mapping cancer markers using boinc. Ideal Pc

[wayland64]

What is the ideal PC for mapping cancer marker and is that process more cpu or gpu dependent?

[ItsTheDuckAgain]

from what I could figure out, since I got interest in it myself, is that it is highly CPU bound and multicore capable. If that is true (not saying that it is though), any machine with shitload of ht-cores should be good enough.

[vanished]

If it's anything like Folding, it can use whatever you have, but a good GPU will net you anywhere from 200k to 500k or 600k PPD, while a typical i7 might get more like 10k, for perspective.

[manikyath]

i dont really have experience, but what i can say is that using a laptop for boinc is a stupid idea, and if you even think that using a laptop wont be a house fire, you need to think again...

 

when i was at the shop to RMA something the dude in line ahead of me had this 4 year old... just beat up laptop.. that's appareantly been running boinc for 4 years, that shit looked painful...

[Imakuni]

6 hours ago, ItsTheDuckAgain said:

from what I could figure out, since I got interest in it myself, is that it is highly CPU bound and multicore capable. If that is true (not saying that it is though), any machine with shitload of ht-cores should be good enough.

6 hours ago, Ryan_Vickers said:

If it's anything like Folding, it can use whatever you have, but a good GPU will net you anywhere from 200k to 500k or 600k PPD, while a typical i7 might get more like 10k, for perspective.

Maping Cancer Markers (which is a World Community Grid subproject) is purely CPU based. It has no GPU app.

 

6 hours ago, manikyath said:

i dont really have experience, but what i can say is that using a laptop for boinc is a stupid idea, and if you even think that using a laptop wont be a house fire, you need to think again...

Not necessarily, newer laptops aren't so bad with crunching. I just got myself one with an i3 5005u and after a couple of hours crunching, it only ever got to 62C. Now granted, it's not that powerful of a CPU (in either number of cores and clocks), and I has no dedicated GPU either.

 

But still, it survived crunching pretty well.

 

 

At any rate, @wayland64, you are looking at high CPU core counts. And supposedly, HT should help as well (though it's impossible to say for sure). As a starting point, I'd upgrade any PC you might have with better CPUs; say, jumping from a Pentium or i3 to an i5/7. If you're already maxed out, then we can think about getting something new.

[manikyath]

2 minutes ago, Imakuni said:

Not necessarily, newer laptops aren't so bad with crunching. I just got myself one with an i3 5005u and after a couple of hours crunching, it only ever got to 62C. Now granted, it's not that powerful of a CPU (in either number of cores and clocks), and I has no dedicated GPU either.

it's appareantly usually the charger that goes belly up first

[Imakuni]

1 minute ago, manikyath said:

it's appareantly usually the charger that goes belly up first

Yeah... the poor thing got hot. Like, REALLY hot. It's probably got 80+ Smoke certification, as far as efficiency goes...

[manikyath]

1 minute ago, Imakuni said:

Yeah... the poor thing got hot. Like, REALLY hot. It's probably got 80+ Smoke certification, as far as efficiency goes...

those power bricks are actually usually pretty good, but no matter how good they are, a fanless design is just limited.

[Ithanul]

Yep, as already stated, WCG at the moment don't support GPUs.  There are a few other BOINC projects that do.

 

In this case, many cores and decent speeds are the way to go, especially 2P/4P setups.  If you want to keep wattage down, the Xeons are a good way to go.  Used E5-2670s tend to float about for a low price since many business off load those when they are upgrading their equipment.  Of course, there are many CPUs floating around to choose from.

[vanished]

4 hours ago, Imakuni said:

Maping Cancer Markers (which is a World Community Grid subproject) is purely CPU based. It has no GPU app.

Then it's hard to imagine they're accomplishing much.  It would be bound by the same factors that influence folding (that is, the same rough ratio of CPU to GPU performance, whether they offer GPU or not), and a lot of folders disable CPU all together since it's such an insignificant boost to the CPU  

[Imakuni]

16 minutes ago, Ryan_Vickers said:

Then it's hard to imagine they're accomplishing much.  It would be bound by the same factors that influence folding (that is, the same rough ratio of CPU to GPU performance, whether they offer GPU or not), and a lot of folders disable CPU all together since it's such an insignificant boost to the CPU  

Folding deals with protein and how they fold; MCM, on the other hand, tries to find potential genes that can cause cancer by matching data from thousands of people.

 

I'm sure that, by this description alone, you can imagine how vastly different the types of workloads for each project are. In F@H's case, GPUs seem to be adequate; in MCM, I suppose the devs decided that GPUs were either unsuited / didn't provide any bonus over running purely on CPU, so they decided to go full CPU mode instead.

