Speed tests show 1 Ghz DDR faster than 1 Ghz pre-DDR

I own a Dual/Ghz/DDR and I do love the thing especially having a machine that is as fast as the previous top of the line for a lot less money. I do believe that it is faster then the previous duals in one situation that is multitasking several programs at once. The more programs you're running on this machine the more it shines because it just doesn't slow down.

That said the tests on the railhead website have a flaw. The tests a) are not multiprocessor aware b) are not altivec aware. This leads to a problem the old G4 has a larger L3 cache which doesn't really come in to play unless you are doing altivec aware tasks. So in my opinion these tests have basically illuminated the L3 cache from the test. That being the case you would expect the new machine to shine it has a faster bus then the previous machines. Which does make a big difference.

Something else that I haven't really been shown is the video performance comparisons between the two machines. This is where I expect the new DDR machines to really shine. The new machines with DMA (Direct Memory Access) directly to the DDR from the AGP and PCI bus should see a boost. If the video drivers were written to take advantage of it the cards should be able to cache directly to the system memory with out much speed loss over there own onboard memory. This is why I don't think there is any speed loss with the new G4MX with only 32MB of ram as compared to the previous model with 64MB of ram. It doesn't need it because it can use the system memory at the same speed as it's own onboard memory.

more meaningless synthetic benchmarks, what a surprise!

for a non altivec, non MP aware test this really seems flawed as any kind of measurement.

So a 933Mhz G4 with PC133, a 2Mb L3, 256K L2 & a 7 stage pipeline, without the aid of altivec has score that makes it 273% faster than a 300Mhz G3 ?

The 733Mhz G4 score is laughable to say the least, it's not even seperating the QS from the digital audio models and there's no way it's 63% faster than a 533Mhz G4 at anything. They both have a 133Mhz FSB and PC133 RAM, 1 has a 4 stage pipeline and 1Mb L2 at a 2:1 ratio, the other has a 7 stage pipeline, a 256K L2 at a 1:1 ratio and a 1Mb L3 at a 3:1 ratio and that cpu didn't use DDR on the L3 either because it was the PPC7450 not the PPC7455.

Re: more meaningless synthetic benchmarks, what a surprise!

Originally posted by barkmonster
for a non altivec, non MP aware test this really seems flawed as any kind of measurement.

So a 933Mhz G4 with PC133, a 2Mb L3, 256K L2 & a 7 stage pipeline, without the aid of altivec has score that makes it 273% faster than a 300Mhz G3 ?

The 733Mhz G4 score is laughable to say the least, it's not even seperating the QS from the digital audio models and there's no way it's 63% faster than a 533Mhz G4 at anything. They both have a 133Mhz FSB and PC133 RAM, 1 has a 4 stage pipeline and 1Mb L2 at a 2:1 ratio, the other has a 7 stage pipeline, a 256K L2 at a 1:1 ratio and a 1Mb L3 at a 3:1 ratio and that cpu didn't use DDR on the L3 either because it was the PPC7450 not the PPC7455.

Click to expand...

The speed difference between the 933Mhz G4 and 300Mhz G3 seems about right if you aren't using Altivec. Also on the speed difference between the 733Mhz G4 and 533Mhz G4 is also very possible correct. You have to take into account that there are several generations of the G4. So it is very possible that it is actually 63% faster. Baically they could be running the same Mhz but because on chip is a later generation some improvements it would still be faster. All in all I think these tests are fairly good at testing all kinds of chips though that's all it does. It really doesn't give you a view of overall system speed because of the missing tests for Altivec and such.

Re: more meaningless synthetic benchmarks, what a surprise!

Originally posted by barkmonster
So a 933Mhz G4 with PC133, a 2Mb L3, 256K L2 & a 7 stage pipeline, without the aid of altivec has score that makes it 273% faster than a 300Mhz G3 ?

