First, there are no perfectly fair benchmarks. Never have been, never will be. All platforms have their strengths and weaknesses, including the G5. You could theoretically create a test that would even allow a Celeron to beat everything out there. However, the G5 is shaping up as perhaps the "best overall" system ever built. Or in other words, a system that doesn't just do one or two things well, but does many things well, comparing favorably with many other systems. But also has the capability to do a few things exceedingly well, beyond the capability of any comparably priced PC out there.
Second, Nasa's test represent their concerns as related to their real world needs. However, they also noted the impact of a proper compiler.
Quote:
Based on an extrapolation of current P4 results, the 2GHz G5 would lag newly announced 3.2GHz P4 systems in Jet3D scalar floating point performance by about 20%, but this kind of comparison is best deferred until G5-aware compiler tools become available (since a 20% performance gain is well within the potential of compiler optimization).
Real world is a combination of Integer, Floating point, system speed, memory access and a little SIMD for spice. A combination that the G5 handles very well with balance and elegance. In the real world, the G5 may very well be the new "king". We will find out soon enough.
Based on raw scalar floating point performance in Jet3D, a 2GHz G5 system can match a 2.66GHz P4 system, and this is a dramatic improvement from earlier tests where G4 systems lagged behind higher clock speed P4 systems. Based on an extrapolation of current P4 results, the 2GHz G5 would lag newly announced 3.2GHz P4 systems in Jet3D scalar floating point performance by about 20%, but this kind of comparison is best deferred until G5-aware compiler tools become available (since a 20% performance gain is well within the potential of compiler optimization).
Parity now...dominance later...
PS I'm not usually a giddy type like LBB, but I have a feeling that I'm going to be using the smiley quite often in the coming months....
Oh, and the Portland FORTRAN compiler they use for the P4 is subpar.
Sorry yall, But Existence is right...kind of. Well, not really. But take a real close look.
In that test the not yet available Dual 2 GHz G5 lost to the oldish 2.66 GHz P4 by an insignificant amount in the scalar tests (which are the most important to this program). By extrapolation, a 3.2 GHz P4 would beat the Dual 2.0 GHz G5 by about 20%.
The vector tests were only done on Altivec, not on SSE/SSE2, apparently because the compiler that would allow SSE/SSE2 optimization degraded scalar P4 performance, more than negating any benefit (since very little is dependent on vector performance). There is no comparison to the P4 here, only to the G4, which the G5 handily beats...but we all knew that. [At least I think I'm reading this thing right...]
Result: Given available compilers, the G5 is inferior to current P4 systems. Given possible future optimizations for both processor-specific issues, dual processors, and 64-bit processing on the G5, and possible future optimizations for vector processor-specific issues and vector processing on the P4, rough parity is envisioned. G5 superiority is not out of the question given the added benefits of dual processors, but remains speculative.
Despite losing by 20% with current compiler technology, this is still good news for the Mac side. Rough parity is all we need...and quite frankly, it's all we can expect.
Engineering is all about trade-offs. Intel only dominated the last few years because of what can be termed "historical accidents," in this case bad business decisions (i.e., Apple's decision to go with Mot and Mot's subsequent-but-unrelated implosion because of it's reliance on and exposure to the telecom bubble of the late 1990s, in which it decided to slash its chip division to the bone). Absent that--which the G5 will moot--both lines will have rough parity over the broad range of implementations.
The x86 line will be better at some things (higher frequency), while the PPC will be better at others (higher efficiency but lower clockrates). The PPC has a better design from an engineering standpoint because it is a newer design, but because of another historical accident Intel happens to have huge amounts of cash to throw at the problem. The only way I see IBM/Apple gaining clear dominance is if they manage to bust open Intel's monopoly of the desktop space to the point where Intel's marketshare crashes to below 50%. (And of course in that case, Intel would go backrupt, unless it happens slowly over a period of decades...which is hard to imagine. It's in the nature of monopolies to manipulate markets to remain in control, which is why anti-trust laws were put into effect a century ago, and which is also why monopolies of all kinds have worked to weaken those anti-trust laws ever since...pretty successfully as the M$ case showed).
