Kickstarter Project Promotes Distributed Computing, BOINC, the Advancement of Science

[deanshea]

We have launched a Kickstarter project for the Advancement of Science. The project promotes distributed computing and the BOINC client. Please check out the link below, participate in BOINC and pledge generously.

http://kck.st/11CrcbM

THANK YOU.

The Sheaffer Family.

 

[ZipSpeed]

While I don't doubt the intentions of wanting to further the advancement of science, how do I guarantee that funds will actually go towards these computers? An honor system? For all I know, these monies will go towards someone's desire for a brand new gaming rig.

My questions aren't intentionally meant to be abrasive. I'm just curious, that's all.

 

[deanshea]

ZipSpeed,

Thank you for your response.

Please see the comments and FAQs on the Kickstarter site.

http://kck.st/11CrcbM

They directly answer your question. One answer is that the updates provided to all donors will include links to various BOINC stats pages where the work units processed by these PCs will be publicly available.

The Sheaffer Family

 

[Ken g6]

Honestly, I think that if the TeAm wants to fund something like this, we should fund our own crack-rack, like back in the olden days.

Meanwhile...Ooh, look, a parts list I can criticize! :twisted:

Case: I'm no case expert, but that's a lot of money for a case. You don't really need a case for a DC "crack-rack". I recall somebody putting one together with several threaded rods, with nuts holding each mobo in place. (Not counting the nut who put it together. )

-$200 + say $50 for parts per PC.
Net: -$150

Mobo: So that's what you're putting four GPUs in. At this price range, you ought to consider an LGA-2011 setup:

- ASRock Fatal1ty X79 Professional
- i7-3930k

50% more cores for less than 25% more $.

Net: +$165

- GPUs: I wouldn't get 680s, when GTX 670s can be had for $360 or less. I'd also suggest you consider getting some AMD cards - 7970s are in a similar price range and can be much better for some projects. (And worse at others.)

Each: -$120
Net: -$480

- PSU: High power, high price, but bad brand. I like the looks of this FSP. 1000W should be sufficient.

Net: -$80

- CPU: See above.

- RAM: You don't want 1.65V RAM - I gather it can burn out the memory controller on the CPU. Fast RAM isn't bad if you can get it, though the quad-channel LGA-2011 chipset alleviates this somewhat. Here is a good 1866 kit for under $200.

Net: -$155

- Hard drive: I wouldn't bother with an SSD. Just get a good standard HDD. Its performance isn't that relevant to DC.

Net: -$140

- Keyboard, monitor: Just get one set. You can connect them to the different machines occasionally, but generally you should just remote desktop or VNC to the other machines when you need to.

Net: -$200 per PC

- Blu-ray burner: You don't need a blu-ray burner for these systems. One DVD drive can work for installing Windows on each of them, then you're done.

Net: -$76 per PC

- Windows: All versions of Windows 8 can access 128GB RAM. So can Win7Pro. You don't need the high-end versions.

Net: -$80

- CPU cooler: Just get an air cooler that fits. If you do the crack-rack thing, you can set the spacing so it fits. A ~$30 Hyper 212 Evo seems to be the best bang-for-the-buck.

Net: -$90

- UPS: You don't need a UPS for these systems. Just make sure they have surge protectors, that you set them up to boot on power received, and that they are set up to resume crunching on boot.

Net: -$420

Total net per PC: $4900 - 1700 = $3200

Now you can afford to build a crack-rack with seven nodes instead of five!

 

[h2co32]

I apologize for the bluntness, but it sounds as if you just want to have your own super computer node or beauwolf cluster, and want others to pay for it. Kick starter is usually for projects that give back to its backers, you having computers to increase your score in DC projects, for the advancement of science or not, is not giving back. In fact, it seems almost like a selfish goal.

A few tips, include more info about you on the page, maybe start a non profit that focuses on donating its own computing resources and advocating for the donation of computational power to DC projects. Otherwise this just seems like a scam to satisfy your own ego.

Again, I apologize for the bluntness. But there are several here that are slowly building their own clusters with our own hard earned money.

