The difference between a synthetic benchmark and a real game is that the real game has all sorts of other things going on. It has to handle user input, much more extensive collision physics, etc.
Yes, that is the difference between a synthetic benchmark and a real game; I never stated otherwise. The problem is, since real games are taking in user input, the variables of the test are constantly being changed and there is no way to duplicate the results a second time on another graphics card. As for "extensive collision physics", that observation invalid since there is no reason a "synthetic benchmark" cannot have much more "extreme collision physics" than a real game, and in fact many do; in addition, "extreme collision physics" doesn't affect the graphics card in the least since physics calculations are done on the CPU and not the VPU/GPU.
Almost all reviewers run timedemos that are almost always created by the game manufacturer, are constantly being used, and have been available for a long time; they then proceed to tout these as "real benchmarks" all the while discrediting 3DMark03 because they are too moronic to understand that in reality there is no difference between the two.
It's much harder to cheat at real benchmarks because they're not static.
If it is not static, then how can the results be duplicated? If one were to do as you propose, which I might add almost no one does, there would be no way of obtaining constant results, let alone verifying the validity of those results. No one uses dynamic benchmarks; they all use prerecorded timedemos.
Most (good) review sites can and do come up with their own demo scripts for popular games.
First, if they come up with a script, then they are not playing a real game, with constant user input, as you stipulated in the previous part of your post. Second, aside from Firingsquad.com, no other review site has recorded their own timedemos, to the best of my knowledge. I might add that Firingsquad has only done so once, and that was in their recent review; unfortunately, no site has yet to implement the procedures I described in one of my previous posts, which would appear to be the most logical course of action.
Just watch the demo, take screenshots, and see if you notice anything odd.
Really? Again, this works if the demo was to be recorded just before the review and it was never publicly available beforehand. As most reviewers use prerecorded timedemos which came with the games themselves, this belief does nothing but add a false sense of security. As I've stated many times in the past, and will continue to state till it is drilled into people's heads, there is no difference between prerecorded timedemo that has been available to the graphics chip manufacturers, and 3DMark03. Since both are static, things such as static clip planes can be inserted by say nVidia without anyone noticing, as there is no way of spotting a static clip plane in a benchmark that is on rails; the only reason nVidia's "tinkering" was exposed in 3DMark03 was because there are versions of 3DMark03 available to certain parties which allow one to move the camera while the demo is running. This, of course, was unbeknownst to nVidia; fortunately for them, no such tools are available in "real benchmarks," so they are free to insert static clip planes with impunity, as "watch[ing] the demo, [and] take[ing] screenshots" cannot be used to expose well-implemented cheats in prerecorded demos, just as they could not be used in 3DMark03.
If you don't notice anything odd, then even if the manufacturer did do something, its an optimization rather than a cheat.
Really? So static clip planes inserted, by say nVidia, at arbitrary points in the benchmark just so that their bloody video card has to do less rendering, are optimizations, and not a cheat? And keep in mind that static clip planes are not exclus
9800 has a faster transform engine, is slightly ahead at lower resolutions.
5900 has a higher fill rate, is slightly ahead at high resolutions.
I'm afraid, my dear chap, that matters are by no means as simple as you present them to be, and attempting to simplify them as such is but an affront to the intelligence of all.
For one thing, the 9800 doesn't have a T&L engine in the manner in which you are thinking of it. Anyways, this is irrelevant, for attempting to dissect a video card in such a manner is a futile effort at best (and moronic at worst); the best course of action is to interpret the information as it is presented and not attempt to extrapolate from it information which cannot be extrapolated.
As for your generalization of which card wins at which resolution, one needs look only at the second page to reveal the falsity of your statement.
Finally, the 5900 does some wonky things, therefore having a varying fill rate at different stages, while the 9800's remains constant throughout; one needs look only at the first page of the review to derive this information. Again, all that matters is the performance that is resultant of the architecture, and not the specs of the architecture.
