A dual 1.6 gig should look slower than a single 1.8 gig on benchmarks that aren't optimized for multiple processors.
"Benchmarks".
Single-threaded benchmarks never (well, rarely) simulate real-world situations, like in a job situation. It may well demonstrate how a given tasks perform when it, and only it is running alone.
How many of you quit all open applications before you open a new one?
If, for example, one is rendering a huge image in Bryce 5 (just to take a single-threaded example). Will you sit there and look at the image as it is rendering=? No, you want to continue working on other projects, or tasks. Another example could be Photoshop. You fire up a huge image, and start a filter/conversion/whatever that takes so much time, that the progress bar shows up, and crawls forward. Sit there and watch? Waste of time, go ahead and work at something else while Photoshop is progressing in the background. Who cares about the 1.8 being faster on unnoticeable, single-threaded tasks, that doesn't even bother displaying a progress bar?
This is where the DP 1.6 would excel over the SP 1.8, and that is what Apple marketing has to convince its market of.
The situation would then be that the mid tower had slower ram then the low end tower. I think that might be a little odd for the consumer. And would then Apple design the 1.8 model as todays 1.8, or more like the single 1.6 (no PCI-X, less RAM slots). I think it would make for a very odd situation to switch the line up from what it is now, to what you want.
I think what they should do is leave the machines as is, and next rev (in 6 months or whatever) go:
PMG5 SP2.0 - 1899 (with option to drop to 1699 with combo drive)
PMG5 DP2.2 - 2399
PMG5 DP2.6 - 2799
All would be "full" machines (8 ram slots, PCI-X, all with superdrives). If they are going to hit 3.0 GHz in 12 months, this next rev will have a good sized MHz jump.
The situation would then be that the mid tower had slower ram then the low end tower.
Why would the mid-end tower have slower RAM than the low end tower?
You're not forced to use dual DDR 333 in a system that uses 1.6GHz G5's. One of those handles dual DDR 400 just fine, but the memory-bus would be a bit faster than what the one processor could support. With two processors, you'll need even more than dual DDR 400 to keep the processors fed, but DDR 400 is the fastest, affordable, JEDEC-standardized RAM you can get, thus the DP 1.6 would have gotten dual DDR 400 and nothing else. For now.
Single-threaded benchmarks never (well, rarely) simulate real-world situations, like in a job situation. It may well demonstrate how a given tasks perform when it, and only it is running alone.
How many of you quit all open applications before you open a new one?
If, for example, one is rendering a huge image in Bryce 5 (just to take a single-threaded example). Will you sit there and look at the image as it is rendering=? No, you want to continue working on other projects, or tasks. Another example could be Photoshop. You fire up a huge image, and start a filter/conversion/whatever that takes so much time, that the progress bar shows up, and crawls forward. Sit there and watch? Waste of time, go ahead and work at something else while Photoshop is progressing in the background. Who cares about the 1.8 being faster on unnoticeable, single-threaded tasks, that doesn't even bother displaying a progress bar?
This is where the DP 1.6 would excel over the SP 1.8, and that is what Apple marketing has to convince its market of.
The problem is that people do use benchmarks to make purchasing decisions, wether we like it or not. Imagine Apple trying to explain that benchmarks used to show the superiority of the G5 over the Pentium 4 can't be used to compare the various G5 models. Confusing. Not good.
Also, the dual 1.6 gig will win in some real life scenarios but will loose in others. For example, say you have a processor intensive app that isn't optimized for multiple processors. If you run that app while also running several non-processor intensive apps (word processor, spread sheet, e-mail, etc), the 1.8 gig machine should do the processor intensive work faster than a dual 1.6 gig. The reason is that the non-processor intensive apps spend most of their time waiting for you to press a key or click the mouse and don't tie up the CPU while doing that. So the processor intensive app has the 1.8 gig CPU almost 100% of the time. In a dual 1.6 gig the same processor intensive app can at most use 100% of a single 1.6 gig CPU, which of course is slower than 100% of a single 1.8 gig CPU.