 

[StitchExperimen]

These people clearly have no idea of what parts to order. The build is in excess. The operating system can be had for free Linux cinna mint with either Nvidia or AMD drivers. the memory they want is 32 gigs and you only need 8. They don't know how to buy a DVD they are paying full retail $79.99. The motherboard is outrageously expensively $388.99. The ASRock OC Formula is $239. And I'm throwing in $80 for Cougar long life 300,000 hour fans. They ordered the wrong Corsair cooler they ordered a H100 and the better newer model model is the H100i. The cost for memory their spending is $350. The SSD doesn't need to be 240GB you can use a 120GB. The backup ups is good for 8 minutes at half load. Warranty is good for only 2 years better to go with UPS and a larger unit that does boost and buck on the input voltage and has the machines shut down after 1 minute of power off time and restart on power resume after X minutes. These larger UPS units use 2 large batteries and you don't have to replace a lot of mini's all at once that can't provide enough power. The big batteries last longer. Also with bigger UPS they can start a generator that runs from natural gas. gab gab gab

I don't know if you can crunch 4 Nvidia cards on a i7 3770K

You can run 4-way CrossFireX most efficiently on a LGA 2011 with 4 channel memory
But with a 8 core Xeon with 16 threads with two 7950s 14+14 for 28 hcc1 it runs fine but with three 7950s sending 8+8+8 for 24 I would get "Computation Errors"
Now the 8 core ran at 3.2GHz and any greater multi cores ran much slower.
So you can't crunch three 7950s even with a 8/16 core processor

A i7 3930@4.2 with 2 7950s and a i7 3770K@4.6or 4.7 with two 7950s are an equal match except one is $333 cheaper. (In PPD)

 

[deanshea]

Thank you for the replies -(blunt or not)

With respect to the hardware suggestions, please see the rather lengthy comment on the kickstarter site. It explains that this is a "typical" list of hardware and explains a few of the different ideas we have for using different components in the optimization; running the Unix OS head to head with WinXP and Win7; and even why we have included a UPS.

http://kck.st/11CrcbM

With respect to the "blunt" reply, we plan on investing hundreds of hours in determining the optimal configuration for BOINC. The project's main goal (beyond running the BOINC client to crunch roughly a gazillion WUs) is to "blue print" and publish the optimized platform as an "open source" document so that anyone can replicate the work we have done and extend our work.

At the end of the project we promise to use the hardware (or value thereof) for the further advancement of science. If this is seen by other's as "selfish", we would love to hear your ideas on how to improve the project, make it more transparent or better benefit the community and science.

We take all of your comments, suggestions and ideas seriously and truly welcome your input.

The Sheaffer Family.

 

[h2co32]

Wuprop@home already has all the data you need. It's a free open source program utilizing boinc to gather its data. The point of DC is to use the communities computing power, so if you want 37k to gather information that was gathered for free, it seems a bit selfish and much like a scam.

And like I said, make a non profit to legitimize your goals in the publics eyes, otherwise it lacks the security that others may expect or want. I get that you want to help science by building computers for compute power. But using other people's money to build what would ultimately be your personal compute cluster is a bit misguided. And it sidesteps the point of DC.

 

[deanshea]

H2co32,

With great respect, we simply disagree with you.

Wuprop@home lists general, and very broad BOINC performance stats for individual components. We know of no source that documents the optimization of a complete computing platform for the BOINC client (by platform we mean hardware, BIOS, operating system, BOINC client software, etc).

Further we disagree that this sidesteps the point of DC. The point of DC is to use the a broad array of computers to process large data sets that cannot be economically processed by other means. The focus of this project is to develop a blue print of an optimized platform. By "optimized" we mean "lowest cost / performance". This is exactly the point of DC.

Finally, yes we could start a non-profit. From personal experience, I assure you this would involve spending thousand of dollars on legal and accounting fees, creating a board of directors and using as substantial percentage of the funds raised to cover overhead and administrative costs.

Instead, we are going to publish a public accounting of how every single cent of this project is spent -- and we know that in doing so we will be criticized in terms of "why didn't you buy x instead of y". We know our donors and the public will be watching what we do.

The work that is done by the platforms will be completely public (as is the case for all PCs running the BOINC client). So donors and the public can see how their "investment" is performing whenever they want - not just when a quarterly or annual report is published as would the the case in a non-profit. There is no overhead, legal, administrative or accounting costs in our plan.

If this is "misguided" or "selfish" we fail to see how.

The Sheaffer Family.

 

[Rudy Toody]

I don't think this is a very good idea.

 

[deanshea]

Rudy,

Thank you for your reply.

