Hello, I'm doing a new system for Rosetta and I'm undecided if buying an E8200 45nm dual core or a Q6600 65nm (G0)

Through some tests I've seen that an E8200 overclocked stable to 3.6/3.8 ghz consumes the same energy of a Q6600 at 2 ghz (underclocked stable)

Someone tells me that at the same frequency the performances of a quad core isn't exactly double than the dual's ones. But I don't know if this "theorem" is valid even with Rosetta

So the question is:

what configuration is more powerful on Rosetta concerning both the work performed and the credit's gain?


thanks

you don't get a linear increase in productivity with additional cores as there is additional overhead for memory access etc.

While I don't dispute the similarity of energy requirements for the two setups you've posted, I would suggest that the Q6600 at stock speed would produce a greater throughput than the O/C'd E8200 for a given power consumption, for the very reason that CPUs have begun to become multi-core.

Energy consumption increases linearly with clock-speed (slightly faster than linearly in practice due to the increased temperatures causing increased electrical resistance). Consumption increases with the square of the voltage though, so an underclocked Q6600 (at stock voltage I assume?) will only use slightly more power at 2.4GHz than at 2GHz. Additionally, you can probably either overclock the Q6600 without increasing the voltage, or undervolt it, which has a massive impact on power consumption (you get some additional gains from reduced heat leading to reduced fan speeds and so on).

Basically, while the 2Ghz Q6600 will produce a similar crunching throughput to the OC'd E8200, the Q6600 at stock speed will be more efficient than it is at 2GHz (assuming the same overheads for the rest of the system).

---

It is not so much an increase in overhead, it is contention for a restricted asset. The memory bandwidth is spread over 4 CPUs rather than 2 ...

Fundamentally, though, cache eases this burden to some extent. That is why it does not "scale" or submit to easy analysis. For long running apps that STAY in cache for the most part will use less memory accesses and thus lower the burden.

Again, this is one of the reasons I have long been an advocate of a "mix" of projects because they will TEND to use slightly different aspects of the computer than other projects. That is why HT systems TEND to be more effective if you are running projects with FP and other projects with INT emphasis ... they use different parts of the CPU and can be more effectively "interleaved" with each other ...

if you only run a FP heavy project you will have more contention for the FPU and the INT unit sits idly by ...

Running FP Heavy with INT heavy allows the core to maintain the pretense of 2 CPUs in that the FP heavy process is using the FPU while the INT unit is being used by the other program.

To the extent possible, unless I am FORCING processing for one reason or another, I try to keep the CPUs all on different projects. On a 4 CPU system I will have at least 4 different projects ... on my 8, I will have 8 or more ... and so on ...

---

Ok, now I know many things, thank you!

One last doubt: with a quad core, 2 Gb of memory in dual channel, will be enough or it requests 4 Gb (but only 3 avaliable due to Win XP restriction)?

Ok, now I know many things, thank you!

One last doubt: with a quad core, 2 Gb of memory in dual channel, will be enough or it requests 4 Gb (but only 3 avaliable due to Win XP restriction)?

Rosetta currently uses around 180MB per thread, so you can expect up to 800MB for the four cores with the current client. That leaves you 1.2GB for whatever you do - so it depends what you do on your PC. If it's just email, web, light office apps etc then 2GB will be plenty. If you're doing something memory intensive then i'd recommend a bit more. 4GB is probably overkill, but there isn't much middle-ground - you won't save much by going for 3x1GB or 2GB + 1GB, so maybe just go for 4GB if you've got the cash.

HTH
Danny

---

Ok, now I know many things, thank you!

One last doubt: with a quad core, 2 Gb of memory in dual channel, will be enough or it requests 4 Gb (but only 3 avaliable due to Win XP restriction)?

Rosetta currently uses around 180MB per thread, so you can expect up to 800MB for the four cores with the current client. That leaves you 1.2GB for whatever you do - so it depends what you do on your PC. If it's just email, web, light office apps etc then 2GB will be plenty. If you're doing something memory intensive then i'd recommend a bit more. 4GB is probably overkill, but there isn't much middle-ground - you won't save much by going for 3x1GB or 2GB + 1GB, so maybe just go for 4GB if you've got the cash.

