CS3
Retouch Pro Test
8bit- 34 seconds
16bit- 48 seconds
Congratulations. Your new mac laptop is almost as fast as my old ibm laptop with a slower processor. "8-bit retouch pro test on CS3 beta for windows on a 2.16ghz core 2 duo with 2gb ram (so same specs as yours only Windows): 32 seconds"
Congratulations. Your new mac laptop is almost as fast as my old ibm laptop with a slower processor. "8-bit retouch pro test on CS3 beta for windows on a 2.16ghz core 2 duo with 2gb ram (so same specs as yours only Windows): 32 seconds"
"A bit depressing when the Activity Monitor registers 796% CPU usage on the 8 core versus 395% usage on the 4 core, yet the total time to complete the task is only 3% faster in the case of Photoshop CS3 and 7% faster in the case of Aperture."
Dollars to doughnuts CS3 takes full advantage of the 8 cores.
OK, David, you pushed my buttons and just to punish you, I'm going to explain Amdahl's Law. Because I like you, I'm going to try to make it very simple. But there will be a pop quiz.
Let's suppose that I have a computer benchmark and that the program which processes it consists of 2 phases, A and B and that A takes 90% of the runtime when the benchmark is run on a single processor. So far so good?
[imgl]http://rutt.smugmug.com/photos/144520243-400x400.jpg[/imgl]
Now suppose that Apple and Adobe do some work and now adding more processors makes A go faster. In fact, let's even say that A gets faster in direct proportion to the number of processors we have. So with 4 processors A goes 4x faster than with a single processor. How fast is our benchmark now?
.25 X 90 + 10 = 32.5
Cool! Now what used to take 100 seconds takes only 32.5! We love it so much we shell out for an 8 processor system. How fast is it on our benchmark?
.125 X 90 + 10 = 21.25
But wait a minute? We doubled the number of processors, but the benchmark only got 35% faster. What happened? Remember that pesky B part of the problem is which took 10% of the initial run. It didn't get any faster when we went to 4 processors. And it still didn't get any faster when we went to 8 processors. At that point it was 1/2 the time for the benchmark. In fact, suppose we had 100 processors.
.01 X 90 + 10 = 10.9
which is only about 2x faster than with 8 processors. After all, even if we made A go away, this benchmark is only going to be 90% faster.
The plot shows performance improvements as more processors are added given the optimistic assumptions above. As you should expect, the improvement curve converges to .1. Additional processors can't make it go any faster than that.
So how is this relevant? Almost all computer problems (and certainly the graphical blurs which make up the bulk of these benchmarks) consist of work which can be done in parallel and work which cannot. If we are very very lucky, our task will be like the hypothetical benchmark I just described and 90% will improve proportionally with the addition of new processors. Even so, adding new processors reaches a point of diminishing returns as above. More likely:
Less than 90% of the benchmark can't be parallelized.
The part that can be parallelized can't be made to scale perfectly with the addition of processors. Perhaps it will only get 80% X Number of Processors faster.
As problems grow in size, the number change. Consider video. Suppose that the numbers for processing each individual frame are as above but that we have an hour worth of frames at 60 frames per second. About a quarter of a million frames. Further suppose that each frame can be processed independently (not always true in video as it happens.) Then we can make the whole benchmark scale with the number of processors until we get to a quarter of a million processors. At that point, Amdahl's Law will start to bite us and quadrupling the number of processors would only make processing 32% faster, etc.
Still with me?
David, this doesn't really reflect badly on either Apple or Adobe. It's more like a law of the universe. Apple and Adobe are doing really well in terms of taking advantage of the hardware Intel makes. If we had some numbers for 1, 2, 4, 8 processor systems of the same speed we could plot the results and figure out exactly how much of the benchmark was parallelized and how much was serial. I'll bet you L glass to donuts that the plot will look roughly like the one here, although probably it will flatten out much faster for any of the benchmarks under consideration.
SO what you're saying John, is that going from 1 to 2 processors makes a big jump in performance, but going from 4 to 8 processors makes quite a bit less improvement.
Makes the price of the 8 core less inviting, doesn't it?
Interesting discussion. Thank you for explaining what is going on in multi-processor systems.
OK, David, you pushed my buttons and just to punish you, I'm going to explain Amdahl's Law. Because I like you, I'm going to try to make it very simple.
SO what you're saying John, is that going from 1 to 2 processors makes a big jump in performance, but going from 4 to 8 processors makes quite a bit less improvement.
Yes and this is inherent in the problem itself, and doesn't reflect negatively on Apple so David can stop cheerleading already.
