nVidia's Jen-Hsun Huang speaks out on X-Box 2 possibility.
- Thread starter Brimstone
- Start date
Oh boy where to start with this . FIrst off you should go check out a sound file of what the geforce fx sounds like. Its louder than a delta black label. Second of all the geforce fx uses top of the line ddr2 ram and that equals alot of money where as ati uses older slower ram and a 256bit bus. The question is which one is more expensive. But lets just say they are both cutting edge and prob about the same price. Then there is the 10 layer board for the r300 vs the 12 layer for the nv30. Last but not least that massive cooler.
If you go to anands and you check out benchmarking of the two cards with aniso and fsaa you will see that in 5 out of 6 tests the radeon smacks down the geforce fx. Code creatures is a special case as it was programed on a geforce 3 and has run slower on radeon cards since the benchmark came out.
On doom 3. Carmack said that the geforce fx while using the nv30 path is faster than the radeon using the arb path (I believe that is the correct name). He also stated that while both using the standerdized extensions (arb) The r300 is twice as fast the nv30. The last thing he said is that when using the r200 path the r300 is faster than using the arb path. Which basicly means there is no optimized path for the r300 to make it look good. He didn't have to go in and program a whole new path just to get decent performance out of it.
All this is a moot point though as the r350 will be out in a month or so and that is where nvidia is screwed. The r350 will be faster unless an act of god does something to the chip. The nv30 will have to drop in price which will hurt margins and it will not b e the highest performance card out there. That will hurt sales esp since the fx takes two slots and sounds like a jet plane .
Okay i just looked over your post and see you mention that no one cares about aa and what not . More polygons aren't going to matter anymore. Next gen where are going to be look at close to a billion polygons of which prob not even a 100million whill be used. Eye candy is going to be important and if card a can do a game at 640x480 with 4xfsaa and card b can do it with 8xfsaa you can bet which one will be more sought after.
If you go to anands and you check out benchmarking of the two cards with aniso and fsaa you will see that in 5 out of 6 tests the radeon smacks down the geforce fx. Code creatures is a special case as it was programed on a geforce 3 and has run slower on radeon cards since the benchmark came out.
On doom 3. Carmack said that the geforce fx while using the nv30 path is faster than the radeon using the arb path (I believe that is the correct name). He also stated that while both using the standerdized extensions (arb) The r300 is twice as fast the nv30. The last thing he said is that when using the r200 path the r300 is faster than using the arb path. Which basicly means there is no optimized path for the r300 to make it look good. He didn't have to go in and program a whole new path just to get decent performance out of it.
All this is a moot point though as the r350 will be out in a month or so and that is where nvidia is screwed. The r350 will be faster unless an act of god does something to the chip. The nv30 will have to drop in price which will hurt margins and it will not b e the highest performance card out there. That will hurt sales esp since the fx takes two slots and sounds like a jet plane .
Okay i just looked over your post and see you mention that no one cares about aa and what not . More polygons aren't going to matter anymore. Next gen where are going to be look at close to a billion polygons of which prob not even a 100million whill be used. Eye candy is going to be important and if card a can do a game at 640x480 with 4xfsaa and card b can do it with 8xfsaa you can bet which one will be more sought after.
BenSkywalker said:
Quieter than a human heartbeat? Where did you read that? You posted links to AnandTech yourself, he talks about volume much louder than a human heartbeat. And Tom's Hardware has MP3's of both the 9700 Pro and FX. The FX is WAY louder... and sounds like it 'whines' rather than hums which ends up being much more annoying even at the same volume.
How about 3DMark2001's Nature test? The Radeon whoops the FX in that one. The Radeon also doubles the FX's Advanced Pixel Shaders speed.
1024 * 768 = 786,432
1280 * 1024 = 1,310,720
1920 * 1080 = 2,073,600
1600 * 1200 = 1,920,000
1080i has more pixels than the maximum sane PC resolution. Between 10x7 and 12x10? That's a laugh. Learn to multiply.
FX's theoretical max at 500MHz is 350MPoly/s, R300's at 325MHz is 325MPoly/s. Not very far apart. Although yes with 8 lights the FX is more efficient...
ATi's RV350 is .13u, it's taped out, and reportedly it'll be widely available before the .15u R350.
And yet the FX needs a 12 layer PCB while the 9700 line needs 10 layers.
You forget that even different cores by the same IHV will handle things slightly differently. To-the-metal GeForce3 code probably wouldn't run quite right on a GF4, and definitely wouldn't run right on an FX...
And even when the core is 'reasonably obsolete' there's still nVidia confidential tech in there which they still use and don't want ATi to see.
No argument here.
CaptainHowdy
Regular
unless a human heartbeat sounds like a Shopvac on steroids, I dont get it either.
BenSkywalker
Regular
For noise levels- http://www.gainward.se/news/030130.pdf
Further reference-
http://www.anandtech.com/news/shownews.html?i=18126&t=pn
Tag-
I am assuming that is a joke and you are just fooling around, with the amount of times it has been discussed about the differences between interlaced and progressive modes there is no way in hell you could not be aware of just how wrong your numbers are.