HTH
Danny

If you use 3*1Gb or 2*1G + 1G, you will lose the dual-channel memory function.

One last note: The Q6600 is a great overclocker and the G0 stepping is the best. I have one running at 3.5GHz for 24 hours a day, 7 days a week. I recently had a failure on that unit but it is back in business now and I will continue to drive it this way until it fails.

The Q9450 is also based on the 45nm technology so if you are going for pure energy effeciency, it will run with the E8200.

All of my new builds will be Q6600 and I plan to overclock all of them.

Good Luck and Thanks for Crunching R@H

---

Thx!

Paul

Ok, now I know many things, thank you!

One last doubt: with a quad core, 2 Gb of memory in dual channel, will be enough or it requests 4 Gb (but only 3 avaliable due to Win XP restriction)?

Rosetta currently uses around 180MB per thread, so you can expect up to 800MB for the four cores with the current client. That leaves you 1.2GB for whatever you do - so it depends what you do on your PC. If it's just email, web, light office apps etc then 2GB will be plenty. If you're doing something memory intensive then i'd recommend a bit more. 4GB is probably overkill, but there isn't much middle-ground - you won't save much by going for 3x1GB or 2GB + 1GB, so maybe just go for 4GB if you've got the cash.

HTH
Danny

If you use 3*1Gb or 2*1G + 1G, you will lose the dual-channel memory function.

True, but how much difference does it make? It might have more effect on Rosetta than other apps - it wouldn't suprise me - but Tomshardware showed that with modern CPUs dual channel makes very little difference.

---

Ok, now I know many things, thank you!

One last doubt: with a quad core, 2 Gb of memory in dual channel, will be enough or it requests 4 Gb (but only 3 avaliable due to Win XP restriction)?

Rosetta currently uses around 180MB per thread, so you can expect up to 800MB for the four cores with the current client. That leaves you 1.2GB for whatever you do - so it depends what you do on your PC. If it's just email, web, light office apps etc then 2GB will be plenty. If you're doing something memory intensive then i'd recommend a bit more. 4GB is probably overkill, but there isn't much middle-ground - you won't save much by going for 3x1GB or 2GB + 1GB, so maybe just go for 4GB if you've got the cash.

HTH
Danny

If you use 3*1Gb or 2*1G + 1G, you will lose the dual-channel memory function.

True, but how much difference does it make? It might have more effect on Rosetta than other apps - it wouldn't suprise me - but Tomshardware showed that with modern CPUs dual channel makes very little difference.

You could test your machine. Put two 1Gs in single-channel and benchmark, then put the same two 1Gs in dual-channel and benchmark. If the difference isn't significant to you, then no issue.

I'll add my voice to those recommending a Quad to a Dual. If your goal is to get more science done for Rosetta, that's the way to go.

---

I'll add my voice to those recommending a Quad to a Dual. If your goal is to get more science done for Rosetta, that's the way to go.

A slower Quad (especially on sale) will out do a faster dual, for the same money ...

---

If it will only be a cruncher, a slower quad would crunch more WUs, but if you are going to use the computer for other purposes, you may want a dual-core with a higher clock.

We are like a bunch of old wives ... "on the other hand ..." :)

That is the problem, you have to balance the issues and then pays your money ...

Over my career I built a bunch of systems where I got the lowest cost items and then "replaced" down as I bought new high end gaming systems ... I would take the slowest BOINC machine and put in a new MB and processor occasionally new HD as the drives went from 100, 133, 300 ... SATA ...

---

I have 2 Q6600 systems, the B4 runs at stock 2.4GHz and the G0 at 3.0GHz at stock voltage and runs SuperPi Orthos and other torture tests rock solid. The difference at Rosetta is readily apparent, 2.4GHz B4 v 3.0GHz G0. Both on 3 hour wu timing.

---

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

I have 2 Q6600 systems, the B4 runs at stock 2.4GHz and the G0 at 3.0GHz at stock voltage and runs SuperPi Orthos and other torture tests rock solid. The difference at Rosetta is readily apparent, 2.4GHz B4 v 3.0GHz G0. Both on 3 hour wu timing.

Um, color me blind as I am not sure I can see any difference ... one has a spread of 6K to 10K sec per task, and so does the other ...