But really, more processors can help some things a lot. Bridge will finish doing it's thing on a big shoot faster. That problem is more like the video problem, each image in pretty much independent of the others so that if you have more than 8 images, you can use 8 processors. If you get i2e to work in batch mode on separate processors (won't be on parallels as I understand it), then you can make it eat through a big shoot fast. But face it, how often to you actually wait for a particular photoshop command to complete on your current system? I wait for surface blur, and I wait for stuff I think is probably disk access, that sort of thing. Well my surface blurs could be faster (and I do use them), but the other stuff isn't going to be helped by more processors.
I do have batch things which could scale with the number of processors, but these particular benchmarks won't capture that.
Gotta love it when people are really good at what they are good at!
But if you have one of those problems which can absorb lots of processors, one of these 8 processor boxes probably isn't the right tool for the job, at least not in terms of, say images processed per hour per dollar. Probably you want more processors without that expensive shared memory bus. I figured this out here before. You can actually get 12 processors on Mac Minis for the same price as one of these Mac Pro 8 things. And that's paying for Apple marketing and software development that you don't really need. Cheaper to build your own PCs and run linux, at least in terms of $/compute. Keep going down that road and the next stop is Pixar and you'll need a warehouse to hold a farm of not quite the latest PCs connected by not quite the fastest network each happily working away on rendering one frame of a movie. At this point, rent, racking, and air conditioning costs figure pretty large in the equation. Next stop, Google, which runs linux on zillions of boxes it builds itself.
That 8 processor thing is basically about bragging rights and perhaps some sort of video editing. True 4 core chips will be out later this year and the prices will drop and become more competitive.
Well, of course the use of the phrase "full advantage" implies that the 8 core will comply with Amdahl's Law, but won't be hampered by niggly little details, spurious little beancounters, and nay-sayers, such as yourself.
Now why don't you go back to being happy with your Mac, and stop asking pesky questions?
One important sense in which the phrase "full advantage" is deceptive in this context is that it implies that the benchmark takes full advantage of the extra 4 processors, i.e., keeps them more or less busy working on the benchmark. But that 8 processor system is spending about 50% of its time (under my optimistic hypothesis about how parallel the benchmarks are) doing the nonparallel part of the problem. When that happens, 7 of the processors are not being used to advantage. So let's see. That means that on average they are 43% idle (roughly.) Since it seems reasonable to hold that 57% is less than full, I'd say it's also reasonable to conclude that the 8 processor system doesn't take full advantage of those processors for this problem.
I'd say you should buy Sid a donut. But then what do I know? Your marketing and legal departments are far better than mine, and that's what really counts here.
Wait! Late breaking news from my legal department: Sid never accepted David's bet. So, David doesn't owe anyone a donut that I can tell.
Well, of course the use of the phrase "full advantage" implies that the 8 core will comply with Amdahl's Law, but won't be hampered by niggly little details, spurious little beancounters, and nay-sayers, such as yourself.
Now why don't you go back to being happy with your Mac, and stop asking pesky questions?
Well, here's the problem, David.
One important sense in which the phrase "full advantage" is deceptive in this context is that it implies that the benchmark takes full advantage of the extra 4 processors, i.e., keeps them more or less busy working on the benchmark. But that 8 processor system is spending about 50% of its time (under my optimistic hypothesis about how parallel the benchmarks are) doing the nonparallel part of the problem. When that happens, 7 of the processors are not being used to advantage. So let's see. That means that on average they are 43% idle (roughly.) Since it seems reasonable to hold that 57% is less than full, I'd say it's also reasonable to conclude that the 8 processor system doesn't take full advantage of those processors for this problem.
One important sense in which the phrase "full advantage" is deceptive in this context is that it implies that the benchmark takes full advantage of the extra 4 processors, i.e., keeps them more or less busy working on the benchmark. But that 8 processor system is spending about 50% of its time (under my optimistic hypothesis about how parallel the benchmarks are) doing the nonparallel part of the problem. When that happens, 7 of the processors are not being used to advantage. So let's see. That means that on average they are 43% idle (roughly.) Since it seems reasonable to hold that 57% is less than full, I'd say it's also reasonable to conclude that the 8 processor system doesn't take full advantage of those processors for this problem.
So I reckon you owe me a doughnut.
Ah. Well I see that your comments are completely incompatible with mine. They couldn't be more contradictory. I suppose that we're just going to have to agree to disagree.
Ah. Well I see that your comments are completely incompatible with mine. They couldn't be more contradictory. I suppose that we're just going to have to agree to disagree.
Sooo... you can get me that doughnut, or an 8 core so I can try it for myself... or I'll trade both to see you crack a smile.