Click to expand...

i can put money on it that my 933 would smoke a 300 mhz G3 in anything, including non altivec tasks. One it is a later gen chip, meaning it has more technological advances...it has an L3 cache...and none of this even mentions the fact that the mhz rating is over 300% of the 300mhz G3 (100% denoting equivalency)...if my 933 is only 273% faster than a 300 mhz...im very disappointed

Re: Re: more meaningless synthetic benchmarks, what a surprise!

Originally posted by BongHits


i can put money on it that my 933 would smoke a 300 mhz G3 in anything, including non altivec tasks. One it is a later gen chip, meaning it has more technological advances...it has an L3 cache...and none of this even mentions the fact that the mhz rating is over 300% of the 300mhz G3 (100% denoting equivalency)...if my 933 is only 273% faster than a 300 mhz...im very disappointed

Click to expand...

Sorry but a 3x faster chip doesn't mean 3x faster performance. There are other bottlenecks such as hard drive speed and ram speed.

Poor Quadra...

P-Worm

Originally posted by P-Worm
Poor Quadra...

P-Worm

Click to expand...

Remember when you would have cut your arm off to have one.

Single 933MHz G4 PM
452.1

Single 867MHz G4 PM
420.1

Dual 867Mhz G4 PM
418.1

Dual 867Mhz DDR G4 PM
417.9

Click to expand...

uh that doesnt sound too promising concidering the single 867 is faster than the dual ddr 867

Originally posted by SilvorX


uh that doesnt sound too promising concidering the single 867 is faster than the dual ddr 867

Click to expand...

By 2.2 that could be the difference in the organization of the drive. Nothing to base a judgement on.

Single 933MHz G4 PM
452.1

Single 867MHz G4 PM
420.1

Dual 867Mhz G4 PM
418.1

Dual 867Mhz DDR G4 PM
417.9

Click to expand...

I don't know why it shows a Dual 867 DDR and a Dual 867. Apple never made a Dual 867 before the DDR version that is out now. They made a single processor 867 and a Dual 800.

MacBandit:

That said the tests on the railhead website have a flaw. The tests a) are not multiprocessor aware b) are not altivec aware. This leads to a problem the old G4 has a larger L3 cache which doesn't really come in to play unless you are doing altivec aware tasks. So in my opinion these tests have basically illuminated the L3 cache from the test. That being the case you would expect the new machine to shine it has a faster bus then the previous machines. Which does make a big difference.

Click to expand...

You are very much wrong. The L3 holds data and instructions and cares not what type of instructions those are. One does not need AltiVec to use L3. It is also not correct to say that AltiVec makes better use of the L3 than straight PPC code would, because AltiVec could very very easily end up working on data set much larger than the L3, at which point the L3 is rendered unimportant and the FSB speed becomes important.

barkmonster:

The 733Mhz G4 score is laughable to say the least, it's not even seperating the QS from the digital audio models and there's no way it's 63% faster than a 533Mhz G4 at anything. They both have a 133Mhz FSB and PC133 RAM, 1 has a 4 stage pipeline and 1Mb L2 at a 2:1 ratio, the other has a 7 stage pipeline, a 256K L2 at a 1:1 ratio and a 1Mb L3 at a 3:1 ratio and that cpu didn't use DDR on the L3 either because it was the PPC7450 not the PPC7455.

Click to expand...

You are very wrong as well. While the longer pipeline CAN hurt performance, there are many many types of programs in which it does not (note how fast P4s can be in the "right" test). There are many things that I can think of computing that would show perfect scaling with clockspeed, and there are certainly things I can think of (as far as instruction mixes) that would show a 745x being far superior to the old G4, even clock-for-clock. The point is that I can pick a test to show the different versions of the G4 however I like, I just need to exploit the strengths and weaknesses of each. Discounting this recenct becnhmark as BS (or whatever) on grounds of it not agreeing with other benckmarks is stupid. This is a real test as much as all the Photoshop tests (or whatever you have in mind that shows the 533 in a more favorable light). The whole point in having a lot of benchmarks to look at is so you can get a better picture of actual performance. One test does not reveal all. Different chips are strong in different areas.

MacBandit:

Sorry but a 3x faster chip doesn't mean 3x faster performance. There are other bottlenecks such as hard drive speed and ram speed.

