I don't see why not. Converting a big endian RISC instruction set into a very different instruction set: I'd guess it takes about 10 instructions to handle each 1 PPC instruction. It's also a very single threaded task. It's possible that they implemented a lookahead buffer in a second thread, but that's not going to really speed that much up aside from perhaps giving you a better JIT compilation.
You forgot perhaps the benchmarks with the developer machines. Of course there is a serious performance hit since translation takes place, but in some tests the developer box matches or exceeds the performance of a dual 2.5 GHz G5.
Quote:
If I said >5x would you still argue? Don't forget that if there's Altivec code in there, It's probably on the order of >40x. Maybe even >80x given that G5s have dual Altivec cores per cpu core.
Altivec support has been added in Rosetta. With the first Intel iMacs already out, we will just have to wait and see what real world situations will reveal.
Those benchmarks seem to average out at around 15-20x. A 2 to 3x improvement seems logical given a little software maturity and the presumed speed benefit of the Yonah. We will ultimately have to wait to see the "real" results, yes, but there's really no way around the fact that PPC software will run MUCH faster on PPCs, even when there is one fewer core.
I'm more peeved about the fact that I could have gotten a $125 rebate on a G5 iMac if I had waited a week, but then again my ADC discount lapsed on the 6th, and I got $300 off for that.
(I see no reason to renew this year, since I'm guessing that it will be 12 more months of Intel Transition stuff.)
Those benchmarks seem to average out at around 15-20x.
What are you talking about? The Xbench result for the dual G5 is around 200, while the single-CPU developer machine got around 70. And this is the crappy Xbench and the first version of Rosetta. I am looking forward to real tests.
I don't see why not. Converting a big endian RISC instruction set into a very different instruction set: I'd guess it takes about 10 instructions to handle each 1 PPC instruction. It's also a very single threaded task. It's possible that they implemented a lookahead buffer in a second thread, but that's not going to really speed that much up aside from perhaps giving you a better JIT compilation.
If I said >5x would you still argue? Don't forget that if there's Altivec code in there, It's probably on the order of >40x. Maybe even >80x given that G5s have dual Altivec cores per cpu core.
I'm not going to argue about how slow / fast it is... but I would like to point out that CISC uses fewer instructions to accomplish things than RISC. This is what people loved about RISC... it was even more lowend than CISC. I'm not totally familiar with RISC assembly... but a MRMW maybe accomplished with 1 instruction on a CISC but may take 2 with a RISC cpu. This maybe help out a bit... *shrugs* just something that crossed my mind
No doubt that a G5 will be faster on PPC code than Rosetta on an Intel Mac. However, I wonder how much real difference there would be between a G4 powerbook (with it's starved FSB) and a nicely spec'd MacBook Pro?
What are you talking about? The Xbench result for the dual G5 is around 200, while the single-CPU developer machine got around 70. And this is the crappy Xbench and the first version of Rosetta. I am looking forward to real tests.
Read the rest of the tests, which are far more real-world.
but a MRMW maybe accomplished with 1 instruction on a CISC but may take 2 with a RISC cpu. This maybe help out a bit... *shrugs* just something that crossed my mind
Not likely. A lot of the instructions in the x86 set are deprecated, and are from the days of 8 and 16bit computers. The Yonah Core has more than one crack stages which synthesize "x86 instructions" into "modern x86 instructions."
Good compilers, such as Intel's Compiler, will completely avoid instructions that are converted into multiple instructions in the crack stage. Rosetta probably does the same. At this level, the PPC ISA isn't much reduced from the Yonah ISA.
You may note that the G5 also does this, since Motorola insisted on a buttload of redundant Register Move Instructions in the PPC ISA. The G5 turns these into POWER instructions in hardware.
Read the rest of the tests, which are far more real-world.
Let me quote explicitly form the article:
The Intel Mac scored 82 on the Thread test, compared to 225 for the G5 machine. In the Computation Thread test, the Pentium machine scored 110, trailing the G5 by only 45 points.
In both the Lock Contention and Memory tests, the G5 significantly exceeded the performance of the Intel Mac, with scores of 420 to 66, and 378 to 214. The Intel Mac managed to exceed the G5's Stream Memory Test score: 351 to 319. The system memory test, however, is a different story. The G5 beat the Intel machine, scoring a 464 compared to 154.
