I've notice that #1 and #2 have quit crunching, only #3 is still going.

#1 has work to crunch but isn't, #2 doesn't have any work left to crunch.

I wonder why that is?

---

maybe some oc going on

A quick look at boincstats states that maybe it isn't quite so bad...


Pos. CPU # CPU Total credit Average credit Credit per CPU Average credit per CPU Average credit per CPU second
159 Intel(R) Core(tm)2 Duo CPU L7500 @ 1.60GHz 10 18,995.90 682.68 1,899.59 68.27 0.004127
160 Quad-Core AMD Opteron(tm) Processor 8347 3 2,490.18 200.08 830.06 66.69 0.004107

As you can see by what's bolded its comparable with a 1.6Ghz core 2 except it does 2x the work (unless in quad fx mode then its 4x) . Granted that there is not much total credit for the AMD's so reserve judgement for later. We shall she how it does.

URL

(hopefully that url stays the same!)

~BoB

EDIT: I have asked for clarification on the credits per cpu second because as it stands it doesn't make sense.

---

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

A quick look at boincstats states that maybe it isn't quite so bad...


Pos. CPU # CPU Total credit Average credit Credit per CPU Average credit per CPU Average credit per CPU second
159 Intel(R) Core(tm)2 Duo CPU L7500 @ 1.60GHz 10 18,995.90 682.68 1,899.59 68.27 0.004127
160 Quad-Core AMD Opteron(tm) Processor 8347 3 2,490.18 200.08 830.06 66.69 0.004107

As you can see by what's bolded its comparable with a 1.6Ghz core 2 except it does 2x the work (unless in quad fx mode then its 4x) . Granted that there is not much total credit for the AMD's so reserve judgement for later. We shall she how it does.

URL

(hopefully that url stays the same!)

~BoB

EDIT: I have asked for clarification on the credits per cpu second because as it stands it doesn't make sense.

Don't think that comparison stands up as the L7500 cpu is a laptop cpu.
Comparing the crunch times at similar AR's with my C2D 6600, they are 2.5 to nearly 4 times slower. Admittedly they are using stock app and I am using Chicken soup.

If all units were VLAR's those Barcelona's with 8 cores on stock app would only have a slightly higher performance than my C2D, using chicken, with only two cores.

Andy

I thought this deserves a new thread...

The first Barcelona on SETI
The second Barcelona on SETI

Andy.

Are there any other Barcelonas running that show better results than the ones here?
The time it takes these machines to earn credit makes it look like the 8 processor Barcelona would likely earn about the same RAC that my Core 2 Duo does in a 24 hour period.

---

When we finally figure it all out, all the rules will change and we can start all over again.

Well, there are 5 on SETI as of now:

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

china_cia seems to like them - he's got two now!

Well, there are 5 on SETI as of now:

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

china_cia seems to like them - he's got two now!

The two 8350 are under the same owner - china cia.

The other three 8347 are under owner - Anonymous. Obviously difficult to tell if Anonymous is the 'same' anonymous for all three although that could be a reasonable assumption.

[edit] I took screen shots of several of china cia's wu's screens just in case they start reporting better results for those same angles, at which case it would be safe to assume overclocking is at work. [eoe]

---

Well, that's more like it. We're getting some data now.

For the sake of the plots, I'm assuming that china_cia'a two 8350s are actually the same machine (same memory, same OS, same benchmark, same version of BOINC, same stock app.....). I'm also assuming that the two anonymous [x86 Family 16 Model 2 Stepping 1] are the same - he's stuck an extra gig of RAM in it since the last run, and upgraded from BOINC v5.10.13 to v5.10.20, but that's all.

So I'm up to about 130 datapoints so far. All a bit bunched on AR at the moment, so I'll let them run overnight, but we should have a good chart to pore over sometime tomorrow.

I'm thinking something is off on these results so far. Maybe these servers are under a load in addition to crunching Seti. That would make the results look worse.

My 2.4 Ghz Athlon X2 beats a Pentium 4 of the same speed in less than half the time.

http://setiathome.berkeley.edu/workunit.php?wuid=157070865

I would expect something similar from the new Quad Core's. They need to upgrade the app to the optimized AMD one and just crunch Seti and nothing else.

---

It looks like all the Barcelonas with results are using the stock cruncher. With little workunits their times are more that twice as slow as those of my Pentium D950s. With regular-size ones, they're less than 2/3 as fast (per core).