Blah blah blah. Anyway, enough philosophy/history/economics. The point is that rough parity is all we should be worried about. Sorry for the ramble.
Sorry yall, But Existence is right...kind of. Well, not really. But take a real close look.
In that test the not yet available Dual 2 GHz G5 lost to the oldish 2.66 GHz P4 by an insignificant amount in the scalar tests (which are the most important to this program). By extrapolation, a 3.2 GHz P4 would beat the Dual 2.0 GHz G5 by about 20%.
-DCQ
I disagree with your conclusion for several reasons :
first you should not compare a dual vs single. The test was only a single CPU test : Note that even though the G4 and G5 systems have dual processors, detailed benchmarks in the present study pertain to a single processor only.
Second, the test choose to compare both computer without altivec or SSE2, because as you stated SSE2 was lame : Note that the higher level of optimization (-O2) and SSE/SSE2 options in the Portland compiler degraded Jet3D performance on the P4 system, and were therefore not used.
It means in the present test even with a single G5 used, the G5 smoke the P4 when altivec is used. Read this : in real world performance the G5 smoke a P4 when you use jet3D. I doubt that the NASA will ever use a P4 to achieve these kind of tests. It would be like to use photoshop without altivec.
I disagree with your conclusion for several reasons :
first you should not compare a dual vs single. The test was only a single CPU test : Note that even though the G4 and G5 systems have dual processors, detailed benchmarks in the present study pertain to a single processor only.
The current compilers available can't optimize for dual chips. So the test represents a real-world test of performance available now. At this moment. Period. Both the author and I recognize that optimization for dual processors would likely result in a substantial but unknown (and therefore irrelevant) increase in performance. But that optimization has not been done, and may never be done if NASA doesn't think it's worth the trouble. They obviously had code running for G4s, but never optimized it for dual processors even though G4s have used duals for over three years now.
Quote:
Originally posted by Powerdoc
Second, the test choose to compare both computer without altivec or SSE2, because as you stated SSE2 was lame : Note that the higher level of optimization (-O2) and SSE/SSE2 options in the Portland compiler degraded Jet3D performance on the P4 system, and were therefore not used. It means in the present test even with a single G5 used, the G5 smoke the P4 when altivec is used. Read this : in real world performance the G5 smoke a P4 when you use jet3D. I doubt that the NASA will ever use a P4 to achieve these kind of tests. It would be like to use photoshop without altivec.
You misunderstood the paper. Vector issues account for a tiny fraction of Jet3D's code. In those small places where it is used, Altivec helped tremendously. SSE/SSE2 probably helped tremendously as well. However, the particular compiler that they used caused degraded performance in the P4's scalar performance, which adversely affected its overall score (because vector processing is such a small part of Jet3D's code). So they didn't optimize for SSE/SSE2, which resulted in increased scalar P4 performance, and thus increased its overall performance. The vast majority of Jet3D's code is scalar, not vector. The author's supposition is that vector performance is relatively irrelevant to overall Jet3D performance. Like the hypothetical future G5 dual compiler optimizations, hypothetical future SSE/SSE2 compiler optimizations would result in better performance of the P4. (Although obviously the dual optimizations would be far more beneficial for the G5 than the SSE/SSE2 optimizations would be for the P4; and that's not mentioning that Altivec is a far superior design to SSE/SSE2).
Absolutely nowhere does it say that a G5 smokes a P4. It illustrates rough parity between a dual 2.0 GHz G5 and a 2.66GHz P4. It extrapolates (with good reason) a 3.2GHz P4 would beat the G5 by 20%. It also goes on to speculate that compiler optimization for the G5 would result in something like a 20% gain for the G5, reestablishing rough parity between a dual G5 and a P4. This is a conservative statement, since we know that optimization for the G5 and for dual processors would probably result in something like 100%-115% performance improvement. but since all of that is speculative, the author left that out of the report.