 

 

Also, it's very rare to find someone on Boinc that uses the GPU, but not the CPU. It's REALLY rare...

[vanished]

1 minute ago, Imakuni said:

Folding deals with protein and how they fold; MCM, on the other hand, tries to find potential genes that can cause cancer by matching data from thousands of people.

 

I'm sure that, by this description alone, you can imagine how vastly different the types of workloads for each project are. In F@H's case, GPUs seem to be adequate; in MCM, I suppose the devs decided that GPUs were either unsuited / didn't provide any bonus over running purely on CPU, so they decided to go full CPU mode instead.

 

 

Also, it's very rare to find someone on Boinc that uses the GPU, but not the CPU. It's REALLY rare...

That is quite different, but surly you see how it's still an extremely parallalizable task - ie, one that benefits from a great many of cores, even if the cores are weaker.  In fact, its own usage of CPUs seems to back this up, as it prefers CPUs with many cores, and even HT apparently helps a lot, as you mentioned yourself.

[Imakuni]

21 minutes ago, Ryan_Vickers said:

That is quite different, but surly you see how it's still an extremely parallalizable task - ie, one that benefits from a great many of cores, even if the cores are weaker.

Well, all of those distributed computing projects are extremely good for parallelism, otherwise they wouldn't be run on people's computers. You have a bunch of tests that can be done independently from each other. However, each specific test might have to be run sequentially, which would make GPUs unsuitable for the task; I assume this would be the case here.

 

But I wouldn't know about that, you'd have to ask the researchers / scavenge the MC forum for an answer.

 

21 minutes ago, Ryan_Vickers said:

In fact, its own usage of CPUs seems to back this up, as it prefers CPUs with many cores, and even HT apparently helps a lot, as you mentioned yourself.

Don't misunderstand "prefers CPUs with many cores to run a single test" to "prefers CPUs with many cores to run multiple different tests at once". Those are very different. The thing I pointed out about upgrading the CPU was more in the sense of increasing core count (to run more tests at once) without having to invest into a new PSU, RAM, Case, etc. that you'd need for building a new computer. It's also easier to just buy 1 component.

 

Oh, and about HT, it may or may not help, we don't really know. At the very least, HT doubles your thread count, which in turn doubles run times per physical hours run and makes you get badges twice as fast, so everyone just leaves it on and forgets about it. I'm pretty sure no one did any sort of test to see if it actually does anything (either for better or for worse), for it's pretty hard to do such thing and people would ignore results and just run with HT on anyways.

[wayland64]

5 hours ago, Imakuni said:

Maping Cancer Markers (which is a World Community Grid subproject) is purely CPU based. It has no GPU app.

 

Not necessarily, newer laptops aren't so bad with crunching. I just got myself one with an i3 5005u and after a couple of hours crunching, it only ever got to 62C. Now granted, it's not that powerful of a CPU (in either number of cores and clocks), and I has no dedicated GPU either.

 

But still, it survived crunching pretty well.

 

 

At any rate, @wayland64, you are looking at high CPU core counts. And supposedly, HT should help as well (though it's impossible to say for sure). As a starting point, I'd upgrade any PC you might have with better CPUs; say, jumping from a Pentium or i3 to an i5/7. If you're already maxed out, then we can think about getting something new.

ok will do thanks

[wayland64]

5 hours ago, Imakuni said:

Maping Cancer Markers (which is a World Community Grid subproject) is purely CPU based. It has no GPU app.

 

Not necessarily, newer laptops aren't so bad with crunching. I just got myself one with an i3 5005u and after a couple of hours crunching, it only ever got to 62C. Now granted, it's not that powerful of a CPU (in either number of cores and clocks), and I has no dedicated GPU either.

 

But still, it survived crunching pretty well.

 

 

At any rate, @wayland64, you are looking at high CPU core counts. And supposedly, HT should help as well (though it's impossible to say for sure). As a starting point, I'd upgrade any PC you might have with better CPUs; say, jumping from a Pentium or i3 to an i5/7. If you're already maxed out, then we can think about getting something new.

So a bunch of old servers would be ideal?

[Imakuni]

1 minute ago, wayland64 said:

So a bunch of old servers would be ideal?

Kinda. While old can get you lots of cores for cheap, they consume quite a bit of power, while not getting that much actual crunching done. A good suggestion would be to hunt for some dirty cheap FX 8xxx CPUs on the used market, they should provide a nice balance of cost, processing power and thread count. For servers (and Intel CPUs in general), anything from the Sandy / Ivy bridge era and above that has HT (again, that badge thingie) is also pretty darn good, whether it be consumer hardware or Xeon stuff.