---

I have 2 Q6600 systems, the B4 runs at stock 2.4GHz and the G0 at 3.0GHz at stock voltage and runs SuperPi Orthos and other torture tests rock solid. The difference at Rosetta is readily apparent, 2.4GHz B4 v 3.0GHz G0. Both on 3 hour wu timing.

Um, color me blind as I am not sure I can see any difference ... one has a spread of 6K to 10K sec per task, and so does the other ...

Granted credit vs claimed credit is not all that different either between the two.

---

Look at the granted credit compared to the CPU time - as I said, they are both crunching 3 hour wu's. The B4 gets mid 40's for 3 hours, the G0 upper 50's.

---

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

Look at the granted credit compared to the CPU time - as I said, they are both crunching 3 hour wu's. The B4 gets mid 40's for 3 hours, the G0 upper 50's.

Ah, I'm awake now ...

4 decoys vs. 8 decoys in the same time period ...

Like I said, color me dense ... I did not see it ...

Of, course, I have long been of the opinion that the only stupid question is the one you did not ask ... so, when I don't get it, I ask ... :)

Thanks ...

I am still trying to figure out how RaH will fit into my plans for the future ... or how much ... at the moment I am working on some trivia while I look at projects and try to decide which are most deserving ... with so many to chose from it is much harder now ...

Especially with so many new ones in the physics side of the house. But, I still want to push several projects over SaH ... at one time that included RaH, but I just don't know anymore ... WCG seems so much more, well, practical and they are doing multiple sub-projects (I have all but one badge I can earn :) )

Oh well, enough of that ... if only I could afford to up-engine the one generic PC I will be keeping ... Maybe later this summer when some other issues settle down and if I am still doing BOINC I can look to see if Frys is selling a reasonably priced quad core and MB (and RAM) that will beat my AMD 4400 (should not be hard, it is not all that fast) ...

---

so, when I don't get it, I ask ... :)

Absolutely right.

If you want to crunch physics, and want a lot of credit fast, Cosmology@Home offers big bucks for the unit bang.

When it comes to a portfolio, my take has always been lead by what I think has the most to give to the society as a whole. Whilst I do have some quota on some machines crunching physics, chemistry and even mathematics, the vast majority of my cycles go to biological/medical projects.

There is the obvious reason, but also a more considered one. Bio/medico is big bucks business. There are money centric businesses trying to do some of the stuff the projects are doing, not with altruistic reasons, but for patents and marketable compounds.

The more that can be done and published in the public domain, the less patenting and profiteering can be acheived, at the expense of peoples suffering, by the companies. Which by circular routing means that the "obvious" reason for crunching bio/medico science has the opportunity to acheive it's goal by denying the alternative.

If a wonder breakthrough is made in the public domain, then the companies strategy is not based on exploiting a patent monopoly, rather by undercutting their rivals getting the wonder out cheaper - this has to be good for everyone.

I agree a lot of the physics, (etc.), projects offer potential benefits, (even SETI), but they are not facing the same merciless profit motivated resources of big business. I know understanding how the universe began is important, but I can't see a long list of companies furiously working to beat each other into the centre of this particular arena.

---

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

so, when I don't get it, I ask ... :)

Absolutely right.

If you want to crunch physics, and want a lot of credit fast, Cosmology@Home offers big bucks for the unit bang.

I was going to add them soon ...

The problem is that the number of PC only projects is far larger than I can really support with only two PC machines and neither is all that new ... 4-5 years old ... so ...

And LHC still is doing some work so ... at the moment I am trying to get PrimeGrid up and QMC then I was going to switch to Cosmology ... while also doing MilkyWay ...

For Bio, I lean towards WCG and need to assign some resources there ... but for the moment, since I did take the break and am still getting "settled in", and am still not sure if I like the overall environment (present company excluded), and may decide the flack from the spouse is not worth the investment and pleasure when added to the flack from the, ahem, um, well, THEM ...

anyway, for the moment I am playing still with my new Mac Pro and love the speed ... and if I stick around I just thought of a plan to gradually move to a pair of quad core PCs ... though I likely will not be able to do a thing about it until late summer or fall ...

I am still on the fence about RaH, this used to be one of my favorite projects, but, I don't know ... there is just something that feels ... odd now ...

Well, thanks for the explanation, I have drifted off topic too much I suppose ...

---