i have to agree... those results don't seem particularly impressive.
talking out of my ass, if ONE 1GHz G4 is 100, I'd love to see what a Dual G4 2GHz would pull off. Maybe very very very slightly less performan than a Dual G5 2GHz result.
i have to agree... those results don't seem particularly impressive.
talking out of my ass, if ONE 1GHz G4 is 100, I'd love to see what a Dual G4 2GHz would pull off. Maybe very very very slightly less performan than a Dual G5 2GHz result.
Again, why shouldn't you be impressed? The G5 is a different beast, and its architecture was designed to give up a little IPC and latency to scale better, but not nearly as dramatically as the P4. So integer math is a bit behind the old G4, the 7450 was slower than the 7400 too...
The main thing is it's faster, quite a bit faster than the previous top of the line 2x1416 MHz Power Mac. And it's faster per clock than the G4 for FP and SIMD...
Again, why shouldn't you be impressed? The G5 is a different beast, and its architecture was designed to give up a little IPC and latency to scale better, but not nearly as dramatically as the P4. So integer math is a bit behind the old G4, the 7450 was slower than the 7400 too...
The main thing is it's faster, quite a bit faster than the previous top of the line 2x1416 MHz Power Mac. And it's faster per clock than the G4 for FP and SIMD...
...in Skidmarks GT anyway.
Yes, 4% faster per clock in Vector and 35% faster in FP. 14% slower in Integer.
Impressive FP, definitely. (But don't ya' think that may have something to do with the G5 having 2 FP Units per chip?)
Vector performance increase is a bit of a joke, though.
Yes, 4% faster per clock in Vector and 35% faster in FP. 14% slower in Integer.
Impressive FP, definitely. (But don't ya' think that may have something to do with the G5 having 2 FP Units per chip?)
Vector performance increase is a bit of a joke, though.
At least we'll be able to scale now.
Actually the only thing that really isn't impressive is FP performance.
Integer: The G5 only has 2 integer units (1+3 for the G4, depending on what you test, the G4 will be faster (simple ops) or slower (complex integer ops) than the G5 at the same clock speed).
SIMD: Should be slower on the G5, but I'm sure in the test (and most real world apps) the vector units aren't used concurrently, so the G5 performing about the same as the G4 seams reasonable.
FP: Since the G5 has two FP units, this should definitely be more than 35% faster.
I hate to dwell on the skidmarks (Apple needs a better name for that test!), but an obvious question concerning performance comes to mind. Do we know whether skidmarks has been recompiled to take better advantage of the G5? Perhaps it's a dumb question but maybe not. It would be interesting to see the results from xBench as a contrast. In any case, is not it more important that the G5 performs so much better in real world applications? As many have said, we don't produce on our machines by running benchmark programs all day.
Do the 970 chips (aka G5s) look as large as they appear on the Apple diagrams? They look positively huge.
I'm a bit off topic perhaps, but because it has to do with speed/benching, I thought I might post it here anyway.....
I went after some prices and specs for the PC3200 DDR400 SDRam modules.
Prices are pretty low, around ?90 for 512Mb, and up to ?165.
Most expensive are the low-latency modules (CAS 2), but I find that there are so many companies offering them (with differing prices/same latency), that I wonder from whom to buy.
As this appears to make a 15% difference in speed, perhaps anyone has good points why to buy (not) from a specific company (Kingston, Corsair, Dane-Elec, TwinMos, etc.)?
Thanks a lot
P.S. Anyone know what the 'HyperX' stands for, related to memory modules from Kingston?
Comments
talking out of my ass, if ONE 1GHz G4 is 100, I'd love to see what a Dual G4 2GHz would pull off. Maybe very very very slightly less performan than a Dual G5 2GHz result.
Hmmm... hope I'm wrong...
Originally posted by ZO
i have to agree... those results don't seem particularly impressive.
talking out of my ass, if ONE 1GHz G4 is 100, I'd love to see what a Dual G4 2GHz would pull off. Maybe very very very slightly less performan than a Dual G5 2GHz result.
Hmmm... hope I'm wrong...
Read the thread again...
Originally posted by Eugene
Read the thread again...