I fail to see the purpose your post serves, for any moron with half a brain who could be bothered to read the bloody article (behaviour which should certainly by encouraged since it leads to a much more intelligent discussion, as well educating the reader) would be able to interpret the information much more accurately than the excuse that is your post.
In all fairness, while a collection of reviews, for there are plenty by this time, would have been preferable, as far as individual reviews go, OCAddiction's review is quite decent. While the manner in which the article arrived on Slashdot is dubious, the article should be judged on its own merit, relative to other video card benchmark articles.
I guess I wouldn't be as pissed if it was a genuinely interesting article, rather than a collection of specs and benchmarks.
As opposed to all the other video card reviews, which are brimming with titillating information? How would you rather the article was written, if not with "specs and benchmarks?"
Alas, the article is as it is, much like its brethren, intended for education, and not necessarily entertainment, although the two are by now means mutually exclusive.
Unfortunately, although the results might be unfairly tipped in nVidia's favour (even though they don't come out on top regardless), if we remove benchmarks all we have left is a guy playing the game under constantly varying circumstances and deriving such results as: "Card X seems to be hyper-fast, while card Y appears to be merely ultra-fast."
All benchmarks done using long established timedemos that have been available to nVidia for a while and are frequently used should be taken with a grain of salt. The best solution at the moment would be for reviewers to record their own demos at the beginning of each benchmark, benchmark the cards, and then released the recorded demo into the public domain so that the user base can verify the results; after all, while nVidia (and ATi) are not to be trusted, neither are reviewers, especially with the likes of Kyle of [H]ardOCP calling themselves as such.
I learned a long ago, from PC Mags 3d benchmarks, that synthetic benchmarks are absolutely useless!
And what exactly differentiates a real benchmark from a synthetic benchmark? While Futuremark does report the fill rate (both single-texturing and multi-texturing), it is simply extraneous information, which is in no way used to determine the resulting 3DMark score; the score is determined by running four game demos, which use engines akin to those used in "real games." The individual game results are reported by 3DMark, multiplied by certain coefficients, and then added together, rendering the result (3DMarks).
The reason 3DMark03 is invalid is not because it is a "synthetic" benchmark, but because nVidia mucked it up with their shenanigans. The frightful truth of the matter is, however, that the same illegitimate "optimizations" (i.e. static clip planes) that were used by nVidia in 3DMark can just as easily be used in any and all timedemo. Hence, your precious "real" benchmarks are just as susceptible, and may be just as compromised and invalid as 3DMark03. To make matters worse, unlike 3DMark03, which offers advanced diagnostic tools that allowed nVidia's dubious actions to be exposed, "real" benchmarks have no such tools. Therefore, exposing cheating in "real" benchmarks is much more difficult; however, just because something cannot be proven does not make it false.
You are of course quite correct. Kyle is merely an idiot with a website, as are many others these days, who is acting as nVidia's PR puppet. He has been ridiculed by many, yet seems intent on maintaining his wicked ways. Unfortunately, thousands of less informed internet users frequent his site, propagating his misinformation.
3DMark2003 added unfair optimizations to their program to make the nvidia card seem better than ATi's
3DMark2003 (or rather Futuremark, since I doubt the program is advanced enough to program itself) did no such thing. nVidia did it all by themselves, and Futuremark conducted an investigation confirming the "optimizations" (cheats i.e. static clip planes inserted by nVidia) and denouncing them. Google for more.
Second, observe the origin as well as the scale of the 3DMark03 graphs: Graph 1, Graph 2
The difference is grossly exaggerated by the graph's peculiar origin (5700 and 3800 instead of 0) and large scale.