Comments
2.33Ghz 17" Macbook Pro, Intel Core 2 Duo w/ 2Gb RAM:
CS3
Retouch Pro Test
8bit- 34 seconds
16bit- 48 seconds
FM Test
8bit - 16 seconds
16bit - 21 seconds
Portfolio • Workshops • Facebook • Twitter
Congratulations. Your new mac laptop is almost as fast as my old ibm laptop with a slower processor. "8-bit retouch pro test on CS3 beta for windows on a 2.16ghz core 2 duo with 2gb ram (so same specs as yours only Windows): 32 seconds"
Portfolio • Workshops • Facebook • Twitter
8 bit: 9 seconds
16 bit: 16 seconds
EDIT: He will re-run when he gets the full version of CS3, and I will post the results here.
Dgrin FAQ | Me | Workshops
Hmmm... more confirmation that the 8 cores don't really help PS that much.
Catapultam habeo. Nisi pecuniam omnem mihi dabis, ad caput tuum saxum immane mittam
http://www.mcneel.com/users/jb/foghorn/ill_shut_up.au
Dollars to doughnuts CS3 takes full advantage of the 8 cores.
Dgrin FAQ | Me | Workshops
Sure it can, but full advantage of all cores does not equal dramatic performance increase. In this test, the cores are maxed yet the rest of the system, outside Photoshop control, prevents more than a few % improvement.
OK, David, 1 donut for one dollar.
Catapultam habeo. Nisi pecuniam omnem mihi dabis, ad caput tuum saxum immane mittam
http://www.mcneel.com/users/jb/foghorn/ill_shut_up.au
Let's suppose that I have a computer benchmark and that the program which processes it consists of 2 phases, A and B and that A takes 90% of the runtime when the benchmark is run on a single processor. So far so good?
[imgl]http://rutt.smugmug.com/photos/144520243-400x400.jpg[/imgl]
Now suppose that Apple and Adobe do some work and now adding more processors makes A go faster. In fact, let's even say that A gets faster in direct proportion to the number of processors we have. So with 4 processors A goes 4x faster than with a single processor. How fast is our benchmark now?
The plot shows performance improvements as more processors are added given the optimistic assumptions above. As you should expect, the improvement curve converges to .1. Additional processors can't make it go any faster than that.
So how is this relevant? Almost all computer problems (and certainly the graphical blurs which make up the bulk of these benchmarks) consist of work which can be done in parallel and work which cannot. If we are very very lucky, our task will be like the hypothetical benchmark I just described and 90% will improve proportionally with the addition of new processors. Even so, adding new processors reaches a point of diminishing returns as above. More likely:
- Less than 90% of the benchmark can't be parallelized.
- The part that can be parallelized can't be made to scale perfectly with the addition of processors. Perhaps it will only get 80% X Number of Processors faster.
As problems grow in size, the number change. Consider video. Suppose that the numbers for processing each individual frame are as above but that we have an hour worth of frames at 60 frames per second. About a quarter of a million frames. Further suppose that each frame can be processed independently (not always true in video as it happens.) Then we can make the whole benchmark scale with the number of processors until we get to a quarter of a million processors. At that point, Amdahl's Law will start to bite us and quadrupling the number of processors would only make processing 32% faster, etc.Still with me?
David, this doesn't really reflect badly on either Apple or Adobe. It's more like a law of the universe. Apple and Adobe are doing really well in terms of taking advantage of the hardware Intel makes. If we had some numbers for 1, 2, 4, 8 processor systems of the same speed we could plot the results and figure out exactly how much of the benchmark was parallelized and how much was serial. I'll bet you L glass to donuts that the plot will look roughly like the one here, although probably it will flatten out much faster for any of the benchmarks under consideration.
Makes the price of the 8 core less inviting, doesn't it?
Interesting discussion. Thank you for explaining what is going on in multi-processor systems.
Moderator of the Technique Forum and Finishing School on Dgrin
Best. Rutt-dig. Ever!
Portfolio • Workshops • Facebook • Twitter
Yes and this is inherent in the problem itself, and doesn't reflect negatively on Apple so David can stop cheerleading already.
But really, more processors can help some things a lot. Bridge will finish doing it's thing on a big shoot faster. That problem is more like the video problem, each image in pretty much independent of the others so that if you have more than 8 images, you can use 8 processors. If you get i2e to work in batch mode on separate processors (won't be on parallels as I understand it), then you can make it eat through a big shoot fast. But face it, how often to you actually wait for a particular photoshop command to complete on your current system? I wait for surface blur, and I wait for stuff I think is probably disk access, that sort of thing. Well my surface blurs could be faster (and I do use them), but the other stuff isn't going to be helped by more processors.