If you think the report states that G5s smoke a P4, you're the one smoking something. (Please pass it around !)
No doubt, given full optimizations for all processors a dual G5 would smoke a P4. I know that. You know that. The author probably knows that. But it was irrelevant for the purposes of the report. (My guess is that NASA officials feel that optimization for dual processors would take NASA far too much time and effort; they probably just want the hardware that will run the current softare the fastest and don't care about anything else hypothetical, which is why the author left all but modest, conservative speculation out of the report.)
The current compilers available can't optimize for dual chips. So the test represents a real-world test of performance available now. At this moment. Period. Both the author and I recognize that optimization for dual processors would likely result in a substantial but unknown (and therefore irrelevant) increase in performance. But that optimization has not been done, and may never be done if NASA doesn't think it's worth the trouble. They obviously had code running for G4s, but never optimized it for dual processors even though G4s have used duals for over three years now.
You misunderstood the paper. Vector issues account for a tiny fraction of Jet3D's code. In those small places where it is used, Altivec helped tremendously. SSE/SSE2 probably helped tremendously as well. However, the particular compiler that they used caused degraded performance in the P4's scalar performance, which adversely affected its overall score (because vector processing is such a small part of Jet3D's code). So they didn't optimize for SSE/SSE2, which resulted in increased scalar P4 performance, and thus increased its overall performance. The vast majority of Jet3D's code is scalar, not vector. The author's supposition is that vector performance is relatively irrelevant to overall Jet3D performance. Like the hypothetical future G5 dual compiler optimizations, hypothetical future SSE/SSE2 compiler optimizations would result in better performance of the P4. (Although obviously the dual optimizations would be far more beneficial for the G5 than the SSE/SSE2 optimizations would be for the P4; and that's not mentioning that Altivec is a far superior design to SSE/SSE2).
Absolutely nowhere does it say that a G5 smokes a P4. It illustrates rough parity between a dual 2.0 GHz G5 and a 2.66GHz P4. It extrapolates (with good reason) a 3.2GHz P4 would beat the G5 by 20%. It also goes on to speculate that compiler optimization for the G5 would result in something like a 20% gain for the G5, reestablishing rough parity between a dual G5 and a P4. This is a conservative statement, since we know that optimization for the G5 and for dual processors would probably result in something like 100%-115% performance improvement. but since all of that is speculative, the author left that out of the report.
If you think the report states that G5s smoke a P4, you're the one smoking something. (Please pass it around !)
No doubt, given full optimizations for all processors a dual G5 would smoke a P4. I know that. You know that. The author probably knows that. But it was irrelevant for the purposes of the report. (My guess is that NASA officials feel that optimization for dual processors would take NASA far too much time and effort; they probably just want the hardware that will run the current softare the fastest and don't care about anything else hypothetical, which is why the author left all but modest, conservative speculation out of the report.)
-DCQ
I think that we both misunderstood the paper
There is two version of Jed3D : a scalar one and a vector one. ( Two versions of Jet3D are tested here: the original double precision scalar code ("scalar"), and a mixed single and double precision vector code with approximate spectral integration ("vector"). In the latter version of the code, key portions of the Jet3D algorithm were rewritten to take advantage of AltiVec for increased in performance. The scalar version of the code is written entirely in FORTRAN (a mix of F77 and F90), while the vector version of the code is a mix of FORTRAN and C. Because of its dependence on AltiVec, the vector version of Jet3D runs on G4 and G5 systems only.)
I don't know if the scalar version is more precise than the vector one (but both are only physical simulations), but the two versions are doing the same job. The vector benchmark is not just small places of jet 3D, it's a different version of jet 3D.
