putting in cpu optimizations now, which the CAT 3 has and the next version some more. Hope to see some awesome results this coming week when the reviews start hitting the street.

[savyj]

Well, here's an interesting perspective. Rumored sources at ATI are concluding that the R350 core VC's will beat the FX by 10%.

Hmmm...does this shed light on the FX performance?

R300 cards run at 325/310[DDR]
R350 cards (rumor) run at 400/400[DDR].

This is 23% faster core and 29% faster memory than the Radeon 9700 Pro. And this will propel the R350 to be 10% faster than the FX(?) Another source says this will allow the R350 to beat the FX in 8 of 10 "selected" benchmarks.

If these are accurate numbers (and this is a big if), all these "on par with the 9700 Pro" statements may be a bit off...unless of course the R350 is only 10% faster than the R300...or is this going to be the case?

[Lezmaka]

At least in UT2K3, a 9700 at 405/400 averages about 25% faster, most tests they ran got at least 20% increase.

http://www.xbitlabs.com/video/radeon9700pro-overclocking/

[Alphy]

Nice find, very good article...Gives a good look to as how the R350 will perform.

[noko]

In 3dMark2001se when I overclock my Radeon 9700 pro 10% I get a 7% increase in benchmark with GPU limiting conditions (4x AA and 8x AF). So a 400mhz R-300 is only a 23% clock rate increase which if the memory isn't overclocked the 23% as well would probably make 3dMark2001 not go up as much as the core speed went up.

In any case if the R-350 is at 400mhz (a 23% faster clock over the R300) and the core hasn't significantly improve efficiency (which I doubt) then I would say the R-350 will probably give no more then 20% increase. I think the R-350 changes are mostly there to get the core speed up and to give a higher yield vice a major improvement in rendering efficiency. This would put the GF FX inbetween the R-300 and R-350 which does sound about right being the GF FX Ultra beats out the R-300 series but slower then the R-350 on average. This is general and I believe certain tests each will have an advantage except the R-300 over the R-350.

[StealthHawk]

Originally posted by noko
Still a game that uses 16bit only, there are a few nowdays would be a good kicker for ATI because ATI official drivers fails to AA in 16 bit :D.

it can do FSAA in 16bit for all D3D games.

[savyj]

Ok, so here's some simple math.

R300 = 100 = baseline
R350 = 100 * 1.25 (25%) = 125
FX * 1.1 = 125
FX = 125 / 1.1
FX = 113.6

This really puts the FX in the same ball park as both the R300 and the R350. Since I believe NVIDIA is still better at crafting video drivers (could change) and NVIDIA is probably going to try to leverage some optimizations in a few popular upcoming games, I think I've just about made up my mind. Only 22 hours and 53 minutes till Monday (MST) :)