I hate to say it (well not really, I like agreeing with Belle ), but Belle is right. Go check some DDR charts in comparison to normal SDRAM and youll find that we dont have the compnents to use DDR yet. Even on high end PCs the difference is at max 5 fps in a game that normally runs several hundred fps (Q3), and little else on everything else.
If you had faster RAM, buses, etc. you might get some chartable performance improvements, but I'm not certain that even two 1GHz G4s would crunch numbers fast enough to saturate it all.</strong><hr></blockquote>
Do the math... if you're crunching floating point on the AltiVec unit then the current bus can read and write about 46 million vectors per second (16 bytes each, ~740 MB/sec). At 1 GHz that means a single G4 can do about 86 operations on each vector (2 instructions per clock, each vector must be read and then written). A dual processor just means the two have to share bandwidth, so that means each processor can spend twice as long on each piece of data. There are always some inefficiences, but that means the processor can spend over a hundred instructions on each vector! To give you some idea, a vector multiplied by a 4x4 matrix (very typical 3D graphics operation) takes about 20 clock cycles. That means that the bus would have to be at least five times faster before it wasn't the bottleneck!
Quake doesn't seen an improvement due to AltiVec optimization because they are only doing a small amount of work per vector, and the floating point unit is fast enough to saturate the bus on this amount of work nevermind the AltiVec unit.
For compute intensive applications, a faster memory bus (and memory) would make an enourmous difference. The 5-10% figures everybody quotes is all from benchmarks where streaming data is not being performed. In media applications, which Apple is fond of, it makes a really big difference.
<strong>SGI has become known lately with their snail-like pace in updating their workstations.
(...)
But even they beat Apple to a 200 Mhz FSB and DDR RAM with their new workstation. Congrats Apple! You are the only remaining graphical workstation vendor to use SDRAM! Yay!
</strong><hr></blockquote>
Um, well, Apple certainly isn't cheap, but putting them into the same class as som $10k+ SGI box seems kinda unfair to me, don't you think?
"even slower than SGI" - I don't know, but that's kinda funny
which states that even a dual 533 is starved of bandwidth and that a single G4/1Ghz is also.</strong><hr></blockquote>
And they will still be bandwidth-limited even with DDR RAM unless they get a faster FSB first (i.e. it's the FSB that's the bottleneck, not the memory bus).
For compute intensive applications, a faster memory bus (and memory) would make an enourmous difference. The 5-10% figures everybody quotes is all from benchmarks where streaming data is not being performed. In media applications, which Apple is fond of, it makes a really big difference.</strong><hr></blockquote>
What really makes me wonder is the fact that, even with the current, limited memory bandwidth, saturating the FSB / memory bus would mean tearing through the G4's whole physical address space (64GB) in little more than one minute.
So unless an algorithm performs a LOT of operations on each and every vector in a data stream, I can't see how it possibly could max out the memory bus (and thus be bandwidth-starved) for any longer period of time.
Um, well, Apple certainly isn't cheap, but putting them into the same class as som $10k+ SGI box seems kinda unfair to me, don't you think?
"even slower than SGI" - I don't know, but that's kinda funny
Bye,
RazzFazz</strong><hr></blockquote>
AAArrrgh! Why am I even trying? Is this really so hard to get?
I've said it before and I'll say it once more. Price is not the issue here. You can also get a cheap-ass Compaq with an Athlon in it for $1200 and it'll have DDR in it.
SGI was relevant however because they're an ailing company that once was great. Part of this is because they haven't been quick to adapt to the changes in the industry. They have been especially slow with developing new graphics hardware lately, and even the graphics subsystem in the new fuel is almost two years old for example. It took them 18 months to give their high end chip a speed bump. All this, and even they were faster than Apple at ditching PC 133.
There are 3 architectures that can run Maya now. One is MIPS, comes with DDR. The other is X86, comes with DDR, RDRAM & even dual channel DDR. You guessed it, the 3rd is PPC/Apple, comes with PC 133. What has this got to do with price?
There are 3 architectures that can run Maya now. One is MIPS, comes with DDR. The other is X86, comes with DDR, RDRAM & even dual channel DDR. You guessed it, the 3rd is PPC/Apple, comes with PC 133. What has this got to do with price?</strong><hr></blockquote>
The point I was making is that I consider it kinda funny to say "Oh look, even this workstation-class, three times as expensive machine outperforms that desktop computer. How dare they."
Fact is, IMO, that SGI's machines have all right to, and *should* probably outperform Apple's offerings, and I find it hardly surprising they do so.
Oh, and I wasn't talking about the fact that PC hardware is fast and a steal compared to both the Mac and the SGI, but *only* about Apple and SGI.