I guess 'whoops' means something different where I come from, at the very least it includes being faster-
http://www.hardocp.com/image.html?image=MTA0MzYyMDg1OTVjVVNkMzFISXhfMl80X2wuZ2lm
9700Pro losing at every resolution is whooping on the GFFX? I guess you could say the R300 whoops the GFFX at CodeCreatures too. And the GFFX whooped the R300 hard in terms of getting to the market using your guidelines
Heard the same thing about the GFFX months before it was.
If nV went in knowing what they needed to do it would, which they obviously would be...
If it could land them a potential $100Million+ contract they may be willing to let a few things slide. Not to mention, when the patents come through admitting to exacting details on certain parts doesn't become all that important.
JVD-
RAM drops in price very quickly, added complexity for the memory bus doesn't.
The board that has been the lead for Doom3 development recently has slower code running on it then one that Carmack just got his hands on. If you want to get in to the ARB path and the features it doesn't address that the NV30 has and how that could give a board with less flexibility working on the current, non final, ARB path which is missing components that the final version will include that the NV30 build currently supports then that is another matter entirely.
What exactly is meant by 'be out'? Paper launch or retail availability? Not saying it won't be retail, but I don't recall the last cycle product(top tier) that hit store shelves within a month of the tecnology announcement(paper launch). This is not criticism, I'm simply stating that all the vendors have been doing a paper launch a month or more in advance of shipping and we have no paper launch on the R350 yet.
Could you point that out, anywhere? I said there wasn't much hype around the anisotropic performance characteristics of the console. Since you didn't make the connection, the XBox stomps the PS2 in texture filtering by a much larger margin then it does in poly throughput and yet we still didn't hear any hype about it. You know why that is? 32tap anisotropic with trilinear filtering at ten times the speed of the competition doesn't quite roll off people's tongues the way 100Million polys/sec does. AA performance is obviously going to be very important next generation, a discussion we have had numerous times here(and in general, the likely narrow gap between any of the platforms in terms of visuals due to limitations increasingly moving to artists and their budget instead of technology).
Simon-
That is true but given the lack of eDRAM or any other details indicating a potential for higher density areas of transistors on the R300 you could just as easily assume that nVidia has a far larger edge then the finer build process would indicate.
Further reference-
http://www.anandtech.com/news/shownews.html?i=18126&t=pn
Tag-
I am assuming that is a joke and you are just fooling around, with the amount of times it has been discussed about the differences between interlaced and progressive modes there is no way in hell you could not be aware of just how wrong your numbers are.
I guess 'whoops' means something different where I come from, at the very least it includes being faster-
http://www.hardocp.com/image.html?image=MTA0MzYyMDg1OTVjVVNkMzFISXhfMl80X2wuZ2lm
9700Pro losing at every resolution is whooping on the GFFX? I guess you could say the R300 whoops the GFFX at CodeCreatures too. And the GFFX whooped the R300 hard in terms of getting to the market using your guidelines
Heard the same thing about the GFFX months before it was.
If nV went in knowing what they needed to do it would, which they obviously would be...
If it could land them a potential $100Million+ contract they may be willing to let a few things slide. Not to mention, when the patents come through admitting to exacting details on certain parts doesn't become all that important.
JVD-
RAM drops in price very quickly, added complexity for the memory bus doesn't.
The board that has been the lead for Doom3 development recently has slower code running on it then one that Carmack just got his hands on. If you want to get in to the ARB path and the features it doesn't address that the NV30 has and how that could give a board with less flexibility working on the current, non final, ARB path which is missing components that the final version will include that the NV30 build currently supports then that is another matter entirely.
What exactly is meant by 'be out'? Paper launch or retail availability? Not saying it won't be retail, but I don't recall the last cycle product(top tier) that hit store shelves within a month of the tecnology announcement(paper launch). This is not criticism, I'm simply stating that all the vendors have been doing a paper launch a month or more in advance of shipping and we have no paper launch on the R350 yet.
Could you point that out, anywhere? I said there wasn't much hype around the anisotropic performance characteristics of the console. Since you didn't make the connection, the XBox stomps the PS2 in texture filtering by a much larger margin then it does in poly throughput and yet we still didn't hear any hype about it. You know why that is? 32tap anisotropic with trilinear filtering at ten times the speed of the competition doesn't quite roll off people's tongues the way 100Million polys/sec does. AA performance is obviously going to be very important next generation, a discussion we have had numerous times here(and in general, the likely narrow gap between any of the platforms in terms of visuals due to limitations increasingly moving to artists and their budget instead of technology).
Simon-
That is true but given the lack of eDRAM or any other details indicating a potential for higher density areas of transistors on the R300 you could just as easily assume that nVidia has a far larger edge then the finer build process would indicate.
Ben - Yes ram does drop in price very quickly. But lets not forget that right now nvidia is the only one buying this ram which will slow the rate at which its price drops. Also in a very short time faster cards from ati will be out and will force the price of the fx down quickly.