Third, 3DMark03 has been rendered an useless benchmark since it is riddled with nVidia "optimizations," which have been deemed illegitimate by Futuremark's own accord. Even the author of the article acknowledges the dubious nature of 3DMark03:3DMark 03 Build 320 - FX 5900 Ultra - Should we include this? Possibly not, however the FX 5900 wins with WHQL Det Drivers
New nVidia optimizations (read: cheats) such as the Anistropic Filtering "optimization," which was exposed by renaming the 3DMark03 executable. Of additional note is that one of the 3DMark03 benchmarks was done using 8x Anisotropic Filtering, deeming it even more illegitimate than the rest of the 3DMark03 benchmarks in the article.
Fourth, the Radeon delivers consistently better image quality throughout, as was acknowledged by the author of the article: 3D Visual Quality - R9800 Pro hands down.
Another thing to be kept in mind is that the Radeon may very well do much better relative to the FX 5900 in all benchmarks since nVidia may very well optimizing for all major timedemos (3DMark03 style) and the reviewed did not record his own demos. Feel free to google if you find you would like more information.
I think he was referring to the fictional computer from The Hitchhiker's Guide to the Galaxy after which the chess computer was named, not the chess computer itself.
"funnily" WTF? You make an informative post, but destroy it by making up words. WTF? No, really, did you think that "funnily" is a word? Please tell me you are not that stupid.
First, all these signs are more or less retarded, but the right sign in the first picture must be one of the stupidest things I have ever seen. It states "Give communism a try! - Free
Linux," but also contains Tux hailing in the manner of a Nazi and stating "Ya Vol!" (sic).
How retarded do these people have to be in order to mix up communism and Nazism, which are polar opposites.
I never meant to imply that nForce boards had fw 800; the first part of my comment was only in relation to hypertransport. Upon rereading my comment, I realized it was a bit ambiguous, and considered posting a clarification, although I hoped that the last part of the comment: "Furthermore, if there is a demand for fw 800 on PCs, they will have it..." (emphasis added)
would dispel any misconceptions regarding the presence of fw 800 on current nForce motherboards. I hope this post has resolved any misunderstandings.
Which wintel motherboards have fw 800 and hypertransport? I'd be interested.
I wasn't really going to comment on this, but since it got modded up 5, Insightful...
nForce, nForce2
You might be further interested in knowing that hypertransport was primarily designed by AMD and is used in all of their Opteron systems, and will also be used in all Athlon64 systems. I guess that's not wintel per se... but it's a PC motherboard nonetheless.
Furthermore, if there is a demand for fw 800 on PCs, they will have it... a small upside of not being at the whim of a single company *cough*Apple*cough*.
Almost as interesting as the article itself is the fact that timothy has finally attained a 1:1 mispelled to correctly spelled word ratio. This is clearly a monumental achievement as it indicates that he is well on his way to reaching the golden ratio (1.618): the ratio at which we will find all of timothy's posts extremely attractive.
Everyone's rooting for you timothy. Don't give up.
I have to admit, I knew the horse-power game was one that Nvidia was going to trip over eventually when the Radeons were *really* pushing the Z-axis occlusion.
That is faulty logic; just because ATi was using Z-axis occlusion and ATi beat nVidia (which is also using z-occlusion nowadays) does not mean that that is why they beat nVidia. The Kyro II used tile based deferred rendering which was far superior to the z-occlusion ATi and nVidia were and are still doing, and look where PowerVR is today.
ATI finally got their hardware on track not by being able to "outshow" Nvidia's muscle, but by outclassing them by being smarter with the bandwidth and 'getting a clue' with drivers.
If being "smarter with the bandwidth" was all that was required, we'd all be using Kyro IIs. Making a successful GPU is not so black and white...
If it wasn't cheating, and they were only looking at shader instructions, then why did the results change when they renamed the file from Quake3.exe to quack3.exe? It is also important to note that the Quake3.exe results were of less quality than the quack3.exe results.
Um... he was talking about 3dMark, not the Quack issue (btw, there are no shaders in Quake3), which although people insist on constantly bringing, up was most likely not a cheat; ATi not only fixed the 5 problematic textures, but also increased the performance past the level of performance attained using the "cheat," in the next driver release.