You will notice also that the vector version is optimised for dual G5. (As before, it is worth noting that the G5 system benchmarked at 5177 MFLOPS and 1.29 MFLOPS/MHz for vector Jet3D performance when two processors were used.)
Same apply for the scalar version (Though dual processor benchmarks are not presented in detail here, it is worth noting that the G5 system benchmarked at 498 MFLOPS and 0.125 MFLOPS/MHz for scalar Jet3D performance when two processors were used.)
Quote:
In that test the not yet available Dual 2 GHz G5 lost to the oldish 2.66 GHz P4 by an insignificant amount in the scalar tests (which are the most important to this program). By extrapolation, a 3.2 GHz P4 would beat the Dual 2.0 GHz G5 by about 20%.
At the light of the nasa article, a dual G5 hits 498 mflops and a P4 2,66 255. for scalar. So the dual win to a 90 % factor in the scalar mode with a non optimized code.
And now if you use the vector code ...
Now i am quite aware that a G5 will not smoke a 3 ghz P4 in any applications, but i think that the G5 is a real performer, and we should be proud of this chip (enfin !).
Hey, me too! Man, hmurchison, you've got some issues to be worked out!
All set man. Got my Zoloft prescription and now things are peeeechy keeen
Honestly though, no disrespect to those on Zoloft, I really think it's going to be best waiting for shipping hardware. I don't think this guy meant for this test to be all conclusive. At any rate the Dual 2Ghz G5 could come stay with me any day.
Yup, shipping hardware rules. However, if it we don't discuss vaporware (defined as anything not shipping), this forums and many others out there would cease to exsit.
The NASA reviewer clearly said he was using non optimized code for the G5. No changes were made. This wasn't meant to be a PC vs Mac test. It was a simple evaluation.
Yeah, but I'm looking at these test results this way: Steve Jobs has said that 3 GHz is attainable in a year's time. Even if the test results were optimized or not, the performance of these G5s aren't going to be stuck for awhile like the G4s started off. I haven't seen Intel climbing the speed latter too fast recently - like they did all last year and the year before. It's taken them approx 5 months to get from 3.06 to 3.2 GHz. So by the time next year rolls around, I think we'll be even closer to Intel's clock speed than we are right now with the top-end G5s. Combined this with the overall better performance of the G5, I think we come out on top no matter what!
Wow, how dare someone try to use his/her real world applications to compare computer systems!!!
The gall to actual think they have a right to evaluate things without the expressed written consent of the AI Forum Commission on Rumor Standards and Practices!!!
How dare someone not spend the time and effort to hide his place of work (esp. a chop shop like NASA) in his email address!!!! Doesn't he realize he would get so much more street cred by setting up a Hotmail account with a legitimate web name like RUaMacFReEk!#$.
Man, it pisses me off when someone tries to make value judgements with hardware that he is working daily with, when everyone knows its better to cruise SpyMac for the real technical specifications. I mean, I've read hundreds of posts that use terms like "gigaflops" and "Vectorized SSE2 code optimized with GCC compilers" to know all about computers...even if I won't even touch a G5 for two more months!!
Heck, I got the insides of the G5 as my monitor wallpaper and the SMP optimization codes for the P4 as my screensaver...and what does he got? Some bogus NASA crap?!
Give me PC Troll diatribes and MacZealot one-liners chocked full of smilies for the real story on POWER4 v. Pentium roadmaps!!
Comments
$500 Xeon, oh yeah.
Originally posted by alcimedes
i think he just likes dumping threads with moronic posters into my forums instead of his.
But I hadn't posted yet
Originally posted by MacsRGood4U
Bo dhi, we can purchase the G5 right now? wow: Where?:
applestore.com
Originally posted by MacsRGood4U
Bo dhi, we can purchase the G5 right now? wow: Where?:
the hardware has been announced and is for sale. Therefore it is not "Future Hardware".
in other words, shut up, and worry about more important things than where a thread is suppose to be.
