CCL Computers - Getting into F@H please help!

[Yewen]

Hi Everyone,

Tom has managed to talk CCL into getting into folding at home in a big way, trouble is my knowledge of distributed computing goes back to SETI many years ago so I am a little out of date.

We will be folding under a CCLOnline account name but part of the Oc3D team!

Initially we are planning to have a display in our showroom highlighting F@H and why our customers should be doing distributed computing. We will have space for probably three machines to be folding on display so I need to get the most performance I can out of the space, preferably with a little nod to energy efficiency so will be using quality PSUs etc.

We also have 100+ office machines that are used for quite basic tasks that are all dual core and above, not planned to roll it out onto the working machines but if the bug hits who knows!

I am hoping to get most of the components for the three show rigs on loan from various manufacturers in return for advertising on a good cause.

The questions I have really are the following:

1) What hardware do I need to look at to get the highest possible output?

2) Is there any reporting I could run on a screen to show what the three machines / CCL were outputting or any fancy graphics that could be run etc?

The processors need to be Intel based so I already have 3x 980X processors sat on my desk ready to be used, just missing the rest of the specifications really!

Any help would be much appreciated.

Tom.

 

[Dewinte]

If you look at the folding at home website I think the clients you'd be using are SMP and GPU2, you have to set the affinity for SMP to leave a thread for GPU2 to use so that SMP has full use of the threads it's assigned to work off (this is based on quick reading). Picking graphics cards is based on what you want from the system... The higher the Nvidia card you pick the higher your PPD but then you also have SLI and someone here who's experienced can comment on SLI 460's vs a 580

but for sure SLI 580's in each rig will be high and mighty. I've seen a couple SR-2 builds with dual xeons and quad 580's do big things. So the GPU choice depends on what you can get from the manufacturers but I'm not sure how the scaling is as in whether two 460s > single 580 or not.

For visuals there is this http://www.legitreviews.com/images/reviews/1461/gtx580-viewer.jpg which you can show what is being done right then and you could also have a page on your website or so that shows the performance of the three machines each day and have that up on one of the screens as well.

I'm sure you'll get plenty of advice from the folders here! They seem ultra knowledgable and may even point out I was wrong

. Good luck and it's a great cause for sure!

Also there are some tweaks with each client such as setting the cores SMP uses and using a command called -bigadv but I'm a noob so I don't know a whole lot, others can comment and be a big help!

 

[hmmblah]

Those 980X will be awesome for running -bigadv. Best bang for the buck GPUs for folding seem to be the GTX460, but any Fermi card is good at it.

As far as monitoring goes, HFM.net is a good tool.

We have stickies in our folding section here on how to set things up.

Welcome aboard!!

Edit: If you have any questions, all of us on the team are willing to help as you set up.

Will these machines be running 24/7??

 

[tinytomlogan]

Those 980X will be awesome for running -bigadv. Best bang for the buck GPUs for folding seem to be the GTX460, but any Fermi card is good at it.

As far as monitoring goes, HFM.net is a good tool.

We have stickies in our folding section here on how to set things up.

Welcome aboard!!

Edit: If you have any questions, all of us on the team are willing to help as you set up.

Will these machines be running 24/7??

Id be worried hmmblah they are after the top team spot

 

[Yewen]

Stickies are great for the setup but the value for money argument of 460 v 580 cards does not really come into it.

All the buying guides I have seen for F@H machines also appear to have a nod towards being sensible, bit different when your one of the largest UK PC retailers looking to make a display in your new showroom with the possibility of sponsorship to an end user buying the components at RRP.

Is it possible to string multiple graphics cards together not in SLI on each machine?

For example would we be able to run 3x 580 cards on each of the three machines, or failing that 480 cards?

And hopefully be running 24/7, just need to look at it for insurance purposes in the showroom.

We should be able to make a bit of an impact with these three rigs, I would just like to see what we could do with over 100 dual core machines folding as they would surely beat the 3 beasties!

 

[hmmblah]

Id be worried hmmblah they are after the top team spot

I won't have it for too long anyway unfortunately. I'll be back down to around 90k PPD in January, some of my machines are only temporary.

I'd love to see us at 1 million PPD avg

. Riot has something planned and now CCL joining us. Oh man this team is flying!

CCL you really need to sell in the US, I'd love to be able to buy from an e-tailer like you! Maybe I should relocate to the UK.

 

[hmmblah]

Stickies are great for the setup but the value for money argument of 460 v 580 cards does not really come into it.