If you think it'll be a problem - "cripple" the 1.8, as Matsu suggested, with dual DDR 333, less RAM-slots and regular PCI-slots. It would make it very obvious that the DP 1.6 really is the better (faster) choice.
They would be very equal at single-threaded benchmarks, and the DP 1.6 would kick the low-ends butt at everything else!
If you think it'll be a problem - "cripple" the 1.8, as Matsu suggested, with dual DDR 333, less RAM-slots and regular PCI-slots. It would make it very obvious that the DP 1.6 really is the better (faster) choice.
So you are suggesting that right now Apple change their line up and completely switch the 1.6 and 1.8 specs? I think that would confuse the heck out of some people (but last week I saw a dual 1.8 on Apples webpage, why does my CompUSA have a single 1.8?) And what happens to existing orders? I have a single 1.8 on order. I sure as hell wouldn't want some crippled 1.8, and I would rather spend the little more and get dual 2.0s and instead of dual 1.6s.
Frankly, Apple isn't going to just up and change the specs now. And if you are talking a future rev here, they definatly wont be using 1.6 and 1.8 GHz parts any longer (if they expect to get to 3.0GHz in 12 months). But if we are sitting here playing "Monday Quaterback" (damn I hope thats the right term, can't think now...), here is what I think should have been done (not sure how feasable, but what the heck):
No, I am not suggesting Apple should turn upside-down on its lineup right now. I'm playing "Monday Quarterback", as you call it, it's what I think should have been done to balance orders and avoid delays and shipping problems.
If you think it'll be a problem - "cripple" the 1.8, as Matsu suggested, with dual DDR 333, less RAM-slots and regular PCI-slots. It would make it very obvious that the DP 1.6 really is the better (faster) choice.
They would be very equal at single-threaded benchmarks, and the DP 1.6 would kick the low-ends butt at everything else!
Slower RAM wouldn't necessarily do the trick due to the L1/L2 caches.
The SPECint_base and SPECfp_base benchmarks use only one CPU.
Perhaps more importantly is that third parties will likely run benchmarks that use only one CPU and then use the results to argue Apple is trying to mislead people. Just look at the SPEC benchmark controversy if you want proof. AFAIK, the SPEC benchmarks don't make any significant use of AltiVec and therefore ignore more than half of the processing power in the 970 chip.
To avoid these problems, I would make sure the clock frequency in a higher end machine is never lower than in a lower end machine.
You're right, I overlooked the SPECfp/int_base-scores, but notice one thing: They only display the SPEC-results of their fastest model. No mention of the SP models. Also, SPECfp/int_rate_base takes advantage of dual processors. No problems with the SPEC-benchmarks at all, IMO.
I also think any self-respecting benchmark-site would, with a dual machine, benchmark with alot of multithreaded benchmarks - even barefeats does this. They'd put empasize on the dual processing capability of the mac, and I believe most of them would at least have in their conclusion that the DP would be best at any multithreaded example, and in any case when multitasking.
I'm probably making a mistake here, by underestimating the stupidity of humans here now.
Consumers aren't dumb enough to ignore the benefits of dual processors. They realize two processors are better than one, and they may even overestimate the advantage by thinking that 1.6GHzX1.6GHz = 2.56GHz. If consumers couldn't tell the difference between a single and a slightly slower dual, then Apple wouldn't have had all those configurations in which the top end MHz machine was a single. (Single 867, Dual 800, etc. . .)
In any case, I'm still wondering whether anyone knows if the 1.8GHz has that vacant second socket. That's really important. The choice between a 1.6 and a 1.8 hinges on that fact for me.
Anyone remember when the top of the line was Dual 800 and the middle was single 867? Apple has no problem with the a cheaper machine having a faster processor.
