Thinking of upgrading a cruncher. Best @ 3930k priceline ?

[Markfw]

On a four core, eight with HT, can Linux in a VM on Windows make better PPD than the Windows client? By how much?

I think you need a 64 bit OS and plenty of memory for that, and I don't have any other 64 bit machines to test that on. Maybe someone else here can answer.

 

[Mr. Pedantic]

On a four core, eight with HT, can Linux in a VM on Windows make better PPD than the Windows client? By how much?

Not much. Intel's 4 cores + hyperthreading no longer counts for bigadv, and it's only with bigadv that running Linux has a benefit - running normal SMP it's better to go with native Windows.

Technically, while 8-core systems still count for bigadv, because Stanford is not really assigning WUs for it any longer in the process of phasing it out completely, if you do run the -bigadv flag, you'll only get SMP units for the most part. They're actually also doing the same for 12-core bigadv; in the future the minimum requirement for bigadv will be a 16-core system, preferably 16 native cores - according to some in FF who have tried it out, 8 Intel cores + HT requires at least 2.7GHz to make the deadline for some of the projects.

This is also why I suggested Mark get dual 6128s with a dual-G34 motherboard (I know I was late, though) - the price would not be that much more than for a single 3930K, and it would last in bigadv much longer and much better.

 

[blckgrffn]

Looks like you can expect ~40-45k ppd on your 3930k even with straight up SMP (-smp 10):

http://www.hardwarecanucks.com/foru...-hwc-folding-home-project-ppd-database-4.html

That 2600k is horrid though, are you running -smp 6 on that guy? 4k is what I got with two cores of a Phenom 2! I would expect to see it at ~25k-35k depending on the WU.

 

[VirtualLarry]

Not much. Intel's 4 cores + hyperthreading no longer counts for bigadv, and it's only with bigadv that running Linux has a benefit - running normal SMP it's better to go with native Windows.

Technically, while 8-core systems still count for bigadv, because Stanford is not really assigning WUs for it any longer in the process of phasing it out completely, if you do run the -bigadv flag, you'll only get SMP units for the most part. They're actually also doing the same for 12-core bigadv; in the future the minimum requirement for bigadv will be a 16-core system, preferably 16 native cores - according to some in FF who have tried it out, 8 Intel cores + HT requires at least 2.7GHz to make the deadline for some of the projects.

This is also why I suggested Mark get dual 6128s with a dual-G34 motherboard (I know I was late, though) - the price would not be that much more than for a single 3930K, and it would last in bigadv much longer and much better.

So much for the notion of doing DC with just "spare cycles" on an ordinary computer. Now you have to splash out for a dual 8-core SMP board, to make any headway in the standings, huh?

 

[Mr. Pedantic]

So much for the notion of doing DC with just "spare cycles" on an ordinary computer. Now you have to splash out for a dual 8-core SMP board, to make any headway in the standings, huh?

Yeah, a lot of people agree that the bigadv points system is unfairly balanced. But in terms of restricting bigadv to higher hardware requirements, that was the original intention. Bigadv was never meant for the likes of highly clocked Core i7 systems, or even really Skulltrail. It was for people who had access to dual or quad-CPU servers or HPC clusters, who would be given the priority for these units and would be given a bonus for their relatively rare hardware. Now, though, the whole thing has become a lesson in unintended consequences of incentives, with people building quad-G34 systems just for folding, because 48-core quad-Magny Cours machines provide so much more ppd/W and even ppd/$ than other systems can.

Of course, the bulk of folding is still contributed, I would say, by people with single-socket systems and GPUs. It's just that now, to be among the top tier of folders, or to significantly advance your team by yourself, you need an inordinate amount of hardware. But that's the way it's always been. I remember when uniprocessor was king, reading stories of admins borging hundreds of systems under their control to contribute points.

By the way, is anyone having problems seeing the last page on this thread?

 

[Markfw]

Jeesh !!! My second unit is a 6904, and its worth over 412,499, and I am getting 111k ppd !!! This cpu is awesome !

 

[biodoc]

Jeesh !!! My second unit is a 6904, and its worth over 412,499, and I am getting 111k ppd !!! This cpu is awesome !

Wow, that's awesome! :thumbsup:

 

[blckgrffn]

Wow, that's awesome! :thumbsup:

YES.

'grats!

And I am waiting for PD before pricing out my Quad G34. It may be a pipe dream, but if it is a) good enough or b)enough people drop their big Interlagos CPUs for cheap then it may become a reality

 

[Diogenes2]

Jeesh !!! My second unit is a 6904, and its worth over 412,499, and I am getting 111k ppd !!! This cpu is awesome !

Can you help me understand what switches I need if I'm running the Linux Virtual box with 3930K

Do I need just -smp or -smp 12 ( or what? )
Do use -bigadv or -advmethods or -big beta or?

Thks in advance...

 

[Markfw]

Can you help me understand what switches I need if I'm running the Linux Virtual box with 3930K

Do I need just -smp or -smp 12 ( or what? )
Do use -bigadv or -advmethods or -big beta or?

Thks in advance...

pretty sure I just did ./fah6 -smp -bigadv -verbosity 9

But I have other settings in the config, like answered "big" to one question.

 

[Diogenes2]

OK... That sounds good..

Thks for the feedback.. It looks like -smp -bigadv pulls in the big work units..
Have a 6903 projecting 115k PPD..