The Intel Mac scored a 125 on the Interface Test, compared to a 380 for the G5, according to one report. Another report claimed the Intel prototype beat a G5 Mac.
The Intel Mac scored well in both the Quartz graphics and OpenGL graphics tests -- matching or exceeding dual-2.5GHz G5 score.
And unless I am missing something, the rest is just parts of the Xbench test. How you get the 15-20x factor is beyond me.
No doubt that a G5 will be faster on PPC code than Rosetta on an Intel Mac. However, I wonder how much real difference there would be between a G4 powerbook (with it's starved FSB) and a nicely spec'd MacBook Pro?
There are people saying that they would be about the same, but I will believe it when I see it.
And unless I am missing something, the rest is just parts of the Xbench test. How you get the 15-20x factor is beyond me.
What's this?
In the CPU test, where G5 systems score between 100 and 200, the Intel Mac only reached the high teens.
In other words, the stuff that's annoying to wait for (i.e. rendering, searching, sorting, any form of computing) blows in Rosetta. Factor in Altivec, which isn't usually part of these tests, and the number goes up.
Everyone likes to hate Altivec now, I guess because it's not in Intel chips and one can't be a fan boy by knowing that orphaned hardware has a leg up on the platform's future. But a lot of stuff used Altivec - more than you'd think - and it's an absolute blessing to anyone who works with graphics.
This is all nice, but it is academic. Those for whom third party PPC code remains the operating environ will be best served by their current PPC macs. Howver the question of upgrading becomes interesting, and it all depends on where you're coming from.
For example. I have an 867Mhz Powerbook. If I loaded my copy of CS1 onto an Intel book to run on Rosetta, how would the performance compare? And not to be discounted is the possibility of running an X86 windows version (which I also have) under a dual boot, or X86 VPC environment. Someone will likely provide a wrapper of some sort that let's you basicallly run a full speed PC off a portion of your HDD. That could work as a solution.
I can afford to wait, so I will, and really, once dual binary versions of the major apps come out, this all becomes academic anyway...
This is all nice, but it is academic. Those for whom third party PPC code remains the operating environ will be best served by their current PPC macs. Howver the question of upgrading becomes interesting, and it all depends on where you're coming from.
For example. I have an 867Mhz Powerbook. If I loaded my copy of CS1 onto an Intel book to run on Rosetta, how would the performance compare? And not to be discounted is the possibility of running an X86 windows version (which I also have) under a dual boot, or X86 VPC environment. Someone will likely provide a wrapper of some sort that let's you basicallly run a full speed PC off a portion of your HDD. That could work as a solution.
I can afford to wait, so I will, and really, once dual binary versions of the major apps come out, this all becomes academic anyway...
In your case, I'd expect the PPC version of Photoshop will run faster on a new intel MacBook than any version on your old PowerBook. The only people who should debate the issue are those with recent PowerBooks and even then its a debate since the new MacBooks are dual core.
The more I think about the new iMacs and MacBooks the more capable I think they are. I'll know more after I go back to Macworld today.
Everyone complaining about buying a G5 iMac in December should feel pity for me. I bought a Color Classic back in the day. But only after I had purchased a Centris 610.
I BOUGHT AN iMAC YESTERDAY! Woot! A Duo, with 2 GB RAM, 256 MB video. When I ordered it, it was 3-5 days shipping. As of this morning, the ship date had slipped to Jan. 19th. Dang it. But it's all good! Apple is alive and well, and frankly, I'd rather be complaining about ship dates and reading rumor boards about this exceptionally innovative company than troubleshooting my XP box in a world where that was the only option.
So, I'm extra excited because this is my first OS X computer... Woot! Woot! Woot! I feel a bit like Vader as he sailed down the chasm after vindicating his son's belief that there was still some good left in him.
I BOUGHT AN iMAC YESTERDAY! Woot! A Duo, with 2 GB RAM, 256 MB video. When I ordered it, it was 3-5 days shipping. As of this morning, the ship date had slipped to Dec. 19th. Dang it. But it's all good! Apple is alive and well, and frankly, I'd rather be complaining about ship dates and reading rumor boards about this exceptionally innovative company than troubleshooting my XP box in a world where that was the only option.
So, I'm extra excited because this is my first OS X computer... Woot! Woot! Woot! I feel a bit like Vader as he sailed down the chasm after vindicating his son's belief that there was still some good left in him.
