<strong>Funny thing is, is that Apple and the Mac Zone catalog refers to is as "Velocity Engine" and I can't any mention of "Altivec" on Apple's site, it's all velocity engine. Is this for a reason? I thought for a while it was referred to as Altivec, if that's changed, maybe because if they go with an IBM with SIMD unit they can still call it the same velocity engine as the G4?</strong><hr></blockquote>
Most of the links from this page also refer to "AltiVec".
"Velocity Engine" is just Apple's name for the technology, and I would guess that there are two possible reasons for them not using Motorola's name: 1) they didn't like "AltiVec", 2) they wanted to leave the door open to non-Motorola implementations but maintain user awareness.
Actually when you apply for a trademark you can put the TM after it, then if the USPTO doesn't find any previous marks that directly conflict with your application, the mark gets published for opposition. The opposition phase lasts about a year, IIRC, and if nobody opposes your application, then you get the (R) instead of the TM.
But like you said, it is usually those with the best paid lawyers that win.
I just found it interesting that neither Apple nor IBM had spent the $$$ to trademark Altivec.
Edit: just reread your post and I think it is true that you can use TM without even applying for the mark, but I'm not sure.
I think you can take a look at 06/20 and 06/21 news in <a href="http://www.macbidouille.com."; target="_blank">www.macbidouille.com.</a> Very interesting info!
<strong>Based on what I've read in the news and boards, I think the G5 chip Apple will be using is a version of the POWER5 which is aimed at the blade sever market and workstation market. It is based on the POWER4. This is a 5th generation version of the POWER (hence G5) and IBM only. This would work with the "No Moto G5 rumors" and "Apple is switching to IBM as a single supplier" and "Apple is working on a G5".
I just don't think we will see them until Jan 03 at MWSF.</strong><hr></blockquote>
I agree, i think quite a few rumors over the past few years have actually been a misunderstanding between the Power4 and the G4. It would explain rumors like the multiple-cores-on-chip G4 of a few years back and also the 4-way SMP G4. No doubt Apple will have SMP boards like that but i doubt they were thinking of it back then.
IBMs POWER# series are impressive machines, it would be good to continue the trend.
Also there was a rumor linked with this Sahara chip a while back that said the whole chip would somehow produce Altivec-like acceleration as opposed to co-processor. Any1 know anything about this?
<strong>I think you can take a look at 06/20 and 06/21 news in <a href="http://www.macbidouille.com."; target="_blank">www.macbidouille.com.</a> Very interesting info!</strong><hr></blockquote>
Anybody care to translate? My French isn't very good and Bablefish usually outputs less than I can guess.
Anybody care to translate? My French isn't very good and Bablefish usually outputs less than I can guess.</strong><hr></blockquote>
The translation has already been done by SYN in
Topic: A few G5 tidbits. What is interesting here is that macbidouille had this morning the name of the Apple person which gave the info. The name later dissapeared. It seems that we are not dealing with rumors this time.
IBMs POWER# series are impressive machines, it would be good to continue the trend.</strong>
A single core Power4 chip would be interesting yes. It'll need to be able to process all of the AltiVec instructions though.
<strong>Also there was a rumor linked with this Sahara chip a while back that said the whole chip would somehow produce Altivec-like acceleration as opposed to co-processor. Any1 know anything about this?</strong>
I'm not getting very good vibes from statements like "AltiVec-like acceleration as opposed to co-processor" and IBM's vague statements on the issue. To me that sounds like what IBM did for the Nintendo Gekko PPC chip, alter the 64 bit FPU to do two 32 bit floats in the same cycle. That's not a good solution at all.
Since the Sahara places much more emphasis on power savings and low heat than the G4 does, it would be sensible to use an AltiVec compatible Sahara in the iBook.</strong>
There would be no point. Sahara is a 0.13 micron product. A Moto 7455 is a 0.18 micron product. Moving the 7455 to a 0.13 micron fab will give it power consumption close to Sahara. The G4 will be more, but it'll be close enough.