<strong>there was a rumor (i think on the register) recently about SGI's interest in the G5 as a MIPS replacement. according to the article, development costs for MIPS was killing SGI...and i suppose that makes sense.
wouldn't it be interesting to see a new AIS alliance of apple, IBM and SGI. IBM could provide the fabrication technologies, SGI could provide some technology from MIPS and their own mobos and apple would provide the market (and marketing). lots of interesting possibilities there....</strong><hr></blockquote>
I saw the same thing. As I recall. Supposedly the R&D cost is taking a huge toll on the company. That and the fact that the proposed specs. (the current White Pater) of the upcoming G5 are extremely impressive. The only problem is from white paper to final implementation, specifications can be added and subtracted like crazy. That's what happened to the G4. They made changes and compromises to get to market sooner and ended up being faced with all sorts of unresearched consequences.
While there may or may not be any true technological reason for Apple to revise it's busses, press wise there are a ton. Perception is everything in our little sound bite society. Steve seems to understand this with his approach to marketing products. If he wants world+dog to buy Macs he has got to address their misgivings in the product to keep his message from being impeded.
We can scream to the mountain top about the MHz myth, but if the average Joe isn't astute enough to understand that for himself I doubt he has the capacity to give more than a cursory glance at the Apple vs. Wintel spec. sheet and say Mac $3000+ sans monitor and no DDR or Rambus(God forbid)vs. Wintel machine $3000 (top o line processor, monitor, and DDR) "Wew hoo...Apple my...). If Apple wants to grow market share, they need that guy to buy a computer.
Apple needs a Wintel comparable base spec. sheet(FSB, memory, subsystems...)in addition to gigaflops, Velocity Engine, iMovie, iDVD & Kitchen sink. Steve's arrogance or stubbornness in this area is a throw back to IBM's stance originally on the GUI; 'If the command line is so difficult for them, they don't need to be buying our computers.'
It's that kind of thinking that got us Microsoft in the first place.
The point I was making is that I consider it kinda funny to say "Oh look, even this workstation-class, three times as expensive machine outperforms that desktop computer. How dare they."
</strong><hr></blockquote>
No, that's not what I said at all.
I said something more like:
"Hey look! Even this bleeding company that recently sold all their 3D patents to Microsoft and hasn't been able to update their hardware for more than 18 months is quicker than Apple at adapting this new technology"
It doesn't matter how great SGI used to be, or how expensive their workstations still are. The fact is that they're not developing anything new, and everybody expects them to wither and die soon. They just lost two of their biggest customers, Pixar got snatched by IBM and DreamWorks/PDI got picked up by HP.
SGI used to be the only game in town for DCC. Just like Apple has been the computer of choice for creative professionals. There's a lesson to be learned here. Shows you what can happen if you sit on your ass doing nothing and expect people to buy your outdated hardware just 'cause they always have.
<strong>SGI used to be the only game in town for DCC. Just like Apple has been the computer of choice for creative professionals. There's a lesson to be learned here. Shows you what can happen if you sit on your ass doing nothing and expect people to buy your outdated hardware just 'cause they always have.</strong><hr></blockquote>
Ok, a couple of things. I've seen those $1500.00 jobs and if you look closely something is always missing when you compare it with a complete package. It might be something you would say was unimportant to you but, we are talking about the general public.
It might be a cheap video card or ethernet card. It might be expandability or ease of use with the case. It's always something and you usually have to look closely because naturally a company wouldn't want to advertise it. Sure you can replace these things ( as my PC friend did ) but, that costs more money.
The other thing is much of the other part of what you say is true, the Dual Gig G4 isn't enough to catch up. Across the board 2, 1 Ghz processors don't equal a clock speed ( or performance ) of 2 Ghz.
The other things also about the speed of the RAM are probably true ( even though they do try to compensate for it ).
It's quite simple, to catch up we need a completely new Powermac. New mother board, new processor, new RAM, etc. We need the G5. I've stated before that the Mhz gap has become so big now that the Mhz myth doesn't matter. That's why this G4 announcement was so quiet. Apple knows this also. I have a feeling this summer things will start to turn around in a big way.
Even though these dual gig G4's are much faster than my now aging G4 450 ( isn't that awful, it's not even 2 years old yet ) I don't think I would spend the money to buy one. But, if someone waived a G5 under my nose I might suddenly be in the market. I don't think I'm alone in that feeling.
Comments
What does your boss do?
<strong>SGI are the only ones for whom the Mhz myth is really true. With apple its a question of how you cook the books.</strong><hr></blockquote>
Apple's MHz myth works when the difference is <500MHz...but the MHz Myth has turned into the GHz Gap!