Click to expand...

I assume that you realize that there are things with will scale perfectly with clock speed, and there are things that will scale perfectly with FSB speed, and there are things that will scale perfectly with hard disk speed. Most things are a combination of all three, but making a blanket statment as you have is plainly wrong.

SilvorX:

uh that doesnt sound too promising concidering the single 867 is faster than the dual ddr 867

Click to expand...

Well the single 867 has 2mb L3 whereas the dual has only 1mb per chip, and the FSB is the same. Remember earlier in this post where I scolded MacBandit about the L3 being useful in non-AltiVec? Well this may be an example.

I was examing the results a bit more closely and I think that there are some "issues" that need to be addressed:

Just strange:
iMac G4-800 > eMac G4-800
iMac G4-700 < eMac G4-700
eMac G4-700 ~= eMac G4-800

733 vs 800:
9.1% higher clock => 18.6% higher performance
(L3 size issue?)

Scaling with 133FSB 2mb L3:
800 DP => 866 SP = 66mhz for 11.0 points
866 SP => 933 SP = 66mhz for 32.0 points
933 SP => 1000 DP = 66mhz for 15.8 points
(Note that going between SP/DP is neither good nor bad.)

Anyway, the new G4 core is clearly very good at this. Note the 550mhz PB compared to the 533mhz DP.

Originally posted by ddtlm

Click to expand...

Thanks for pointing this stuff out. Though I did know this already. My point on Altivec is it seems to be a bandwith hog so it excels when it can use the L3 cache because of the very high bandwith there. I wasn't saying the L3 Cache was not being used at all it's just not being used to fullest.


I also realize that some tests will show that a chip with twice the Mhz is actually 10x as fast this is known. Overall though most of the time when you analyze a complete system the speed does not go up equally with the Mhz speed of the cpu.

I don't disregard these tests at all in that they give you a very good break down of how your system might perform doing the tasks it was tested in. That's all a benchmark tells you anyhow. It just tells you what your system can do with those specific tests it has very little to do with real life and real applications most of the time.

Originally posted by SilvorX


uh that doesnt sound too promising concidering the single 867 is faster than the dual ddr 867

Click to expand...

According to what people have been saying, though, this is a single processor test only so a single 867 with 2MB L3 is going to perform better than one processor in a DP system with 1MB L3 Cache.

Originally posted by ddtlm
I was examing the results a bit more closely and I think that there are some "issues" that need to be addressed:

Just strange:
iMac G4-800 > eMac G4-800
iMac G4-700 < eMac G4-700
eMac G4-700 ~= eMac G4-800

733 vs 800:
9.1% higher clock => 18.6% higher performance
(L3 size issue?)

Scaling with 133FSB 2mb L3:
800 DP => 866 SP = 66mhz for 11.0 points
866 SP => 933 SP = 66mhz for 32.0 points
933 SP => 1000 DP = 66mhz for 15.8 points
(Note that going between SP/DP is neither good nor bad.)

Anyway, the new G4 core is clearly very good at this. Note the 550mhz PB compared to the 533mhz DP.

Click to expand...

The problem with tests like this is that they are made to give an example of where the machines are at. The only way to make it accurate would be to factor in a few statistical equasions.

I do not study statistics to go into too much detail but to get a close to accurate study it is necessary to run your results through equasions to factor out the errors in life.

For example, it is possible that only certain people will send in their scores. It is necessary to factor in that people with "faster" machines are MORE likely only to send in good results where as people with machines that are considered to be slower are MORE likely to send in any result.

You also have to factor in the fact that newer machines or more newly formatted are MORE likely to be more effecient where as machines that have not been formatted or partitioned are likely to perform less effeciently and show off a score less than maximum.

It is a VERY complex process and the results should only be used as a guide.

Dual processors...