Second, Nasa's test represent their concerns as related to their real world needs. However, they also noted the impact of a proper compiler.
Based on an extrapolation of current P4 results, the 2GHz G5 would lag newly announced 3.2GHz P4 systems in Jet3D scalar floating point performance by about 20%, but this kind of comparison is best deferred until G5-aware compiler tools become available (since a 20% performance gain is well within the potential of compiler optimization).
Real world is a combination of Integer, Floating point, system speed, memory access and a little SIMD for spice. A combination that the G5 handles very well with balance and elegance. In the real world, the G5 may very well be the new "king". We will find out soon enough.
Originally posted by alcimedes
i think he just likes dumping threads with moronic posters into my forums instead of his.
$500 Xeon, oh yeah.
He was talking about each processor. At least I hope he was.
I say it is time to let NASA do the bake-offs during keynotes and let the Windoze lackeys flame at them!!
Yeah Linux vs. OSX ... I'd take finishing neck and neck with them just fine!!!
Applenut ignoring an opportunity to dis a head-to-head comparison showing positive Mac results?!?!?
Based on raw scalar floating point performance in Jet3D, a 2GHz G5 system can match a 2.66GHz P4 system, and this is a dramatic improvement from earlier tests where G4 systems lagged behind higher clock speed P4 systems. Based on an extrapolation of current P4 results, the 2GHz G5 would lag newly announced 3.2GHz P4 systems in Jet3D scalar floating point performance by about 20%, but this kind of comparison is best deferred until G5-aware compiler tools become available (since a 20% performance gain is well within the potential of compiler optimization).
Parity now...dominance later...
PS I'm not usually a giddy type like LBB, but I have a feeling that I'm going to be using the
...
Originally posted by Existence
Huh? Kicks ass? The obsolete 2.66GHz P4 beats the 2GHz G5. You can readily build or buy a 2.66GHz for <$500.
http://arstechnica.infopop.net/OpenT...4&m=1210972375
Oh, and the Portland FORTRAN compiler they use for the P4 is subpar.
Sorry yall, But Existence is right...kind of. Well, not really. But take a real close look.
In that test the not yet available Dual 2 GHz G5 lost to the oldish 2.66 GHz P4 by an insignificant amount in the scalar tests (which are the most important to this program). By extrapolation, a 3.2 GHz P4 would beat the Dual 2.0 GHz G5 by about 20%.
The vector tests were only done on Altivec, not on SSE/SSE2, apparently because the compiler that would allow SSE/SSE2 optimization degraded scalar P4 performance, more than negating any benefit (since very little is dependent on vector performance). There is no comparison to the P4 here, only to the G4, which the G5 handily beats...but we all knew that. [At least I think I'm reading this thing right...]
Result: Given available compilers, the G5 is inferior to current P4 systems. Given possible future optimizations for both processor-specific issues, dual processors, and 64-bit processing on the G5, and possible future optimizations for vector processor-specific issues and vector processing on the P4, rough parity is envisioned. G5 superiority is not out of the question given the added benefits of dual processors, but remains speculative.
Despite losing by 20% with current compiler technology, this is still good news for the Mac side. Rough parity is all we need...and quite frankly, it's all we can expect.
Engineering is all about trade-offs. Intel only dominated the last few years because of what can be termed "historical accidents," in this case bad business decisions (i.e., Apple's decision to go with Mot and Mot's subsequent-but-unrelated implosion because of it's reliance on and exposure to the telecom bubble of the late 1990s, in which it decided to slash its chip division to the bone). Absent that--which the G5 will moot--both lines will have rough parity over the broad range of implementations.