Another reason why we hear nothing but crickets and the wind swirling through the trees these days, regarding the G5 "benchmarks-as-bogus-and-fixed" controversy is because they were run using SUB-optimal code. In fact they were run using code specifically designed to take advantage of the G4; 90% of which is noting but crap anyway compared to what the PC-people have with respect to effort and code-optimization development.
The developers blamed Apple for a lot of things, but the reality is that developers did little to take full advantage of Apple's hardware and OS. Sure there were bugs; however, Apple debunked a lot of that when they started releasing their own software to show what could *really* be done if developers took the time and made the effort. The G5 takes all their crappy code and runs it that much faster. There should be no excuses now. Another thing regarding SPEC in relation to the benchmarks is the compiler.
Regarding GCC, we've got a quote directly from one of the chief architects of the 970 as well as an architect working on GCC. The bottom line is that GCC on x86 has had far more lead time and contains vastly more optimization than it does for GCC on PPC. What's more, GCC is hobbled EVEN FURTHER on the 970...
Quote from what Hannibal learned during his interview:
Quote:
One interesting fact that I learned about gcc is that the Power4 and 970's peculiar group dispatch scheme and issue queue structure doesn't quite fit with gcc's internal model of what a processor should look like. As a result, the L1 cache latency number (and I think some of the other numbers, as well) in the machine description file had to be altered from their true value in order to get the best performance out of gcc.
OK, now for David Edelsohn's direct comment:
Quote:
David Edelsohn: ...and gcc is modeling this [i.e. the L1 cache latency] with one more additional cycle because we've seen a benefit... The gcc scheduler is not really designed ideally for a processor like the 970 and the Power4 and others, and that's a lot of what the IBM and Apple teams have worked on, due to the complexity of the processor with the dispatch groups and the whole way it dispatches and issues and what parts are in-order and what parts are out-of-order, trying to better instruct the compiler how to arrange code to match that. So there are certain places where we give it [i.e. gcc] information that's more ideal for what it needs to generate than for exactly describing the processor. So again that's a lot of what IBM and Apple have been working on... that was one of the things that we're continuing to work on to try to get the best performance out of the processor.
In other words, they are trying to improve GCC performance on the 970 by making it better architected for this new processor. They've only just begun to fix GCC so it will work better with the 970. PC people sure jumped to conclusions a little too quick that time though.
In many cases a dual processor G4 will excel over a single processor G5 at 1.6 or 1.8. There is nothing better than SMP do improve responsivenees and performance.
Sure you will not see a peformance increase with single thread applications, but that is only one way to measure performance. Responsiveness of the system with respect to the user or the network are also valid measures of performance. As in the abiltiy of the machine ot handle utility tasks while the program doing the gross work is running.
For Apple to have come up short with SMP based G5's is a bit hard to understand after conditioning everyone with the SMP based G4's. I'm sure beyond any doubt that if Apple had released a middle range SMP machine (at a reasonable price) it would be selling as well as the current high end machine. Considering how optimized MAC/OS X and many of the key MAC applications are for SMP I can not see a great deal of interest in single processor G5 machines. Lets face it for the non pro Apple still haves the iMac line which would perform fine in many applications. The markets that Apple plays in, on the professional scale, often benefit from the performance that SMP brings.
Dave
Quote:
Originally posted by Zapchud
This is where the DP 1.6 would excel over the SP 1.8, and that is what Apple marketing has to convince its market of.
Comments
Originally posted by Tidris
A dual 1.6 gig should look slower than a single 1.8 gig on benchmarks that aren't optimized for multiple processors.
"Benchmarks".
Single-threaded benchmarks never (well, rarely) simulate real-world situations, like in a job situation. It may well demonstrate how a given tasks perform when it, and only it is running alone.
How many of you quit all open applications before you open a new one?
