Home |
Search |
Today's Posts |
#41
|
|||
|
|||
Apple defends tests
Mighta been the early OS ports or some sort of internal bandwidth issue
then, because they (6100s as I recall) were dog slow at database and office stuff we were integrating to. The 6100 had some quirks, the 8100's were much better. MS Office in particular sucked due to the way they applications were built. (look up p-code someday). About two years later, we started using the DEC PC's with Alpha 233 cards in them and they smoked the crap out of all sorts of things... Yup, we had a couple of them at one job. That NT re-compiler thing was pretty weird. -- Dr. Nuketopia Sorry, no e-Mail. Spam forgeries have resulted in thousands of faked bounces to my address. |
#42
|
|||
|
|||
Apple defends tests
SSE was a complete ****ing joke compared to the vector engine in the G4.
No kidding. It resoundingly smoked the intel family on best-case optimized code that applied well to this kind of processing. Interesting. Did this include SSE2? How about AMD CPUs? -S |
#43
|
|||
|
|||
Apple defends tests
Yeah, that's right. When an Altivec app can pretty much own the whole
machine, it is very much superior technology to SSE2 and 3DNow! The original SSE was more marketing than legitimate go-juice. If you take a look at the non-Altivec version vs the Altivec version of the Distributed.net RC5 client, it's astounding. Altivec improves performance by about 250%. Nowhere, on no app, does SSE2 come close to that boost, AFAIK. But it does seem to me that Altivec does not play nice at all on the current Macs in multitasking scenarios. Seems like the restricted FSB bandwidth and memory speeds, plus the way the G4 registers are designed creates pretty horrific inefficiencies in many cases. Thus, relatively minor improvement was realized on DAWs and plugins, where the scenario is inherently multitasking in a big way. If the G5 can overcome those issues, Altivec could finally have a big impact of DAWs. Without improved Altivec when multitasking, the G5 will still be a big boost for the Mac and bring + / - parity with Wintel world. With Altivec running like it should, G5 could actually prove to be a worldbeater. Unfortunately, I'm not too optomistic that IBM/Apple/OSX/compilers/developers will all line up to make it so. Brilliant technology, with mediocre realworld success to date for the most part, certainly for DAWs. Regards, Brian T nuke wrote: How did it compare when the scales were evened out and the code also supported the SSE and SSE2 SIMD sets? SSE was a complete ****ing joke compared to the vector engine in the G4. No kidding. It resoundingly smoked the intel family on best-case optimized code that applied well to this kind of processing. Like BLAST and the other protein folding number cruncher applications out there, they run killer on alti-vec. The only challenge is feeding them with data. -- Dr. Nuketopia Sorry, no e-Mail. Spam forgeries have resulted in thousands of faked bounces to my address. |
#44
|
|||
|
|||
Apple defends tests
Yeah, that's right. When an Altivec app can pretty much own the whole
machine, it is very much superior technology to SSE2 and 3DNow! The original SSE was more marketing than legitimate go-juice. Interesting, then, that Waves was able to get far, far more out of SSE than out of Altivec in their optimizations. If you take a look at the non-Altivec version vs the Altivec version of the Distributed.net RC5 client, it's astounding. Altivec improves performance by about 250%. Nowhere, on no app, does SSE2 come close to that boost, AFAIK. It does seem, however, that this is a one-of-a-kind example. I haven't seen anything besides this one thing show such a disparity. But it does seem to me that Altivec does not play nice at all on the current Macs in multitasking scenarios. Seems like the restricted FSB bandwidth and memory speeds, plus the way the G4 registers are designed creates pretty horrific inefficiencies in many cases. Thus, relatively minor improvement was realized on DAWs and plugins, where the scenario is inherently multitasking in a big way. If the G5 can overcome those issues, Altivec could finally have a big impact of DAWs. Perhaps... but I don't think the Distributed.net results can be used as evidence of that. If there was more evidence to be seen, I might believe it. Without improved Altivec when multitasking, the G5 will still be a big boost for the Mac and bring + / - parity with Wintel world. With Altivec running like it should, G5 could actually prove to be a worldbeater. Hmmm... running like it should? What evidence do we have to determine how it should run? -S |
#45
|
|||
|
|||
Apple defends tests
I just did a quick search through the distributed.net FAQ.
