LHC@home

Hi,

anyone seen this before:

http://lhcathome.cern.ch/lhcathome/workunit.php?wuid=2504551

My AMD Quad core computed the WU in a "normal" amount of time whereas the Intel based systems ran very short successfull outcomes.

I think that results need to be scrutinized as this looks weird to me.

Cheers

PJ

---

Though they are SUPPOSED to be the same, the math units of the Intel and AMD are different. Though it is better than what we had in the past, the standard that the FPUs are built to, well, is actually pretty "loose" ...

So, two "identical", as far as the standard is concerned, FPU can return different results. In this case, it looks like you ran the balls around all day long and the other guys had the beam hit the wall in a very short time.

Not sure how it will work out, but if the consensus is that the beam should have hit the wall, well, oops ...

But, this is the point of running the simulations ...

---

Hello!

Today I got a 1 million turns WU.It is now 2 hours crunching and s 20% progress. Time remaining is about 4 H 45m.

But I calculated the remaining time to nearly 8 hours.

How could this be?

Below link to my machine.

http://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=6613436

Hello!

Today I got a 1 million turns WU.It is now 2 hours crunching and s 20% progress. Time remaining is about 4 H 45m.

But I calculated the remaining time to nearly 8 hours.

How could this be?

Below link to my machine.

http://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=6613436

Time to complete is ALWAYS an estimate.

One of the problems with LHC is that many of the tasks end early which, if all the million turn tasks on you computer have ended early, this will bias the time calculator to short times. I know it looks odd, but it is normal.

What is not normal, at least for LHC is for the progress to "freeze" ... as long as it is clicking up, and if you have a version that shows the graphics you can see the turn indicator ... and it is going up ... all is well in LHC-Land ...

---

Hi,

anyone seen this before:

My AMD Quad core computed the WU in a "normal" amount of time whereas the Intel based systems ran very short successfull outcomes.

Actually, yes. I have.

Several years ago, when BOINC was still a gleam in someone's eye and SETI@home was the only game in town, I had two PC's. One processor was a 486-66 Intel and the other had an AMD Pentium-class processor. The Intel-based machine would routinely process as many as three SETI@home work units to every one the AMD chip could process. I ran both machines for a couple of years before my dissatisfaction over other things finally prompted me to mothball the AMD machine. To this day, I won't even consider an AMD processor-based machine.

Memory doesn't serve me well enough to be able to quote exact clock speeds and whatnot, but I do remember that I came to the same conclusion as others on this thread... that the mathmatic units on the two supposedly-comparable chips were wildly out-of-sync, with the Intel chip being the better of the two, by far.

Cheers

Bob in Boise

---

Hi,

anyone seen this before:

My AMD Quad core computed the WU in a "normal" amount of time whereas the Intel based systems ran very short successfull outcomes.

Actually, yes. I have.

Several years ago, when BOINC was still a gleam in someone's eye and SETI@home was the only game in town, I had two PC's. One processor was a 486-66 Intel and the other had an AMD Pentium-class processor. The Intel-based machine would routinely process as many as three SETI@home work units to every one the AMD chip could process. I ran both machines for a couple of years before my dissatisfaction over other things finally prompted me to mothball the AMD machine. To this day, I won't even consider an AMD processor-based machine.

Memory doesn't serve me well enough to be able to quote exact clock speeds and whatnot, but I do remember that I came to the same conclusion as others on this thread... that the mathmatic units on the two supposedly-comparable chips were wildly out-of-sync, with the Intel chip being the better of the two, by far.

Cheers

Bob in Boise

Evidently SUN differs, since has recently announced Opteron based blade servers. And my last LHC result has been crunched by three AMD cpus. Cheers.
Tullio

I ran both machines for a couple of years before my dissatisfaction over other things finally prompted me to mothball the AMD machine. To this day, I won't even consider an AMD processor-based machine.

Memory doesn't serve me well enough to be able to quote exact clock speeds and whatnot, but I do remember that I came to the same conclusion as others on this thread... that the mathmatic units on the two supposedly-comparable chips were wildly out-of-sync, with the Intel chip being the better of the two, by far.

Cheers

Bob in Boise

On some projects, the L2 cache size plays a very big factor in WU runtimes, even when FLOPs in benchmark test are similar. This could be another example of that.

---

I ran both machines for a couple of years before my dissatisfaction over other things finally prompted me to mothball the AMD machine. To this day, I won\'t even consider an AMD processor-based machine.

Memory doesn\'t serve me well enough to be able to quote exact clock speeds and whatnot, but I do remember that I came to the same conclusion as others on this thread... that the mathmatic units on the two supposedly-comparable chips were wildly out-of-sync, with the Intel chip being the better of the two, by far.

Cheers

Bob in Boise

On some projects, the L2 cache size plays a very big factor in WU runtimes, even when FLOPs in benchmark test are similar. This could be another example of that.

AMD\'s have so little of this, becuase of the way they communicate with the chipset. They dont actually need as much L2 Cache as Intel Processors.

The procs do flip flop some in terms of performance. In the Pentium days, arguably Intel had the upper hand, until the Athlons. AMD then pulled the upper hand through the Athlon classics (K7) and the A64, as the Pentium 4s were hell under-performing P3s even. Then Curusoe came out. In a competitive market, this should be expected.

There\'s also errata on any given CPU or core stepping, of which the FDIV design flaw is probably one of the more noted (early Pentium or )5 cores). This can also be in one stepping of a core and not another (ala 2 dif core steppings of a Pentium II. If an errata comes up, it could result in a miscalculation or erroring out, where even another core stepping of the same CPU doesn\'t exhibit the same behaviour.

---

Hi,

....One processor was a 486-66 Intel and the other had an AMD Pentium-class processor. The Intel-based machine would routinely process as many as three SETI@home work units to every one the AMD chip could process....

Sounds like you had an AMD K6-II, they had hideously slow FPUs; the K6-III corrected this.


TTFN - Pete.

---

hi,

all the explanations given here don\'t give any good reason why an intel CPU is (!!!) 20,000 times faster than an equivalent AMD CPU...

intel CPUs may be 1,5 - 2 times faster, but 20,000x, it\'s irrationnal!!!

I agree with PJ, these WUs have to be analized...

(sorry for bad english),

BD

Gee -- 20,000 times? (Makes note to self--> buy Q6600 immediately)

---

If I've lived this long, I've gotta be that old

This sure sounds like one of the WU\'s \"hit the wall\" early on, and the other one completed all it\'s turns. (still 100,000, I believe).

A fair number of WU\'s on this project complete in just a few seconds - the beam hits the wall! There\'s no error in the WU processing, and it\'s valid data.

---