---

It looks like all the Barcelonas with results are using the stock cruncher. With little workunits their times are more that twice as slow as those of my Pentium D950s. With regular-size ones, they're less than 2/3 as fast (per core).

I noticed the same thing and it doesn't make sense. An Athlon is faster than that. Something is definitely off here.

---

Well, here are the results of my charting so far:


(direct link)

I've combined the plots for hosts 3768873 and 3794140 into a single 'Opteron 8350' sequence, and likewise 3773162 and 3787191 into 'Opteron 8347': in both cases, I'm pretty sure that they're the same machine, and the owner is just testing them out, installing and re-installing other software (I saw a lot of restarted WUs among the data).

I've also left some pre-existing data from my 2 x E5320 host 2901600, because I reckon an 8-way server-class CPU setup is a better point of comparison than the consumer-grade duos and quads. My data was collected between August 17 and 26 using the stock app (while no optimised app was available for MS Vista), so I hope you'll agree it's a fair comparison. My Xeons run at stock 1.86GHz, and have 4 x 1GB FB DIMM memory: all I know about the Barcelonas is what you can see on the host summary pages.

To put some hard figures on it:
Xeon E5320 ... average 15.0877 credits per core per hour (158 results)
Barca 8350 ... average 14.0284 credits per core per hour (60 results)
Barca 8347 ... average 15.4495 credits per core per hour (103 results)
Q6600 (at stock) ... average 23.4943 credits per core per hour (199 results)

To put some hard figures on it:
Xeon E5320 ... average 15.0877 credits per core per hour (158 results)
Barca 8350 ... average 14.0284 credits per core per hour (60 results)
Barca 8347 ... average 15.4495 credits per core per hour (103 results)
Q6600 (at stock) ... average 23.4943 credits per core per hour (199 results)

Richard, I'm having difficulty making sense of the graph without having to think a lot... Could you try with time on the x-axis, as per convention?

ps. I'm getting about 50 creds per core per hour from my best quad, and 43 for the other one.

To put some hard figures on it:
Xeon E5320 ... average 15.0877 credits per core per hour (158 results)
Barca 8350 ... average 14.0284 credits per core per hour (60 results)
Barca 8347 ... average 15.4495 credits per core per hour (103 results)
Q6600 (at stock) ... average 23.4943 credits per core per hour (199 results)

Richard, I'm having difficulty making sense of the graph without having to think a lot... Could you try with time on the x-axis, as per convention?

ps. I'm getting about 50 creds per core per hour from my best quad, and 43 for the other one.

As a work-around you could take a screenie, stuff it into PhotoEd and rotate it by 90 deg :o

Richard, I'm having difficulty making sense of the graph without having to think a lot... Could you try with time on the x-axis, as per convention?

ps. I'm getting about 50 creds per core per hour from my best quad, and 43 for the other one.

Well, it depends what you mean by time. No, really - bear with me.

I plot by 'date reported' when it matters when something happened:



tells me that my CPDN models speeded up when Simon released his 2.2B 'chicken' optimisations.

But it doesn't work when there are two independent variables.

Here, we're interested in whether the Barcy's are better than a reference standard under various SETI processing conditions. For any SETI WU (independent variable, plotted by AR, X-axis, log scale), which is better, Xeon or Barcy?

I'm plotting duration, so the further north (up-screen) the blob, the worse the performance. The further south (down-screen) the better. Does that help?

{viewers in the Antartican hemisphere may wish to reverse those compass bearings}

I guess I'm just used to seeing time in the x-axis, even durations...
I'll get used to it!

I did have a brain, somewhere... I think I must have mislaid it while researching load-balancing failover mysql clustering and high availability apache server farming... @-)

Richard, I'm having difficulty making sense of the graph without having to think a lot... Could you try with time on the x-axis, as per convention?

I plot by 'date reported' when it matters when something happened:

But it doesn't work when there are two independent variables.

I think the standard convention is to plot the independent variable (stimulus) on the X-axis, and the dependent variable (response) on the Y-axis.

So I think Richard followed that convention with his choice here.

Time of observation is pretty nearly never a dependent variable, so we get used to seeing it on the x axis, but that is not what is plotted here.

---

I have never seen AR versus CPU time graphed any other way, If so someone please link.

---

"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.

I have never seen AR versus CPU time graphed any other way, If so someone please link.

That's probably because only Tony and I have graphed it!

That's probably because only Tony and I have graphed it!

Well graph paper IS expensive you know ! :D

---

"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.