They mention that the fact that PowerPC and some Intel CPUs score well is because they impliment 32-bit rotate functions in hardware. (who knows how much, if at all, this particular thing would benefit an audio plugin). Interestingly... the P4 dropped this capability... and so all of the "modern" Intel CPUs fell way behind at that point. I think we have a freak case here that relies very heavily on a very specific set of circumstances. -S "Brian Tankersley" wrote in message ... Yeah, that's right. When an Altivec app can pretty much own the whole machine, it is very much superior technology to SSE2 and 3DNow! The original SSE was more marketing than legitimate go-juice. If you take a look at the non-Altivec version vs the Altivec version of the Distributed.net RC5 client, it's astounding. Altivec improves performance by about 250%. Nowhere, on no app, does SSE2 come close to that boost, AFAIK. But it does seem to me that Altivec does not play nice at all on the current Macs in multitasking scenarios. Seems like the restricted FSB bandwidth and memory speeds, plus the way the G4 registers are designed creates pretty horrific inefficiencies in many cases. Thus, relatively minor improvement was realized on DAWs and plugins, where the scenario is inherently multitasking in a big way. If the G5 can overcome those issues, Altivec could finally have a big impact of DAWs. Without improved Altivec when multitasking, the G5 will still be a big boost for the Mac and bring + / - parity with Wintel world. With Altivec running like it should, G5 could actually prove to be a worldbeater. Unfortunately, I'm not too optomistic that IBM/Apple/OSX/compilers/developers will all line up to make it so. Brilliant technology, with mediocre realworld success to date for the most part, certainly for DAWs. Regards, Brian T nuke wrote: How did it compare when the scales were evened out and the code also supported the SSE and SSE2 SIMD sets? SSE was a complete ****ing joke compared to the vector engine in the G4. No kidding. It resoundingly smoked the intel family on best-case optimized code that applied well to this kind of processing. Like BLAST and the other protein folding number cruncher applications out there, they run killer on alti-vec. The only challenge is feeding them with data. -- Dr. Nuketopia Sorry, no e-Mail. Spam forgeries have resulted in thousands of faked bounces to my address. |
#46
|
|||
|
|||
Apple defends tests
In article , Scott
Reams wrote: Yeah, that's right. When an Altivec app can pretty much own the whole machine, it is very much superior technology to SSE2 and 3DNow! The original SSE was more marketing than legitimate go-juice. Interesting, then, that Waves was able to get far, far more out of SSE than out of Altivec in their optimizations. Because according to scott reams nothing that says the G4's were better could possibly be true. If you take a look at the non-Altivec version vs the Altivec version of the Distributed.net RC5 client, it's astounding. Altivec improves performance by about 250%. Nowhere, on no app, does SSE2 come close to that boost, AFAIK. It does seem, however, that this is a one-of-a-kind example. I haven't seen anything besides this one thing show such a disparity. Because according to scott reams nothing that says the G4's were better could possibly be true. But it does seem to me that Altivec does not play nice at all on the current Macs in multitasking scenarios. Seems like the restricted FSB bandwidth and memory speeds, plus the way the G4 registers are designed creates pretty horrific inefficiencies in many cases. Thus, relatively minor improvement was realized on DAWs and plugins, where the scenario is inherently multitasking in a big way. If the G5 can overcome those issues, Altivec could finally have a big impact of DAWs. Perhaps... but I don't think the Distributed.net results can be used as evidence of that. If there was more evidence to be seen, I might believe it. Because according to scott reams nothing that says the G5's were better could possibly be true. Without improved Altivec when multitasking, the G5 will still be a big boost for the Mac and bring + / - parity with Wintel world. With Altivec running like it should, G5 could actually prove to be a worldbeater. Hmmm... running like it should? What evidence do we have to determine how it should run? -S Because according to scott reams nothing that says the G5's were better could possibly be true. you're a hell of an open minded guy there scott |
#47
|
|||
|
|||
Apple defends tests
In article , Scott
Reams wrote: I just did a quick search through the distributed.net FAQ. They mention that the fact that PowerPC and some Intel CPUs score well is because they impliment 32-bit rotate functions in hardware. (who knows how much, if at all, this particular thing would benefit an audio plugin). Interestingly... the P4 dropped this capability... and so all of the "modern" Intel CPUs fell way behind at that point. I think we have a freak case here that relies very heavily on a very specific set of circumstances. -S Because according to scott reams if anything apple scores better ever its a freak case. because scott is so open minded. "Brian Tankersley" wrote in message ... Yeah, that's right. When an Altivec app can pretty much own the whole machine, it is very much superior technology to SSE2 and 3DNow! The original SSE was more marketing than legitimate go-juice. If you take a look at the non-Altivec version vs the Altivec version of the Distributed.net RC5 client, it's astounding. Altivec improves performance by about 250%. Nowhere, on no app, does SSE2 come close to that boost, AFAIK. But it does seem to me that Altivec does not play nice at all on the current Macs in multitasking scenarios. Seems like the restricted FSB bandwidth and