<strong>
If you had faster RAM, buses, etc. you might get some chartable performance improvements, but I'm not certain that even two 1GHz G4s would crunch numbers fast enough to saturate it all.</strong><hr></blockquote>
Do the math... if you're crunching floating point on the AltiVec unit then the current bus can read and write about 46 million vectors per second (16 bytes each, ~740 MB/sec). At 1 GHz that means a single G4 can do about 86 operations on each vector (2 instructions per clock, each vector must be read and then written). A dual processor just means the two have to share bandwidth, so that means each processor can spend twice as long on each piece of data. There are always some inefficiences, but that means the processor can spend over a hundred instructions on each vector! To give you some idea, a vector multiplied by a 4x4 matrix (very typical 3D graphics operation) takes about 20 clock cycles. That means that the bus would have to be at least five times faster before it wasn't the bottleneck!
Quake doesn't seen an improvement due to AltiVec optimization because they are only doing a small amount of work per vector, and the floating point unit is fast enough to saturate the bus on this amount of work nevermind the AltiVec unit.
For compute intensive applications, a faster memory bus (and memory) would make an enourmous difference. The 5-10% figures everybody quotes is all from benchmarks where streaming data is not being performed. In media applications, which Apple is fond of, it makes a really big difference.
<strong>SGI has become known lately with their snail-like pace in updating their workstations.
(...)
But even they beat Apple to a 200 Mhz FSB and DDR RAM with their new workstation. Congrats Apple! You are the only remaining graphical workstation vendor to use SDRAM! Yay!
</strong><hr></blockquote>
Um, well, Apple certainly isn't cheap, but putting them into the same class as som $10k+ SGI box seems kinda unfair to me, don't you think?
"even slower than SGI" - I don't know, but that's kinda funny
Bye,
RazzFazz
<strong>I'd also refer you to this thread
<a href="http://arstechnica.infopop.net/OpenTopic/page?q=Y&a=tpc&s=50009562&f=8300945231&m=336095374 3&p=3" target="_blank">http://arstechnica.infopop.net/OpenTop ic/page?q=Y&a=tpc&s=50009562&f=8300945231&m=336095374 3&p=3</a>
which states that even a dual 533 is starved of bandwidth and that a single G4/1Ghz is also.</strong><hr></blockquote>
And they will still be bandwidth-limited even with DDR RAM unless they get a faster FSB first (i.e. it's the FSB that's the bottleneck, not the memory bus).
Bye,
RazzFazz
<strong>
For compute intensive applications, a faster memory bus (and memory) would make an enourmous difference. The 5-10% figures everybody quotes is all from benchmarks where streaming data is not being performed. In media applications, which Apple is fond of, it makes a really big difference.</strong><hr></blockquote>
What really makes me wonder is the fact that, even with the current, limited memory bandwidth, saturating the FSB / memory bus would mean tearing through the G4's whole physical address space (64GB) in little more than one minute.
So unless an algorithm performs a LOT of operations on each and every vector in a data stream, I can't see how it possibly could max out the memory bus (and thus be bandwidth-starved) for any longer period of time.
What am I missing?
Bye,
RazzFazz
<strong>
Um, well, Apple certainly isn't cheap, but putting them into the same class as som $10k+ SGI box seems kinda unfair to me, don't you think?
"even slower than SGI" - I don't know, but that's kinda funny
Bye,
RazzFazz</strong><hr></blockquote>
AAArrrgh! Why am I even trying? Is this really so hard to get?
I've said it before and I'll say it once more. Price is not the issue here. You can also get a cheap-ass Compaq with an Athlon in it for $1200 and it'll have DDR in it.
SGI was relevant however because they're an ailing company that once was great. Part of this is because they haven't been quick to adapt to the changes in the industry. They have been especially slow with developing new graphics hardware lately, and even the graphics subsystem in the new fuel is almost two years old for example. It took them 18 months to give their high end chip a speed bump. All this, and even they were faster than Apple at ditching PC 133.
There are 3 architectures that can run Maya now. One is MIPS, comes with DDR. The other is X86, comes with DDR, RDRAM & even dual channel DDR. You guessed it, the 3rd is PPC/Apple, comes with PC 133. What has this got to do with price?
<strong>
There are 3 architectures that can run Maya now. One is MIPS, comes with DDR. The other is X86, comes with DDR, RDRAM & even dual channel DDR. You guessed it, the 3rd is PPC/Apple, comes with PC 133. What has this got to do with price?</strong><hr></blockquote>
The point I was making is that I consider it kinda funny to say "Oh look, even this workstation-class, three times as expensive machine outperforms that desktop computer. How dare they."
Fact is, IMO, that SGI's machines have all right to, and *should* probably outperform Apple's offerings, and I find it hardly surprising they do so.
Oh, and I wasn't talking about the fact that PC hardware is fast and a steal compared to both the Mac and the SGI, but *only* about Apple and SGI.
