Optimized clients

[Bok]

Sticky thread for posting links to optimized clients.

I'll add in a how-to later today for compiling, for now this is just a few links. Feel free to add more links and I'll merge them into the sticky..

Very fast windows clients based on the 5.3.1 source

More windows based clients

Compiling your own

Fast AMD linux clients

Bok

[Angus]

Thanks , Bok.

I think it's important to point out that the BOINC client does not actually do any crunching. It is only your PC's interface with all the various BOINC application servers.

However, it DOES affect your credits since the BOINC client runs the benchmarks which are used to calculate credit.


The optimized science applications (your second link) DO affect the crunching time. If you crunch faster without changing the benchmarks, you will actually get less credit for a specific work unit. Then it becomes important to run optimized BOINC clients to get the benchmarks up to get more credits.

Currently, there is no optimized Rosetta science application, so running the the optimized BOINC client to improve your benchmarks should generate more credits per given WU.

[PCZ]

We need some optimised version 5 linux clients.

Version 4 is no problem but version 5 has been a pain to build and the scores are way down.

[Bok]

I still can't get scores anywhere close to the ones I have on 4.72 optimized clients with the 5.31 source. I'm doing exactly the same thing so I don't understand why either. I'm hoping someone over at the seti forums releases some soon...

Bok

[Mustard]

You got an url for downloading them?

[Bok]

See first post,

I updated it with the fastest linux AMD clients I have found.

Or did you mean the 5.x.x linux ones I have compiled ? I'm not using them personally.

Bok

[Mustard]

Originally posted by Bok
See first post,

I updated it with the fastest linux AMD clients I have found.

Or did you mean the 5.x.x linux ones I have compiled ? I'm not using them personally.

Bok

I was after the 4.7x boinc. Thank you!!

[Digital Parasite]

I noticed that the Rosetta home page has 5.2.8 as the recommended version but 5.3.1 optimized has been out for a while. Why wouldn't Rosetta be using the latest, or what is the latest?

Is it worthwhile to upgrade to 5.x clients or should I just stick with my 4.x since they seem to be working ok?

[EvoDude]

Bok,
As you know I'm not the quickest learner on the planet and just wanted to double check I got this right!

I dl'ed the optimized 5.3.1 client for my Athlon 64 from your sticky. I then, just in case, moved the boinc file I already had into a backup folder and dropped your version in where it was and restarted my Boinc manager.

Is that it, almost seems too simple? With just a single 3500+ rig running a gig of ram my RAC score is about 260 - what sort of difference can I expect from this alteration?

Thanks for your time and patience

[Marky-UK]

That's all you have to do!

Look at your benchmark before and after you change the client and you'll see the difference it makes. The truXoft client gets you 10-20% I've found on a P4/Althlon XP/Althlon 64.

If you want a client that gives an even higher benchmark, try these clients: http://www.guntec.de/Crunch3r/boincx86.html

I've found those give benchmarks about 10% higher than the truXoft client.

[EvoDude]

Thanks Marky.

When I tried to install the other files you linked to it tells me that some of the files are in use. I had shut down Boinc as instructed but still says they are in use.

Should I remove Boinc Manager from my start up and reboot, then try again?

[Marky-UK]

Try closing BOINC and BOINC Manager (making sure they're not in the task list anymore), nothing should be using the files then. I run BOINC as a service, so just had to stop the service and could copy the files over fine.

If you still can't copy the files over, maybe you need something like Process Explorer from Sysinternals to find out what process is holding the DLLs open.

[EvoDude]

Gotcha Marky - never thought to check Task Manager for Boinc still running. Sorted now, leave it running undisturbed and see how it goes.

Benchmark now showing:-

2549 MIPS Whetstone
7697 MIPS Dhrystone

Unfortunately I didn't find the benchmark tab before the change so will have to trust this is better than I was getting.

Again, thanks for time and patience!

[Digital Parasite]

I just tried upgrading from BOINC 4.x to BOINC 5.x. I installed the 5.2.13 client from the Rosetta home page for Windows which went fine. I then tried to use the Crunch3r optimized client but I get this error when I try to run boinc.exe:
"This application failed to start because the application configuration is incorrect"

Any idea what is wrong? Using the truxoft optimized client is working fine. I went from 1862.91/3756.6 with the 4.x optimized client to 1560/5849 with the 5.x optimized on. The regular 5.x client was 1884/1403.

[Bok]

Did you copy in all the dll's from the package too ?

Bok

[Digital Parasite]

Originally posted by Bok
Did you copy in all the dll's from the package too ?

Yes I did.

[EvoDude]

I posted some information and a link to these optimised clients on DCZone forum where some of you already post. Now I am being accused of cheating. I take this as a slur on my character and find it rather insulting. So much so that I have decided to stay away from DCZone. I don't need that sort of abuse.