The Opteron will be available in a desktop version. The Xeon is not, and will not be.
How so? The Opteron is to the Athlon64 ("the desktop version") as the current-generation Xeon is to the Pentium 4. In fact, the Pentium 4 is much closer to the Xeon than the Athlon64 is to the Opteron.
Tom didn't enable the second memory channel
The Athlon64 will not have a second memory channel regardless...
The Opteron has an integrated memory controller. Therefore, its "Front Side Bus" runs at the same speed as the actual processor.
Proof: With the memory controller integrated onto the AMD Opteron processor, the front side bus (interface to memory) runs at the speed of the processor
Cg is not an API, it is a programming language. It compiles to Direct3D and OpenGl. As for proprietary GL extensions, ATi has EXT_VERTEX_SHADER and ATI_FRAGMENT_SHADER.
nVidia is not ahead of ATi at all. nVidia and ATi both manufacture their chips at TSMC. The supposed reason that it took nVidia so long with the FX was that TSMC was having trouble with the 0.13 micron process. Not that they ironed out all the problems, there is no reason they can't manufacture ATi's chips using the same process.
In fact, considering that nVidia's 0.13 micron card runs hotter and slower than nVidia's previous generation 0.18 micron card (I'm not even talking about the 9800 here) really shows that nVidia is really behind this round.
Cg ("Their own graphics language") is a programming language.
Glide was an API.
Cg compiles to OpenGL or Direct3D, and it could very well be used on ATi cards with slight modifications. The two are nothing alike.
Slashdot Mirror
For those of you that missed it, the comic is in reference to this article.
The difference between a synthetic benchmark and a real game is that the real game has all sorts of other things going on. It has to handle user input, much more extensive collision physics, etc.
Yes, that is the difference between a synthetic benchmark and a real game; I never stated otherwise. The problem is, since real games are taking in user input, the variables of the test are constantly being changed and there is no way to duplicate the results a second time on another graphics card. As for "extensive collision physics", that observation invalid since there is no reason a "synthetic benchmark" cannot have much more "extreme collision physics" than a real game, and in fact many do; in addition, "extreme collision physics" doesn't affect the graphics card in the least since physics calculations are done on the CPU and not the VPU/GPU.
Almost all reviewers run timedemos that are almost always created by the game manufacturer, are constantly being used, and have been available for a long time; they then proceed to tout these as "real benchmarks" all the while discrediting 3DMark03 because they are too moronic to understand that in reality there is no difference between the two.
It's much harder to cheat at real benchmarks because they're not static.
If it is not static, then how can the results be duplicated? If one were to do as you propose, which I might add almost no one does, there would be no way of obtaining constant results, let alone verifying the validity of those results. No one uses dynamic benchmarks; they all use prerecorded timedemos.
Most (good) review sites can and do come up with their own demo scripts for popular games.
First, if they come up with a script, then they are not playing a real game, with constant user input, as you stipulated in the previous part of your post. Second, aside from Firingsquad.com, no other review site has recorded their own timedemos, to the best of my knowledge. I might add that Firingsquad has only done so once, and that was in their recent review; unfortunately, no site has yet to implement the procedures I described in one of my previous posts, which would appear to be the most logical course of action.
Just watch the demo, take screenshots, and see if you notice anything odd.
Really? Again, this works if the demo was to be recorded just before the review and it was never publicly available beforehand. As most reviewers use prerecorded timedemos which came with the games themselves, this belief does nothing but add a false sense of security. As I've stated many times in the past, and will continue to state till it is drilled into people's heads, there is no difference between prerecorded timedemo that has been available to the graphics chip manufacturers, and 3DMark03. Since both are static, things such as static clip planes can be inserted by say nVidia without anyone noticing, as there is no way of spotting a static clip plane in a benchmark that is on rails; the only reason nVidia's "tinkering" was exposed in 3DMark03 was because there are versions of 3DMark03 available to certain parties which allow one to move the camera while the demo is running. This, of course, was unbeknownst to nVidia; fortunately for them, no such tools are available in "real benchmarks," so they are free to insert static clip planes with impunity, as "watch[ing] the demo, [and] take[ing] screenshots" cannot be used to expose well-implemented cheats in prerecorded demos, just as they could not be used in 3DMark03.