The only thing Sahara might be useful for is an Oqo style computer, or perhaps an ultra-portable. All of Apple's current Macs will be fine with 0.8 to 1.4 GHz G4 CPUs.
Most of the links from this page also refer to "AltiVec".
"Velocity Engine" is just Apple's name for the technology, and I would guess that there are two possible reasons for them not using Motorola's name: 1) they didn't like "AltiVec", 2) they wanted to leave the door open to non-Motorola implementations but maintain user awareness.</strong><hr></blockquote>
I was mostly referring to the public side of things. I just found it interesting that Altivec was no longer used on the public side, maybe because they don't intend to keep using Alitvec but they do want a SIMD unit and Velocity Engine is Apple's name, right?
Since the Sahara places much more emphasis on power savings and low heat than the G4 does, it would be sensible to use an AltiVec compatible Sahara in the iBook.</strong>
There would be no point. Sahara is a 0.13 micron product. A Moto 7455 is a 0.18 micron product. Moving the 7455 to a 0.13 micron fab will give it power consumption close to Sahara. The G4 will be more, but it'll be close enough.
The only thing Sahara might be useful for is an Oqo style computer, or perhaps an ultra-portable. All of Apple's current Macs will be fine with 0.8 to 1.4 GHz G4 CPUs.[/QB]<hr></blockquote>
Well the idea came from the Apple employee who said that the iBook would stay with the G3 for the foreseeable future. Perhaps there is something about future G3s that we don't know yet?
I think if IBM wants Apple's business they jump on the AltiVec bandwagon... they know the Gekko (or something like it) isn't a good solution -- it was done specifically to fill Nintendo's need. If they did that for them, then they'll probably add the "Velocity Engine" for Apple. IBM seems pretty aggressive about customer-specific processor development.
I love how AI received so much inside info that they know what's up but they stopped being a rumors site. They were on top and then they quit. Some how I tend to believe the well ran dry.
Y'know Programmer, I'd say 'shove' Altivec if it gave Apple blistering alround speed.
The G3 did alright at the time without Altivec. I mean...
...if integration into a 'Power 4' desktop variant is all thats holding it up. Probably not though. I'd take alround raw power.
AMD/Intel have SIMD units but if stuff doesn't take advantage then so what? The CPUs have enough raw power without getting too concerned.
Apple needs a better alround CPU. Period.
Yeesh, the G4 is sticking like **** on a stick.
Humph.
Still, in light of your 'bandwidth' comments...it'll be interesting how a DDR and hypertransport enabled G4 holds up to the opposition. Personally, I think Clawhammer will still kick the crap out of the G4 on Hypertransport...because...won't AMD also be using Hypertransport? Here's hoping they can stick another fpu or so onto the G4 while we die of holded breath for that Gthingie.
Perhaps there is something about future G3s that we don't know yet?</strong><hr></blockquote>
Call me crazy, but I would love to see the return of the G3. An 'ultra elite' version, way cranked up with some wild new cache. I have a quicksilver G4 and a Blue & White G3. I still insist that the G3 is a better computer. Regardless of the rom versions, os updates/reloads, or configurations, the G3 is more stable, and just 'runs better'. What would you think if G3's were overhauled and replaced G4s? <img src="graemlins/lol.gif" border="0" alt="[Laughing]" /> Like I said, sounds crazy, but I wouldn't mind that much. The G3 has always been good to me, and there are many days that it just seems faster. I think my 400 B&W got a IBM chip (during the great G3 chip swap). The Quicksilver has a Moto if I'm not mistaken. Coincidence? Or a message from God about which company makes a better chip?
The Programmer is right, the main market of Apple is graphic, multimedia, video, and audio. All software in this aera take advantage of the G4 altivec stuff.
Superdrive on a G3 even a 2 ghz will be much slower for encoding MPeG 2 than a single one ghz G4.
So no future without altivec.
The comparison with the other simd unit of AMD and INTEL are not relevant because :
- they are less efficient than altivec
- there is too many of them , mmx, 3Dnow, sse, SSe2 : a real nightmare for programmers.