memory speeds, plus the way the G4 registers are designed creates pretty horrific inefficiencies in many cases. Thus, relatively minor improvement was realized on DAWs and plugins, where the scenario is inherently multitasking in a big way. If the G5 can overcome those issues, Altivec could finally have a big impact of DAWs. Without improved Altivec when multitasking, the G5 will still be a big boost for the Mac and bring + / - parity with Wintel world. With Altivec running like it should, G5 could actually prove to be a worldbeater. Unfortunately, I'm not too optomistic that IBM/Apple/OSX/compilers/developers will all line up to make it so. Brilliant technology, with mediocre realworld success to date for the most part, certainly for DAWs. Regards, Brian T nuke wrote: How did it compare when the scales were evened out and the code also supported the SSE and SSE2 SIMD sets? SSE was a complete ****ing joke compared to the vector engine in the G4. No kidding. It resoundingly smoked the intel family on best-case optimized code that applied well to this kind of processing. Like BLAST and the other protein folding number cruncher applications out there, they run killer on alti-vec. The only challenge is feeding them with data. -- Dr. Nuketopia Sorry, no e-Mail. Spam forgeries have resulted in thousands of faked bounces to my address. |
#48
|
|||
|
|||
Apple defends tests
It does seem, however, that this is a one-of-a-kind example. I haven't
seen anything besides this one thing show such a disparity. Because according to scott reams nothing that says the G4's were better could possibly be true. It's true. The fact that it is the only example is what brings the question. If you have other examples, please share them. Perhaps... but I don't think the Distributed.net results can be used as evidence of that. If there was more evidence to be seen, I might believe it. Because according to scott reams nothing that says the G5's were better could possibly be true. Nothing says either way yet. The systems aren't available. My point was that I don't think Altivec will save the day from what I've seen. The G5 itself could save the day, however. Hmmm... running like it should? What evidence do we have to determine how it should run? Because according to scott reams nothing that says the G5's were better could possibly be true. Because, unlike yourself, I need more than one solitary piece of evidence to draw a conclusion. It doesn't matter of it's a G5, a P5, or a BMW M5. Also... unlike yourself, I don't believe anything any manufacturer claims about an unreleased product. you're a hell of an open minded guy there scott Sure am. You might try it one day. -S |
#49
|
|||
|
|||
Apple defends tests
I just did a quick search through the distributed.net FAQ.
They mention that the fact that PowerPC and some Intel CPUs score well is because they impliment 32-bit rotate functions in hardware. (who knows how much, if at all, this particular thing would benefit an audio plugin). Interestingly... the P4 dropped this capability... and so all of the "modern" Intel CPUs fell way behind at that point. I think we have a freak case here that relies very heavily on a very specific set of circumstances. Because according to scott reams if anything apple scores better ever its a freak case. because scott is so open minded. Because... unlike yourself... one solitary piece of evidence -is- a freak case until there is at least one more piece of evidence to confirm it as meaningful. You might try being more thorough when trying to understand performance... and not lean on -any- singular piece of evidence to draw conclusions from. You'll get there some day. I believe in you, MusikGuitarboy. -S |
#50
|
|||
|
|||
Apple defends tests
In article , Brian Tankersley
wrote: Unfortunately, I'm not too optomistic that IBM/Apple/OSX/compilers/developers will all line up to make it so. Brilliant technology, with mediocre realworld success to date for the most part, certainly for DAWs. Mediocre realworld success for DAW's? Huh?? David Correia Celebration Sound Warren, Rhode Island www.CelebrationSound.com |
#51
|
|||
|
|||
Apple defends tests
Oh screw it, dude. You are so full of crap I just cannot deal with it
any more. It seems to me you just posted a huge, complex, self-indulgent equivelent of "nanee-nanee-boo-boo". Why, oh why, do I do this to myself? I do not have time to deal with this pointlessness. I already vowed to never check the Sonar forum again. Here I am, sucked into the insanity yet again here on RAP. Next....over.....done. No kidding, never again will I subject myself to the aggravation of interaction with you. At this point, I don't even care who's right. In fact, you can be right. Yes, you are completely right and know pretty much everything there is to know. OK? Rock on, dude, and enjoy your life greatly. Regards, Brian T Scott Reams wrote: We've been over this before and I just don't have the energy to do it all again. But OTOH, the inconsistency in your assessment begs for comment. Does a Waves plugin ever get to "own" the whole machine? Read my post again about Altivec and multitasking. The test results I'm referring to regarding Waves plugins involved testing one plugin at a time and measuring CPU usage. There is no multitasking going on in the benchmark. The plugin "owns" the machine in these tests. I pointed out that Altivec has seen marginal results on DAWs for that reason. OTOH, Altivec yeilds very good results on Altiverb (notice the name) which just happens to prefer to "own" a whole machine, with many Mac users basically dedicating a machine to it, pretty much stand alone. How can we come to any conclusions with a plugin that is a) not available for any platform other than Mac, and b) not available in a non-Altivec optimized version. There is no