If you don't notice anything odd, then even if the manufacturer did do something, its an optimization rather than a cheat.
Really? So static clip planes inserted, by say nVidia, at arbitrary points in the benchmark just so that their bloody video card has to do less rendering, are optimizations, and not a cheat? And keep in mind that static clip planes are not exclus
9800 has a faster transform engine, is slightly ahead at lower resolutions. 5900 has a higher fill rate, is slightly ahead at high resolutions.
I'm afraid, my dear chap, that matters are by no means as simple as you present them to be, and attempting to simplify them as such is but an affront to the intelligence of all.
For one thing, the 9800 doesn't have a T&L engine in the manner in which you are thinking of it. Anyways, this is irrelevant, for attempting to dissect a video card in such a manner is a futile effort at best (and moronic at worst); the best course of action is to interpret the information as it is presented and not attempt to extrapolate from it information which cannot be extrapolated.
As for your generalization of which card wins at which resolution, one needs look only at the second page to reveal the falsity of your statement.
Finally, the 5900 does some wonky things, therefore having a varying fill rate at different stages, while the 9800's remains constant throughout; one needs look only at the first page of the review to derive this information. Again, all that matters is the performance that is resultant of the architecture, and not the specs of the architecture.
I fail to see the purpose your post serves, for any moron with half a brain who could be bothered to read the bloody article (behaviour which should certainly by encouraged since it leads to a much more intelligent discussion, as well educating the reader) would be able to interpret the information much more accurately than the excuse that is your post.
In all fairness, while a collection of reviews, for there are plenty by this time, would have been preferable, as far as individual reviews go, OCAddiction's review is quite decent. While the manner in which the article arrived on Slashdot is dubious, the article should be judged on its own merit, relative to other video card benchmark articles.
I guess I wouldn't be as pissed if it was a genuinely interesting article, rather than a collection of specs and benchmarks.
As opposed to all the other video card reviews, which are brimming with titillating information? How would you rather the article was written, if not with "specs and benchmarks?"
Alas, the article is as it is, much like its brethren, intended for education, and not necessarily entertainment, although the two are by now means mutually exclusive.
Unfortunately, although the results might be unfairly tipped in nVidia's favour (even though they don't come out on top regardless), if we remove benchmarks all we have left is a guy playing the game under constantly varying circumstances and deriving such results as: "Card X seems to be hyper-fast, while card Y appears to be merely ultra-fast."
All benchmarks done using long established timedemos that have been available to nVidia for a while and are frequently used should be taken with a grain of salt. The best solution at the moment would be for reviewers to record their own demos at the beginning of each benchmark, benchmark the cards, and then released the recorded demo into the public domain so that the user base can verify the results; after all, while nVidia (and ATi) are not to be trusted, neither are reviewers, especially with the likes of Kyle of [H]ardOCP calling themselves as such.
I learned a long ago, from PC Mags 3d benchmarks, that synthetic benchmarks are absolutely useless!
And what exactly differentiates a real benchmark from a synthetic benchmark? While Futuremark does report the fill rate (both single-texturing and multi-texturing), it is simply extraneous information, which is in no way used to determine the resulting 3DMark score; the score is determined by running four game demos, which use engines akin to those used in "real games." The individual game results are reported by 3DMark, multiplied by certain coefficients, and then added together, rendering the result (3DMarks).