We are always looking for ways to improve the project. Could you help us understand why you don't think this is a very good idea? We welcome everyone's thoughts.

We truly hope to get input from as many perspectives as possible prior to beginning the builds. Other people's perspectives help us refine our thinking and build consensus.

We will incorporate your thinking (and that of any one else) and/or explain why we take the position we do. We believe in having open, public and constructive dialogue and debate in order to best fulfill the objective of the project.

The Sheaffer Family

 

[VirtualLarry]

I agree with Rudy, this just doesn't seem like a good project for kickstarter. Instead of "Won't you do it for the children", it has become, "Won't you do it for science?".

I agree with the comments to start a non-profit DC research firm. Then you can run donation drives regularly, and keep your fleet upgraded.

 

[deanshea]

Thanks VirtualLarry,

We appreciate and respect your perspective. If the Kickstarter project fails to fund, we will certainly look at other options.

Are there folks on this forum that would be interested in helping to create a not-for-profit entity to fund and operate a project like this?

The Sheaffer Family.

 

[Rudy Toody]

http://www.rechenkraft.net/wiki/index.php?title=Rechenkraft:English_visitors

 

[deanshea]

Per the link to rechenkraft.net:

"use our total annual budget of 2833 €."

Which equates to about $3,700 per year in available funding. It would take 9 years at this funding rate to have sufficient assets to launch the project -- assuming zero cost for legal, accounting, administration, fundraising and overhead.

 

[Vertaxis]

I've been reviewing your parts list for a node and have been scratching my head on this. There are ways to save on the per node cost that you haven't explored.

1. Buy an 8-port KVM so that you don't need to buy 5 monitors, 5 keyboards, and 5 mice.
2. Go headless. These are compute nodes, so they don't really need any form of keyboard, monitor, or mouse.
3. Buy Win7 Pro and save the $50.00 per node licensing. You don't need all the bells and whistles that Ultimate offers for compute nodes.
4. If you have the skills, install Linux instead of Windows. Linux distros are free to download, and you don't have to pay Microsoft for a license for the OS. Also, Linux uses fewer of the machine's resources so the Boinc projects will have more resources available on the same hardware.
5. Don't install an optical drive in a compute node. They'll only idle an suck power for nothing. Likely, an optical drive will be used for the installation of the OS, drivers, and AV. Once the system is built, you will never use it again in that system. Instead, put 1 DVD drive in an external 5.25" USB enclosure and boot the installation media off that.
6. Don't use SSDs. Compute nodes for Boinc don't need high performance drives. To save on power, get a 2.5" notebook drive like the Western Digital Scorpio Black WD2500BEKT 250GB 7200 RPM http://www.newegg.com/Product/Produc...82E16822136279 $65.00 each vs $219.00 each. Otherwise, use a WD Blue or Black 7200 rpm 3.5" drive.

Just a few ideas to stretch the money.

Vertaxis

 

[Rudy Toody]

Per the link to rechenkraft.net:

"use our total annual budget of 2833 €."

Which equates to about $3,700 per year in available funding. It would take 9 years at this funding rate to have sufficient assets to launch the project -- assuming zero cost for legal, accounting, administration, fundraising and overhead.

They only serve up projects---yoyo@home and rna@home. They support science by providing a place for scientists to run projects.

 

[deanshea]

Vertaxis,

Thank you for the reply.

I am going to cut and paste the rather lengthy comment on the Kickstarter site to help explain. Please let me know if you have further questions, ideas or comments.

++++++++++++++++++++++++++++++++++++++++++++++++

Jonathan,

Thank you for the comment.

Below is an except of an email response from me to a researcher in the Netherlands who is working on a similar project. I hope this answers some of your questions.

===========================================

Thank you for your kind words of advice and for sharing your experiences in such a similar project to our Kickstarter project. The guidance you have provided will advance our project significantly; we are truly thankful.

The below is intended to further describe the intent of our project. We do not go into this level of detail in the Kickstarter description because we believe that it will simply be too technical for most potential donors.

The schedule of "typical" components we have provided in the Newegg wish list is intended only to be an example of one possible solution to be used in determining the optimal configuration of hardware, OS, BIOS and BOINC client software. It is fully our intent to build five very different platforms in order to help determine the optimal configuration. Actually, we will likely "swap" components between systems in our attempt to determine the "optimal" configuration in terms of cost/performance. The optimized configuration will eventually be "blue printed" and provided as an "open source" document.

