Please, go to toyota, and ask them to deliver a 4runner with a Ford firmware for the engine management, and see the answer. if you can get this done, i ll agree with you and release all my soft for every CPU.

thanks

who?

So, you want martin to perform some task, and that'll get you to do what you've promised on many occasions to do already? You started with "I'll release my code". Then failed to deliver. So you promised the exact same thing again, then failed to deliver. When questioned about it, you then changed to "I'll release a Binary only", and have failed to do that. Now, it's I'll release an SSE4 only version. That ofcourse won't help any AMD users, but also won't help the majority of Intel users even if you make good on your word. Up[edit]un[/edit]til now it's just been "Ooooh, Look at me and my toys (both hardware and software) that you don't have and can't get" Bragging. '

So by now that we're using MB, you've promise two LF source codes, one LF binary, all of which at this point won't help squat, so you'll never be able to live up to your word on those. The Project has moved on and your opportunity is gone.

Now we have a promise of the Source Code once again, but only in SSE4. This I actually believe you might do, but it won't be enough to make up for the past lost opportunities to save your credibility.

ok, just read the thread and stop repeating again and again the same stuff ...
all the answers of your questions are below.

thanks, from now on, i ll just ignore this kind of posting, from fan boys, and the TOPIC here is BARCELONA on SETI
!!!!!!!

---

who?
Skulltrail D5400XS

Do I understand right? Is Phenom scheduled for release shortly ? (That AMD page says September 25th) any reviews apart from the Anandtech preview one yet?

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Do I understand right? Is Phenom scheduled for release shortly ? (That AMD page says September 25th) any reviews apart from the Anandtech preview one yet?

The link says:

Like Mother Nature, AMD has a dark side — and on September 25, 2007
it will be revealed to the world.

Exactly what "revealed to the world" means is anyones guess. Anyone know anything more?

Please, go to toyota, and ask them to deliver a 4runner with a Ford firmware for the engine management, and see the answer. if you can get this done, i ll agree with you and release all my soft for every CPU.

thanks

who?

So, you want martin to perform some task, and that'll get you to do what you've promised on many occasions to do already? You started with "I'll release my code". Then failed to deliver. So you promised the exact same thing again, then failed to deliver. When questioned about it, you then changed to "I'll release a Binary only", and have failed to do that. Now, it's I'll release an SSE4 only version. That ofcourse won't help any AMD users, but also won't help the majority of Intel users even if you make good on your word. Up[edit]un[/edit]til now it's just been "Ooooh, Look at me and my toys (both hardware and software) that you don't have and can't get" Bragging. '

So by now that we're using MB, you've promise two LF source codes, one LF binary, all of which at this point won't help squat, so you'll never be able to live up to your word on those. The Project has moved on and your opportunity is gone.

Now we have a promise of the Source Code once again, but only in SSE4. This I actually believe you might do, but it won't be enough to make up for the past lost opportunities to save your credibility.

ok, just read the thread and stop repeating again and again the same stuff ...
all the answers of your questions are below.

thanks, from now on, i ll just ignore this kind of posting, from fan boys, and the TOPIC here is BARCELONA on SETI
!!!!!!!

Actually, the topic seems to be Barcelona-on-SETI-and-how-Intel-kicks-Barcelona-Butt.

I'd prefer it be about Barcelona, and not about wonderfully optimized science apps for unreleased processors.

---

Please don't quote in biblical proportions what is only two messages away!
Instead, pick out sentences to refer to, if necessary.

The original off-topic message is but a click away!

Trying to get back "on topic", a rather strangely mixed note here:

Intel claims silicon crown despite AMD Barcelona

Intel is confident, with a few benchmark exceptions, that it will still have the fastest x86 processor in the market following today's launch of AMD's Barcelona Opteron processor

[...]

AMD released a set of nine benchmarks on Monday including SpecFP_rate2006. All benchmark data was based on a 2GHz Barcelona.

Compared to the best published benchmark score for Intel's quad-core Xeon, AMD outperformed the competition on all benchmarks.

[...]

But AMD puts itself at a potential disadvantage as it will ship only 2GHz Opterons at the launch, while its competitor already offers 3GHz models.

[...]


So... It's still a question of which processor and what price-point is best for your application for when you can buy whichever processor.

