I am guessing the dual 1.42GHz G4 will be faster than the single 1.6GHz G5 for most things. Are there any benchmark results out there confirming or denying this?
I am guessing the dual 1.42GHz G4 will be faster than the single 1.6GHz G5 for most things. Are there any benchmark results out there confirming or denying this?
I wouldn't necessarily say that. The Dual 1.42 has an advantage on Dual Processor apps of course. But the FPU of the G5 are so much better along with better Memory Bandwidth don't be suprised to see a Single 1.6 G5 beat a Duallie in quite a few tests.
I am guessing the dual 1.42GHz G4 will be faster than the single 1.6GHz G5 for most things. Are there any benchmark results out there confirming or denying this?
I have seen comparisons between the dual 1,42 and the former dual 1,25 (the first dual 1,25 released), the dual 1,42 was not really more fast than his elder. The memory bandwitch starvation, probabily was the reason for this.
With the correct compilers, the G5 1,6 will probabily been faster than the dual 1,42 except for altivec involving small files that can be locked in the L3 cache. The G5 scale lineary for the altivec stuff with the G4, according to the Nasa tests. However this test do not take into account large files, and therefore the bus and the memory bandwitch of the main memory.
It will be interesting to see some bench, but i would not buy a dual 1,42 knowing his huge price and his disapointing performances.
I wouldn't necessarily say that. The Dual 1.42 has an advantage on Dual Processor apps of course. But the FPU of the G5 are so much better along with better Memory Bandwidth don't be suprised to see a Single 1.6 G5 beat a Duallie in quite a few tests.
However floating point intensive applications are rare. Also the G4 has the L3 cache that helps to compensate for the lower memory bandwitdh.
The only benchmark results I have seen so far that shed some light on this are the Skidmarks numbers at thinksecret.com. The numbers compare a single 1GHz G4 against a single 2GHz G5 if I understand them correctly.
Adjusting for clock frquencies, a single 1.6GHz G5 should score 172 * 1.6 / 2.0 = 138 versus 142 for a single 1.42GHz G4 in integer operations. For floating point the single G5 should score 270 * 1.6 / 2.0 = 216 versus 142 for the single G4. For vector operations the single G5 should score 208 * 1.6 / 2.0 = 166 versus 142 for the single G4. However, the dual G4 has two G4s, so assuming an 80% multiprocessing efficiency we would be looking at a total rating of 142 * 2 * 0.8 = 227 for the dual G4 in all the categories. That suggests the dual 1.42GHz G4 is faster than a single 1.6GHz G5 for most things that can use both processors simultaneously.
However floating point intensive applications are rare. Also the G4 has the L3 cache that helps to compensate for the lower memory bandwitdh.
The only benchmark results I have seen so far that shed some light on this are the Skidmarks numbers at thinksecret.com. The numbers compare a single 1GHz G4 against a single 2GHz G5 if I understand them correctly.
Adjusting for clock frquencies, a single 1.6GHz G5 should score 172 * 1.6 / 2.0 = 138 versus 142 for a single 1.42GHz G4 in integer operations. For floating point the single G5 should score 270 * 1.6 / 2.0 = 216 versus 142 for the single G4. For vector operations the single G5 should score 208 * 1.6 / 2.0 = 166 versus 142 for the single G4. However, the dual G4 has two G4s, so assuming an 80% multiprocessing efficiency we would be looking at a total rating of 142 * 2 * 0.8 = 227 for the dual G4 in all the categories. That suggests the dual 1.42GHz G4 is faster than a single 1.6GHz G5 for most things that can use both processors simultaneously.
Skidmark is lame for comparing a G4 with a G5.
You surestimate the performance of a dual system over a single. The G5 is more SMP than the G4. The G5 can reach 80 % efficiency, the G4 won't.
Many software dual optimized, are in fact optimized for only some tasks. A dual 1,42 ghz will be slower than an hypothetical single 2,5 G4.
here some benchmarks of a dual 1 ghz versus a single 1 ghz (powerlogix cards on a G4 466) :
speedmark 3,2 : single 133, dual 139
i movie 2,13 rendering : single 0,43, dual 0,40
i tune 3 : single 0,42,dual 0,33
The software who take more advantage of SMP is obviously here I tune, and it's only a 27% boost. You are way beyond the supposed 80 % ratio. That's why i claim (we have to wait for new benchmarks) that the G5 1,6 will be faster than the dual 1,42 in many applications (better mobo, better HD, better video card).
Forget the G4, it's a lame antique ready for the garbage. The sooner Apple ditches it the better.
I tend to side with this opinion. I think the G5 will just keep going, when the G4 starts struggling under heavy workloads. The performance difference may not come to light in benchmark tests as well as it does in real work applications.
As for floating point intensive operation not being commonplace - that might be true, unless you've bought the machine to do 3D or scientific work, in which case it's your whole raison d'etre.
I would say stay away from the 1,42*2s. The 1,25 dual is a much better buy. They wer the ones we should compare to the 1,6 G5s with regards to both price and performance.
Comments
Originally posted by Tidris
I am guessing the dual 1.42GHz G4 will be faster than the single 1.6GHz G5 for most things. Are there any benchmark results out there confirming or denying this?
I wouldn't necessarily say that. The Dual 1.42 has an advantage on Dual Processor apps of course. But the FPU of the G5 are so much better along with better Memory Bandwidth don't be suprised to see a Single 1.6 G5 beat a Duallie in quite a few tests.