One specific example is that we intend to build comparable platforms based on X79 and Z77 motherboards to help determine if available PCI-E bandwidth creates a material bottleneck in this particular application -- and how such a choice may limit processor selection (e.g currently there is only a very limited selection of LGA 2011 processors) and the corollary effect of this limited processor selection on cost/performance.

We will also calculate the cost/performance benefit of running fewer GPUs across a larger number of PCs (as you suggest) and are considering testing a dual processor motherboard. Both Intel and AMD processors will be evaluated.
Further, it is our intent to test Windows vs. Unix operating systems. Oddly, most current "top producing PCs" for the BOINC client run either the Win7 or WinXP OS. We are unsure why this should be so and hope to determine the reason through this "head to head" testing.

Similarly, we intend to test 8GB, 16GB, 32GB RAM configurations (with various "speeds" and latencies); SSD vs. various hard drive configurations; and different GPUs. For example we may run two GTX 690s against four GTX680s against four GTX580s and will certainly compare nVidia vs.AMD/ATI performance.

We are indeed concerned about power requirements and heat dissipation. Our intent is attempt this project from our home so that we may have more time and freedom to modify the systems on a constant basis -- while keeping our day jobs .

....

I certainly agree that the liquid cooling option is not required under normal operating conditions, however we intend to determine if overclocking the CPUs has a material advantage and our experience leads us to believe that enhanced CPU cooling can lead to greater stability / lower temps at high processor speeds/voltages. Also, LGA 2011 processors do not come with a "stock" cooling fan.

With regard to the other components (monitor, mice, optical drive, etc), we do understand that these are not absolutely necessary for the builds, but they do provide convenience (and maybe efficiency) as we run and rebuild these systems many times over the course of the year.

One thing that we have tried to make clear is that the $25,000 for hardware is a budget; our actual cost may be more or less. If it is less, the "extra" funds will be used to operate the project for a longer period of time; if the cost is more, we will meet our one year operating commitment, at our expense.

I would very much like to hear your thoughts on this more detailed description of the intent of the project and our overall plan. I welcome any further advice you are willing to offer.

Best Regards,

The Sheaffer Family
====================================================
To answer some of your additional questions:

The PCs will run nothing other than the BOINC client. with the possible exception of running benchmarking programs from time to time to assist in the optimization.

After the project is complete, we determine the disposition of the hardware. If funds allow, we may continue to run them indefinitely. Alternatively we may sell the components to build new platforms based on newer technology. Or we may donate them to other scientific pursuits. In any case the hardware (or value thereof) will be used exclusively for the advancement of science...we are simply unsure of what the best use may be a year and a half from now.

Our intent is to run the platforms from our home; however we do have real concerns about power consumption and heat dissipation. The advantage to running from our home is easy access and additional time for optimization. The downside is total power draw and power fluctuations inherent in our residential grid. - hence the UPS in the sample schedule of hardware.

However, I have access to a regional computing center that hosts mainframes and server farms and am certain that I can install the platforms in this facility for little or no cost beyond power consumption fees. So we do have a "Plan B". The downside is that that we will have to go to this facility to make any mods to the hardware, so optimization will clearly take longer.

I think the rest of your questions were answered in the cut and paste above, but if not, please just ask. We are happy to respond.

Thank you again for your interest in the project, for being a BOINC participant, and for your pledge.

The Sheaffer Family

++++++++++++++++++++++++++++++++++++++++++

 

[Vertaxis]

Will you be joining projects that require Single Precision or Double Precision computing? This will greatly affect your choice in GPUs.

 

[deanshea]

Indeed. Some projects also do not support GPU processing at all or only on specific architecture. Some support specific operating systems. Optimization will be based on running a variety of projects and reporting cost/performance on each "flavor" of project.

 

[deanshea]

They only serve up projects---yoyo@home and rna@home. They support science by providing a place for scientists to run projects.

Rudy,

Yes, I agree that they provide an exemplary service.

However, I'm not following the connection to our Kickstarter project. They serve up projects; our intent is to blueprint an optimal platform for processing project data.

 

[petrusbroder]

Well, I have a few times considered to construct a BOINC-crunching "super computer".

