Ti4600 vs 9700 Pro: who's the image quality king in FS

[Guest WorkingStiff]

>Hey Stiffy, >>If you are not careful, you are going to dribble all over >yourself. So without looking like a complete idiot, why >don't you fill us in on some of that logic of yours? >>It is not unreasonable to think that the NV30 will best the >9700 X2 is some benchmarks. Some benchmarks....LOL...nice dodge.That's certainly NOT what you said originally. To quote you in post #4, "There were benchmarks out of Tawain recently on either Tomshardware.com or Anandtech.com that showed the NV30 having twice the performance of the 9700. No kidding."Even the charts you posted did not show "twice" the performance except for Pixel Shading, and you cannot authenticate the source of the charts.>Currently my card does not utilize anistrophic filtering, so >I was going to try and 1> see what it looks like when >implimented in FS2002 and 2> What kind of card I should >invest in before upgrading both my CPU and video card if I >decide to do those things. >Anyone with half a clue knows that Anistropic (note the spelling) filtering is implemented at the system level. So unless you're running some antique or non-standard video card, the function can be accessed either through the driver or a third-party tweaker.>So homey, if you are feeling froggy, jump on in on this >discussion. I'm willing to discuss this or any other topic >on any level you choose. **YAWN**

[Guest]

Remember that the NV30 is designed around a 0.13-micron manufacturing process, meaning that there's significant room for NVIDIA to include additional features in the chip as well as the ability to ramp up clock speeds well beyond that of the 9700. With the addition of clock speed up to 1G and possibly beyond, the NV30 will use DDR II memory which will act like a quad speed pump to the DDR's double speed. This is a similar type core as will be utilized by the upcoming Intel Prescott Pentium 4/5. As the article I referenced in a previous post noted the NV30 will "blow ATI's offering out of the water." Again, the foundry referrs to the casting of the chip, and by the middle of August when the first benchmarks were released, the NV30 obviously was already taped out. I agree. I think we have gone over all this before.

[Guest]

Hey Stiffy,Whaz up? Yes, I believe those benchmarks were leaked. If you want to get technical about it, the pixel shading benchmark you noted was about 200% greater for the NV30 than the 9700's. I guess I underestimated the NV30s actual performance, but, if you can infer, I was using recollection and did not have the information in front of me at the time I made the first post.As for my GeForce256 card, it currently does not utilize anistropic filtering with my current drivers nor do I have a third-party tweaker to help me impliment them. In addition, I am not interested in implimenting them on my current card because it is slow enough as it is. There is nothing logically inconsistent with my sematics here. I am not sure what you were getting at.In that article, I asked for examples to see the effects so that I could begin to research the issue of what card to upgrade to next. As for you statement, "Does anyone really think the nVidia NV30 will be twice as fast as a Radeon 9700? It's certainly not logical onclusion for a number of reasons..," <-- notice the spelling and the lack of an indefinite article, I think it will be for the following reasons:Remember that the NV30 is designed around a 0.13-micron manufacturing process, meaning that there's significant room for NVIDIA to include additional features in the chip as well as the ability to ramp up clock speeds well beyond that of the 9700. With the addition of clock speed up to 1G and possibly beyond, the NV30 will use DDR II memory which will act like a quad speed pump to the DDR's double speed. This is a similar type core as will be utilized by the upcoming Intel Prescott Pentium 4/5. As the article I referenced in a previous post noted the NV30 will "blow ATI's offering out of the water." Again, the foundry referrs to the casting of the chip, and by the middle of August when the first benchmarks were released, the NV30 obviously was already taped out. I agree. I think we have gone over all this before. See ya Spanky or Stiffy rather...