I entirely agree with the Programmer : if IBM wants the Apple market they will do a SIMD unit ativec compatible (no need of a license for that see AMD ...). I will add that sahara is supposed to have a SIMD unit without any further precisions.
The AltiVec unit isn't to blame for Apple's problems, it is the performance saviour despite the problems... without it Apple's performance problems would be considerably worse. The G3 has only now been pushed up to 1 GHz, well after the G4 has made it to that level.
The big problem with SIMD execution units is that they are still being coded in assembly language. For conventional processors that pretty much went out the window in the 70's and 80's. SIMD is so new that developers haven't yet figured out how to make effective use of them easily. That doesn't mean its not possible -- it just hasn't been done yet. Since there are so many of them in the market now, its only a matter of time... and processors which aren't equipped SIMD units when this stuff does get figured out will be at a real disadvantage. AltiVec is the best of the lot, so Apple stands to gain the most and it wouldn't surprise me if they are being fairly aggressive with research in this area.
Hmm.. SIMD is not that difficult to program, and the latest gcc (in the dev tools) does a descent job at vectorizing your code. There are also some code vectorizer out there that do a good job.
Altivec can be programmed from C. The only test I did is this one. I needed to compute sum(x^2 log x^2), an l2 entropy measure not with complete accuracy.
Standard c code took 1 second using log(), it took 1/3 the time using a lookup table for log() it took 1/21 the time using altivec (yes, 21x faster)
<strong>Hmm.. SIMD is not that difficult to program, and the latest gcc (in the dev tools) does a descent job at vectorizing your code. There are also some code vectorizer out there that do a good job.
Altivec can be programmed from C. The only test I did is this one. I needed to compute sum(x^2 log x^2), an l2 entropy measure not with complete accuracy.
Standard c code took 1 second using log(), it took 1/3 the time using a lookup table for log() it took 1/21 the time using altivec (yes, 21x faster)
That is a fairly simple case, and done using the most advanced SIMD unit in the consumer market. If you are writing a lot of software that has to run on a wide variety of machines then you aren't going to want to support all of the different SIMD units. The auto-vectorizing is a start, but C/C++ just doesn't lend itself well to vector code unless you put in custom extensions. The AltiVec extensions are fine, but they only work with the AltiVec. There is a lot more potential gain to be had, but it will require a new approach to development.
The auto-vectorizing is a start, but C/C++ just doesn't lend itself well to vector code unless you put in custom extensions. The AltiVec extensions are fine, but they only work with the AltiVec. There is a lot more potential gain to be had, but it will require a new approach to development.</strong><hr></blockquote>
I remember programming in FORTAN on a CRAY. It was one of the best vectorizing compilers available. Even then, you could get up to a 2x improvement by coding it using the vector primitives yourself instead of letting the compiler do it.
Comments
<a href="http://www.uspto.gov/"; target="_blank">http://www.uspto.gov/</a>;
<strong>Funny thing is, is that Apple and the Mac Zone catalog refers to is as "Velocity Engine" and I can't any mention of "Altivec" on Apple's site, it's all velocity engine. Is this for a reason? I thought for a while it was referred to as Altivec, if that's changed, maybe because if they go with an IBM with SIMD unit they can still call it the same velocity engine as the G4?</strong><hr></blockquote>
<a href="http://developer.apple.com/hardware/ve/index.html"; target="_blank">http://developer.apple.com/hardware/ve/index.html</a>;
Most of the links from this page also refer to "AltiVec".
"Velocity Engine" is just Apple's name for the technology, and I would guess that there are two possible reasons for them not using Motorola's name: 1) they didn't like "AltiVec", 2) they wanted to leave the door open to non-Motorola implementations but maintain user awareness.
But like you said, it is usually those with the best paid lawyers that win.
I just found it interesting that neither Apple nor IBM had spent the $$$ to trademark Altivec.
Edit: just reread your post and I think it is true that you can use TM without even applying for the mark, but I'm not sure.
