Can BOINC be setup to run one project on one core and a second project on the second core? Does it run more efficiently if it were to operate like this?

---

My dual core machines run Rosetta on one core and Einstein on the other. However, only those two projects are attached to those computers.

---

Two projects with the same 'Resource Share' seem to run just fine together. We also have run R@H and E@H this way without any loss of productivity...(^^).

---

However, you will not have a guarantee that one processor is always running (say) Einstein and the other (say) Rosetta. I have a couple of dual core setups in this type of configuration, and every now and again they will run 100% Rosetta or 100% Einstein. But only for some short periods of time (maybe a day or so) - in the much longer perspective, you get 50% on each (assuming equal resource share).

--
Mats

---

As Mats says, you will see one project running on both CPUs occaisionally. It just depends on how the scheduling is setup and how the reletive debt works out. Here's a QA with some comments as well.

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

If you run a long-duration project on one core (say, CPDN), set 'no more work' against that project, and pretty much it'll be one project for one core, and the other for the second.

---

Can BOINC be setup to run one project on one core and a second project on the second core? Does it run more efficiently if it were to operate like this?

Remember, they still share the same floating point resources? Anybody?

---

If you run a long-duration project on one core (say, CPDN), set 'no more work' against that project, and pretty much it'll be one project for one core, and the other for the second.

...and be SURE to set "no new tasks" on that long running project. This was over a year ago, and BOINC improves every release, but I did just that, withOUT the no new tasks settings, and BOINC decided it would be a GREAT idea to download a second climate WU to run on the second hyperthreaded CPU... for all of an hour or something and then go back to the normal crunching of two different projects side by side.

Yes, carp, they share some common resources, including floating point CPU time. To address one of the original questions, I've seen others post that running two threads on a hyperthreaded Intel gives roughly 10-15% more throughput then running a single thread. Although I believe this figure was assuming you are crunching Rosetta on both cores. If you crunch a project that is less float intensive then Rosetta on the other core, then you may see greater production from the second core.

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

Can BOINC be setup to run one project on one core and a second project on the second core? Does it run more efficiently if it were to operate like this?

Remember, they still share the same floating point resources? Anybody?

Not on dual core systems (they do on Hyperthreaded systems, but they are not dual core)

---

Team mauisun.org

Can BOINC be setup to run one project on one core and a second project on the second core? Does it run more efficiently if it were to operate like this?

Remember, they still share the same floating point resources? Anybody?

Not on dual core systems (they do on Hyperthreaded systems, but they are not dual core)

Exactly. It's the same as having two completely independent CPU's in their own sockets, except it only takes up one socket.

There is really no sharing going on. Of course, the OS may decide that task0 should run on Core0 at one point, with task 1 running on Core1, and then switch task1 to Core0 and Task0 to Core1, which causes all of the cache content in both cores to be completely useless - but that's hopefully rare enough that it doesn't make a whole lot of difference. Some OS's are better than others at keeping the same CPU/Core for the same task...

--
Mats

---

Can BOINC be setup to run one project on one core and a second project on the second core? Does it run more efficiently if it were to operate like this?

Remember, they still share the same floating point resources? Anybody?

Not on dual core systems (they do on Hyperthreaded systems, but they are not dual core)

Exactly. It's the same as having two completely independent CPU's in their own sockets, except it only takes up one socket.

There is really no sharing going on. Of course, the OS may decide that task0 should run on Core0 at one point, with task 1 running on Core1, and then switch task1 to Core0 and Task0 to Core1, which causes all of the cache content in both cores to be completely useless - but that's hopefully rare enough that it doesn't make a whole lot of difference. Some OS's are better than others at keeping the same CPU/Core for the same task...

--
Mats

That is incorrect! (Only true on Aging unshared K8 architecture)
On Core 2, the cores are sharing the L2 cache. using a L2 cache friendly and a non L2 cache friendly workload can improve your performance.

With a shared L2, you can get up to 4Mb on Core 2 for a thread, if the 2nd one does not requires memory. That improve dramatically most of the memory intensive workload.


who?

Would that only be true for the Core2 Duo and not the Core2 Quad though, since the Quad needs to share the caches (between the individual die-cores) over the FSB/Mem ?

---

Team mauisun.org

Would that only be true for the Core2 Duo and not the Core2 Quad though, since the Quad needs to share the caches (between the individual die-cores) over the FSB/Mem ?

you need to pair the workload a little bit smarter in the case of the quad core, and play with affinity a little more, but the general idea stay the same. Because the L2 share adapt to the "needs" of your algorythms, you can get some pretty nice tricks. for example, you can be running a "linpack" like algorythm without disturbing the other tread working with 3.5MB of L2 cache.
(the algorythm of Linpack is use to take a lot of memory bandwith , but the algorythm was reworked, and it is now very good at data locality, it even gets a 100% success rate on the L2 cache reads, because the prefecters of Core 2 works)

are you aware of any 4x4 grand Father on the TOP list?

who?

