The PPCNUX-Team ran a benchmark on various Macs and operating systems, including the new MacPro. “When I tested the iMac Core Duo back in May the results were far better than I expected. That given, the Core2 disappoints me. Yes, it is faster than the original CoreDuo; but with all its announced enhancements I expected more. Its not the major leap ahead I expected.”
Testing a Mac Pro, PowerMac G5, iMac CoreDuo
About The Author
Follow me on Twitter @thomholwerda
2006-08-30 7:58 pmTuishimi
Ha. You edited your message as I was entering mine, below.
I don’t think so? Altho’ the article was a but difficult to read at times…
When it tested the iMac CoreDuo back in May this year the results were a far better than I expected. That given the Core2 disappoints me. Yes it is faster than the original CoreDuo. But with all its announced enhancements I expected more (see Der Core, Intels großer Wurf?, sorry only in German). Its not the major leap ahead I expected.”
Edited 2006-08-30 19:59
Does the result of this test simply mean that so far, Athlon64 CPU are the best processor, if we’re looking at performance? His website seems to want to advance PowerPC architecture and *N*X, but when Athlon64 comes out best, I guess it might be rather objective.
What I’d be interested in too, is how does this performance relate to the use of energy. Obviously, a 100 Mhz PowerMac wouldn’t perform so well, but it hopefully uses less Watts than a G5. What CPU really performs best per watt? If all of us used that CPU, it could make the difference of quite a few power plants less needed.
I’ve always liked the idea behind the RISC/PowerPC architecture. It makes sense to turn on a transistor only if it’s needed (if I’m not talking nonsense here, because I don’t know much about CPUs beyond this). Too bad the G5s got too hot to be continued by Apple, I love the PPC G4 for hardly ever needing the fan on this iBook.
2006-08-30 9:05 pmormandj
His test shows his one very specific benchmark runs best on A64 cpus. How he jumps from this benchmark to blanket statements about performance of the chips/machines is absolutely beyond me.
Anybody who can program worth a salt can write a program that runs better on one particular kind of chip than another, even only using high-level programming languages. He may not have intentionally done this, but that’s quite obviously what’s going on. His benchmark results completely conflict with just about everything else posted on the web, showing 10-20% increase over the old core duos, and the intel procs beating even AMD FX series processors.
Bad, bad benchmark. Better written review than most of the recent ones I’ve seen on this site, however!
2006-08-30 10:49 pmrayiner
Regarding performance, the conclusions are a bit iffy. The Athlon 64 might very well be faster for his particular code sequence. However, its generally slower overall, according to SPEC (which tests may different types of code sequences). The precise reason why would require more information. Perhaps the code in question is helped by the Athlon’s lower latency memory controller, or perhaps the Core 2 is new enough that GCC hasn’t been optimized for it yet. The latter case is quite likely, as there is not a lot of information out there about the nature of the Core 2’s FPU, nor does the author describe precisely what compilation settings he is using in his test. Its possible that, like the P4, the Core2 isn’t that great at running x87 FPU code, and he should’ve compiled it to use SSE for scaler FP instead.
As for power, it’s not a huge deal. A CPU uses about as much power as a single incandescent lightbulb, and then only at full-load. The average person is likely to save much more power just by remembering to turn off the bathroom light than by switching to a more engergy-efficient CPU. Indeed, if the lower-energy CPU is much slower, using it might even increase your energy use. It’s better for a 100W CPU to complete a task in 10 minutes and idle the rest of the time than for a 30W CPU to complete a task in 45 minutes, fully-loaded the whole time.
The reason people are concerned about power efficiency is not because computers use particularly large amounts of power in absolute terms, but because more energy-efficient CPUs allow for laptops with longer battery lives, server rooms with lower air-conditioning requirements, etc. In terms of power efficiency, PPC/RISC versus x86 is largely meaningless. There is nothing in PPC/RISC that says unused transistors should be turned off. It’s a function of the implementation, not the instruction set or architecture. Core 2 actually has one of the most sophisticated set of power-saving techniques out there. The new 970MP, used in the dual and quad-core PowerMacs, does frequency-scaling and clock gating (cutting the power-hungry clock-signal to idle parts of the chip). The Core 2 takes things a step further by very fine-grained clock-gating, as well as the ability to turn-off idle functional units. Furthermore, all the circuits in the Core 2 are optimized for power usage (the circuits in the 970MP are optimized for density and frequency). What the Intel designers did with the Core 2 (and the Core Duo and the P-M) was to define a specific frequency target, and design all the components for that target. In traditional designs, each component is designed for maximum frequency, which means that you end up with some components that have the headroom to run at higher frequencies, but can’t because the rest of the chip wouldn’t be able to keep up. In the P-M design, that unusable excess frequency headroom is traded for lower power usage.
When you look at the X2 pricing now, at least in Europe, it looks like the price performance ratio is exceptional. Socket 939 boards can be had for about Euro 30, and then the lower end x2s are really cheap. You can put together what would have been a super high end system not long ago for a few hundred.
Though, I read comments on a LUG list recently that you can save Euro 200 a year or so by moving to a lower power system…
So maybe they are more cost effective to buy than to run?