Bye,
RazzFazz
<strong>there was a rumor (i think on the register) recently about SGI's interest in the G5 as a MIPS replacement. according to the article, development costs for MIPS was killing SGI...and i suppose that makes sense.
wouldn't it be interesting to see a new AIS alliance of apple, IBM and SGI. IBM could provide the fabrication technologies, SGI could provide some technology from MIPS and their own mobos and apple would provide the market (and marketing). lots of interesting possibilities there....</strong><hr></blockquote>
I saw the same thing. As I recall. Supposedly the R&D cost is taking a huge toll on the company. That and the fact that the proposed specs. (the current White Pater) of the upcoming G5 are extremely impressive. The only problem is from white paper to final implementation, specifications can be added and subtracted like crazy. That's what happened to the G4. They made changes and compromises to get to market sooner and ended up being faced with all sorts of unresearched consequences.
While there may or may not be any true technological reason for Apple to revise it's busses, press wise there are a ton. Perception is everything in our little sound bite society. Steve seems to understand this with his approach to marketing products. If he wants world+dog to buy Macs he has got to address their misgivings in the product to keep his message from being impeded.
We can scream to the mountain top about the MHz myth, but if the average Joe isn't astute enough to understand that for himself I doubt he has the capacity to give more than a cursory glance at the Apple vs. Wintel spec. sheet and say Mac $3000+ sans monitor and no DDR or Rambus(God forbid)vs. Wintel machine $3000 (top o line processor, monitor, and DDR) "Wew hoo...Apple my...). If Apple wants to grow market share, they need that guy to buy a computer.
Apple needs a Wintel comparable base spec. sheet(FSB, memory, subsystems...)in addition to gigaflops, Velocity Engine, iMovie, iDVD & Kitchen sink. Steve's arrogance or stubbornness in this area is a throw back to IBM's stance originally on the GUI; 'If the command line is so difficult for them, they don't need to be buying our computers.'
It's that kind of thinking that got us Microsoft in the first place.
[ 01-31-2002: Message edited by: ArkAngel ]</p>
<strong>
The point I was making is that I consider it kinda funny to say "Oh look, even this workstation-class, three times as expensive machine outperforms that desktop computer. How dare they."
</strong><hr></blockquote>
No, that's not what I said at all.
I said something more like:
"Hey look! Even this bleeding company that recently sold all their 3D patents to Microsoft and hasn't been able to update their hardware for more than 18 months is quicker than Apple at adapting this new technology"
It doesn't matter how great SGI used to be, or how expensive their workstations still are. The fact is that they're not developing anything new, and everybody expects them to wither and die soon. They just lost two of their biggest customers, Pixar got snatched by IBM and DreamWorks/PDI got picked up by HP.
SGI used to be the only game in town for DCC. Just like Apple has been the computer of choice for creative professionals. There's a lesson to be learned here. Shows you what can happen if you sit on your ass doing nothing and expect people to buy your outdated hardware just 'cause they always have.
<strong>SGI used to be the only game in town for DCC. Just like Apple has been the computer of choice for creative professionals. There's a lesson to be learned here. Shows you what can happen if you sit on your ass doing nothing and expect people to buy your outdated hardware just 'cause they always have.</strong><hr></blockquote>
Good point.
Ok, a couple of things. I've seen those $1500.00 jobs and if you look closely something is always missing when you compare it with a complete package. It might be something you would say was unimportant to you but, we are talking about the general public.
It might be a cheap video card or ethernet card. It might be expandability or ease of use with the case. It's always something and you usually have to look closely because naturally a company wouldn't want to advertise it. Sure you can replace these things ( as my PC friend did ) but, that costs more money.
The other thing is much of the other part of what you say is true, the Dual Gig G4 isn't enough to catch up. Across the board 2, 1 Ghz processors don't equal a clock speed ( or performance ) of 2 Ghz.
The other things also about the speed of the RAM are probably true ( even though they do try to compensate for it ).
It's quite simple, to catch up we need a completely new Powermac. New mother board, new processor, new RAM, etc. We need the G5. I've stated before that the Mhz gap has become so big now that the Mhz myth doesn't matter. That's why this G4 announcement was so quiet. Apple knows this also. I have a feeling this summer things will start to turn around in a big way.
Even though these dual gig G4's are much faster than my now aging G4 450 ( isn't that awful, it's not even 2 years old yet ) I don't think I would spend the money to buy one. But, if someone waived a G5 under my nose I might suddenly be in the market. I don't think I'm alone in that feeling.
<strong>
No, that's not what I said at all.
</strong><hr></blockquote>
Well, it's just how it sounded to me.
Obviously, though, I was misinterpreting you there, so sorry 'bout that.
Bye,
RazzFazz