The reason 3DMark03 is invalid is not because it is a "synthetic" benchmark, but because nVidia mucked it up with their shenanigans. The frightful truth of the matter is, however, that the same illegitimate "optimizations" (i.e. static clip planes) that were used by nVidia in 3DMark can just as easily be used in any and all timedemo. Hence, your precious "real" benchmarks are just as susceptible, and may be just as compromised and invalid as 3DMark03. To make matters worse, unlike 3DMark03, which offers advanced diagnostic tools that allowed nVidia's dubious actions to be exposed, "real" benchmarks have no such tools. Therefore, exposing cheating in "real" benchmarks is much more difficult; however, just because something cannot be proven does not make it false.
You are of course quite correct. Kyle is merely an idiot with a website, as are many others these days, who is acting as nVidia's PR puppet. He has been ridiculed by many, yet seems intent on maintaining his wicked ways. Unfortunately, thousands of less informed internet users frequent his site, propagating his misinformation.
3DMark2003 added unfair optimizations to their program to make the nvidia card seem better than ATi's
3DMark2003 (or rather Futuremark, since I doubt the program is advanced enough to program itself) did no such thing. nVidia did it all by themselves, and Futuremark conducted an investigation confirming the "optimizations" (cheats i.e. static clip planes inserted by nVidia) and denouncing them. Google for more.
Interesting that in 3DMARK, the FX 5900 ran away with it.
The FX 5900 ran away with nothing.
First, the Radeon won in 3DMark01.
Second, observe the origin as well as the scale of the 3DMark03 graphs: Graph 1, Graph 2
The difference is grossly exaggerated by the graph's peculiar origin (5700 and 3800 instead of 0) and large scale.
Third, 3DMark03 has been rendered an useless benchmark since it is riddled with nVidia "optimizations," which have been deemed illegitimate by Futuremark's own accord. Even the author of the article acknowledges the dubious nature of 3DMark03: 3DMark 03 Build 320 - FX 5900 Ultra - Should we include this? Possibly not, however the FX 5900 wins with WHQL Det Drivers
New nVidia optimizations (read: cheats) such as the Anistropic Filtering "optimization," which was exposed by renaming the 3DMark03 executable. Of additional note is that one of the 3DMark03 benchmarks was done using 8x Anisotropic Filtering, deeming it even more illegitimate than the rest of the 3DMark03 benchmarks in the article.
Fourth, the Radeon delivers consistently better image quality throughout, as was acknowledged by the author of the article: 3D Visual Quality - R9800 Pro hands down.
Another thing to be kept in mind is that the Radeon may very well do much better relative to the FX 5900 in all benchmarks since nVidia may very well optimizing for all major timedemos (3DMark03 style) and the reviewed did not record his own demos. Feel free to google if you find you would like more information.
Or when Linus gets the commercial and media attention Steve Jovs gets.
Well, I can't imagine this Jovs guy gets a lot of commercial and media attention, seeing as I've never heard of him...
I think he was referring to the fictional computer from The Hitchhiker's Guide to the Galaxy after which the chess computer was named, not the chess computer itself.
"funnily" WTF? You make an informative post, but destroy it by making up words. WTF? No, really, did you think that "funnily" is a word? Please tell me you are not that stupid.
funniÂly adv.
First, all these signs are more or less retarded, but the right sign in the first picture must be one of the stupidest things I have ever seen. It states "Give communism a try! - Free Linux," but also contains Tux hailing in the manner of a Nazi and stating "Ya Vol!" (sic).
How retarded do these people have to be in order to mix up communism and Nazism, which are polar opposites.
Wie sind die Borg...
:P
Yeah, how are the Borg?
(Wie = How, Wir = We)
I never meant to imply that nForce boards had fw 800; the first part of my comment was only in relation to hypertransport. Upon rereading my comment, I realized it was a bit ambiguous, and considered posting a clarification, although I hoped that the last part of the comment:
"Furthermore, if there is a demand for fw 800 on PCs, they will have it..." (emphasis added)
would dispel any misconceptions regarding the presence of fw 800 on current nForce motherboards. I hope this post has resolved any misunderstandings.