I would be very interested in Free-DC members views on whether the use of optimised clients is actually cheating. My own thoughts are that I am only getting a fair return for a fair days crunching. I could swap my Athlon 64 3500+ for several smaller rigs that would actually do less work but get a lot more points. What do members here think?

Edit:- Mods - you may want to move this post or create a new thread for it, no problem with that.

[ladypcer]

EvoDude, this has been brought up in several forums, and the consensus is that using the optimized Linux version corrects a known bug, but that using the Optimized Windows version only gives you a falsely inflated benchmark and a falsely inflated claimed credit.
Shrug..

I understand they are working on a different way to award credit earned, so the benchmark will soon mean nothing.

[Bok]

My personal feelings are that it is not cheating. If you were to change the code to produce a static high benchmark then yes, that would be, but all I see this doing is optimizing your PC. Somewhat akin to overclocking your pc in a software sense I suppose though that's not a great analogy.

The clients are available to anyone who wants to put in the effort to use them /shrug

Bok

[ladypcer]

I kind of thought that at first too, Bok. Like using flag optimizations in F@H, but it was explained to me that in F@H, running flags makes your cpu crunch better (via SSE2, etc), but in Boinc (windows), all it does is inflate the benchmark and does nothing to really make your cpu run the program better or faster, so you're just claiming a higher bench, but not doing any extra work.

It would be one thing if everyone ran the optimized version, everyone would be on an even footing.

Maybe it would be better if they just gave xxx credit for each finished wu, and forgot the benchmarks altogether.

I know that some programs are asking users NOT to run the optimized Windows version because it skews the credits/results.

I'm still up in the air on it, and waiting for the projects input on the matter.

[EvoDude]

So what you're saying is that it is thought to be unfair that someone with a faster PC running an optimised client shouldn't be allowed to claim extra credit, despite the fact that they are already doing more work.

At DC projects like WCG for instance any PC running 24/7 can achieve 960 points per day regardless of the throughput. If this is the case we would all be better to just buy old slow cheap rigs and do very little work while picking up loads of points. Kinda defeats the aim of getting the work done. Or am I still reading this situation wrong?

[ladypcer]

I think their point is that you are not actually doing more work. You are seeing higher credit due to an inflated benchmark, not due to extra work processed.
At least that was how I understood it.

I'm not judging you or any others, I'm just trying to explain what was told to me, and why others perceive running optimized windows boinc as "cheating".

Oh, I agree with you about slow pc's getting big points because of the time it takes them to run a wu. It doesn't seem right to me either.
Like I said, they just need to give wu's a certain point structure and award points by "finished" work.

[Marky-UK]

An optimised BOINC client can be useful when the project is using optimised software too. SETI@home is probably the most (in)famous in this respect - if you're running an optimised SETI@home client you also need an optimised BOINC client to stop claiming lower credit for a WU than an average PC would.

Other projects are, I think, rumoured to have at least some optimisations in them. I've heard that Einstein@home has some, and have heard talk about some other projects too.

If a project is using optimisations, and the standard BOINC client doesn't, this will cause processors which support the optimisations to claim less credit than processors which don't. An optimised BOINC client levels things up a bit.

Just my 2p anyway...

[EvoDude]

I think their point is that you are not actually doing more work. You are seeing higher credit due to an inflated benchmark, not due to extra work processed.
At least that was how I understood it.

I'm not judging you or any others, I'm just trying to explain what was told to me, and why others perceive running optimized windows boinc as "cheating".

Oh, I agree with you about slow pc's getting big points because of the time it takes them to run a wu. It doesn't seem right to me either.
Like I said, they just need to give wu's a certain point structure and award points by "finished" work.

Thanks ladypcer. You've hit the nail right on the head I think. When projects get round to attaching scores to work throughput then we will all be on a level playing field.

In the end the scores don't matter so much as the level of contribution to the science we each choose to assist. The points just make life interesting while we do it.

Thanks to all contributers here for an informative debate on the subject.

[Paladin*]

Right now it's nothing but a big sham over @ the Rosetta Project with the Credits. I quit running the Project when I started seeing Host's getting over 2000 Credits for 1 3 hour WU & other Hosts averaging between 400-500 Credits per hour of processing ... I figured whats the point ...

[ladypcer]

I decided to run eOn, DPAD and WCG for now, with a little Climate thrown in.
Now all I need is for eOn to quit sleeping, and DPAD to connect to server to send results.

[Angus]

I decided to run eOn, DPAD and WCG for now, with a little Climate thrown in.
Now all I need is for eOn to quit sleeping, and DPAD to connect to server to send results.