Comments
Originally posted by Splinemodel
I don't see why not. Converting a big endian RISC instruction set into a very different instruction set: I'd guess it takes about 10 instructions to handle each 1 PPC instruction. It's also a very single threaded task. It's possible that they implemented a lookahead buffer in a second thread, but that's not going to really speed that much up aside from perhaps giving you a better JIT compilation.
You forgot perhaps the benchmarks with the developer machines. Of course there is a serious performance hit since translation takes place, but in some tests the developer box matches or exceeds the performance of a dual 2.5 GHz G5.
If I said >5x would you still argue? Don't forget that if there's Altivec code in there, It's probably on the order of >40x. Maybe even >80x given that G5s have dual Altivec cores per cpu core.
Altivec support has been added in Rosetta. With the first Intel iMacs already out, we will just have to wait and see what real world situations will reveal.
I'm more peeved about the fact that I could have gotten a $125 rebate on a G5 iMac if I had waited a week, but then again my ADC discount lapsed on the 6th, and I got $300 off for that.
(I see no reason to renew this year, since I'm guessing that it will be 12 more months of Intel Transition stuff.)
Originally posted by Splinemodel
Those benchmarks seem to average out at around 15-20x.
What are you talking about?
Originally posted by Splinemodel
I don't see why not. Converting a big endian RISC instruction set into a very different instruction set: I'd guess it takes about 10 instructions to handle each 1 PPC instruction. It's also a very single threaded task. It's possible that they implemented a lookahead buffer in a second thread, but that's not going to really speed that much up aside from perhaps giving you a better JIT compilation.
If I said >5x would you still argue? Don't forget that if there's Altivec code in there, It's probably on the order of >40x. Maybe even >80x given that G5s have dual Altivec cores per cpu core.
I'm not going to argue about how slow / fast it is... but I would like to point out that CISC uses fewer instructions to accomplish things than RISC. This is what people loved about RISC... it was even more lowend than CISC. I'm not totally familiar with RISC assembly... but a MRMW maybe accomplished with 1 instruction on a CISC but may take 2 with a RISC cpu. This maybe help out a bit... *shrugs* just something that crossed my mind
Originally posted by PB
What are you talking about?
Read the rest of the tests, which are far more real-world.
Originally posted by emig647
but a MRMW maybe accomplished with 1 instruction on a CISC but may take 2 with a RISC cpu. This maybe help out a bit... *shrugs* just something that crossed my mind
Not likely. A lot of the instructions in the x86 set are deprecated, and are from the days of 8 and 16bit computers. The Yonah Core has more than one crack stages which synthesize "x86 instructions" into "modern x86 instructions."
Good compilers, such as Intel's Compiler, will completely avoid instructions that are converted into multiple instructions in the crack stage. Rosetta probably does the same. At this level, the PPC ISA isn't much reduced from the Yonah ISA.
You may note that the G5 also does this, since Motorola insisted on a buttload of redundant Register Move Instructions in the PPC ISA. The G5 turns these into POWER instructions in hardware.
Originally posted by Splinemodel
It's reasonable to believe that Rosetta uses only the modern x86 ISA, which is not much less RISC than the POWER ISA.
Sorry I don't understand the last part of that sentence... Are you saying the instructions are bit more low level in the new x86 instruction sets?
Originally posted by Splinemodel
Read the rest of the tests, which are far more real-world.
Let me quote explicitly form the article:
The Intel Mac scored 82 on the Thread test, compared to 225 for the G5 machine. In the Computation Thread test, the Pentium machine scored 110, trailing the G5 by only 45 points.
In both the Lock Contention and Memory tests, the G5 significantly exceeded the performance of the Intel Mac, with scores of 420 to 66, and 378 to 214. The Intel Mac managed to exceed the G5's Stream Memory Test score: 351 to 319. The system memory test, however, is a different story. The G5 beat the Intel machine, scoring a 464 compared to 154.
The Intel Mac scored a 125 on the Interface Test, compared to a 380 for the G5, according to one report. Another report claimed the Intel prototype beat a G5 Mac.
The Intel Mac scored well in both the Quartz graphics and OpenGL graphics tests -- matching or exceeding dual-2.5GHz G5 score.
And unless I am missing something, the rest is just parts of the Xbench test. How you get the 15-20x factor is beyond me.
Originally posted by Matsu
No doubt that a G5 will be faster on PPC code than Rosetta on an Intel Mac. However, I wonder how much real difference there would be between a G4 powerbook (with it's starved FSB) and a nicely spec'd MacBook Pro?