The x86 line will be better at some things (higher frequency), while the PPC will be better at others (higher efficiency but lower clockrates). The PPC has a better design from an engineering standpoint because it is a newer design, but because of another historical accident Intel happens to have huge amounts of cash to throw at the problem. The only way I see IBM/Apple gaining clear dominance is if they manage to bust open Intel's monopoly of the desktop space to the point where Intel's marketshare crashes to below 50%. (And of course in that case, Intel would go backrupt, unless it happens slowly over a period of decades...which is hard to imagine. It's in the nature of monopolies to manipulate markets to remain in control, which is why anti-trust laws were put into effect a century ago, and which is also why monopolies of all kinds have worked to weaken those anti-trust laws ever since...pretty successfully as the M$ case showed).
Blah blah blah. Anyway, enough philosophy/history/economics. The point is that rough parity is all we should be worried about. Sorry for the ramble.
-DCQ
Originally posted by DCQ
Sorry yall, But Existence is right...kind of. Well, not really. But take a real close look.
In that test the not yet available Dual 2 GHz G5 lost to the oldish 2.66 GHz P4 by an insignificant amount in the scalar tests (which are the most important to this program). By extrapolation, a 3.2 GHz P4 would beat the Dual 2.0 GHz G5 by about 20%.
-DCQ
I disagree with your conclusion for several reasons :
first you should not compare a dual vs single. The test was only a single CPU test : Note that even though the G4 and G5 systems have dual processors, detailed benchmarks in the present study pertain to a single processor only.
Second, the test choose to compare both computer without altivec or SSE2, because as you stated SSE2 was lame : Note that the higher level of optimization (-O2) and SSE/SSE2 options in the Portland compiler degraded Jet3D performance on the P4 system, and were therefore not used.
It means in the present test even with a single G5 used, the G5 smoke the P4 when altivec is used. Read this : in real world performance the G5 smoke a P4 when you use jet3D. I doubt that the NASA will ever use a P4 to achieve these kind of tests. It would be like to use photoshop without altivec.
Originally posted by Powerdoc
I disagree with your conclusion for several reasons :
first you should not compare a dual vs single. The test was only a single CPU test : Note that even though the G4 and G5 systems have dual processors, detailed benchmarks in the present study pertain to a single processor only.
The current compilers available can't optimize for dual chips. So the test represents a real-world test of performance available now. At this moment. Period. Both the author and I recognize that optimization for dual processors would likely result in a substantial but unknown (and therefore irrelevant) increase in performance. But that optimization has not been done, and may never be done if NASA doesn't think it's worth the trouble. They obviously had code running for G4s, but never optimized it for dual processors even though G4s have used duals for over three years now.
Originally posted by Powerdoc
Second, the test choose to compare both computer without altivec or SSE2, because as you stated SSE2 was lame : Note that the higher level of optimization (-O2) and SSE/SSE2 options in the Portland compiler degraded Jet3D performance on the P4 system, and were therefore not used. It means in the present test even with a single G5 used, the G5 smoke the P4 when altivec is used. Read this : in real world performance the G5 smoke a P4 when you use jet3D. I doubt that the NASA will ever use a P4 to achieve these kind of tests. It would be like to use photoshop without altivec.
You misunderstood the paper. Vector issues account for a tiny fraction of Jet3D's code. In those small places where it is used, Altivec helped tremendously. SSE/SSE2 probably helped tremendously as well. However, the particular compiler that they used caused degraded performance in the P4's scalar performance, which adversely affected its overall score (because vector processing is such a small part of Jet3D's code). So they didn't optimize for SSE/SSE2, which resulted in increased scalar P4 performance, and thus increased its overall performance. The vast majority of Jet3D's code is scalar, not vector. The author's supposition is that vector performance is relatively irrelevant to overall Jet3D performance. Like the hypothetical future G5 dual compiler optimizations, hypothetical future SSE/SSE2 compiler optimizations would result in better performance of the P4. (Although obviously the dual optimizations would be far more beneficial for the G5 than the SSE/SSE2 optimizations would be for the P4; and that's not mentioning that Altivec is a far superior design to SSE/SSE2).