...*could* actually hurt performance under certain circumstances. If the system overhead is negligable and is the only thing being passed off to the second processor, then three things could hurt performance: the benchmark task swapping back and forth between processors (unlikely if it's high enough priority), extra bus bandwidth eaten up by second processor, extra time needed to snoop in the other processor's cache. You'd have to work pretty hard, but this would result in single processor machines beating/equaling duals. In real life (or even most benchmarks), of course, the duals would almost invariably win.

Catfish_Man:

I think the SMP overhead is actually high enough that single-CPU machines are better at single-CPU tasks, although it's not something I worry about much (small effect).

Having both an iMac G4 and a TiBook 800, I have seen noticable speed differences between the two.

Any disk intensive activity tends to be faster on the iMac which has a faster disk.

Any CPU bound processes tend to be faster on the TiBook which has the L3 cache.

There are any number of factors (some of which have been stated above) that can cause speed differences. And for a chart to be useful, these differences need to be listed as well.

For reference the TiBook is at 10.2 and the iMac at 10.1.5, I have a couple of pieces of software that broke on 10.2 and am awaiting updates for them before upgrading the iMac.

Sorry, I dont agree at all! All of the articles I read say that the old 1Ghz DP powermac is quicker than the current dual 1ghz systems. The reason, the extra cache in the old systems. Sorry to all those people with the new DDRAM system!

There is obviously not much in terms of speed differnce. However, in the benchmarks I have seen in various places around the web, the older model is quicker. One place I saw the benchmarks were for Macworld Magazine to name just one. I dont believe everything I read, just ones that are backed up over in many different independant sources.

Originally posted by Falleron
Sorry, I dont agree at all! All of the articles I read say that the old 1Ghz DP powermac is quicker than the current dual 1ghz systems. The reason, the extra cache in the old systems. Sorry to all those people with the new DDRAM system!

Click to expand...

you know, who cares if it's a bit slower? the benefits outweigh the losses. you gain More HD bays, another optical bays, and another on board HD Bus. I own a dual 867 and obviously chose it over the old dual gig for these reasons.

Originally posted by madamimadam
The problem with tests like this is that they are made to give an example of where the machines are at. The only way to make it accurate would be to factor in a few statistical equasions.
I do not study statistics to go into too much detail but to get a close to accurate study it is necessary to run your results through equasions to factor out the errors in life.

Click to expand...

I don't trust these guys' data - some of it just seems contradictory, and while they know little about cpus I think though know even less about statistical errors. I love the idea of 'factoring out the errors in life.' The random errors I think you're talking about (while you can't discount) should in a test of like this be very minimal. Random errors occur from human inability to reproduce results among other things, the distribution will be strictly normal and an equally weighted least squares adjustment of the data will yield simple average, my point is though - there shouldn't be random errors. These are calculations performed based on code that doesn't change. The same machine should produce the same results in the same test under the same conditions. If it doesn't than there's a flaw in the design of the experiment (hard drive being accessed, other applications, or complications). There will always be outliers, I understand that, but this should be really freaking consistent.
You also mention user bias but if these guys' data is worth ANYTHING at all, than they should have checked that **** at the door.
As far as systematic errors, sure, the chip frequency could be buffed up a little or could be running a little hot. There are discrepancies in chip manufacturing, each chip is different, but under the same test it should result the same. These can be modeled though again by taking a larger sample of the same machines, the distribution of errors for these machines I have to believe is going to be minimal. That's an entire different test though, we wouldn't even be measuring the speed of the computers at that point - we'd only be looking at the distribution of errors in speed of the same machine. In effect, measuring Apple's ability to produce the same machine.
I don't like these guys, their experiment designs don't seem fair to all the machines, but that's just my opinion. Anyone else?

On a side note, I went out with this amazing girl last night and I'm just waiting on myself to screw it up... I really like her but jesus, I know I'm going to **** it up.

Originally posted by Jimong5


you know, who cares if it's a bit slower? the benefits outweigh the losses. you gain More HD bays, another optical bays, and another on board HD Bus. I own a dual 867 and obviously chose it over the old dual gig for these reasons.

Click to expand...

Good point. The only reason I was defending the old dual ghz was because I have one! What does it matter, its not like anyone will notice the speed difference.