On to what carmack said . Well that is just what he said , the fx is the fastest while runing the nv30 path with the radeon 9700 slower in most cases but pulls ahead sometimes whie it is runing the ARB path. The radeon9700pro is twice as fast as the nv30 while both are running the ARB path. THe radeon is faster whie running the r200 path over the ARB path. The ARB is the highest quality path made. What i take from this is carmack couldn't get good performance out of the nv30 . So he made its own path. The radeon could use the bestquality one and ran it fast ... almost as fast as the nv30 runs its special code.
Well what imean by out is bench marks on the web. Since i was in compusa today and saw no geforce fxs and the one i ordered on the net hasn't shipped yet . Come on if ati released cards to anands and it showed it spanking the fx by 30% in ever bench mark and was the same price who would by the fx. Its like buying a 2003 model car for 20grand when next years model will be out in a month for 20grand
[/b]
On to what carmack said . Well that is just what he said , the fx is the fastest while runing the nv30 path with the radeon 9700 slower in most cases but pulls ahead sometimes whie it is runing the ARB path. The radeon9700pro is twice as fast as the nv30 while both are running the ARB path. THe radeon is faster whie running the r200 path over the ARB path. The ARB is the highest quality path made. What i take from this is carmack couldn't get good performance out of the nv30 . So he made its own path. The radeon could use the bestquality one and ran it fast ... almost as fast as the nv30 runs its special code.
Well what imean by out is bench marks on the web. Since i was in compusa today and saw no geforce fxs and the one i ordered on the net hasn't shipped yet . Come on if ati released cards to anands and it showed it spanking the fx by 30% in ever bench mark and was the same price who would by the fx. Its like buying a 2003 model car for 20grand when next years model will be out in a month for 20grand
[/b]
BenSkywalker said:
I'll believe Gainward's claims when I see it. They're an exception.
You really think in interlaced the core does half the pixels or something?
The core is still rendering (hopefully) 60 fields per second at 1920x1080.
I guess 'whoops' means something different where I come from, at the very least it includes being faster-
http://www.hardocp.com/image.html?image=MTA0MzYyMDg1OTVjVVNkMzFISXhfMl80X2wuZ2lm
9700Pro losing at every resolution is whooping on the GFFX? I guess you could say the R300 whoops the GFFX at CodeCreatures too. And the GFFX whooped the R300 hard in terms of getting to the market using your guidelines
My mistake, I never read H's preview, I thought I saw 3DMark2001 Nature results at AnandTech but I was wrong.
About the advanced pixel shaders though:
http://www.anandtech.com/video/showdoc.html?i=1779&p=6
EMBM, Vertex Shader, and Advanced Shader are the Radeon's domain...
And it isn't far behind at DOT3 and Pixel Shader.
Curiously enough it looks to me like the GF FX rocks the house at fixed-function work but at this point pretty much keels over and dies the moment shaders are enabled... could have something to do with FP32 but I'm not sure.
You cut out the second half of what I said.
------------
Well I guess they could pull a GF FX and have a full fixed-function pipeline alongside the shader pipe...
Revealing core functions could lend them a contract somehow? Explain?
And I'm sure it's already patented, silly, this is an NV2x core we're talking about. It's still proprietary nVidia tech.
Tagrineth said:
Wait a second. EMBM is the Radeon's domain (R300 + 3fps), but it "isn't far behind at DOT3 (NV30 + 17fps) or Pixel/Frag Shading (NV30 + 23fps).
I also fail to see how the Vertex Shader of the R300 is superior. It might be faster; but architecturally, the NV30 owns it in elegence.
Steve Dave Part Deux
Newcomer
To be valid, architectural comparisons between two cards must be done clock for clock. Clock for clock R300 is significantly faster.
CaptainHowdy
Regular
yup, and being the R350 will be out first(and Rumors of Nvidia pulling out on the NV30 are all over right now) , its no contest.Steve Dave Part Deux said:
Steve Dave Part Deux said:
Bollocks.
Pipeline length influence clock speed. And pipeline length is a (micro) architectural parameter.
Longer pipeline == shorter individual pipestages => higher clock speed.
A valid comparison must be done in the same process and have similar die size.
In real life we'll *always* be comparing apples to oranges. In the end the only valid comparison is that of the market.
Cheers
Gubbi
Gubbi said:
well Gubbi. If chip a does 3million polygons at 325mhz and card b does 3million polygons at 400mhz . Then chip a is faster than chip b at doing polygons. Simple logic would tell you that if chip a is upped to 400mhz it would do more than 3 million polygons and if chip b is reduced to 325mhz it would do less than 3 million polygons.
jvd said:
You missed the point completely. Because chip a is made with a shorter pipeline, it can't clock at 400MHz.
Show me a Radeon 9700 clocked at 400MHz
Show me a Radeon 9700 Pro clocked at 500MHz
And overclocks don't count.
This is *exactly* the same discussion the CPU geeks have over Athlon vs. P4
You compare products on absolute performance, not clock speed
Cheers
Gubbi
Gubbi said:
Exactly, well said.
Or I suppose that when comparing the R300 against the Parhelia - they could be forced to disable 4 of the R300s pipelines as it skews its per unit preformance and they should be benched with equal standings. Except people don't realise that just as the number of computational units in a design is a function of the overall preformance, so is clockspeed.