Absolutely nowhere does it say that a G5 smokes a P4. It illustrates rough parity between a dual 2.0 GHz G5 and a 2.66GHz P4. It extrapolates (with good reason) a 3.2GHz P4 would beat the G5 by 20%. It also goes on to speculate that compiler optimization for the G5 would result in something like a 20% gain for the G5, reestablishing rough parity between a dual G5 and a P4. This is a conservative statement, since we know that optimization for the G5 and for dual processors would probably result in something like 100%-115% performance improvement. but since all of that is speculative, the author left that out of the report.
If you think the report states that G5s smoke a P4, you're the one smoking something. (Please pass it around
No doubt, given full optimizations for all processors a dual G5 would smoke a P4. I know that. You know that. The author probably knows that. But it was irrelevant for the purposes of the report. (My guess is that NASA officials feel that optimization for dual processors would take NASA far too much time and effort; they probably just want the hardware that will run the current softare the fastest and don't care about anything else hypothetical, which is why the author left all but modest, conservative speculation out of the report.)
-DCQ
Originally posted by DCQ
The current compilers available can't optimize for dual chips. So the test represents a real-world test of performance available now. At this moment. Period. Both the author and I recognize that optimization for dual processors would likely result in a substantial but unknown (and therefore irrelevant) increase in performance. But that optimization has not been done, and may never be done if NASA doesn't think it's worth the trouble. They obviously had code running for G4s, but never optimized it for dual processors even though G4s have used duals for over three years now.
You misunderstood the paper. Vector issues account for a tiny fraction of Jet3D's code. In those small places where it is used, Altivec helped tremendously. SSE/SSE2 probably helped tremendously as well. However, the particular compiler that they used caused degraded performance in the P4's scalar performance, which adversely affected its overall score (because vector processing is such a small part of Jet3D's code). So they didn't optimize for SSE/SSE2, which resulted in increased scalar P4 performance, and thus increased its overall performance. The vast majority of Jet3D's code is scalar, not vector. The author's supposition is that vector performance is relatively irrelevant to overall Jet3D performance. Like the hypothetical future G5 dual compiler optimizations, hypothetical future SSE/SSE2 compiler optimizations would result in better performance of the P4. (Although obviously the dual optimizations would be far more beneficial for the G5 than the SSE/SSE2 optimizations would be for the P4; and that's not mentioning that Altivec is a far superior design to SSE/SSE2).
Absolutely nowhere does it say that a G5 smokes a P4. It illustrates rough parity between a dual 2.0 GHz G5 and a 2.66GHz P4. It extrapolates (with good reason) a 3.2GHz P4 would beat the G5 by 20%. It also goes on to speculate that compiler optimization for the G5 would result in something like a 20% gain for the G5, reestablishing rough parity between a dual G5 and a P4. This is a conservative statement, since we know that optimization for the G5 and for dual processors would probably result in something like 100%-115% performance improvement. but since all of that is speculative, the author left that out of the report.
If you think the report states that G5s smoke a P4, you're the one smoking something. (Please pass it around
No doubt, given full optimizations for all processors a dual G5 would smoke a P4. I know that. You know that. The author probably knows that. But it was irrelevant for the purposes of the report. (My guess is that NASA officials feel that optimization for dual processors would take NASA far too much time and effort; they probably just want the hardware that will run the current softare the fastest and don't care about anything else hypothetical, which is why the author left all but modest, conservative speculation out of the report.)
-DCQ
I think that we both misunderstood the paper
There is two version of Jed3D : a scalar one and a vector one. ( Two versions of Jet3D are tested here: the original double precision scalar code ("scalar"), and a mixed single and double precision vector code with approximate spectral integration ("vector"). In the latter version of the code, key portions of the Jet3D algorithm were rewritten to take advantage of AltiVec for increased in performance. The scalar version of the code is written entirely in FORTRAN (a mix of F77 and F90), while the vector version of the code is a mix of FORTRAN and C. Because of its dependence on AltiVec, the vector version of Jet3D runs on G4 and G5 systems only.)
