Stridder44
Universal Traveler
Isn't there already a thread on this same subject? 
Powermac G5 SPEC numbers cooked?
- Thread starter malexgreen
- Start date
malexgreen
Contributor and PB User
I'm not saying they lied intentially. I'm saying that Veritest screwed up the benchmark. I think that for Veritest to quote SPEC numbers for machines that are less than the SPEC numbers published for those same machines by those machines' OEMs is shady business. Why not just shoot to beat what's already published instead of trying to create some worst-case scenario?
Why would you use GCC when you could get better results using a vendor compiler? What was bull about haxial's write-up?
lurk
Mitä?
You are assuming that the SPEC numbers that are published are somehow more honest. The truth of the matter is that they are all bogus to a significant degree. The reason for this is that compiler optimization is not a black and white thing. With the exception of a reletivly small set of optimizations most changes that a compiler makes may or may not inprove the speed of the compiled program. More often than not an optimization will make an arbitrary program run slower. That is the reason that there are bookshelves full of clever optimizations which will never make it out of research compilers.
Now if you are trying to get a good SPEC for your marketing materials you have an advantage. You can try out the different optimizations on the benchmark and see which ones make it go faster. This can really boost the score on any benchmark and it basically eliminates any usefullness of that score for making useful comparisons. I am not going to be running benchmarks on my machine so using the help one hurt the other rule of optimization I can basically assume that it I compiled my code using the same optimization settings it may well be slower than traditionally optimized code.
These numbers are 100% about who can pee higher on the wall and there is only a fairly minor correspondance to "real" performance. But as we have seen in the fracas Apple has caused people do get their knickers in a knot over these things. Which comes to what I think is the overriding reason the super optimizing vendor compilers were not used. They weren't used because the G5 would have still probably won and IBM would not like the G5 to be outshining its bigger Power4 brother.
Now that is not a very unlikely result when you consider that the SPEC tests are very compute bound single threaded benchmarks. The added cache and other on die processors in the Power4 chip really would not help that much. Given that the G5 is running at a higher clock I really would not be suprised to see it come out on top. Does that make is a faster chip than the Power4? Certianly not when you consider that each Power4 basically contains 4 G5s. But that fact is not captured in this benchmak...
Looks like I got ahead of myself and answered that above.
Several things, he attacks turing off hyperthreading as unfair when that actually improves Intel's performance. Dell does the same when they compute their numbers. He asserts that SSE2 was not enabled when it was. That was just off the top of my head and I really don't want to take the time to go back and read that tripe again to find the other flaws.
Have Fun,
-Eric
I have a G5 on order so at this point I can only point out a couple of items. I do biomedical imaging and I need to address in memory more than 2Gb of RAM thus need a 64 bit machine. I have tested the HP and Compaq version of the itanium as well as dual xenon for our purposes and cannot manipulate >2Gb files and when I do some tricky programming the damn things thrash so badly that I can't use them. I got a quote from IBM for 64 bit machines (that uses similar cpu's) and it was going to cost 50K per machine with 8Gb RAM with a 3K per year maintenance contract.
Additionally, comparision of gcc to intel compilers is not always a good comparison as the intel compiler (and some others) is very good at optimization whereas the gcc compiler is, in my own possibly uninformed opinion, not as well optimized for the G5 chip as intel is for the intel chip. Thus in many ways Apple is not putting their best foot forward either, claims aside.
To me the only fair comparision is based upon what you actually do with a computer. My 7 year old is much faster at pattern recognition than my intel cluster.
Bottom line:
For biomedical imaging needs the G5 does things that cost much more on other machines and that cannot be done on the intel machines that are being talked about with windows (linux is a different issue, where drivers for latest video/dvd hardware etc for our needs make it cost inefficient).
64 bit addressing
low cost
low maintence (thus low cost)
with throughput that exceeds the best windows machines.
if that were not the case we would switch, if windows offered a fast 64 bit operating system that worked on intel boxes. No such machines exist.
--Bill Ditto
Chair, Biomedical Engineering Dept.
University of Florida
Additionally, comparision of gcc to intel compilers is not always a good comparison as the intel compiler (and some others) is very good at optimization whereas the gcc compiler is, in my own possibly uninformed opinion, not as well optimized for the G5 chip as intel is for the intel chip. Thus in many ways Apple is not putting their best foot forward either, claims aside.
To me the only fair comparision is based upon what you actually do with a computer. My 7 year old is much faster at pattern recognition than my intel cluster.
Bottom line:
For biomedical imaging needs the G5 does things that cost much more on other machines and that cannot be done on the intel machines that are being talked about with windows (linux is a different issue, where drivers for latest video/dvd hardware etc for our needs make it cost inefficient).
64 bit addressing
low cost
low maintence (thus low cost)
with throughput that exceeds the best windows machines.
if that were not the case we would switch, if windows offered a fast 64 bit operating system that worked on intel boxes. No such machines exist.
--Bill Ditto
Chair, Biomedical Engineering Dept.
University of Florida
Cool. Good to hear from someone who actually will put the G5 to use(not like WC3 isn't good )
Please come back and let us know how the G5 performes for you when you get it.
)Please come back and let us know how the G5 performes for you when you get it.
Randman
HA! HA! HA!
The numbers are just marketing, much like estimated mpg in cars; it's just to try and hook a potential buyer.
Show me a G5, no really, please, I want one.
Anyway, show me a G5 and fetch me a few, but not many, of these idiots who go around saying it'll underperform and we'll test it out ourselves. Sheesh.
Show me a G5, no really, please, I want one.
Anyway, show me a G5 and fetch me a few, but not many, of these idiots who go around saying it'll underperform and we'll test it out ourselves. Sheesh.
SoulCollector
Registered
Well have you guy read the NASA benchmarks of the G5 and the Pentium4 and G4??? Pretty crazy results especially from NASA you would think ...they wouldnt need a computer like the G5..cuz they probably have a comparable G30 or somthing....Check out .Macnn.com..
Stridder44
Universal Traveler
This thread is old news...we'll all just have to wait till the G5 is in the consumers hands to really see.