There's a manualsend.bat in the DPAD folder that will force the upload if your results.txt file is over 10K in size.

Otherwise - it seems to have a mind of it's own for sending.

[devzero]

I see there are new optimized linux clients available from:

http://calbe.dw70.de/boinclx86.html

Seems to give better benchmarks than before.

[Bok]

I started testing them yesterday and indeed for athlon XP's they seem faster, the SSE2 ones though seem about the same as the equivalent 5.3.6 ones. I'm goign to try the 64bit SSE3 one later today on a Dual xeon with EMT64..

DC opterons don't have SSE3 do they? Fro some reason I thought they did, but




[root@dcopt17001 ~]# cat /proc/cpuinfo
processor : 0
vendor_id : AuthenticAMD
cpu family : 15
model : 35
model name : Dual Core AMD Opteron(tm) Processor 170
stepping : 2
cpu MHz : 2409.649
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 2
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt lm 3dnowext 3dnow pni lahf_lm cmp_legacy
bogomips : 4828.94
TLB size : 1024 4K pages
clflush size : 64
cache_alignment : 64
address sizes : 40 bits physical, 48 bits virtual
power management: ts fid vid ttp


says otherwise...

Bok

[TeeJay]

CPUZ v1.16
x2-4200...
MMX, Extended MMX, 3DNow!, Extended 3DNow!, SSE, SSE2, X86-64

270...
MMX, Extended MMX, 3DNow!, Extended 3DNow!, SSE, SSE2, X86-64

That is what I get here Bok... So, no SSE3 .

>>TeeJay

I started testing them yesterday and indeed for athlon XP's they seem faster, the SSE2 ones though seem about the same as the equivalent 5.3.6 ones. I'm goign to try the 64bit SSE3 one later today on a Dual xeon with EMT64..

DC opterons don't have SSE3 do they? Fro some reason I thought they did, but




[root@dcopt17001 ~]# cat /proc/cpuinfo
processor : 0
vendor_id : AuthenticAMD
cpu family : 15
model : 35
model name : Dual Core AMD Opteron(tm) Processor 170
stepping : 2
cpu MHz : 2409.649
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 2
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt lm 3dnowext 3dnow pni lahf_lm cmp_legacy
bogomips : 4828.94
TLB size : 1024 4K pages
clflush size : 64
cache_alignment : 64
address sizes : 40 bits physical, 48 bits virtual
power management: ts fid vid ttp


says otherwise...

Bok

[Shish]

OK, last op finished ND iM MORE OR LESS sensible for a coupla hours so ...
I tested the new crunch whatever clients and got some very odd, mixxed results that may have been affecceted by UVNCing into home server.
Mosstly about the same as normal or 2.56Ghz Opt 165@280Mhxx3bus of around 10500 but occasional (depending on what eklse running of upto 14500. n ot sure whats happening with them.
Soorry bout the mispelling but i aint got a typist here just now an im flat on muy back with the leg up rigid int ractionand using a modified book reader and pageturner so im poking at keys from below sort of and cant go back or hit some keys on this stpid mobil;e keyboard.
Off for a 3 day sleep soon after a good meal and then back and forth to Aaberdeen for hp oxygen therapy to speed up healing otherwise i d be in a body cast for 6 months and a chair for bout the same..this means i missout on the chair thank god but i`ve got a nice power one on order (coulda built 2 servers for the damn price of it ...sheesh).an thts me knaclered, steaks on the way (got the nurses horrified at steak an chips an salsald at 0530am ;-)...smoiles easier to ype than find off the pagbe somewhere. keep it up guys, cant let XS get it too easy :-).
rgards
shish
mucho gracias Lauren catch U on t flip side.

[Shish]

OK thtas breakfast/lunch/dinner for the next couple days...just found th efile onmy home desktop of comparisons on bonk/rosetta from standard/updated version and crunch3rs clients and also with nothing else hogging cpu (all services stopped inc 4 vms running dimes service) and sorry if file is as bit long....
but i cleaned a lot out i think..
mebbe this what they talkin about?
I always,when I remember, ruin manua l thingy...s
sorry puutiing me out aguin

[cwhyl]

Hi Bok
In /cpuinfo SSE3 is labled by it's codename "pni", short for "prescott new instructions"
(I have the same chip )

[Bok]

Ah,

thanks for that info.

Bok

[Jkusuda]

What is the recommended optimized client I should be using for my 4800x2 under WinXP? A couple of the sites listed in this thread don't work. I'm thinking my benchmarks are too low.

Measured floating point speed 2298.2 million ops/sec
Measured integer speed 4238.48 million ops/sec

Thanks

[(_KoDAk_)]

C2D have
3241 floating point MIPS (Whetstone) per CPU
7158 integer MIPS (Dhrystone) per CPU
)))