Yes I think that's one of his...

Benher – If you’re going to talk about your optimized SETI client, why don’t you release the binary? Especially since nobody can seem to compile the client (or the optimized version on sourceforge) for a Win32 machine.

Also, I’m assuming you modified the BOINC client source code to produce your very high floating point and integer measurements. If you modify your benchmarks, what’s to stop others from doing the same?

---

I supplied the code to Eric Korpela, as part of the process to include in the general release science client. He thought it would be good to put it on sourceforge to have additional programers contribute. This allowed JavaLizard to post his Altivec code and Erik Heien posted his FFTW3 version.

That decision is up to Eric and David.

> Also, I’m assuming you modified the BOINC client source code to produce your
> very high floating point and integer measurements. If you modify your
> benchmarks, what’s to stop others from doing the same?

The benchmarks have long been a bone of contention. Even with the release version of BOINC, the benchmarks vary wildly between operating systems and CPUs. My AMD XP at 2.1 GHz benchmarks at about 1938 Mflops under Windows with version 4.13, (and claims close to 40 credits per WU), but benchmarks at 1050 Mflops under CYGWIN (unoptimized) and about 1300 optimized. Under Linux, it benchmarked at 1055 Mflops (version 4.13). My Pentium machines seem to come in on the low side as far as benchmarks and requested credit (based on averages).

The problem as to what to do with the bemchmarks is interesting. Right now, WU's are generating roughly 30 credits on average (so it seems about 1500 would be the "correct" benchmark for my Athlon). If a newer "optimized" client comes out that allows an existing machine to do a WU in 30% less time, should that machine be awarded 30% more credit? Most likely, it's actually doing 30% fewer FLOPS due to the optimized code. So it should technically be awarded 30% LESS credit (per WU) so that it will get exactly the same amount of credit as it did before using the optimized client. It's not doing any more work than it did before, it's just doing it more efficiently. Of course, then those without the optimized client will not be happy, as these new optimized clients will effectively lower the average per WU being requested.

It's certainly a dilemma, but at this point, a few homegrown clients are not going to impact what others are awarded for credit. In my research (into Hans' scsreamers ;), those with homegrown clients aim to request the "correct" amount of credit based on the existing release of BOINC. Yes, they're going to get more credit per day with the optimized clients...that's probably OK considering the effort they're putting into development for the ultimate good of the project.

Just my two cents (ok, maybe 10 cents).

---

Hopefully at least the benchmark of linux vs Windows will be more close when they release BOINC 4.6x.

From BOINC CVS checkin notes in November 2004:

David 2 Nov 2004
- split Dhrystone source into 2 files (from Peter Smithson).
This supposedly prevents compiler optimizations on Win
that give inflated results

From BOINC_dev mailing List (Peter Smithson):

The original dhrystone source for revison 2.0 was split intentionally to
avoid over optimisation (see comments in original dhrystone test source
code). The BOINC uses merged source which causes the benchmark to run
unusually quickly when compiled with MSVC 7.1 due to better optimisation
possible. Presumably the optimiser is taking out most of the test and not
giving a result consistent with a real application. MSVC 6.0 did not do
this.
Link

---

SETI@home classic workunits: 1,985 CPU time: 24,567 hours



Professor Desty Nova
Researching Karma the Hard Way

Feel the Power of Athlon XP ;-)

http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=535675

---

<img src="http://www.boincstats.com/stats/banner.php?cpid=a391dacd6238aec627931362f1150623">

> Feel the Power of Athlon XP ;-)
>
> http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=535675
>

Impressive times there! Crunch on!

---

Now I know for sure I am depressed ... I want to go out and buy one ...

Toxino, what is the clock speed on that processor?

---

Its a Athlon Mobile 2600+ @ 2800 Mhz (real) Ram G-Skill PC4800(DDR600) @ 230Mhz 6,3,3,2.5 Rating ~ 4400+
Changing Ram from Geil to GSkill to get a Higher FSB has given me 7 minutes lower Crunching Time . Both Ram are 2x512 MB.

---

<img src="http://www.boincstats.com/stats/banner.php?cpid=a391dacd6238aec627931362f1150623">

If you think that's impressive, take a look at this:

http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=36273

He/she is certainly knocking out those WUs. ;)

---

Be lucky

Neil

Yeah its impressiv. There is a little difference:

Athlon XP 1 CPU : 2600 million ops/sec
6226.11 million ops/sec

Xeon with 4 CPU´s: 2990.31 million ops/sec
7992.03 million ops/sec

so u can see urself if i have 4 CPU´s............

be lucky too ;-)

---

<img src="http://www.boincstats.com/stats/banner.php?cpid=a391dacd6238aec627931362f1150623">

> Yeah its impressiv. There is a little difference:
>
> Athlon XP 1 CPU : 2600 million ops/sec
> 6226.11 million ops/sec
>
> Xeon with 4 CPU´s: 2990.31 million ops/sec
> 7992.03 million ops/sec
>
> so u can see urself if i have 4 CPU´s............
>
> be lucky too ;-)
>

I assume the CPU benchmarks of the XEON system is for ONE CPU only, since the recent average credit is four! times higher than yours, at ~about the same CPU-Power than your Athlon mobile.

---

> > Yeah its impressiv. There is a little difference:
> >
> > Athlon XP 1 CPU : 2600 million ops/sec
> > 6226.11 million ops/sec
> >
> > Xeon with 4 CPU´s: 2990.31 million ops/sec
> > 7992.03 million ops/sec
> >
> > so u can see urself if i have 4 CPU´s............
> >
> > be lucky too ;-)
> >
>
> I assume the CPU benchmarks of the XEON system is for ONE CPU only, since the
> recent average credit is four! times higher than yours, at ~about the same
> CPU-Power than your Athlon mobile.
>

I also believe that rig is 4 real CPUs (no hyperthreading). A hyperthreaded Xeon 3.06 should be doing WU's in 11,000 seconds or so; this PC routinely gets times in the 6000's.

---

> I also believe that rig is 4 real CPUs (no hyperthreading). A hyperthreaded
> Xeon 3.06 should be doing WU's in 11,000 seconds or so; this PC routinely gets
> times in the 6000's.

My MSDN subscription lapsed a long time ago, but, as I recall XP Pro will only wrk for up to 2 CPUs, unless that has been changed. if that is true, that would still indicate a HT Xeon.

But those may simply be "old" xeons based on the P III core, which rumor has as being more efficient in processing work.

When I get real depressed I go shopping for a new computer and I was looking hard at building a dual Xeon ... That will probably happen after I get this years Power Mac (or I save up past $4K) so maybe I will be able to tell you more early next year.

---