If, for example, one is rendering a huge image in Bryce 5 (just to take a single-threaded example). Will you sit there and look at the image as it is rendering=? No, you want to continue working on other projects, or tasks. Another example could be Photoshop. You fire up a huge image, and start a filter/conversion/whatever that takes so much time, that the progress bar shows up, and crawls forward. Sit there and watch? Waste of time, go ahead and work at something else while Photoshop is progressing in the background. Who cares about the 1.8 being faster on unnoticeable, single-threaded tasks, that doesn't even bother displaying a progress bar?
This is where the DP 1.6 would excel over the SP 1.8, and that is what Apple marketing has to convince its market of.
As I commented before, the real key is to unlock the bus ratios so that Apple can mix and match procs and mobo's to better suit their marketing needs.
Originally posted by Matsu
I think a good lineup would look like this:
CUBOID redux SP1.6 from 1299-1599
PMG5 SP1.8 1999
PMG5 DP1.6 2499
PMG5 DP2ghz 2999
The situation would then be that the mid tower had slower ram then the low end tower. I think that might be a little odd for the consumer. And would then Apple design the 1.8 model as todays 1.8, or more like the single 1.6 (no PCI-X, less RAM slots). I think it would make for a very odd situation to switch the line up from what it is now, to what you want.
I think what they should do is leave the machines as is, and next rev (in 6 months or whatever) go:
PMG5 SP2.0 - 1899 (with option to drop to 1699 with combo drive)
PMG5 DP2.2 - 2399
PMG5 DP2.6 - 2799
All would be "full" machines (8 ram slots, PCI-X, all with superdrives). If they are going to hit 3.0 GHz in 12 months, this next rev will have a good sized MHz jump.
Originally posted by kupan787
The situation would then be that the mid tower had slower ram then the low end tower.
Why would the mid-end tower have slower RAM than the low end tower?
You're not forced to use dual DDR 333 in a system that uses 1.6GHz G5's. One of those handles dual DDR 400 just fine, but the memory-bus would be a bit faster than what the one processor could support. With two processors, you'll need even more than dual DDR 400 to keep the processors fed, but DDR 400 is the fastest, affordable, JEDEC-standardized RAM you can get, thus the DP 1.6 would have gotten dual DDR 400 and nothing else. For now.
Originally posted by Zapchud
"Benchmarks".
Single-threaded benchmarks never (well, rarely) simulate real-world situations, like in a job situation. It may well demonstrate how a given tasks perform when it, and only it is running alone.
How many of you quit all open applications before you open a new one?
If, for example, one is rendering a huge image in Bryce 5 (just to take a single-threaded example). Will you sit there and look at the image as it is rendering=? No, you want to continue working on other projects, or tasks. Another example could be Photoshop. You fire up a huge image, and start a filter/conversion/whatever that takes so much time, that the progress bar shows up, and crawls forward. Sit there and watch? Waste of time, go ahead and work at something else while Photoshop is progressing in the background. Who cares about the 1.8 being faster on unnoticeable, single-threaded tasks, that doesn't even bother displaying a progress bar?
This is where the DP 1.6 would excel over the SP 1.8, and that is what Apple marketing has to convince its market of.
The problem is that people do use benchmarks to make purchasing decisions, wether we like it or not. Imagine Apple trying to explain that benchmarks used to show the superiority of the G5 over the Pentium 4 can't be used to compare the various G5 models. Confusing. Not good.
Also, the dual 1.6 gig will win in some real life scenarios but will loose in others. For example, say you have a processor intensive app that isn't optimized for multiple processors. If you run that app while also running several non-processor intensive apps (word processor, spread sheet, e-mail, etc), the 1.8 gig machine should do the processor intensive work faster than a dual 1.6 gig. The reason is that the non-processor intensive apps spend most of their time waiting for you to press a key or click the mouse and don't tie up the CPU while doing that. So the processor intensive app has the 1.8 gig CPU almost 100% of the time. In a dual 1.6 gig the same processor intensive app can at most use 100% of a single 1.6 gig CPU, which of course is slower than 100% of a single 1.8 gig CPU.