[ 06-21-2002: Message edited by: foamy ]</p>
<strong>Hmm... I think this may be the source of the Register's November 2001 PPC G5 rumor (as seen in JYD's AI <a href="http://forums.appleinsider.com/cgi-bin/ultimatebb.cgi?ubb=get_topic&f=1&t=001805"; target="_blank">thread</a> some time ago). Or at least something along the same lines. It is from <a href="http://www.digit-life.com:"; target="_blank">www.digit-life.com:</a>
[ 06-19-2002: Message edited by: THT ]</strong><hr></blockquote>
I think you can take a look at 06/20 and 06/21 news in <a href="http://www.macbidouille.com."; target="_blank">www.macbidouille.com.</a> Very interesting info!
<strong>Based on what I've read in the news and boards, I think the G5 chip Apple will be using is a version of the POWER5 which is aimed at the blade sever market and workstation market. It is based on the POWER4. This is a 5th generation version of the POWER (hence G5) and IBM only. This would work with the "No Moto G5 rumors" and "Apple is switching to IBM as a single supplier" and "Apple is working on a G5".
I just don't think we will see them until Jan 03 at MWSF.</strong><hr></blockquote>
I agree, i think quite a few rumors over the past few years have actually been a misunderstanding between the Power4 and the G4. It would explain rumors like the multiple-cores-on-chip G4 of a few years back and also the 4-way SMP G4. No doubt Apple will have SMP boards like that but i doubt they were thinking of it back then.
IBMs POWER# series are impressive machines, it would be good to continue the trend.
Also there was a rumor linked with this Sahara chip a while back that said the whole chip would somehow produce Altivec-like acceleration as opposed to co-processor. Any1 know anything about this?
[ 06-21-2002: Message edited by: Toast ]</p>
<strong>I think you can take a look at 06/20 and 06/21 news in <a href="http://www.macbidouille.com."; target="_blank">www.macbidouille.com.</a> Very interesting info!</strong><hr></blockquote>
Anybody care to translate? My French isn't very good and Bablefish usually outputs less than I can guess.
<strong>
Anybody care to translate? My French isn't very good and Bablefish usually outputs less than I can guess.</strong><hr></blockquote>
The translation has already been done by SYN in
Topic: A few G5 tidbits. What is interesting here is that macbidouille had this morning the name of the Apple person which gave the info. The name later dissapeared. It seems that we are not dealing with rumors this time.
IBMs POWER# series are impressive machines, it would be good to continue the trend.</strong>
A single core Power4 chip would be interesting yes. It'll need to be able to process all of the AltiVec instructions though.
<strong>Also there was a rumor linked with this Sahara chip a while back that said the whole chip would somehow produce Altivec-like acceleration as opposed to co-processor. Any1 know anything about this?</strong>
I'm not getting very good vibes from statements like "AltiVec-like acceleration as opposed to co-processor" and IBM's vague statements on the issue. To me that sounds like what IBM did for the Nintendo Gekko PPC chip, alter the 64 bit FPU to do two 32 bit floats in the same cycle. That's not a good solution at all.
Since the Sahara places much more emphasis on power savings and low heat than the G4 does, it would be sensible to use an AltiVec compatible Sahara in the iBook.</strong>
There would be no point. Sahara is a 0.13 micron product. A Moto 7455 is a 0.18 micron product. Moving the 7455 to a 0.13 micron fab will give it power consumption close to Sahara. The G4 will be more, but it'll be close enough.
The only thing Sahara might be useful for is an Oqo style computer, or perhaps an ultra-portable. All of Apple's current Macs will be fine with 0.8 to 1.4 GHz G4 CPUs.
<strong>
<a href="http://developer.apple.com/hardware/ve/index.html"; target="_blank">http://developer.apple.com/hardware/ve/index.html</a>;
Most of the links from this page also refer to "AltiVec".