baseline to compare to... and no way to decide what Altivec's impact is. And I reach my conclusions based upon a number of sources. Some hard facts, some totally subjective personal impressions built up over time, and all points in between. I've seen Photoshop filters, Be specific... and point out what leads you to the conclusion that Altivec was meaningfully involved in the result. the RC5 client, Logic, Specifically? You said yourself that I back up what I say with numbers when I can. All I'm asking is the same of you. a handful of audio plugins and a few other apps in the before/after Altivec cases. Specifically? Frankly Scott, you continue the habit of giving much greater credence to your own presumptions than those of others, which is only human nature I suppose. I present specifics when I can... and use those specifics to form opinions. I would happily invite any specifics you can provide that might help shape my opinion. You know that's how my mind works. If you want to convince me of something... show me. Your mentions of Logic and Photoshop are a bit nebulous. Why do you presume that it is SSE/SSE2 or Altivec alone that are wholly or primarily responsible for the improvement in Waves code that was ported from TDM/Moto 56K origins? I don't. I just remember what I was told by the Waves staff regarding their take on Altivec and what it was doing for them. I'll also reiterate... these plugins were benchmarked one at a time... no multitasking. Pure CPU usage per-plugin. The rift between the performance of the Altivec-enabled Waves plugins and the SSE/SSE2/3dNow! enabled plugins was gargantuan, and still is. The Waves plugs on the Mac side had -no- Altivec optimizations before 3.5... and hardly gained when optimizations were added. You'd think, from your own statements, that Altivec should have saved the day... unless of course Waves' programmers are incompetent. A major piece of this whole thread was the lame compilers Apple chose for the P4. It's been beaten to death here. The vast difference in the disparate P4 Spec scores between Apple and everybody else was the compilers, and that's not just about SSE2, my man. Of course not. I'm not discussing that at this point. What if Waves, being new to releasing x86 based plugins at rev 1.0 once upon a time, just picked a crappy compiler, did a poor job of porting, etc, etc, and a big part of the improvement was simply cleaning up the code and using a newer/better compiler? What if hitting the SSE2 Enable switch in the compiler was good for about 30-40%? Could be. Can you prove otherwise? Do you think the Waves PR Dept was going to say, "Hey, the original Native port kinda sucked, plus we used a lame compiler", even if they really did? My point, again... pre-3.5 Waves plugins on Mac took no advantage of Altivec. 3.5 and beyond does take advantage, and Waves had 1.5 years to figure it out. Waves 3.5 plugins are hardly an improvement on Apple systems over previous versions. Do you think this is because Waves didn't understand Altivec? Or could it be that this is the most one can expect from Altivec when optimizing audio plugins? There must be -some- explanation Waves couldn't get much more out of the Apple systems. Is, perhaps, RC5 the -only- benchmark in existence where the G4 blows Intel and AMD away? Show me otherwise. Show me it isn't a fluke. I'm open to that. So you appear to have cranked some circular logic into your assessment methodologies. Altivec *alone* (the only change in the RC5 client), yielded about 250%. But you say that's likely a "one-of-a-kind" scenario. Why you come to that conclusion, I don't know. But fine. Can you give us chapter and verse details on other cases where SSE or SSE2 *alone* has yielded the same 200% that you quote in the article? You mention evidence. Cool. Bring it. Can you give a single specific example besides RC5 of a G4 blowing away a modern PC... regardless of how much impact Altivec had? If Altivec can provide a 250% improvement to something besides RC5, where is the example? Where is there a single cross-platform audio plugin running more efficiently on G4 than P4/Athlon to support your theory? -Somebody- must have figured out how to get the magic out of Altivec in the audio world by now if it is there to be had. Who has? Demonstrate with hard facts and figures that your own "200% improvement" quote on the Waves website is due, as you have said, to SIMD instructions being employed I never claimed that improvement was due to SIMD. My point is that if Waves has optimized as best they can using SSE/SSE2/3dNow!/Altivec... along with whatever else... why aren't the G4s performing far better than the Intel/AMD systems as they do in the RC5 benchark? Could it be that the RC5 results are one of a kind? All I want to see is one other example... hopefully one that relates to audio. One cross-platform plugin. That's all. AND that similar improvements can be found in enough other specific, well documented cases to assure that yours is a reasonable assumption and cannot be a "one-of-a-kind" case. Just to be clear, your quote is: " Interesting, then, that Waves was able to get far, far more out of SSE than out of Altivec in their optimizations." Pure presumption on your part, by your own standards. So I suggest you post according to the standards you place on others. We await your facts and figures. I await yours. I've seen RC5... and I've seen some casual mention of Photoshop filters and Logic Performance. I think I've delivered a bit more than that. Show me something else. I am honestly interested. -S |
#52
|
|||
|
|||
Apple defends tests
Altivec, not Macs. Read carefully what I said. Ask Logic users how much