Which wintel motherboards have fw 800 and hypertransport? I'd be interested.
I wasn't really going to comment on this, but since it got modded up 5, Insightful...
nForce, nForce2
You might be further interested in knowing that hypertransport was primarily designed by AMD and is used in all of their Opteron systems, and will also be used in all Athlon64 systems. I guess that's not wintel per se... but it's a PC motherboard nonetheless.
Furthermore, if there is a demand for fw 800 on PCs, they will have it... a small upside of not being at the whim of a single company *cough*Apple*cough*.
Almost as interesting as the article itself is the fact that timothy has finally attained a 1:1 mispelled to correctly spelled word ratio. This is clearly a monumental achievement as it indicates that he is well on his way to reaching the golden ratio (1.618): the ratio at which we will find all of timothy's posts extremely attractive.
Everyone's rooting for you timothy. Don't give up.
I have to admit, I knew the horse-power game was one that Nvidia was going to trip over eventually when the Radeons were *really* pushing the Z-axis occlusion.
That is faulty logic; just because ATi was using Z-axis occlusion and ATi beat nVidia (which is also using z-occlusion nowadays) does not mean that that is why they beat nVidia. The Kyro II used tile based deferred rendering which was far superior to the z-occlusion ATi and nVidia were and are still doing, and look where PowerVR is today.
ATI finally got their hardware on track not by being able to "outshow" Nvidia's muscle, but by outclassing them by being smarter with the bandwidth and 'getting a clue' with drivers.
If being "smarter with the bandwidth" was all that was required, we'd all be using Kyro IIs. Making a successful GPU is not so black and white...
If it wasn't cheating, and they were only looking at shader instructions, then why did the results change when they renamed the file from Quake3.exe to quack3.exe? It is also important to note that the Quake3.exe results were of less quality than the quack3.exe results.
Um... he was talking about 3dMark, not the Quack issue (btw, there are no shaders in Quake3), which although people insist on constantly bringing, up was most likely not a cheat; ATi not only fixed the 5 problematic textures, but also increased the performance past the level of performance attained using the "cheat," in the next driver release.
The Opteron will be available in a desktop version. The Xeon is not, and will not be.
How so? The Opteron is to the Athlon64 ("the desktop version") as the current-generation Xeon is to the Pentium 4. In fact, the Pentium 4 is much closer to the Xeon than the Athlon64 is to the Opteron.
Tom didn't enable the second memory channel
The Athlon64 will not have a second memory channel regardless...
or use DDR400
Can't blame him for that. AMD does not officially support DDR400.
the 800mhz bus Opteron
The Opteron has an integrated memory controller. Therefore, its "Front Side Bus" runs at the same speed as the actual processor.
Proof: With the memory controller integrated onto the AMD Opteron processor, the front side bus (interface to memory) runs at the speed of the processor
Cg is not an API, it is a programming language. It compiles to Direct3D and OpenGl. As for proprietary GL extensions, ATi has EXT_VERTEX_SHADER and ATI_FRAGMENT_SHADER.
13(?) micrometer technology
It is actually 0.13.
nVidia is not ahead of ATi at all. nVidia and ATi both manufacture their chips at TSMC. The supposed reason that it took nVidia so long with the FX was that TSMC was having trouble with the 0.13 micron process. Not that they ironed out all the problems, there is no reason they can't manufacture ATi's chips using the same process.
In fact, considering that nVidia's 0.13 micron card runs hotter and slower than nVidia's previous generation 0.18 micron card (I'm not even talking about the 9800 here) really shows that nVidia is really behind this round.
Cg ("Their own graphics language") is a programming language.
Glide was an API.
Cg compiles to OpenGL or Direct3D, and it could very well be used on ATi cards with slight modifications. The two are nothing alike.
I guess he decided to install Longhorn on his server...