Intel Performance question. 3.0 GHZ P4 versus 3.2 GHZ P4
26 Mar 2005 20:57:06 UTC
Topic 188751
(moderation:
I have Einstein@home running on two computers.
Here are the benchmarks for these two computers taken from the Einstein@home webpage.
3.0 GHZ P4 running XP Pro SP1
----------------------------------------
Number of CPUs 1
Measured floating point speed 1546.38 million ops/sec
Measured integer speed 3088.95 million ops/sec
3.2 GHZ P4 running XP Pro SP1
-----------------------------------------
Number of CPUs 2
Measured floating point speed 876.72 million ops/sec
Measured integer speed 791.93 million ops/sec
An Einstein@home work unit typically completes on the 3.0 GHZ computer at around 28,000 CPU seconds while the 2nd presumably faster 3.2 GHZ CPU takes approximately 69,000 seconds.
I decided to experiment and I changed the BOINC account options to process only one work unit at a time on the 3.2 GHZ hyperthreaded computer and the time went down to 50,000 seconds.
Now I can see that the floating point speed on the 3.2 GHZ processor is a bit more than twice as fast as that of the 3.0 GHZ. Taking into account hyperthreading, I guess this would be expected.
But why the huge discrepancy with integer speed? More to the point why is Einstein@home slower on the 3.2 GHZ machine? 28,000 * 2 = 56,000 is still less than 69,000.
Maybe I am totally misinterpreting something here but I’m mystified. I would appreciate it if someone could explain this.
--Gary
Intel Performance question. 3.0 GHZ P4 versus 3.2 GHZ P4
)
It appears to me that your 3.2gHz machine is not running at 3.2 gHz. Your bench are very low.
here are my bench marks for a 2.6 gHz overclocked to 2.73 gHz.
Measured floating point speed 1164.56 million ops/sec
Measured integer speed 1788.88 million ops/sec
Take a look at your Programs\\accessories\\system tools\\sysinfo to see at what speed your processor is running.
Perhaps you can try turning
)
Perhaps you can try turning HT off. You might be running one instance at only 50%. Look at your task manager.
> It appears to me that your
)
> It appears to me that your 3.2gHz machine is not running at 3.2 gHz. Your
> bench are very low.
>
> here are my bench marks for a 2.6 gHz overclocked to 2.73 gHz.
> Measured floating point speed 1164.56 million ops/sec
> Measured integer speed 1788.88 million ops/sec
I knew that Intel is a bit slower, but I never realized that they are that slow.
Benchmark of my notebook with mobile AMD64 2800+(1.6GHz real):
Measured floating point speed 1516.82 million ops/sec
Measured integer speed 4136.82 million ops/sec
faeshn
When boinc benchmarks a
)
When boinc benchmarks a hyperthreaded P4 it divides the benchmarks in half when there are 2 virtual cpu's.
Measured floating point speed 2330 million ops/sec
Measured integer speed 3576 million ops/sec
My p4 completes 2 work units in 46000 or one every 23000 seconds. I would admit that it is not blazingly fast but it is probably on a par with your AMD64 2800.
I also have an Opteron 146 clocked at 2.1 gHz that knocks em out in 21200 seconds.
> 3.2 GHZ P4 running XP Pro
)
> 3.2 GHZ P4 running XP Pro SP1
> -----------------------------------------
> Number of CPUs 2
> Measured floating point speed 876.72 million ops/sec
> Measured integer speed 791.93 million ops/sec
These values looks to me quite low. Here are mine, XP Pro SP2, P4 HT, 512MB Ram:
Measured floating point speed 1276.57 million ops/sec
Measured integer speed 1075 million ops/sec
When running the benchmark, be sure there are no heavy software running.
The numbers I give are the one right after the upgrade of the CC, running fully automaticly, meaning all programs running in the background left like they were.
Greetings from Belgium
Thierry
> When boinc benchmarks a
)
> When boinc benchmarks a hyperthreaded P4 it divides the benchmarks in half
> when there are 2 virtual cpu's.
Thanks. I never understood where the power went away.
> Measured floating point speed 2330 million ops/sec
> Measured integer speed 3576 million ops/sec
> My p4 completes 2 work units in 46000 or one every 23000 seconds. I would
> admit that it is not blazingly fast but it is probably on a par with your
> AMD64 2800.
It takes for one WU on my notebook 28200 seconds, 3 WUs per day, now since a month.
> I also have an Opteron 146 clocked at 2.1 gHz that knocks em out in 21200
> seconds.
The mobile AMD64 1.6Ghz equals the Opteron 142. Your Opteron slowed down would need same time.
faeshn,
off now, here in Germany we have this night change to summertime and I have to get out very early.
Benchmarks are not relevant.
)
Benchmarks are not relevant. Only processing time ... :)
If you look at the averages I have built up you can see a general trend:
P4 2.8 GHz (18 results) => ~30K Seconds, or ~8:15
P4 3.0 GHz HT (55 results) => ~39K Seconds, or ~10:50
P4 3.2 GHz HT (62 results) => ~40K Seconds, or ~11:12
Now, these are Skewed for the 3.2 and don't reflect current numbers, but, one of my 3.2 MB did not have dual channel memory and typically took an hour or more per model than did the other 3.2 GHz processor ...
The HT processors take more time per WU, but produce more work per unit time ... You can look in the Glossary for the words Hyper-Threading, look at the link for the page that discusses benchmarks and HT on vs off ...
The short answer is that HT distributes work amongst internal capbilities of the CPU and we don't see a "doubling" because we spend most time in code places where we don't see the gains ... it *MAY* happen that some of the effort of the optimization folks will pay off and we may see more improvement. In particular if we can get smarter about the scheduling of work within the Science Applications so we use more of the resources more of the time ...
it looks very slow to
)
it looks very slow to me
my P4 3.2 with HT and two 512 MB Twinmos Syster
27.03.2005 00:09:37||Benchmark results:
27.03.2005 00:09:37|| Number of CPUs: 2
27.03.2005 00:09:37|| 1370 double precision MIPS (Whetstone) per CPU
27.03.2005 00:09:37|| 1201 integer MIPS (Dhrystone) per CPU
Greetings from Germany NRW
Ulli S@h Berkeley's Staff Friends Club m7 ©
Everybody: Thanks for all
)
Everybody:
Thanks for all of the responses.
A couple more specs. The 3.2 GHZ processor is on a Gigabyte 8KNXP motherboard with 2 512 Meg Corsair Low Latency twinx memory sticks. I have not overclocked this board and all of the relevant settings are at factory defaults.
Paul Buck’s averages are interesting. The faster the Intel processor, the longer the work unit processing time? Now that is counterintuitive to say the least. That explains this issue somewhat.
But comparing my 3.2 GHZ benchmarks with Sir Ulli’s is rather interesting. His benchmarks are significantly faster. There might be room for improvement with my configuration. I would think that CPU benchmarks would be closer than this even though we are probably using different motherboards, memory and so forth.
I’ll examine my rig more closely this week and recheck the bios settings. I’ll post here if I figure out what’s wrong (if anything).
--Gary
Ok, to avoid comparing apples
)
Ok, to avoid comparing apples and oranges...when a question like this is asked, people should ask the questioner (he's already answered now) the following.
CPU Vendor
CPU Speed listed Mhz
CPU Actual Mhz (could be overclocked)
CPU Core version - Can use a program like CPU-Z or Everest Home Edition to find this out.
CPU L2 Cache size
RAM Speed 2100, 2700, 3200, etc.
RAM stick Quantity - Ie can they be interleaved