[JimmiG 244]

>Remember that the NV30 is designed around a 0.13-micron manufacturing process, meaning that there's significant room for NVIDIA to include additional features in the chip as well as the ability to ramp up clock speeds well beyond that of the 9700.>Not significant room. An additional 10M transistors, which isn't that much at all. Especially if they go with an 8x2 engine (vs R9700 8x1), they will already have used up much of the space. Clockspeed will be about 70MHz higher than the Radeon 9700.DDR-II is still DDR, which stands for Double Data Rate. It is tweaked to be able to reach higher clockspeeds (500+ MHz x2 in the case of NV30), which is the only major advantage it offers over conventional DDR memory. It is quad pumped internally, with half the data width externally. In addition to that, NV30 will only feature a 128-bit bus. So internally, four times the speed of DDR. Half of that is lost when it talks TO the memory bus, it's yet again cut in half when it reaches the narrow 128-bit bus of NV30. So out of your original vast bandwidth, you are left with about 16GB/sec.You really need to learn how to interpret your sources. The original article with the benchmarks was not posten ON or BY Anandtech. It was just a link to another site. It had nothing to do with Anandtech at all.Read the Inq article again:"DDR II memory at 1000MHz -- almost 400 MHz faster than the memory used on the Radeon 9700 PRO -- are key elements of the NV30's architecture that Nvidia expects will help it blow ATi's offering out of the water."It sais that "NVIDIA EXPECTS"..."WILL HELP"..it to blow ATI's offering out of the water. Nvidia EXPECTS...get it? It doesn't say it will blow the R9700 away. It sais Nvidia expects it will. What Nvidia expects is not relevant. What they expect is only what they HOPE will happen. They wouldn't say that they don't expect the NV30 to blow the R300 out of the water, would they?"Again, the foundry referrs to the casting of the chip, and by the middle of August when the first benchmarks were released, the NV30 obviously was already taped out. I agree. I think we have gone over all this before."Please no! There did not exist a NV30 when the benchmarks were published! Nvidia Codename NV30 taped-out somewhere between the 5th and 9th of August. From tape-out to having a working chip takes at least two weeks, and the benchmarks were originally posted on the 6th of August. It's not even certain that the A0 silicon will even work. If it didn't, it would have been another MONTH before there existed a working NV30.On top of that, the benchmark results are, like I've tried to explain before, COMPLETELY ABSURD.The Radeon9700 running Quake3 is CPU Limited. Even at 1280x1024 with 4x AA! It's completely and utterly impossible for any chip to score almost 600 FPS in this game. If you can't grasp why, I suggest you do some reading on how the CPU and videocard, as well as other subsystems like memory and AGP work together to draw the scene.It has to do with how Quake3, using an old 3d engine, is programmed so that it let's the CPU do most of the geometry calculations, while the videocard does the textures and lightmaps.The engine had some rudimentary support for T&L (or just the L if I remember correctly), but nothing that would offload the CPU to the extent that your "benchmarks" would suggest.http://firingsquad.gamers.com/hardware/r300/page21.aspWhy do you think the Radeon 8500LE and Geforce4 Ti4600 score virtually identical results in this benchmark in the lower resolutions? If there was a card that was twice as fast as the R9700, do you think it would score 224.2FPS at 1600x1200? It wouldn't! It would score about 115FPS just like all the other cards. do at the lower resolutions. A perfect example of a CPU limited game-engine.

[JimmiG 244]

Pointing out grammar and spelling errors on a messageboard is a very good way to look very bad.1. Most of us don't have time to read through our posts and correct every single grammar and spelling error before posting2. The Internet is international - many of us don't have english as our first language.3. Sooner or later, you will make an error in a post. If you have complained about other's spelling and grammar, you'll look very stupid.

[Guest]

The transistors you are referring to are mainly in the cache of of the videocard memory like they are on a CPU. The small number of transistors that additional features will require will be there. In addition, the NV30 will cut out a lot of redundant features of older cards. I can interpret my sources just fine. The NV30 benchmarks posted on Anandtech.com were secondary sources, I realize. Most of the benchmarks on Anandtech.com come from secondard sources. I suspect what happened is that after the chip taped out, a functional model was used to acquire those benchmarks and then were subsequently leaked. http://www6.tomshardware.com/graphic/02q2/...acharts-03.htmlDo you notice the variation in the frames per second using Quake III here? They were obtained by using the following system:AMD AthlonXP 2000+ASUS A7V266-EVIA KT266A133 MHz FSB256MB PC-266 MHz RAM CL2 (2x128MB)Because of the significant variation noted, that tells us that Quake is a very video card reliant program unlike our beloved FS2002. If FS2002 were used, you would not see nearly the variation noted in Quake because of FS2002's CPU dependence. There was once a website called FSBench.com and it had a number of videocards running FS2000. What it showed was the older cards (even PCI cards) out performing (albeit only slightly) the newer cards. This was direct evidence suggesting that the program, unlike using Quake, was very CPU dependent. As far as your comment, "It sais that "NVIDIA EXPECTS"..."WILL HELP"..it to blow ATI's offering out of the water. Nvidia EXPECTS...get it? It doesn't say it will blow the R9700 away. It sais Nvidia expects it will. What Nvidia expects is not relevant. What they expect is only what they HOPE will happen. They wouldn't say that they don't expect the NV30 to blow the R300 out of the water, would they?"I am sure Nvidia knows full well that the NV30 will blow the pants off of the 9700. You are getting caught up in the third party interpretations of possible events and events.I guess we will have to see when more benchmarks come out, but I suspect you won't believe them either...