difference Altivec made in the realworld on their DAW. Not a lot, unfortunately. Brian T david wrote: In article , Brian Tankersley wrote: Unfortunately, I'm not too optomistic that IBM/Apple/OSX/compilers/developers will all line up to make it so. Brilliant technology, with mediocre realworld success to date for the most part, certainly for DAWs. Mediocre realworld success for DAW's? Huh?? David Correia Celebration Sound Warren, Rhode Island www.CelebrationSound.com |
#53
|
|||
|
|||
Apple defends tests
Yawn
"Musikboy" wrote in message .. . In article , Scott Reams wrote: snip I await yours. I've seen RC5... and I've seen some casual mention of Photoshop filters and Logic Performance. I think I've delivered a bit more than that. Show me something else. I am honestly interested. -S Listen scott, if albert einstein and carl sagan came back from the dead and told you that mac's were faster you would say the same thing because you are ultimately biased and want to sell you services amking and setting up inferior PC daws which rule the amatuer world btw (more digi 001's live in PC's than mac's) but just dont cut it in the real pro world. |
#54
|
|||
|
|||
Apple defends tests
Brian,
It's unfortunate you see it that way. You are pressing me to put up the numbers... which, as you know, is something I've been doing for some time. I am asking you to do the same. I know you are a smart person... and I know you probably have good reason for coming to the conclusions you do. I just want you to show me the specifics. nothing wrong with that. Oh well. -S "Brian Tankersley" wrote in message news Oh screw it, dude. You are so full of crap I just cannot deal with it any more. It seems to me you just posted a huge, complex, self-indulgent equivelent of "nanee-nanee-boo-boo". Why, oh why, do I do this to myself? I do not have time to deal with this pointlessness. I already vowed to never check the Sonar forum again. Here I am, sucked into the insanity yet again here on RAP. Next....over.....done. No kidding, never again will I subject myself to the aggravation of interaction with you. At this point, I don't even care who's right. In fact, you can be right. Yes, you are completely right and know pretty much everything there is to know. OK? Rock on, dude, and enjoy your life greatly. Regards, Brian T Scott Reams wrote: We've been over this before and I just don't have the energy to do it all again. But OTOH, the inconsistency in your assessment begs for comment. Does a Waves plugin ever get to "own" the whole machine? Read my post again about Altivec and multitasking. The test results I'm referring to regarding Waves plugins involved testing one plugin at a time and measuring CPU usage. There is no multitasking going on in the benchmark. The plugin "owns" the machine in these tests. I pointed out that Altivec has seen marginal results on DAWs for that reason. OTOH, Altivec yeilds very good results on Altiverb (notice the name) which just happens to prefer to "own" a whole machine, with many Mac users basically dedicating a machine to it, pretty much stand alone. How can we come to any conclusions with a plugin that is a) not available for any platform other than Mac, and b) not available in a non-Altivec optimized version. There is no baseline to compare to... and no way to decide what Altivec's impact is. And I reach my conclusions based upon a number of sources. Some hard facts, some totally subjective personal impressions built up over time, and all points in between. I've seen Photoshop filters, Be specific... and point out what leads you to the conclusion that Altivec was meaningfully involved in the result. the RC5 client, Logic, Specifically? You said yourself that I back up what I say with numbers when I can. All I'm asking is the same of you. a handful of audio plugins and a few other apps in the before/after Altivec cases. Specifically? Frankly Scott, you continue the habit of giving much greater credence to your own presumptions than those of others, which is only human nature I suppose. I present specifics when I can... and use those specifics to form opinions. I would happily invite any specifics you can provide that might help shape my opinion. You know that's how my mind works. If you want to convince me of something... show me. Your mentions of Logic and Photoshop are a bit nebulous. Why do you presume that it is SSE/SSE2 or Altivec alone that are wholly or primarily responsible for the improvement in Waves code that was ported from TDM/Moto 56K origins? I don't. I just remember what I was told by the Waves staff regarding their take on Altivec and what it was doing for them. I'll also reiterate... these plugins were benchmarked one at a time... no multitasking. Pure CPU usage per-plugin. The rift between the performance of the Altivec-enabled Waves plugins and the SSE/SSE2/3dNow! enabled plugins was gargantuan, and still is. The Waves plugs on the Mac side had -no- Altivec optimizations before 3.5... and hardly gained when optimizations were added. You'd think, from your own statements, that Altivec should have saved the day... unless of course Waves' programmers are incompetent. A major piece of this whole thread was the lame compilers Apple chose for the P4. It's been beaten to death here. The vast difference in the disparate P4 Spec scores between Apple and everybody else was the compilers, and that's not just about SSE2, my man. Of course not. I'm not discussing that at this point. What if Waves, being new to releasing x86 based plugins at rev 1.0 once upon a time, just picked a crappy compiler, did a poor job of porting, etc, etc, and a big part of the improvement was simply cleaning up the code and using a newer/better compiler? What if hitting the SSE2 Enable switch in the compiler was good for about 30-40%? Could be. Can you prove otherwise? Do you think the Waves PR Dept was going to say, "Hey, the original Native port kinda sucked, plus we used a lame compiler", even if they really did? My point, again... pre-3.5 Waves plugins on Mac took no advantage of Altivec. 