They would be very equal at single-threaded benchmarks, and the DP 1.6 would kick the low-ends butt at everything else!
The benchmarks used on Apples Powermac Performance site would display the DP 1.6 as faster, except for the BLAST one perhaps.
Put some heavy emphasize on that duals kick ass, as his Steveness would say it.
Problem should be solved!
Originally posted by Zapchud
If you think it'll be a problem - "cripple" the 1.8, as Matsu suggested, with dual DDR 333, less RAM-slots and regular PCI-slots. It would make it very obvious that the DP 1.6 really is the better (faster) choice.
So you are suggesting that right now Apple change their line up and completely switch the 1.6 and 1.8 specs? I think that would confuse the heck out of some people (but last week I saw a dual 1.8 on Apples webpage, why does my CompUSA have a single 1.8?) And what happens to existing orders? I have a single 1.8 on order. I sure as hell wouldn't want some crippled 1.8, and I would rather spend the little more and get dual 2.0s and instead of dual 1.6s.
Frankly, Apple isn't going to just up and change the specs now. And if you are talking a future rev here, they definatly wont be using 1.6 and 1.8 GHz parts any longer (if they expect to get to 3.0GHz in 12 months). But if we are sitting here playing "Monday Quaterback" (damn I hope thats the right term, can't think now...), here is what I think should have been done (not sure how feasable, but what the heck):
Single 1.6GHz/4 RAM Slots/256 MB RAM/PCI/80 GB PATA/Combo Drive - $1599
Dual 1.6GHz/4 RAM Slots/256 MB RAM/PCI X/80GB SATA/Combo Drive - $1899
Dual 1.8GHz/6 RAM Slots/512 MB RAM/PCI X/120GB SATA/SuperDrive - $2399
Dual 2.0GHz/8 RAM Slots/512 MB RAM/PCI X/160GB SATA/SuperDrive - $2799
Originally posted by Zapchud
If you think it'll be a problem - "cripple" the 1.8, as Matsu suggested, with dual DDR 333, less RAM-slots and regular PCI-slots. It would make it very obvious that the DP 1.6 really is the better (faster) choice.
They would be very equal at single-threaded benchmarks, and the DP 1.6 would kick the low-ends butt at everything else!
The benchmarks used on Apples Powermac Performance site would display the DP 1.6 as faster, except for the BLAST one perhaps.
Slower RAM wouldn't necessarily do the trick due to the L1/L2 caches.
The SPECint_base and SPECfp_base benchmarks use only one CPU.
Perhaps more importantly is that third parties will likely run benchmarks that use only one CPU and then use the results to argue Apple is trying to mislead people. Just look at the SPEC benchmark controversy if you want proof. AFAIK, the SPEC benchmarks don't make any significant use of AltiVec and therefore ignore more than half of the processing power in the 970 chip.
To avoid these problems, I would make sure the clock frequency in a higher end machine is never lower than in a lower end machine.
I also think any self-respecting benchmark-site would, with a dual machine, benchmark with alot of multithreaded benchmarks - even barefeats does this. They'd put empasize on the dual processing capability of the mac, and I believe most of them would at least have in their conclusion that the DP would be best at any multithreaded example, and in any case when multitasking.
I'm probably making a mistake here, by underestimating the stupidity of humans here now.
In any case, I'm still wondering whether anyone knows if the 1.8GHz has that vacant second socket. That's really important. The choice between a 1.6 and a 1.8 hinges on that fact for me.
The developers blamed Apple for a lot of things, but the reality is that developers did little to take full advantage of Apple's hardware and OS. Sure there were bugs; however, Apple debunked a lot of that when they started releasing their own software to show what could *really* be done if developers took the time and made the effort. The G5 takes all their crappy code and runs it that much faster. There should be no excuses now. Another thing regarding SPEC in relation to the benchmarks is the compiler.