[JimmiG 244]

So GPU's come with cache memory now? That's new to me :-lol I thought Bitboys were working on a design like that a couple of years ago, but it never turned into any real products.Exactly what significant additional features can you add with only a small number of transistors? An extra vertex shader? How about 4 times the number of maximum instructions in the vertex shader compared to R300? One more TMU/pipe? Most features will easily use up 10M transistors or more.They already dropped the fixed-function T&L engine when they moved to NV20 (NV20 emulates T&L with a vertex program and I think NV25 does that also). What features are they going to drop exactly, and how do they plan to make it compatible with older applications and the DirectX standard if they just drop some features that they don't consider important?The benchmarks in articles posted on Anandtech come from Anandtech's testing lab. Any articles they link to in the News section is external and has nothing to do with Anandtech. Is that what you meant?Also note the spelling error. This is why you should not point out errors in other's posts (yeah I know I just did just that..sorry :) )edit: I swear it comes right back at you :-lol I'm going to leave my grammar error uncorrected and let this be the perfect example of why you should not point out other's grammar and spelling errors.Actually I notice a surprisingly small variation. Ti4600 scores only about 10% better than Ti4200. In the Aquanox benchmarks in the same article, the difference is over 25%. 3DMark falls somewhere between Q3A and Aquanox with a difference of 11%. They made 3DMark 2001 so that it would also stress the CPU and memory subsystems, because 3DMark 2000 was often found to be too videocard dependant.Jedi Knight II is yet again extremely CPU dependant, with the entire GF4Ti family scoring virtually identical results. This game, being newer than Q3A, but using the same engine, features levels with more geometry, and since the CPU takes care of the geometry, it becomes overloaded and can't keep up with the videocard.You are correct in that FS2002 is another example of a very CPU dependant application, and unless you use lots of anisotropic filtering, FSAA and a very high resolution, most videocards would score identical results.There's no doubt that NV30 will be faster than the R9700. Nvidia are in a situation where they simply can't release the card unless it outperforms the Radeon 9700. If it doesn't, they'll tweak it until it does, before releasing it. How much faster it will be remains to be seen, but looking at leaked specs it is possible to guesstimate the performance.Development of NV30 and R300 started at about the same time, maybe 2 years ago. What was possible for Nvidia to implement was also possible for ATI to implement. The difference was that ATI decided to play it safe and go with a .15um design while Nvidia took the chance, hoping that the .13micron process would be ready in time. There are several articles on the advantages and disadvantages of each, and I suggest you read those. Going with a .13micron process does not autmoatically solve all the problems in the world.ATI recently announced the use of DDR-II and GDDR3 memory on their future products. With a memory bus twice as wide as NV30, ATI are in a very interesting position, being able to release a Radeon 9700 powered by DDR-II memory by the time Nvidia announce the NV30. ATI has also demonstrated a R300 using DDR-II memory.

[Guest]

My 9700 scored 276.5 on the Quake III test. The reason I went with the ATI this time was cause I've never owned an nvidia card that lived up to its hype. In other words, I've never owned an nvidia card where you could use all the features that came with it without killing the framerate. Until now. The 9700 is designed to run with everything turned on and the picture quality is top notch. About the time the NV30 comes out, ATI will be introducing its next card. Even if the NV30 lives up to the hype, it still won't be the king. ATI has the market cornered for quite a while IMHO. Course, competition is great. Us users end up the winners. Right now, the winner is ATI by a wide margin. We'll just have to wait and see if nvidia can close the gap. I hope they can!!