I've been editing a previous post, and I'll post a link to another thread that I've tried to go into more detail on these results:
http://forums.appleinsider.com/showt...0&pagenumber=3
Originally posted by audiopollution
I've been editing a previous post, and I'll post a link to another thread that I've tried to go into more detail on these results:
http://forums.appleinsider.com/showt...0&pagenumber=3
Again, why shouldn't you be impressed? The G5 is a different beast, and its architecture was designed to give up a little IPC and latency to scale better, but not nearly as dramatically as the P4. So integer math is a bit behind the old G4, the 7450 was slower than the 7400 too...
The main thing is it's faster, quite a bit faster than the previous top of the line 2x1416 MHz Power Mac. And it's faster per clock than the G4 for FP and SIMD...
...in Skidmarks GT anyway.
Originally posted by Eugene
Again, why shouldn't you be impressed? The G5 is a different beast, and its architecture was designed to give up a little IPC and latency to scale better, but not nearly as dramatically as the P4. So integer math is a bit behind the old G4, the 7450 was slower than the 7400 too...
The main thing is it's faster, quite a bit faster than the previous top of the line 2x1416 MHz Power Mac. And it's faster per clock than the G4 for FP and SIMD...
...in Skidmarks GT anyway.
Yes, 4% faster per clock in Vector and 35% faster in FP. 14% slower in Integer.
Impressive FP, definitely. (But don't ya' think that may have something to do with the G5 having 2 FP Units per chip?)
Vector performance increase is a bit of a joke, though.
At least we'll be able to scale now.
Originally posted by audiopollution
Yes, 4% faster per clock in Vector and 35% faster in FP. 14% slower in Integer.
Impressive FP, definitely. (But don't ya' think that may have something to do with the G5 having 2 FP Units per chip?)
Vector performance increase is a bit of a joke, though.
At least we'll be able to scale now.
Actually the only thing that really isn't impressive is FP performance.
Integer: The G5 only has 2 integer units (1+3 for the G4, depending on what you test, the G4 will be faster (simple ops) or slower (complex integer ops) than the G5 at the same clock speed).
SIMD: Should be slower on the G5, but I'm sure in the test (and most real world apps) the vector units aren't used concurrently, so the G5 performing about the same as the G4 seams reasonable.
FP: Since the G5 has two FP units, this should definitely be more than 35% faster.
Originally posted by Eugene
No, it isn't. It's about right. My Dual 1 GHz G4 scores 100/100/100 exactly in Skidmarks GT. I thought I already explained the benchmarks...
Thanks for this info it was not written in Think Secret that skidmarks do not take advantage of dual.
But this results are quite strange anyway : spec int is 50 % more at equal ghz on a G5 compared to a G4 and it's 24 % down with Skidmarks.
Do the 970 chips (aka G5s) look as large as they appear on the Apple diagrams? They look positively huge.
Originally posted by Kecksy
Plus, the Vector unit on the G5 benifits from the fast frontside bus, so it should beat the G4 clock for clock.
In real world apps yes.
However, I don't know how Skidmarks GT works, the test probably wasn't designed to hit the main memory that often.
Originally posted by Powerdoc
Thanks for this info it was not written in Think Secret that skidmarks do not take advantage of dual.
But this results are quite strange anyway : spec int is 50 % more at equal ghz on a G5 compared to a G4 and it's 24 % down with Skidmarks.
Not all benchmarks are created equal, are compiled equal, or leverage the same type of computations.
Comparing a very simple benchmark against the Spec suite is pretty useless. Even using the suite of Spec benchmarks alone is largely pointless.
Originally posted by Eugene
Not all benchmarks are created equal, are compiled equal, or leverage the same type of computations.
Comparing a very simple benchmark against the Spec suite is pretty useless. Even using the suite of Spec benchmarks alone is largely pointless.
You are right, we should only take in consideration real applications performance.
Screw the lame benchmarks
I say, if you dont like the results then dont buy one EVER, even when you find out you were wrong!
I went after some prices and specs for the PC3200 DDR400 SDRam modules.
Prices are pretty low, around ?90 for 512Mb, and up to ?165.
Most expensive are the low-latency modules (CAS 2), but I find that there are so many companies offering them (with differing prices/same latency), that I wonder from whom to buy.
As this appears to make a 15% difference in speed, perhaps anyone has good points why to buy (not) from a specific company (Kingston, Corsair, Dane-Elec, TwinMos, etc.)?
Thanks a lot
P.S. Anyone know what the 'HyperX' stands for, related to memory modules from Kingston?