"Velocity Engine" is just Apple's name for the technology, and I would guess that there are two possible reasons for them not using Motorola's name: 1) they didn't like "AltiVec", 2) they wanted to leave the door open to non-Motorola implementations but maintain user awareness.</strong><hr></blockquote>
I was mostly referring to the public side of things. I just found it interesting that Altivec was no longer used on the public side, maybe because they don't intend to keep using Alitvec but they do want a SIMD unit and Velocity Engine is Apple's name, right?
<strong>[qb]Originally posted by Programmer:
Since the Sahara places much more emphasis on power savings and low heat than the G4 does, it would be sensible to use an AltiVec compatible Sahara in the iBook.</strong>
There would be no point. Sahara is a 0.13 micron product. A Moto 7455 is a 0.18 micron product. Moving the 7455 to a 0.13 micron fab will give it power consumption close to Sahara. The G4 will be more, but it'll be close enough.
The only thing Sahara might be useful for is an Oqo style computer, or perhaps an ultra-portable. All of Apple's current Macs will be fine with 0.8 to 1.4 GHz G4 CPUs.[/QB]<hr></blockquote>
Well the idea came from the Apple employee who said that the iBook would stay with the G3 for the foreseeable future. Perhaps there is something about future G3s that we don't know yet?
I think if IBM wants Apple's business they jump on the AltiVec bandwagon... they know the Gekko (or something like it) isn't a good solution -- it was done specifically to fill Nintendo's need. If they did that for them, then they'll probably add the "Velocity Engine" for Apple. IBM seems pretty aggressive about customer-specific processor development.
The G3 did alright at the time without Altivec. I mean...
...if integration into a 'Power 4' desktop variant is all thats holding it up. Probably not though. I'd take alround raw power.
AMD/Intel have SIMD units but if stuff doesn't take advantage then so what? The CPUs have enough raw power without getting too concerned.
Apple needs a better alround CPU. Period.
Yeesh, the G4 is sticking like **** on a stick.
Humph.
Still, in light of your 'bandwidth' comments...it'll be interesting how a DDR and hypertransport enabled G4 holds up to the opposition. Personally, I think Clawhammer will still kick the crap out of the G4 on Hypertransport...because...won't AMD also be using Hypertransport? Here's hoping they can stick another fpu or so onto the G4 while we die of holded breath for that Gthingie.
Lemon Bon Bon
<strong>
Perhaps there is something about future G3s that we don't know yet?</strong><hr></blockquote>
Call me crazy, but I would love to see the return of the G3. An 'ultra elite' version, way cranked up with some wild new cache. I have a quicksilver G4 and a Blue & White G3. I still insist that the G3 is a better computer. Regardless of the rom versions, os updates/reloads, or configurations, the G3 is more stable, and just 'runs better'. What would you think if G3's were overhauled and replaced G4s? <img src="graemlins/lol.gif" border="0" alt="[Laughing]" /> Like I said, sounds crazy, but I wouldn't mind that much. The G3 has always been good to me, and there are many days that it just seems faster. I think my 400 B&W got a IBM chip (during the great G3 chip swap). The Quicksilver has a Moto if I'm not mistaken. Coincidence? Or a message from God about which company makes a better chip?
Superdrive on a G3 even a 2 ghz will be much slower for encoding MPeG 2 than a single one ghz G4.
So no future without altivec.
The comparison with the other simd unit of AMD and INTEL are not relevant because :
- they are less efficient than altivec
- there is too many of them , mmx, 3Dnow, sse, SSe2 : a real nightmare for programmers.
I entirely agree with the Programmer : if IBM wants the Apple market they will do a SIMD unit ativec compatible (no need of a license for that see AMD ...). I will add that sahara is supposed to have a SIMD unit without any further precisions.
The big problem with SIMD execution units is that they are still being coded in assembly language. For conventional processors that pretty much went out the window in the 70's and 80's. SIMD is so new that developers haven't yet figured out how to make effective use of them easily. That doesn't mean its not possible -- it just hasn't been done yet. Since there are so many of them in the market now, its only a matter of time... and processors which aren't equipped SIMD units when this stuff does get figured out will be at a real disadvantage. AltiVec is the best of the lot, so Apple stands to gain the most and it wouldn't surprise me if they are being fairly aggressive with research in this area.