Originally posted by Tidris
I am guessing the dual 1.42GHz G4 will be faster than the single 1.6GHz G5 for most things. Are there any benchmark results out there confirming or denying this?
I have seen comparisons between the dual 1,42 and the former dual 1,25 (the first dual 1,25 released), the dual 1,42 was not really more fast than his elder. The memory bandwitch starvation, probabily was the reason for this.
With the correct compilers, the G5 1,6 will probabily been faster than the dual 1,42 except for altivec involving small files that can be locked in the L3 cache. The G5 scale lineary for the altivec stuff with the G4, according to the Nasa tests. However this test do not take into account large files, and therefore the bus and the memory bandwitch of the main memory.
It will be interesting to see some bench, but i would not buy a dual 1,42 knowing his huge price and his disapointing performances.
Originally posted by hmurchison
I wouldn't necessarily say that. The Dual 1.42 has an advantage on Dual Processor apps of course. But the FPU of the G5 are so much better along with better Memory Bandwidth don't be suprised to see a Single 1.6 G5 beat a Duallie in quite a few tests.
However floating point intensive applications are rare. Also the G4 has the L3 cache that helps to compensate for the lower memory bandwitdh.
The only benchmark results I have seen so far that shed some light on this are the Skidmarks numbers at thinksecret.com. The numbers compare a single 1GHz G4 against a single 2GHz G5 if I understand them correctly.
Adjusting for clock frquencies, a single 1.6GHz G5 should score 172 * 1.6 / 2.0 = 138 versus 142 for a single 1.42GHz G4 in integer operations. For floating point the single G5 should score 270 * 1.6 / 2.0 = 216 versus 142 for the single G4. For vector operations the single G5 should score 208 * 1.6 / 2.0 = 166 versus 142 for the single G4. However, the dual G4 has two G4s, so assuming an 80% multiprocessing efficiency we would be looking at a total rating of 142 * 2 * 0.8 = 227 for the dual G4 in all the categories. That suggests the dual 1.42GHz G4 is faster than a single 1.6GHz G5 for most things that can use both processors simultaneously.
Originally posted by Powerdoc
It will be interesting to see some bench, but i would not buy a dual 1,42 knowing his huge price and his disapointing performances.
I wouldn't buy a new dual 1.42GHz G4 either, but I think the price of used dual 1.42GHz machines can get very interesting.
Originally posted by Tidris
However floating point intensive applications are rare. Also the G4 has the L3 cache that helps to compensate for the lower memory bandwitdh.
The only benchmark results I have seen so far that shed some light on this are the Skidmarks numbers at thinksecret.com. The numbers compare a single 1GHz G4 against a single 2GHz G5 if I understand them correctly.
Adjusting for clock frquencies, a single 1.6GHz G5 should score 172 * 1.6 / 2.0 = 138 versus 142 for a single 1.42GHz G4 in integer operations. For floating point the single G5 should score 270 * 1.6 / 2.0 = 216 versus 142 for the single G4. For vector operations the single G5 should score 208 * 1.6 / 2.0 = 166 versus 142 for the single G4. However, the dual G4 has two G4s, so assuming an 80% multiprocessing efficiency we would be looking at a total rating of 142 * 2 * 0.8 = 227 for the dual G4 in all the categories. That suggests the dual 1.42GHz G4 is faster than a single 1.6GHz G5 for most things that can use both processors simultaneously.
Skidmark is lame for comparing a G4 with a G5.
You surestimate the performance of a dual system over a single. The G5 is more SMP than the G4. The G5 can reach 80 % efficiency, the G4 won't.
Many software dual optimized, are in fact optimized for only some tasks. A dual 1,42 ghz will be slower than an hypothetical single 2,5 G4.
here some benchmarks of a dual 1 ghz versus a single 1 ghz (powerlogix cards on a G4 466) :
speedmark 3,2 : single 133, dual 139
i movie 2,13 rendering : single 0,43, dual 0,40
i tune 3 : single 0,42,dual 0,33
The software who take more advantage of SMP is obviously here I tune, and it's only a 27% boost. You are way beyond the supposed 80 % ratio. That's why i claim (we have to wait for new benchmarks) that the G5 1,6 will be faster than the dual 1,42 in many applications (better mobo, better HD, better video card).
knowing his huge price and his disapointing performances.
Just wouldn't be AI without you Powerdoc!
The 1.6 G5 will DECIMATE the dual 1.42 GHz G4.
Forget the G4, it's a lame antique ready for the garbage. The sooner Apple ditches it the better.
Originally posted by Junkyard Dawg
One word: Bandwidth.
The 1.6 G5 will DECIMATE the dual 1.42 GHz G4.
Forget the G4, it's a lame antique ready for the garbage. The sooner Apple ditches it the better.
I tend to side with this opinion. I think the G5 will just keep going, when the G4 starts struggling under heavy workloads. The performance difference may not come to light in benchmark tests as well as it does in real work applications.
As for floating point intensive operation not being commonplace - that might be true, unless you've bought the machine to do 3D or scientific work, in which case it's your whole raison d'etre.
Originally posted by Aquatic
Just wouldn't be AI without you Powerdoc!
Quite right
I would say stay away from the 1,42*2s. The 1,25 dual is a much better buy. They wer the ones we should compare to the 1,6 G5s with regards to both price and performance.