3.5 and beyond does take advantage, and Waves had 1.5 years to figure it out. Waves 3.5 plugins are hardly an improvement on Apple systems over previous versions. Do you think this is because Waves didn't understand Altivec? Or could it be that this is the most one can expect from Altivec when optimizing audio plugins? There must be -some- explanation Waves couldn't get much more out of the Apple systems. Is, perhaps, RC5 the -only- benchmark in existence where the G4 blows Intel and AMD away? Show me otherwise. Show me it isn't a fluke. I'm open to that. So you appear to have cranked some circular logic into your assessment methodologies. Altivec *alone* (the only change in the RC5 client), yielded about 250%. But you say that's likely a "one-of-a-kind" scenario. Why you come to that conclusion, I don't know. But fine. Can you give us chapter and verse details on other cases where SSE or SSE2 *alone* has yielded the same 200% that you quote in the article? You mention evidence. Cool. Bring it. Can you give a single specific example besides RC5 of a G4 blowing away a modern PC... regardless of how much impact Altivec had? If Altivec can provide a 250% improvement to something besides RC5, where is the example? Where is there a single cross-platform audio plugin running more efficiently on G4 than P4/Athlon to support your theory? -Somebody- must have figured out how to get the magic out of Altivec in the audio world by now if it is there to be had. Who has? Demonstrate with hard facts and figures that your own "200% improvement" quote on the Waves website is due, as you have said, to SIMD instructions being employed I never claimed that improvement was due to SIMD. My point is that if Waves has optimized as best they can using SSE/SSE2/3dNow!/Altivec... along with whatever else... why aren't the G4s performing far better than the Intel/AMD systems as they do in the RC5 benchark? Could it be that the RC5 results are one of a kind? All I want to see is one other example... hopefully one that relates to audio. One cross-platform plugin. That's all. AND that similar improvements can be found in enough other specific, well documented cases to assure that yours is a reasonable assumption and cannot be a "one-of-a-kind" case. Just to be clear, your quote is: " Interesting, then, that Waves was able to get far, far more out of SSE than out of Altivec in their optimizations." Pure presumption on your part, by your own standards. So I suggest you post according to the standards you place on others. We await your facts and figures. I await yours. I've seen RC5... and I've seen some casual mention of Photoshop filters and Logic Performance. I think I've delivered a bit more than that. Show me something else. I am honestly interested. -S |
#55
|
|||
|
|||
Apple defends tests
In article ,
"Scott Reams" wrote: Brian, It's unfortunate you see it that way. You are pressing me to put up the numbers... which, as you know, is something I've been doing for some time. I am asking you to do the same. I know you are a smart person... and I know you probably have good reason for coming to the conclusions you do. I just want you to show me the specifics. nothing wrong with that. Oh well. -S If you're really interested,(you have claimed you are) in such numbers, it'll only take a few minutes on the web to come up with examples of Altivec's abilities. Contrary to your claim, it's not a one time 'fluke'. Steve |
#56
|
|||
|
|||
Apple defends tests
If you're really interested,(you have claimed you are) in such numbers,
it'll only take a few minutes on the web to come up with examples of Altivec's abilities. Contrary to your claim, it's not a one time 'fluke'. I have looked up quite a few examples... but have seen no other case where Altivec provided a 250% improvement that catapulted G4 performance beyond any other CPU. I am happy to be wrong about this... but I haven't seen an example. Just people saying it isn't a fluke. Assume moment that I simply am not able to find the examples you speak of... and point them out. I am open to any information on this as long as it can be found. -S |
#57
|
|||
|
|||
Apple defends tests
Scott Reams wrote:
Because... unlike yourself... one solitary piece of evidence -is- a freak case until there is at least one more piece of evidence to confirm it as meaningful. I just posted a link to a NASA tst of the G5. I found it interesting, though as usual, it may not mean squat to a DAW. I try to restrain laminar airflow calcs in my music. g See "NASA Tests Apple G5" in this forum. Or not. Whatever. -- hank alrich * secret mountain audio recording * music production * sound reinforcement "If laughter is the best medicine let's take a double dose" |
#58
|
|||
|
|||
Apple defends tests
david wrote:
In article , Brian Tankersley wrote: Unfortunately, I'm not too optomistic that IBM/Apple/OSX/compilers/developers will all line up to make it so. Brilliant technology, with mediocre realworld success to date for the most part, certainly for DAWs. Mediocre realworld success for DAW's? Huh?? On the one hand, PT on the Mac rules the market; on the other hand, Brian's rig completely outstrips any known PT rig in terms of simultaneous hardware I/O, number of tracks, plugin instantiation, etc. It's that old "Well, he can't sing for **** but his face is everywhere" versus "Man, how come nobody has ever heard of _this_ guy, who sings loops barrel rolls around the rest of the pack??" -- hank alrich * secret mountain audio recording * music production * sound reinforcement "If laughter is the best medicine let's take a double dose" |
#59
|
|||
|
|||
Apple defends tests
Yeah... I read that one. Very interesting article.