Altivec can be programmed from C. The only test I did is this one. I needed to compute sum(x^2 log x^2), an l2 entropy measure not with complete accuracy.
Standard c code took 1 second using log(), it took 1/3 the time using a lookup table for log() it took 1/21 the time using altivec (yes, 21x faster)
Here is the part of the code FYI.
\tc \t = vec_ld(0, signal);\t\t\t\t/* Load 4 values.\t\t\t\t\t\t\t\t*/
\tsave = c;
\tc \t = vec_madd(c, c, zero);\t\t\t/* Square them.\t\t\t\t\t\t\t\t\t*/
\tl \t = vec_loge(c);\t\t\t\t\t\t/* Take the base 2 log of the squares.\t\t\t*/
\tb \t = vec_cmple(c, zero);\t\t\t\t/* Look for 0 valued coeffs.\t\t\t\t\t*/
\tl = vec_andc(l, b);\t\t\t\t\t/* Set l to zero if c was zero (from -inf)\t\t*/
\tl \t = vec_madd(l, log2, zero);\t\t\t/* Make it a base e log.\t\t\t\t\t\t*/
\tc \t = vec_madd(c, l, zero);\t\t\t/* Multiply c2 by log c2.\t\t\t\t\t\t*/
\tvec_st(c, 0, v0);
\ttheCost = -(*v0 + *v1 + *v2 + *v3);
<strong>Hmm.. SIMD is not that difficult to program, and the latest gcc (in the dev tools) does a descent job at vectorizing your code. There are also some code vectorizer out there that do a good job.
Altivec can be programmed from C. The only test I did is this one. I needed to compute sum(x^2 log x^2), an l2 entropy measure not with complete accuracy.
Standard c code took 1 second using log(), it took 1/3 the time using a lookup table for log() it took 1/21 the time using altivec (yes, 21x faster)
Here is the part of the code FYI.
\tc \t = vec_ld(0, signal);\t\t\t\t/* Load 4 values.\t\t\t\t\t\t\t\t*/
\tsave = c;
\tc \t = vec_madd(c, c, zero);\t\t\t/* Square them.\t\t\t\t\t\t\t\t\t*/
\tl \t = vec_loge(c);\t\t\t\t\t\t/* Take the base 2 log of the squares.\t\t\t*/
\tb \t = vec_cmple(c, zero);\t\t\t\t/* Look for 0 valued coeffs.\t\t\t\t\t*/
\tl = vec_andc(l, b);\t\t\t\t\t/* Set l to zero if c was zero (from -inf)\t\t*/
\tl \t = vec_madd(l, log2, zero);\t\t\t/* Make it a base e log.\t\t\t\t\t\t*/
\tc \t = vec_madd(c, l, zero);\t\t\t/* Multiply c2 by log c2.\t\t\t\t\t\t*/
\tvec_st(c, 0, v0);
\ttheCost = -(*v0 + *v1 + *v2 + *v3);</strong><hr></blockquote>
That is a fairly simple case, and done using the most advanced SIMD unit in the consumer market.
<strong>
The auto-vectorizing is a start, but C/C++ just doesn't lend itself well to vector code unless you put in custom extensions. The AltiVec extensions are fine, but they only work with the AltiVec. There is a lot more potential gain to be had, but it will require a new approach to development.</strong><hr></blockquote>
I remember programming in FORTAN on a CRAY. It was one of the best vectorizing compilers available. Even then, you could get up to a 2x improvement by coding it using the vector primitives yourself instead of letting the compiler do it.
C is just not easy to vectorize. There are attempt to solve this with new languages (i.e. <a href="http://dynamo.ecn.purdue.edu/~hankd/SWAR/Scc.html)," target="_blank">http://dynamo.ecn.purdue.edu/~hankd/SWAR/Scc.html),</a> but performance is so tied to the actual hardware, that it may not be possible.