Regarding GCC, we've got a quote directly from one of the chief architects of the 970 as well as an architect working on GCC. The bottom line is that GCC on x86 has had far more lead time and contains vastly more optimization than it does for GCC on PPC. What's more, GCC is hobbled EVEN FURTHER on the 970...
Quote from what Hannibal learned during his interview:
One interesting fact that I learned about gcc is that the Power4 and 970's peculiar group dispatch scheme and issue queue structure doesn't quite fit with gcc's internal model of what a processor should look like. As a result, the L1 cache latency number (and I think some of the other numbers, as well) in the machine description file had to be altered from their true value in order to get the best performance out of gcc.
OK, now for David Edelsohn's direct comment:
David Edelsohn: ...and gcc is modeling this [i.e. the L1 cache latency] with one more additional cycle because we've seen a benefit... The gcc scheduler is not really designed ideally for a processor like the 970 and the Power4 and others, and that's a lot of what the IBM and Apple teams have worked on, due to the complexity of the processor with the dispatch groups and the whole way it dispatches and issues and what parts are in-order and what parts are out-of-order, trying to better instruct the compiler how to arrange code to match that. So there are certain places where we give it [i.e. gcc] information that's more ideal for what it needs to generate than for exactly describing the processor. So again that's a lot of what IBM and Apple have been working on... that was one of the things that we're continuing to work on to try to get the best performance out of the processor.
In other words, they are trying to improve GCC performance on the 970 by making it better architected for this new processor. They've only just begun to fix GCC so it will work better with the 970. PC people sure jumped to conclusions a little too quick that time though.
--
Ed
Sure you will not see a peformance increase with single thread applications, but that is only one way to measure performance. Responsiveness of the system with respect to the user or the network are also valid measures of performance. As in the abiltiy of the machine ot handle utility tasks while the program doing the gross work is running.
For Apple to have come up short with SMP based G5's is a bit hard to understand after conditioning everyone with the SMP based G4's. I'm sure beyond any doubt that if Apple had released a middle range SMP machine (at a reasonable price) it would be selling as well as the current high end machine. Considering how optimized MAC/OS X and many of the key MAC applications are for SMP I can not see a great deal of interest in single processor G5 machines. Lets face it for the non pro Apple still haves the iMac line which would perform fine in many applications. The markets that Apple plays in, on the professional scale, often benefit from the performance that SMP brings.
Dave
Originally posted by Zapchud
This is where the DP 1.6 would excel over the SP 1.8, and that is what Apple marketing has to convince its market of.
Originally posted by Bigc
uh oh, the AI cardinal sin, using the term MAC, naughty naughty....but point well taken.
MAC=Macs Are Computers.
cMac 900 G3 No display $499/ cMac 1.6Ghz G5 $1699
eMac 1Ghz G3 $819/ eMac 1Ghz G4 $949
iMac 1.1Ghz G3 $1399/ iMac 1.4 Ghz G5 $1799/
PM G5 Dual 1.6 $2299/ PM G5 Dual 1.8 $2699/ PM G5 Dual 2.0 $2999
Yes, a bit messy, but probably the best lineup. A cMac ("cheap Mac") would have no display and look like the bottom part of an iMac.
Of course they would never do this, that's just my opinion.
By the time the G5 ships I'm getting the feeling I may as well await the rev B...
I can't wait to see a G5 in the flesh. They do look stunning.
Apple must be quite busy now...finishing Panther.
But what of the iBook, iMac2 and Powerbook updates?
And maybe another DLD to build upon the success of the iPod?
Then...X-serve? Surely they aint going to sell anymore of them until the G5 enters the line..?
Random thoughts.
Lemon Bon Bon
Originally posted by hmurchison
Don't count on it. How many people have 4 SATA drives.
I need 10 drives, plus NAS.