I actually emailed Craig (the author) with some questions about his methods and found a detailed reply this morning. Nice guy. -S "LeBaron & Alrich" wrote in message ... Scott Reams wrote: Because... unlike yourself... one solitary piece of evidence -is- a freak case until there is at least one more piece of evidence to confirm it as meaningful. I just posted a link to a NASA tst of the G5. I found it interesting, though as usual, it may not mean squat to a DAW. I try to restrain laminar airflow calcs in my music. g See "NASA Tests Apple G5" in this forum. Or not. Whatever. -- hank alrich * secret mountain audio recording * music production * sound reinforcement "If laughter is the best medicine let's take a double dose" |
#60
|
|||
|
|||
Apple defends tests
LeBaron & Alrich wrote:
Scott Reams wrote: Because... unlike yourself... one solitary piece of evidence -is- a freak case until there is at least one more piece of evidence to confirm it as meaningful. I just posted a link to a NASA tst of the G5. I found it interesting, though as usual, it may not mean squat to a DAW. I try to restrain laminar airflow calcs in my music. g See "NASA Tests Apple G5" in this forum. Or not. Whatever. CFD jobs are funny, because they are basically the same calculations over and over again. They parallelize really well, and they can really fill up a vector pipe very efficiently. They also have a lot of integer stuff going on, more so than an FFT for example. So it's a little different than most audio jobs. Most of the NASA CFD guys are going to large arrays of Linux machines, just because the bang for the buck is so good and since the jobs parallelize so well, they can take advantage of large arrays of cheap machines. This is harder with audio jobs. But the general discussions about floating point speed and vector pipe performance are pretty relevant. --scott -- "C'est un Nagra. C'est suisse, et tres, tres precis." |
#61
|
|||
|
|||
Apple defends tests
I have looked up quite a few examples... but have seen no other case
where Altivec provided a 250% improvement that catapulted G4 performance beyond any other CPU. You might be interested in the following link, from another (new) RAP thread: http://members.cox.net/craig.hunter/g5/ While it is headlined as a benchmark of the G5, they do provide floating point benchmarks for G4, G5 and P4 (with and without using AltiVec for G4 and G5). Also note that they've only compiled and optimized the code for G4 and P4 as they don't, yet, have a Fortran compiler that compiles for G5. Some numbers: P4 2.66GHz Scalar: 255 MFLOPS, 0.096 MFLOPS/MHz G4 1.25GHz Scalar: 129 MFLOPS, 0.103 MFLOPS/MHz G5 2GHz, Scalar: 254 MFLOPS, 0.127 MFLOPS/MHz G4 1.25GHz Vector: 1612 MFLOPS, 1.290 MFLOPS/MHz G5 2GHz Vector: 2755 MFLOPS, 1.378 MFLOPS/MHz To me it seems that AltiVec gave an *enormous* improvement in this application (the jet noise prediction tool Jet3D). I can't vouch for the validity or integrity of the test. I just thought it fit well in this thread. Regards /Jonas |
#62
|
|||
|
|||
Apple defends tests
Right.
I read that and found it interesting. Even had a conversation with the author about his methods. To me it seems that AltiVec gave an *enormous* improvement in this application (the jet noise prediction tool Jet3D). It is a bit tough to draw a conclusion here... as the "vector" version was run on the Altivec machines only. The "scalar" version may not be directly comparable because the code is so much different. I'm also curious as to whether or not the "vector" version would run on an SSE/SSE2-enabled machine if so coded. The author doesn't say. Time was only spent coding the vector version for Altivec. If Altivec-enabled CPUs are the only ones capable of running a vector version under any circumstances, then the numbers are quite valid... but does that translate to audio? It's an interesting question. I'd like to see someone take advantage of it (this should be possible even on a G4) if this is the case. It is all interesting... but does raise questions. -S "Jonas Eckerman" wrote in message ... I have looked up quite a few examples... but have seen no other case where Altivec provided a 250% improvement that catapulted G4 performance beyond any other CPU. You might be interested in the following link, from another (new) RAP thread: http://members.cox.net/craig.hunter/g5/ While it is headlined as a benchmark of the G5, they do provide floating point benchmarks for G4, G5 and P4 (with and without using AltiVec for G4 and G5). Also note that they've only compiled and optimized the code for G4 and P4 as they don't, yet, have a Fortran compiler that compiles for G5. Some numbers: P4 2.66GHz Scalar: 255 MFLOPS, 0.096 MFLOPS/MHz G4 1.25GHz Scalar: 129 MFLOPS, 0.103 MFLOPS/MHz G5 2GHz, Scalar: 254 MFLOPS, 0.127 MFLOPS/MHz G4 1.25GHz Vector: 1612 MFLOPS, 1.290 MFLOPS/MHz G5 2GHz Vector: 2755 MFLOPS, 1.378 MFLOPS/MHz To me it seems that AltiVec gave an *enormous* improvement in this application (the jet noise prediction tool Jet3D). I can't vouch for the validity or integrity of the test. I just thought it fit well in this thread. Regards /Jonas |
#63
|
|||
|
|||
Apple defends tests
I guess real world success isn't measured in number of installed
systems in pro studios. Are you saying that all those systems are coded for the AltiVec instruction set? /Jonas |
#64
|
|||
|
|||
Apple defends tests
Interesting. Did this include SSE2?
How about AMD CPUs? -S Yes and yes. Trashed them all, using best case, most optimized code on all platforms. The G4 family vector engine is pretty darn good. The only way to grind it any faster was to spend (a lot) more money on bigger-iron hardware. -- Dr. Nuketopia Sorry, no e-Mail. Spam forgeries have resulted in thousands of faked bounces to my address. |
#65
|
|||
|
|||
Apple defends tests
Trashed them all, using best case, most optimized code on all platforms.
