Possible cheater? Anyone know P4s?

[uOpt]

So, after reporting they seemed to have deleted that computer that was having astronomical benchmarks for a P4 2.4 GHz.

Today this new box pops up, scoring more than my 8-core:
http://stats.free-dc.org/new/allhost...direction=DESC

http://boinc.bakerlab.org/rosetta/sh...?hostid=201192
CPU type GenuineIntel
Intel(R) Pentium(R) 4 CPU 3.20GHz
Number of CPUs 2
Operating System Microsoft Windows XP
Professional Edition, Service Pack 1, (05.01.2600.00)
Memory 2045.73 MB
Cache 976.56 KB
Measured floating point speed 39998.54 million ops/sec
Measured integer speed 39998.71 million ops/sec

Looks like hyperthreaded P4. 2 CPUs but no Xeon.

I don't know whether this guy has clocked it at 4.13 GHz but for 32 bit Windows that looks fishy.

I know that Netburst CPUs do very well for drystone and wetstone, and if he had a 4-CPU Xeon, fine, but I am not sure I am ready to believe his for a hyperthreaded Pentium-4 or just a dual-core Netburst.

I see a 3.6 GHz dual-core P4 should be around 5928 Whetstone 10440 Dhrystone.

[eddieate]

yeah must be cheating/unintentionally glitching, I've never seen Measured floating point speed and Measured integer speed so close before or indeed that high.

[ex2cib]

probably the same person cheating again, they need to just ban the guys IP from registering, if thats possible

[Movieman]

It's a BS score. Impossibly high and you will never see the wet and drystone that close..never!
Time to report again!

[Mc Intosh]

How did you find this guy, maybe we can all keep an eye out.

this is the kind of sh*it that will make memebers leave if this starts to happen on a large scale..

[uOpt]

They really need to improve the benchmarking.

It would be easy to send all hosts a small chunk of "real" work, the same chunk, and measure how fast that is being returned with correct results. Also kicks out people with machines that are so overclocked that they don't compute correctly.

From my experience with drystone and wetstone I can also tell that they don't have much to do with real-world performance, much less scientific computing real-world performance.

[mad mikee]

The R@H people know of it, but I haven't seen anything in response

[Movieman]

They really need to improve the benchmarking.

It would be easy to send all hosts a small chunk of "real" work, the same chunk, and measure how fast that is being returned with correct results. Also kicks out people with machines that are so overclocked that they don't compute correctly.

From my experience with drystone and wetstone I can also tell that they don't have much to do with real-world performance, much less scientific real-world performance.

The last time I talked to David Kim at Baker Labs I suggested that they kill the benchmark altogether. Set a value for the particular work unit,say 100 points as an example. Then send it out. If it takes X machine 2 hours to crunch it he gets 100 points for 2 hours work. If it goes to " Holy Cripes the world's monster machine" and it takes him 10 mins to crunch it he gets 100 points for 10 mins work..
This totally eliminates the cheating...And it takes care of the issue of giving more points to faster( in theory more expensive and therefore a bigger donation to the project) machines.

[ex2cib]

How did you find this guy, maybe we can all keep an eye out.

this is the kind of sh*it that will make memebers leave if this starts to happen on a large scale..

usually, you have to check the top computers after you give the RAC time to build

then you just compare

intel's suck in r@h for one thing.
you might think he has a multi cpu system, he can't, as only 1 core gets benchmarked, which is the same score given to all of the work. benchmark scores dont add to each other. you would need about a 15-20ghz p4 with quad channel dd3 to pull numbers like that

its a BOINC problem, not an R@H problem

[Movieman]

usually, you have to check the top computers after you give the RAC time to build

then you just compare

intel's suck in r@h for one thing.
you might think he has a multi cpu system, he can't, as only 1 core gets benchmarked, which is the same score given to all of the work. benchmark scores dont add to each other. you would need about a 15-20ghz p4 with quad channel dd3 to pull numbers like that

its a BOINC problem, not an R@H problem

Hey,Hey! Let's be nice to my Intels! They've been good to me!
The guy in question has a P4-3200 w/HT..Nice PC but no world beater. The benches are bogus as all hell.
This is from my DX3600 that's a heck of a lot better system than that guy is using!
CPU type GenuineIntel
Intel(R) Xeon(TM) CPU 3.60GHz
Number of CPUs 4
Operating System Microsoft Windows XP
Professional Edition, Service Pack 1, (05.01.2600.00)
Memory 2046.95 MB
Cache 976.56 KB
Swap space 3943.74 MB
Total disk space 68.94 GB
Free Disk Space 18.4 GB
Measured floating point speed 2375.29 million ops/sec
Measured integer speed 3449.24 million ops/sec
I agree, it is a BOINC thing, not a Rosetta thing..