Happy crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Yes, back on track: can someone summarize the current barc performance on seti yet, or is it too early to tell quantitatively, and fairly?

Yes, back on track: can someone summarize the current barc performance on seti yet, or is it too early to tell quantitatively, and fairly?

Over the past 24 hours, I've seen my RAC go from 125 to 165.

Makes one think we need to see quite a few results before we start drawing conclusions...

---

Yes, back on track: can someone summarize the current barc performance on seti yet, or is it too early to tell quantitatively, and fairly?

I tried to do that downthread (here) - about 10 days ago, but I haven't seen anything suggesting that they've speeded up since then.

Came out rather slower than my Xeon 5320s (1.86GHz).

Here is the latest count:

Quad-Core AMD Opteron(tm) Processor 8347
Quad-Core AMD Opteron(tm) Processor 8346 HE [AMD64 Family 16 Model 2 Stepping 1]
Quad-Core AMD Opteron(tm) Processor 8346 HE [AMD64 Family 16 Model 2 Stepping 1]
Quad-Core AMD Opteron(tm) Processor 8350
Quad-Core AMD Opteron(tm) Processor 8347 [AMD64 Family 16 Model 2 Stepping 1]
Quad-Core AMD Opteron(tm) Processor 8347 [x86 Family 16 Model 2 Stepping 1]
Quad-Core AMD Opteron(tm) Processor 8347 [x86 Family 16 Model 2 Stepping 1]
Quad-Core AMD Opteron(tm) Processor 8350
Quad-Core AMD Opteron(tm) Processor 8347 [AMD64 Family 16 Model 2 Stepping 1]
Quad-Core AMD Opteron(tm) Processor 8346 HE [AMD64 Family 16 Model 2 Stepping 1]
Quad-Core AMD Opteron(tm) Processor 8347 [Family 16 Model 2 Stepping 1]

Ok, I must be pretty confused then. I believe that the version of seti one uses and the collection of wu's one receives can make a difference in the computed rac. Also, the rac can be expected to vary in time due to the statistical mechanics of the quorem granting mechanism.

But I just looked at the 'measured floating point speed' and 'measured integer speed' of two arbitrarily selected computers, one a Q and the other a B. The numbers were 2600 and 5700 M ops/sec for the Q and 1400 and 2600 M Ops/sec for the B, respectively.

So unless running a boinc derived benchmark is totally specious, in which case one wonders why the benchmark exists at all, then at least for seti, and independent of the rac, I'd conclude that the Q was a solid 2x faster than the B.

Hence, why on earth would we want to debate the B's value at all, at least on these discussion boards?

But I just looked at the 'measured floating point speed' and 'measured integer speed' of two arbitrarily selected computers, one a Q and the other a B. The numbers were 2600 and 5700 M ops/sec for the Q and 1400 and 2600 M Ops/sec for the B, respectively.

take a look @ host id 3823002 vs 3787191 the cpu is the same but the benchmarks are way out...

Also when comparing benchmarks across some of the pc's doing the same wu's the barcalona was slower yet the benchmarks say its faster... Almost makes me think there is more to this than what meets the eye.

~BoB

---

Do you Good Search for Seti@Home? http://www.goodsearch.com/?charityid=888957
Or Good Shop? http://www.goodshop.com/?charityid=888957

No point at all in discussing the benchmarks.
They have long been broken, and can't be trusted at all as they give
different results depending on the time of day and phase of the moon.

No point at all in discussing the benchmarks.
They have long been broken, and can't be trusted at all as they give
different results depending on the time of day and phase of the moon.

Now that's just got to be one of the better quotes! :-)

(The benchmarks measure what they measure. That often isn't representative of what s@h does with the hardware...)

Happy crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

What you seem to be saying is that the boinc architects and benchmark authors are bone heads because they implemented such bogus technology. Is that right?

If so, then what else about this technology project is so flawed that it makes one wonder why we are here at all?

I've been here at seti a long time and am getting a little tired of the constant apologies for server issues, errant scripts, and now it seems bad client software. I don't plan on slitting my wrists yet, but these issues are getting a little stale. Plus I don't see any (publishable) analysis/science coming out despite all our efforts.

So today I feel a sort of a double whammy due to your feedback below.