I don't know if the scalar version is more precise than the vector one (but both are only physical simulations), but the two versions are doing the same job. The vector benchmark is not just small places of jet 3D, it's a different version of jet 3D.
You will notice also that the vector version is optimised for dual G5. (As before, it is worth noting that the G5 system benchmarked at 5177 MFLOPS and 1.29 MFLOPS/MHz for vector Jet3D performance when two processors were used.)
Same apply for the scalar version (Though dual processor benchmarks are not presented in detail here, it is worth noting that the G5 system benchmarked at 498 MFLOPS and 0.125 MFLOPS/MHz for scalar Jet3D performance when two processors were used.)
In that test the not yet available Dual 2 GHz G5 lost to the oldish 2.66 GHz P4 by an insignificant amount in the scalar tests (which are the most important to this program). By extrapolation, a 3.2 GHz P4 would beat the Dual 2.0 GHz G5 by about 20%.
At the light of the nasa article, a dual G5 hits 498 mflops and a P4 2,66 255. for scalar. So the dual win to a 90 % factor in the scalar mode with a non optimized code.
And now if you use the vector code ...
Now i am quite aware that a G5 will not smoke a 3 ghz P4 in any applications, but i think that the G5 is a real performer, and we should be proud of this chip (enfin !).
Originally posted by Telomar
But I hadn't posted yet
Hey, me too!
Originally posted by klinux
Hey, me too!
All set man. Got my Zoloft prescription and now things are peeeechy keeen
Honestly though, no disrespect to those on Zoloft, I really think it's going to be best waiting for shipping hardware. I don't think this guy meant for this test to be all conclusive. At any rate the Dual 2Ghz G5 could come stay with me any day.
Originally posted by hmurchison
The NASA reviewer clearly said he was using non optimized code for the G5. No changes were made. This wasn't meant to be a PC vs Mac test. It was a simple evaluation.
Yeah, but I'm looking at these test results this way: Steve Jobs has said that 3 GHz is attainable in a year's time. Even if the test results were optimized or not, the performance of these G5s aren't going to be stuck for awhile like the G4s started off. I haven't seen Intel climbing the speed latter too fast recently - like they did all last year and the year before. It's taken them approx 5 months to get from 3.06 to 3.2 GHz. So by the time next year rolls around, I think we'll be even closer to Intel's clock speed than we are right now with the top-end G5s. Combined this with the overall better performance of the G5, I think we come out on top no matter what!
i have no idea so eve if we'll say 4GHz...being generous i think
the 3-4GHz isn't as impressive as 2-3Ghz...thast a 150% increase as opposed to intels 133.3%
970 should do apple good both speed wise and in the eyes of the market (clockspeed = speed)
Originally posted by AirSluf
The guys at NASA...bunged it up pretty good.
The gall to actual think they have a right to evaluate things without the expressed written consent of the AI Forum Commission on Rumor Standards and Practices!!!
How dare someone not spend the time and effort to hide his place of work (esp. a chop shop like NASA) in his email address!!!! Doesn't he realize he would get so much more street cred by setting up a Hotmail account with a legitimate web name like RUaMacFReEk!#$.
Man, it pisses me off when someone tries to make value judgements with hardware that he is working daily with, when everyone knows its better to cruise SpyMac for the real technical specifications. I mean, I've read hundreds of posts that use terms like "gigaflops" and "Vectorized SSE2 code optimized with GCC compilers" to know all about computers...even if I won't even touch a G5 for two more months!!
Heck, I got the insides of the G5 as my monitor wallpaper and the SMP optimization codes for the P4 as my screensaver...and what does he got? Some bogus NASA crap?!
Give me PC Troll diatribes and MacZealot one-liners chocked full of smilies for the real story on POWER4 v. Pentium roadmaps!!
(edited for clarity)