@Who?

Not necessarily.

I'm not taking any "sides". I have AMD systems, but I think the Intel Core2Duo/Quad is the "current" champ.

I am not a super-big techie, so please excuse any ignorance. But isn't it true that Intel is following the lead of AMD, in that in 2007 or 2008 Intel will abandon having two dual cores slapped together and instead have "native" quad-cores?

AMD Demos Native Quad-Core Processor

"At Thursday's demonstration, AMD showed off a quad-core processor running four x86 processing cores on a single die of silicon. Intel uses two dual-core processors and connects them so they work together."

"Intel will have a monolithic quad-core, very much like the AMD design, down the road."

"AMD demonstrated a server with a four-socket system powered by four yet-to-be-released Quad-Core AMD Opteron processors (codenamed Barcelona) manufactured on 65-nm silicon-on-insulator process technology. The server used all 16 cores."

"...AMD will get a better performance boost with its quad-core processors than Intel did."

are you aware of any 4x4 grand Father on the TOP list?

who?

I have a tangentially related question to all of this. It looks like I will be upgrading one of my boxes to the new core 2 duo configuration. It will be an E6400 2.13GHz core 2 Duo with 2 MB cache. How will this box compare with a 3.2 GHz single core processor that I now have? By this I mean how much faster will the new box crunch a related work unit. Addtionally, I have often wondered if it were possible to copy and paste an identicle WU and run it on two different machines to compare the different times. Is this even possible?

---

(Click for detailed stats)

By this I mean how much faster will the new box crunch a related work unit. Addtionally, I have often wondered if it were possible to copy and paste an identicle WU and run it on two different machines to compare the different times. Is this even possible?

Each decoy evidently takes different amounts of time - on the same machine; so you need to run 24hours worth of a WU to generate enough decoys to make a decent comparison. Comparing the time to create decoys might not be a fast and accurate way to tell the difference.

Comparing a 2Ghz Athlon 64, a 3+ Ghz P4, and a ~2Ghz Core duo were about the same when dealing with a week's worth of uploaded data. (comparing ~186,000 secs of work between the 3). The Core 2 cpus were supposed to be even better - so your new lower clocked system will be at least 2 times faster than your higher clocked older system.

---

Out of interest i just did a really rough calculation using some results from your 3.2GHz P4 and a 1.83GHz core2duo and after multiplying it up to take account of the 2.14GHz clock speed it came out at 2.7x more productive.

---

Out of interest i just did a really rough calculation using some results from your 3.2GHz P4 and a 1.83GHz core2duo and after multiplying it up to take account of the 2.14GHz clock speed it came out at 2.7x more productive.

Wow! That's (3.2) X (1.83 X 2.14) = 2.7. Sound like that million man math of a few years ago!! You guys who are really good at these computers really know your math!. ;-)

On an even better note, there is a good possibility that the computer that the new box will replace (my old P III 863MHz), will continue to crunch for R@H. I'm sellling it dirt cheap to my neighbor who I'm certain will let me run R@H when he's not using it for his grandkids. I am now a happy cruncher!!

---

(Click for detailed stats)

looks like my maths is better than yours: (3.2) / (1.83 X 2.14) = 2.7 ;-)

But for that calculation to make sense you'd have to double it seeing as there's two cores.

actually it was:

(average of 6 core2duo 1.83GHz results / runtime) / (average of 20 of your results / runtime) / 1.83 * 2.14 * 2

It gave 2.667.

---

@Who?

Not necessarily.

I'm not taking any "sides". I have AMD systems, but I think the Intel Core2Duo/Quad is the "current" champ.

I am not a super-big techie, so please excuse any ignorance. But isn't it true that Intel is following the lead of AMD, in that in 2007 or 2008 Intel will abandon having two dual cores slapped together and instead have "native" quad-cores?

AMD Demos Native Quad-Core Processor

"At Thursday's demonstration, AMD showed off a quad-core processor running four x86 processing cores on a single die of silicon. Intel uses two dual-core processors and connects them so they work together."

"Intel will have a monolithic quad-core, very much like the AMD design, down the road."

"AMD demonstrated a server with a four-socket system powered by four yet-to-be-released Quad-Core AMD Opteron processors (codenamed Barcelona) manufactured on 65-nm silicon-on-insulator process technology. The server used all 16 cores."

"...AMD will get a better performance boost with its quad-core processors than Intel did."

are you aware of any 4x4 grand Father on the TOP list?

who?

I read on internet as well that chicken have teeth, and there is aliens everywhere. it is nice to have 4 processors, it is an other task to produce millions for the billions of users!
Bottom line: for more than 8 months, no Quad core ... just speaking about it! and do you really think intel will sit and look?
They are way a head on process technology, this will make a huge difference.
I noticed that the answer your are posting here was posted in many different forums around the planete ... hummm hummm .... fudding?

who?