There are people saying that they would be about the same, but I will believe it when I see it.
Originally posted by PB
Let me quote explicitly form the article:
And unless I am missing something, the rest is just parts of the Xbench test. How you get the 15-20x factor is beyond me.
What's this?
In the CPU test, where G5 systems score between 100 and 200, the Intel Mac only reached the high teens.
In other words, the stuff that's annoying to wait for (i.e. rendering, searching, sorting, any form of computing) blows in Rosetta. Factor in Altivec, which isn't usually part of these tests, and the number goes up.
Everyone likes to hate Altivec now, I guess because it's not in Intel chips and one can't be a fan boy by knowing that orphaned hardware has a leg up on the platform's future. But a lot of stuff used Altivec - more than you'd think - and it's an absolute blessing to anyone who works with graphics.
Originally posted by Splinemodel
What's this?
In the CPU test, where G5 systems score between 100 and 200, the Intel Mac only reached the high teens.
Sorry, I am not native english speaker. What "high teens" stands for?
Originally posted by PB
Sorry, I am not native english speaker. What "high teens" stands for?
Young adults using narcotics.
For example. I have an 867Mhz Powerbook. If I loaded my copy of CS1 onto an Intel book to run on Rosetta, how would the performance compare? And not to be discounted is the possibility of running an X86 windows version (which I also have) under a dual boot, or X86 VPC environment. Someone will likely provide a wrapper of some sort that let's you basicallly run a full speed PC off a portion of your HDD. That could work as a solution.
I can afford to wait, so I will, and really, once dual binary versions of the major apps come out, this all becomes academic anyway...
Originally posted by Matsu
This is all nice, but it is academic. Those for whom third party PPC code remains the operating environ will be best served by their current PPC macs. Howver the question of upgrading becomes interesting, and it all depends on where you're coming from.
For example. I have an 867Mhz Powerbook. If I loaded my copy of CS1 onto an Intel book to run on Rosetta, how would the performance compare? And not to be discounted is the possibility of running an X86 windows version (which I also have) under a dual boot, or X86 VPC environment. Someone will likely provide a wrapper of some sort that let's you basicallly run a full speed PC off a portion of your HDD. That could work as a solution.
I can afford to wait, so I will, and really, once dual binary versions of the major apps come out, this all becomes academic anyway...
In your case, I'd expect the PPC version of Photoshop will run faster on a new intel MacBook than any version on your old PowerBook. The only people who should debate the issue are those with recent PowerBooks and even then its a debate since the new MacBooks are dual core.
The more I think about the new iMacs and MacBooks the more capable I think they are. I'll know more after I go back to Macworld today.
Originally posted by Xool
The more I think about the new iMacs and MacBooks the more capable I think they are. I'll know more after I go back to Macworld today.
And, please, run some tests, if possible, for the poor rest of us.
Anyway, check out http://abs.apple.com/transition/.
But the biggest news of all, is....
I BOUGHT AN iMAC YESTERDAY! Woot! A Duo, with 2 GB RAM, 256 MB video. When I ordered it, it was 3-5 days shipping. As of this morning, the ship date had slipped to Jan. 19th. Dang it. But it's all good! Apple is alive and well, and frankly, I'd rather be complaining about ship dates and reading rumor boards about this exceptionally innovative company than troubleshooting my XP box in a world where that was the only option.
So, I'm extra excited because this is my first OS X computer... Woot! Woot! Woot! I feel a bit like Vader as he sailed down the chasm after vindicating his son's belief that there was still some good left in him.
[Edit to fix ship date]
Originally posted by PB
Sorry, I am not native english speaker. What "high teens" stands for?
Originally posted by godrifle
I BOUGHT AN iMAC YESTERDAY! Woot! A Duo, with 2 GB RAM, 256 MB video. When I ordered it, it was 3-5 days shipping. As of this morning, the ship date had slipped to Dec. 19th. Dang it. But it's all good! Apple is alive and well, and frankly, I'd rather be complaining about ship dates and reading rumor boards about this exceptionally innovative company than troubleshooting my XP box in a world where that was the only option.
So, I'm extra excited because this is my first OS X computer... Woot! Woot! Woot! I feel a bit like Vader as he sailed down the chasm after vindicating his son's belief that there was still some good left in him.
Congrats for your purchase and welcome!