[wayland64]

2 hours ago, Imakuni said:

Kinda. While old can get you lots of cores for cheap, they consume quite a bit of power, while not getting that much actual crunching done. A good suggestion would be to hunt for some dirty cheap FX 8xxx CPUs on the used market, they should provide a nice balance of cost, processing power and thread count. For servers (and Intel CPUs in general), anything from the Sandy / Ivy bridge era and above that has HT (again, that badge thingie) is also pretty darn good, whether it be consumer hardware or Xeon stuff.

Is there a website that shows benchmarks for which CPUs complete boinc projects the best?

[Imakuni]

45 minutes ago, wayland64 said:

Is there a website that shows benchmarks for which CPUs complete boinc projects the best?

BOINC is a platform, not a project; unlike folding, you can't have a generic benchmark tool. If you want to know which CPUs are nice for a given project (and subproject), you'd have to ask for that specific project, either here or (preferably) on their own forum.

[wayland64]

26 minutes ago, Imakuni said:

BOINC is a platform, not a project; unlike folding, you can't have a generic benchmark tool. If you want to know which CPUs are nice for a given project (and subproject), you'd have to ask for that specific project, either here or (preferably) on their own forum.

OK thanks

[Ithanul]

7 hours ago, Ryan_Vickers said:

That is quite different, but surly you see how it's still an extremely parallalizable task - ie, one that benefits from a great many of cores, even if the cores are weaker.  In fact, its own usage of CPUs seems to back this up, as it prefers CPUs with many cores, and even HT apparently helps a lot, as you mentioned yourself.

The difference in F@H and BOINC projects WUs is F@H runs one WU across all available threads and BOINC projects run a WU per thread.  This the reason 2P/4P systems are preferred in BOINC.  Though, back in the early days of F@H and bigadv, 2P/4P rigs were king as well (back when a GTX580 with its 17-20K PPD was considered nice).

 

Here is a good article from the Team USA BOINC team on the matter:  http://usa.lanex.com/?page_id=110

[Bleedingyamato]

On February 17, 2017 at 8:34 PM, Ithanul said:

The difference in F@H and BOINC projects WUs is F@H runs one WU across all available threads and BOINC projects run a WU per thread.  This the reason 2P/4P systems are preferred in BOINC.  Though, back in the early days of F@H and bigadv, 2P/4P rigs were king as well (back when a GTX580 with its 17-20K PPD was considered nice).

 

Here is a good article from the Team USA BOINC team on the matter:  http://usa.lanex.com/?page_id=110

You mean 2 or 4 CPU computers?

 

That article was saying a lot of BOINC projects can use a GPU rather well it seemsed.  So I'm confused, for BOINC are your CPU and GPU both useful or just one?

[Ithanul]

10 minutes ago, Bleedingyamato said:

You mean 2 or 4 CPU computers?

 

That article was saying a lot of BOINC projects can use a GPU rather well it seemsed.  So I'm confused, for BOINC are your CPU and GPU both useful or just one?

Varies by project.  There is some that use GPU but not all.  And, not all projects support all types of GPUs.

 

The few projects that I know that can use GPU are Einstein@Home, Milkyway@Home (DP required), GPUGrid, Collatz Conjecture.  There was Peom@Home, but they finished up and shut down the project.

 

2P/4P - two processor / four processor.

[Bleedingyamato]

5 minutes ago, Ithanul said:

Varies by project.  There is some that use GPU but not all.  And, not all projects support all types of GPUs.

 

2P/4P - two processor / four processor.

Hmmm..  That could be an issue for me.  I was thinking I could use my secondary desktop for either folding or BOINC but it's got a Pentium G3258 and a 4GB GTX 960 in it.

[Ithanul]

3 minutes ago, Bleedingyamato said:

Hmmm..  That could be an issue for me.  I was thinking I could use my secondary desktop for either folding or BOINC but it's got a Pentium G3258 and a 4GB GTX 960 in it.

Nah, you can use a single CPU.  It just 2P/4P systems are great systems to do BOINC (the butt ton of threads).

 

Heck, I still see peeps along with me rocking 1090Ts in BOINC projects.  I even ran my lappy's i3 during the Pent.  What is nice about BOINC projects, most don't lose points when you pause WUs.

[Bleedingyamato]

1 minute ago, Ithanul said:

Nah, you can use a single CPU.  It just 2P/4P systems are great systems to do BOINC (the butt ton of threads).

Is it ok if I pm you about F@H/BOINC if I decide to get into it?

 

I was thinking it could be a nice use for my secondary desktop when I'm not using it.  Most of the time it's just sitting turned off.