Confused by Rosetta RAC of 641.35 for AMD 64 3400+

641.35 15,753.55 AuthenticAMD AMD Athlon™ 64 Processor 3400+ Microsoft Windows XP Home Edition, Service Pack 2, (05.01.2600.00)
Can someone explain how is it possible for an AMD 64 3400+ to attain such a high RAC ?

I’m guessing optimised client :nod: what are the benchmarks ?


No idea. I just picked the 4th machine from Nightlords list of PCs.

Just wondering what I’m doing “wrong” ?!?! :eek:

ah - yup, would appear to be using the optimised client :deal:

There is a thread in the crunchers cafe on the best ones to use - they really make a massive difference :slight_smile:


That an optimised client or optimised science app?

Well just upped my main rig to a Optimised version of 5.2.13

Amazed in the jump in Interger speed…3000 ish to 8000 :eek:

optimised CORE client, means a version of the boinc.exe optimised to the type of processor you have and the extensions it supports, rather than using a “support all”.

The only one I know of that is optimised “cruncher app” is seti. I think the others are restricted source.


The opteron 165 under Windows gets almost 10K (9743) on integer on each CPU with a optimize BOINC client. :eek:

Rosetta will be releasing their source according to Dr. Kim in the fairly near future. They want to get everything as “perfect” as possible first. Main reason for the release is for optimizing to hardware for those so inclined.

What one’s are fastest though?


And are they only for Pentium 3 upwards? Cos I could sure use one on a Pentium 2 :stuck_out_tongue:

use CPU-Z to identify the extensions that your CPU supports, for your P2, I’d think running the XP/P3 one should give a small boost.

Check out the post in the cafe, was most excellent work :agree:


It is indeed my machine.

I use a Boinc Core client from TruXoft. This core client is not particularly optomised - it has no SSE3 specifics. It attempts to adjust the claimed credit itteratively to match the expected claimed credit against a known workunit.

The known WU is the standard test unit at Seti. This WU should claim 32 credit based on the number of flops required. Most machines under claim due to problems with the benchmarking system. TruXoft released his Boinc Client to try to overcome this problem. In essence, after self calibrating over several units, the client should claim “fair” credit based on performance against the standard unit at Seti.

Other additions to this client are the ability to return results immediately and flagging up rogue WU that are stuck. Both of these features will add to the RAC by maintaining an even flow of results to the project.

Be careful however running Boinc clients from Crunch3r as they do not attempt to moderate the claimed credit and are considered by some to be over-inflating the claimed credit, with obvious problems on Rosetta. On Seti that client is a monster, but dangerous on Rosetta.

The machine in question, like most of the others, was switched back on to Rosetta a few weeks back after I had detached it previously. When I re-attached I merged the old and new host together. Rosetta added the RAC from the old and new hosts together resulting in a RAC of about 620 iirc.

I thought that was odd at the time, but since then it has slowly climbed to stand at 640 (in fact 648 according to BoincStats).

It is overclocked from std 2.4GHz to 2.58GHz with HTT at 215MHz on stock cooler. It also has 1GB of Kingston DDR400. I find that on Rosetta, especially recently with shorter run times I get several older units or suspended locked up units pre-empted in memory - effectively a memory leak. So I run 1GB when the machine warrants it. That machine returns about 600-700 credit per day on Rosetta. On Seti, with the validation process, it used to return about 800-900 credit per day.

In fact all my machines have taken a big hit in RAC by moving to Rosetta. I used to average 8000-9000 credit per day on Seti with the validation process. After switching to Rosetta, I achieved 4500 per day, now increased to about 5200-5500 with two additional crunchers.

It is painfully easy to inflate the claimed credit and hence inflate your score on Rosetta, but that would be cheating. The TruXoft client at least attempts to claim fair credit for work done.

Hope that helps a bit.

[edited to add some data on memory on machine in question, and I can’t spell!]

Cheers Nightlord.
After reading that, I think that might be more of a better optimized client than the other one I installed earlier.
If it’s more fair I’m up for it.

Oh and can you check the XML file? Think I’ve got it set right…espcaillay as it appears in messaes in BOINC…

AIming to turn on calibration, the return immediatly and the flag suspect WU’s…(Saved here as a .txt but XML on machine)

Yup, the xml is fine. You may get some false alarms with check_max_time at 2.0. But no harm in starting from there - mine range between 2.0 and 3.0. Beware, if a unit is suspended due to triggering the max_check_time flag it will be pre-empted and drain more memory.

I also added the following:


which has never happened, but with the wacky due dates on Rosetta recently, some of the older units might run overdue.

The output in the message screen is pretty difficult to interpret. Reading his site and This thread over at Seti helps explain some of the messages (but it is a huge thread!)

Installed it and running fine here.

Does it work on Rosetta and SIMAP?

Last SIMAP result said:

08/02/2006 00:12:40|boincsimap|CC calibration: blocked [negative calibration limit] 10.44 >> 4.00 (time: 1890s >> 747s / Gfpops: 2.12 >> 1.83)

Guessing it will take more units to drag the score down to about 4 where it should be? (other result submitted for WU is 4.83)

It should be ok on all Boinc projects, except of course CPDN. Even there it will run perfectly ok, but just not do anything to calibrate credits!

It takes quite a few units to self calibrate. Also remember that you may have been “under claiming” relative to the (seti) std unit before. So it may not match your previous claimed credit, but perhaps somewhere in between.

Anyone else got any thoughts on this TruXoft calibrating client, 'cos I’m kind of aware that I’m presenting a one sided view - maybe others have a different view.

I’ll use it some more and see how it fairs…

Didn’t realise that the Crunch3r ones where frowned upon in Rosetta but this one should be fine I guess…

My personal view is that on the windows boxes I have, I have absolutely no worries about using the cruncher client, even for Rosetta. It evens out the massive loss I take running on linux diskless. I think I’ve said somewhere before, the exact same hardware (nearly everything is barton2500’s and asus a7n8x variants), using standard clients the difference is output is massive. I could take a slight difference on the chin, but after running the boinc client standard for five days I have no problem “reclaiming” my losses on linux by using a crunch3r client.


I’m really glad I asked the original question now, thanks very much for all the info.


I agree, I use the Crunch3r client on my 2 linux boxes to offset the appallingly bad credits claimed by the stock linux client.

If I had more boxes on linux, then I would also consider it fair to run some Windows boxes on that client too.

The TruXoft client sometimes adjusts the credit, sometimes it doesn’t. Either way, it’s trying to match a standard benchmark. For me it’s a personal choice to be as even-handed as possible.

Opteron 270 running linux and crunch3r:
Measured floating point speed 2857.99 million ops/sec
Measured integer speed 4577.78 million ops/sec

3.4 Ghz pentium4 with hyperthreading running Windows and TruXoft (Nightlord)
Measured floating point speed 1917.65 million ops/sec
Measured integer speed 3810.07 million ops/sec

Looks pretty balanced to me. I would expect a 2 Ghz AMD64 to out run half a 3.4Ghz P4