The G4 family vector engine is pretty darn good. What I'm seeing so far seems to support this... The question is, can the vector engine be utilized in an audio environment... and if so, why hasn't it been? -S |
#66
|
|||
|
|||
Apple defends tests
In article , Scott Reams
wrote: Trashed them all, using best case, most optimized code on all platforms. The G4 family vector engine is pretty darn good. What I'm seeing so far seems to support this... The question is, can the vector engine be utilized in an audio environment... and if so, why hasn't it been? -S Ummm hello it has been. logic uses it thats how you can get so many of it's plugins going. i think DP uses it. I even think pro tools le uses it. a lot of games have to have a G4 bcause of the altivec engine. |
#67
|
|||
|
|||
Apple defends tests
BTW...
My point was this: If the vector engine gives such huge gains in specific scenarios (RC5, the NASA tests)... and if the same gains are possible in the audio world... any Altivec-enabled CPU should be hands-down the highest performing out there in this field. -S "Musikboy" wrote in message .. . In article , Scott Reams wrote: Trashed them all, using best case, most optimized code on all platforms. The G4 family vector engine is pretty darn good. What I'm seeing so far seems to support this... The question is, can the vector engine be utilized in an audio environment... and if so, why hasn't it been? -S Ummm hello it has been. logic uses it thats how you can get so many of it's plugins going. i think DP uses it. I even think pro tools le uses it. a lot of games have to have a G4 bcause of the altivec engine. |
#68
|
|||
|
|||
Apple defends tests
Hi,
"LeBaron & Alrich" wrote: I just posted a link to a NASA tst of the G5. I found it interesting, though as usual, it may not mean squat to a DAW. I try to restrain laminar airflow calcs in my music. g See "NASA Tests Apple G5" in this forum. Or not. Whatever. I did find certain aspects of the NASA test curious. For example, in specifying the P4 systems, neither motherboard was identified, nor were other systemic issues such as the chipset. It reflects the notion of those that do such tests that P4s are "generic" and comparable, which is far from the truth of the matter. The conclusion that I drew from the tests is that NASA may be able to justify replacing their G4s with G5s, but, they could more easily justify dumping all of their Macs for 3.x GHz P4s, which are hardly exotic. Curious that the author didn't arrive at the same conclusion. Neil |
#69
|
|||
|
|||
Apple defends tests
Scott Reams wrote:
If the vector engine gives such huge gains in specific scenarios (RC5, the NASA tests)... and if the same gains are possible in the audio world... any Altivec-enabled CPU should be hands-down the highest performing out there in this field. It depends entirely on how well the vector pipe gets used. At my undergrad school, they were using a Cyber 850 vector machine to run big cobol jobs for administrative MIS work. Needless to say the performance was not very good because it just wasn't designed for that. --scott -- "C'est un Nagra. C'est suisse, et tres, tres precis." |
#70
|
|||
|
|||
Apple defends tests
The conclusion that I drew from the tests is that NASA may be able to
justify replacing their G4s with G5s, but, they could more easily justify dumping all of their Macs for 3.x GHz P4s, which are hardly exotic. Curious that the author didn't arrive at the same conclusion. I tend to agree in a broader sense... but if you take a good look at the article, you'll see that the author is showing that this particular code, when specifically coded for Altivec, runs far, far better on an Altivec-enabled CPU than anything else (nearly 10x faster). If all the code NASA ever ran was the same code they used in this test... and if no further optimizations were available for non-Altivec systems... then the obvious choice would be to use an Apple system. That said... I think the truth has a wider scope... and that if you factor in everything NASA does with these systems, the gap probably disappears. -S |
#71
|
|||
|
|||
Apple defends tests
In article , "Scott
Reams" wrote: when specifically coded for Altivec, runs far, far better on an Altivec-enabled CPU than anything else (nearly 10x faster). If all the code NASA ever ran was the same code they used in this test... and if no further optimizations were available for non-Altivec systems... then the obvious choice would be to use an Apple system. That said... I think the truth has a wider scope... and that if you factor in everything NASA does with these systems, the gap probably disappears. I expect NASA would have different machines for different tasks, so they could happily get the G5s for this task and other departments could get whatever works best for them. In the past it was sometimes said that it's tough to have both PCs and Macs in one place, but that's long outdated. It's not difficult to support both and network all the machines if you want to (Nasa may have reason to keep some machines off the network). I've seen it done many times without problems - except for careless users that create problems no matter what machine you give them. -- Jay Frigoletto Mastersuite Los Angeles www.promastering.com |
Reply |
|
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Forum | |||
Seeing/hearing and sighted/blind tests | High End Audio | |||
Yet another DBT post | High End Audio | |||
Note to the Idiot | Audio Opinions | |||
Audiophile glossary | High End Audio | |||
fileABX, an new ABX utility that helps performing ABX tests with any hardware or software player | High End Audio |