What you seem to be saying is that the boinc architects and benchmark authors are bone heads because they implemented such bogus technology. Is that right?

If so, then what else about this technology project is so flawed that it makes one wonder why we are here at all?

... So today I feel a sort of a double whammy due to your feedback below.

Sorry, but your summary there is wrong.


There is a lot of good history about the original benchmarks that were (are) used. The name "Cobblestone" is a very diliberately chosen name with multiple descriptive puns. (Look up the words cobble, cobbled, and cobblestone. Also whetstone and dhrystone (with and without the "h") and how blades are sharpened.)

Simply, Boinc is far too broad and runs on far too wide a variety of hardware to ever hope to have an all-encompasing accurate single benchmark. Just take a look at the Intel vs AMD benchmarking arguments, there's a fistful of various benchmarks all being argued over. And those benchmarks are just for the CPUs! Add in the vagaries of whatever applications, IO bandwidth, user's usage, and other aspects including the time of day and phase of the moon (metaphors for external influences for those literally minded), and you should start to see the problems for using such very specific benchmarks.

Just one example of how the science applications can perform differently on different hardware is for my resource share compared to my RACs between projects. They very clearly don't match.

Is the system broken because of all that?... Certainly not.

Can it be improved?... Certainly so.


Note that all of this is an "experiment". If we knew how best to do all this, then we wouldn't be using Boinc at all!

Unfortunately, the perfect successor to Boinc doesn't exist (yet).

Welcome to the experiment, and welcome to real development and real Science.

We are indeed doing something very useful with all this. And all very nearly "for free".

Happy crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Well said Martin.

All software is (or should be) in a constant state of
development and improvement, and those with ideas and
inspiration should be listened to.

In that way we can evolve the system to better suit the
needs of the science and please everyone involved.

Not to be argumentative, but you really aren't saying much here. Perhaps you can't. Fine.

The "systems" v "cpu" argument is well understood and not informative. It would be nice to have a clean benchmark designed for seti-type numerical applications to differentiate between cpu's to address the B v. Q arguments, ideally one that distinguishes between the various aspects of system design. But I'm not volunteering to write it.

And regarding the more important issue, that of science output, I really think things are beginning to smell here. Nothing (but PR) published on the specific SETI topic for four years or more. Huge amounts of computational energy and the concomittant results have been returned to Berkeley, but nothing has been synthesized. I don't think the structure of the ultimate analysis has been promulgated (at least publicly). It would be nice to believe that the scientific analysis is more subtle and mathematical than having skymaps and results-returned tables.

EOM

What you seem to be saying is that the boinc architects and benchmark authors are bone heads because they implemented such bogus technology. Is that right?

If so, then what else about this technology project is so flawed that it makes one wonder why we are here at all?

I've been here at seti a long time and am getting a little tired of the constant apologies for server issues, errant scripts, and now it seems bad client software. I don't plan on slitting my wrists yet, but these issues are getting a little stale. Plus I don't see any (publishable) analysis/science coming out despite all our efforts.

So today I feel a sort of a double whammy due to your feedback below.

Benchmarking is a black art. Always has been.

The idea is to create some kind of "synthetic" job that can be taken from computer to computer, that will predict how well it will do real jobs.

That's great as long as the instruction mix (add, divide, cos(), etc.) are the same in the benchmark as in the actual application.

For BOINC, that is likely never to be the case. We have projects like SETI that do complex math, and HashClash which does very little arithmetic (and no floating point at all).

So, the benchmark gives a rough idea -- good enough to download some work, and the duration correction factor corrects for the differences between the benchmark prediction, and reality.

So, I wouldn't call any of this perfect by any means, much of it is based on predicting the future (i.e. the -9 work units that complete very early).

But sometimes "good enough" really is, and until someone invents a "psychic benchmark" it is what we have.

---

Boinc devs have already brought up the idea of each project creating a "benchmark" wu for each partipant to run, so that credit could be adjusted between projects using fpops. However, Many projects don't want to do this. Some don't have the source code for the apps they run, so fpops counting is impractical. Not to mention, what would be the interval that users would need to download and run it? every 5 days like the current benchmark runs??

It's impossible to find one benchmark system that works for all projects.

This has already been discussed.