Edited 2006-08-30 20:06
Has made me interested to see all those Core 2 Duo benchmarks redone with 64bit software though. I knew from experience that AMD64 was a little bit snappier than IA32 generally. Generally by less than 10%.
It could just be that EMT64 != AMD64, or as stated that his software is biased to the AMD architecture.
Still, given most stuff is still 32bit, Core 2 Duo seems to be the way to go atm.
when you don’t bench against systems to rule out Mhz.
Pick 5 systems. Same Ram (speed/amount). Same GPU. Same Mhz and Same Compiler.
Run a series of compiler tests that reveal the pros/cons and where the compiler is optimized/non-optimized.
Run a series of apps across all the chips.
Run a series tests that measure features in the CPU common across the CPUs.
Until we see these standards everything else is pointless.
There are many factors that affects performance on a system, let alone the performance as the judging factor of a system purchase.
For Example, my recent move to core2 Duo was because of heat that became out of control.
My P4 @3.0 GHz (HT enabled, no overclocking) generates heat of 53 degrees celcius on the best fan colled Zalman heat sink in the market. When I moved to Core 2 Duo E6300 @1.86 GHz the heat of CPU became just 35 degrees on stock intel heat sink.
Now imagine the above CPU heat + 6 HDDs heat (including one with 10,000rpm disk) + 1GHz Memory (2 modules) + intel 875 chipset + BFG nvidia GPU 6600 GT (overclocked by manufacturer) + gigabit ethernet + DVD writter + Creative Soundblaster audigy 2ZS all in one system, even with 9 fans installed it didn’t prevent HDD death match from stopping, loosing 1 HDD every 2 years in average.
If I moved to AMD I know it would be better than my current heat situation, but the CPU will be at 45 degrees with performance so difficult to match from Intel unless you pay alot for it!
Besides, Apple now could include 2 more HDDs to the system (total of 4)than the previous unprofessional amount of 2, all because of heat reduction. It would not be possible even with AMD to achieve this. And you know adding more disks also means improving the HDD subsystem if you decide to run RAID 0.
So, these benchmarks are not a factor to consider when buying your system even if performance is the main reason.
I am the one who did the benchmark and published the article (BTW: why don’t you comment on the article at ppcnux.com?)
I know, that this benchmark is somewhat specific. But as I stated it is a very good indication on performance in my working environments today and in the past (eg. HD performance matters only little, Graphics performace doesn’t matter at all).
As I tried to express with my first sentence, everybody should do the benchmark which reflects best his/her environment and needs. Or you should collect benchmark results that come near to it.
And why should I not share the results with others. Like with any other news one must check with benchmarks what is relevant to me and what not.
That said let me briefly go through your comments:
As far as FPU-number-crunching is concerned, I would consider the Athlon64 best, followed by the G5 and the Woodcrest. However, Woodcrest may come close at 3GHz depending how it scales with the clock (I tested th 2GHZ model). For Quad systems it may look a little different. To my surprise Opterons 2xx performed worse than Athlon64 in the past (see my benchmark page att http://www.r-goetz.de/RGBench).
BTW: The G5 is a far smaller chip (in area) than the Pentium 4. Hence using even less power (100W vs 130W) it has a higher heat density. I think its the heat density that make the problem with the G5 cooling.
I coded this program back in 1997 when we had to made a decision at the MPIKGF which new machine to buy. It turns out that it was very representative for the mix of programs running there (all number crunching type). So it is not intentionally coded to blame and CPU design.
About almost the rest of the benches in web: Almost all compare Athlon and Core2-CPUs in IA32 mode. This is truly the relevant reference for a lot of users which are bond the 32bit-Windows. But this means to through away major capibilities of the Athlons in AMD64 mode. So if you compile your code yourself you should use each CPU in the condition they offer and compare than. This is what I done. And Athlons in AMD64 mode are fastest, but not by a lot. Athlons in IA32 mode on the other hand are outperformed by the Core2 CPU (as also can be seen in the tables)
Do you supply me with this systems? 🙂 I like to do this benchmarks. Actually I try to run my benchmark on all systems I can get my hands on. But normally I have no influence on there components.
As stated also by rayliner. As far as desktop and (compute-)servers are concerned heat is not that major issue as long as sufficient cooling is supplied and the noise level stay within reasonable range. The latter depends strongly on the person deciding upon it. To me even the Quad G5 is not to noising (though is is not the most silent one).
And you won’t stuff a lot of Disks and high end graphics card, sound cards etc. into a Number-crunching system. I am always happy with the smallest (non-UMA)-GPU I can get.
One word on SPEC95/98/2000.
SPEC was always I good indication on FPU-Performance unless you looked at Intel numbers. To me they always seem to be exaggerated, comparing them with other scientific benchmark (not only mine 😉 ). Unfortunately I didn’t bookmark the references for the latter. But I remember that to bioinformatics benches to first G5 perform a lot better than the Pentium 4, though they were close in SPEC2000. Hence, I tend to not trust SPECmarks if it come to Intel-CPUs (maybe this changes with SPEC2006. We will see).
Sorry, for this long post. and sorry for not going into all details in my article. But the article was already quite long.
Sorry, scratch that, he is testing the Core 2 Duo chips, I’d forgotten that the Mac Pro had the new CPUs in them. I’m an idiot
Edited 2006-08-30 19:59