I do have batch things which could scale with the number of processors, but these particular benchmarks won't capture that.
Not sure that's possible
Portfolio • Workshops • Facebook • Twitter
Told ya so.
Dgrin FAQ | Me | Workshops
Gotta love it when people are really good at what they are good at!
But if you have one of those problems which can absorb lots of processors, one of these 8 processor boxes probably isn't the right tool for the job, at least not in terms of, say images processed per hour per dollar. Probably you want more processors without that expensive shared memory bus. I figured this out here before. You can actually get 12 processors on Mac Minis for the same price as one of these Mac Pro 8 things. And that's paying for Apple marketing and software development that you don't really need. Cheaper to build your own PCs and run linux, at least in terms of $/compute. Keep going down that road and the next stop is Pixar and you'll need a warehouse to hold a farm of not quite the latest PCs connected by not quite the fastest network each happily working away on rendering one frame of a movie. At this point, rent, racking, and air conditioning costs figure pretty large in the equation. Next stop, Google, which runs linux on zillions of boxes it builds itself.
That 8 processor thing is basically about bragging rights and perhaps some sort of video editing. True 4 core chips will be out later this year and the prices will drop and become more competitive.
Catapultam habeo. Nisi pecuniam omnem mihi dabis, ad caput tuum saxum immane mittam
http://www.mcneel.com/users/jb/foghorn/ill_shut_up.au
Tell me which part of this statement is wrong, and I'll buy you your friggin' doughnut!
Dgrin FAQ | Me | Workshops
Sid, now you ask him to define "full advantage".
Catapultam habeo. Nisi pecuniam omnem mihi dabis, ad caput tuum saxum immane mittam
http://www.mcneel.com/users/jb/foghorn/ill_shut_up.au
Well, of course the use of the phrase "full advantage" implies that the 8 core will comply with Amdahl's Law, but won't be hampered by niggly little details, spurious little beancounters, and nay-sayers, such as yourself.
Now why don't you go back to being happy with your Mac, and stop asking pesky questions?
Dgrin FAQ | Me | Workshops
I'd say you should buy Sid a donut. But then what do I know? Your marketing and legal departments are far better than mine, and that's what really counts here.
Wait! Late breaking news from my legal department: Sid never accepted David's bet. So, David doesn't owe anyone a donut that I can tell.
Time to move on.
One important sense in which the phrase "full advantage" is deceptive in this context is that it implies that the benchmark takes full advantage of the extra 4 processors, i.e., keeps them more or less busy working on the benchmark. But that 8 processor system is spending about 50% of its time (under my optimistic hypothesis about how parallel the benchmarks are) doing the nonparallel part of the problem. When that happens, 7 of the processors are not being used to advantage. So let's see. That means that on average they are 43% idle (roughly.) Since it seems reasonable to hold that 57% is less than full, I'd say it's also reasonable to conclude that the 8 processor system doesn't take full advantage of those processors for this problem.
So I reckon you owe me a doughnut.
Catapultam habeo. Nisi pecuniam omnem mihi dabis, ad caput tuum saxum immane mittam
http://www.mcneel.com/users/jb/foghorn/ill_shut_up.au
Ah. Well I see that your comments are completely incompatible with mine. They couldn't be more contradictory. I suppose that we're just going to have to agree to disagree.
Dgrin FAQ | Me | Workshops
Catapultam habeo. Nisi pecuniam omnem mihi dabis, ad caput tuum saxum immane mittam
http://www.mcneel.com/users/jb/foghorn/ill_shut_up.au
Sid, is my humor too dry for you?
Dgrin FAQ | Me | Workshops
Catapultam habeo. Nisi pecuniam omnem mihi dabis, ad caput tuum saxum immane mittam
http://www.mcneel.com/users/jb/foghorn/ill_shut_up.au
Anyway, here are the stats for my new system:
CS2
FredMiranda - 21 seconds
Retouch Pro (8 bit) - 51 seconds
Even if the extra cores don't do much... 9 seconds is still pretty freakin' fast compared to my piddly 51.
I know that when you get older you get slower, but have you gotten around to that yet?
16 bit - 22 Seconds
8 bit - 15 Seconds (That's a 5 second improvement now that PS is a universal binary as opposed to running under Rosetta.)
Charlie
Macbook Pro, 2.4 GHz Core 2 Duo
Teeny bit faster
Retouch Pro Test
8bit- 32 seconds
16bit- 46 seconds
FM Test
8bit - 13 seconds
16bit - 19 seconds
Portfolio • Workshops • Facebook • Twitter