When my farm was most productive (more than 1 M points/day) I had a total for 112 CPU-cores (CPUs with 2 and 4 and 8 cores/threads) and 11 GPUs - and that was 2 years ago. At that time all the computers were "out of date" machines, and I got them for free (because some parts were broken, and I repaired them) or for cheap.
The lessons I learned were:
1. get a high quality PSU with sufficient power.
2. a small HDD with just average performance is sufficient. If the computer boots in 15 seconds or in 2 minutes does not matter if the computer crunches 24/7/365. You dont need a lot of space either; most of my comps had 120 - 200Gbyte drives.
3. Optical drives are not necessary; make a bootable USB-stick ( I use 16 Gbyte) and install from there.
4. Cooling of the CPU and GPUs is very important; but higher temps can be tolerated. I have never had a GPU or CPU fail from overheating.
5. The CPU has to have sufficient power to feed the GPUs. If you use four GPUs you need at least an 8-thread CPU because - depending on the project - the GPU will use somewhere between 2% and 25% of one thread, some projects even more. If the GPU does not use that much of the CPUs power, then the power will be used for crunching.
6. Have a healthy mix of NVIDIA and Radeon (preferrably double precision) GPUs. Some projects prefer NVIDIA GPUs others Radeon GPUs ... and the difference may be enormous.
7. Have sufficient RAM - I think 2 Gbytes/core or thread. Many projects do not use that much RAM, but some do, and they really slow down if you have too little RAM.
8. a case has to be very good (and thus expensive) if it would get better cooling than an open case. I used the least expensive I could get and they worked great with 2 fans in each. Just use common sense and some thought in placing the fans ...
9. use KVMs to connect the computers to one monitor, keyboard and mouse. I got six cheap ones connected 4 to the 5th, and one chained to one of the four and had 21 computers connected to an old LCD screen, a 8 year old keyboard and a cheap mouse. They never failed ...
10. all computers generate heat - and you would do best if you reused the heat generated for other purposes; I heated my old house with the crunchers. My system had an efficiency of 80 - 85% of extracting the heat - and it was sufficient to heat my 380 square meter house (4090 square foot) if the outside temps were above -17ºC (1,5ºF); thus the energy bill was acceptable. There is a thread about this somewhere in the forum...
11. use common sense when doing this stuff, be very scientific in evaluationg performance vs. costs vs. need to invest. Please do not think that the people in this forum are not knowledgable or experienced - some of us have been crunching för 10 years or more, most for more than 3 - 6 years and we have had quite a few discussions about hardware, software etc.

Once you have created a "super cruncher" it will be average within a year and more or less obsolete after 2 years. Also it is not possible to "blue print" one system for all projects; you need at least two systems, preferrably three (one for the projects preferring NVIDIA-GPUs, one for those projects preferring Radeon GPUs and one with most CPU-cores and NVIDIA-GPUs - a "Folding@Home Cruncher", where some projects give very high ppd only if you have at least 8 cores/threads, preferrably 16 - 24 cores/threads).

I will not contribute to this project because I agree with my team mates in their comments above and because I think that you have not sufficient competence (as seen from your pats list) to run and maintain such a project.

 

[ZipSpeed]

Not to be a Debbie-downer but collecting $86 of the $37,000 goal with 49 days left doesn't inspire much confidence for this project. And this is after asking people on technical forums that know a thing or two about DC to make a contribution. Explaining what DC is to the Joe Sixpacks will completely fly over their heads and I doubt you'll get much money from that crowd either.

Believe me, I would love to have more high end rigs in my small little fleet but we all make do with what we have, are given or can afford. And to reiterate what someone pointed out earlier, part of the fun is the accumulation of points. I rather fund my own computers and earn points that are mine. Yes, the points are meaningless in the grand scheme of things but human nature shows that we all like to collect. Without the ability to collect points, I doubt there would be as many donors to DC.

 

[Rudy Toody]

Well stated, Peter!

1) and 7) are very important.
2) could be an SSD when crunching a project that has frequent check-pointing. (However, SSDs can become corrupted with excessive writes.) 40-60GB is fine for crunching.

 

[VirtualLarry]

SSDs are a bad idea for crunchers. I had one in a BOINC cruncher, with the default checkpoint frequency, and after a few months, I had 9TB in writes, and the SSD health was only at 74%.

Granted, it was only a 30GB OCZ Agility SSD, which is known for high write amplification, but still.