All the buying guides I have seen for F@H machines also appear to have a nod towards being sensible, bit different when your one of the largest UK PC retailers looking to make a display in your new showroom with the possibility of sponsorship to an end user buying the components at RRP.

Is it possible to string multiple graphics cards together not in SLI on each machine?

For example would we be able to run 3x 580 cards on each of the three machines, or failing that 480 cards?

And hopefully be running 24/7, just need to look at it for insurance purposes in the showroom.

We should be able to make a bit of an impact with these three rigs, I would just like to see what we could do with over 100 dual core machines folding as they would surely beat the 3 beasties!

The 580s are best of course if money is no object. SLI has no effect on folding. Each card will fold individually. You can have a mix of cards if you wanted. Only reason I have SLI enabled on my 470 is to game on, otherwise it would be off.

 

[Yewen]

Thats interesting, so we could run for example a 595, 580 and 570 on each machine?

Would there be a way of seeing the difference in performance each card was offering while folding?

 

[hmmblah]

Thats interesting, so we could run for example a 595, 580 and 570 on each machine?

Would there be a way of seeing the difference in performance each card was offering while folding?

Yep. HFM.net will show the PPD of each client so you would be able to compare the folding performance of each card.

 

[Yewen]

Are there any versions of F@H available that show a visual representation of what the machine is doing?

Currently at work so can't install or play around with any of the clients on my work machine.

 

[AMD_PBz]

First off: I think this is an awesome thing that you guys are doing at CCL so thanks to everyone involved for getting the CCL team onboard!

Secondly: Just remember that Prime 95, 3dmark vantage, Kombuster etc... stable means absolutely nothing when it comes to Folding as Stanford introduces new GPU3 Work units and I'm glad Afterburner has 5 sockets to preset overclocks as I've got just about all 5 used up for each of my Cards!

Like a visual representation other than just a chart with numbers?

Cheers

Paul

 

[hmmblah]

Are there any versions of F@H available that show a visual representation of what the machine is doing?

Currently at work so can't install or play around with any of the clients on my work machine.

The graphical F@H client can show a protein folding. I believe there is a mild performance hit by having it showing though. Most of us here run the console only client or use GPU tracker, neither of which will show a visual representation AFAIK.

Edit: Our resident F@H salesman beat me to it!

 

[AMD_PBz]

The graphical F@H client can show a protein folding. I believe there is a mild performance hit by having it showing though. Most of us here run the console only client or use GPU tracker, neither of which will show a visual representation AFAIK.

Edit: Our resident F@H salesman beat me to it!

I don't know what you are talking about

 

[Yewen]

If the machines are going in the showroom through a KVM at least one of them has to be showing something happening for customers to look at.

Other two will be setup for performance, just got to work out all the wonderful specifics of how we are going to get this together and running in a customer friendly way.

The logic behind CCL getting into folding is it is an interesting example of what high powered computing can achieve. Great for a shop trying to sell high powered computing components while appealing to people who might not be into gaming.

Do CAS latencies on memory have an impact on performance for folding any more? (might be showing how long ago I folded here!)

 

[stevehat1]

We should be able to make a bit of an impact with these three rigs, I would just like to see what we could do with over 100 dual core machines folding as they would surely beat the 3 beasties!

The 980,s with SMP -bigadv and 3 good cards (570's, 580's) could be good for as much 100K points per day each.

The dual cores if they are later cpu's should do around 2K+ points per day......I would think that the three beasties would provide more points than the 100 dual cores....

 

[Yewen]

Dual cores in the office are of the Core 2 Duo and Pentium Dual Core variety with the older machines being AMD Athlon X2 3800-4400 chips.

My work machine is now folding, fear its power! (It struggles to run MSN Messenger with all my other programs open

)

Cheers for the advice guys, will start throwing up spec ideas once I have gotten some manufacturers involved.

 

[Xyphox]

Welcome onboard guys!

Your service has been great when I've ordered from you... match that with the folding yeah?

 

[SnW]

Welcome to the Team CCL (Yewen) great to have you aside of us

 

[lasher]

Tom,

In your sales are set up a single monitor and run on one rig the visual element of F@H.

Something like this

http://www.youtube.com/watch?v=qk35_ZFC1qM&feature=related

You can KVM the thee rigs together so if people want to compare PPD production your sales guys can switch between the machines and show them in less graphical form with the programmes already mentioned.

FahMon or HFM.net

 

[kup]

Hey Tom and welcome to the Team.

First of all, thank you for the support, it's awesome to have such a big name helping us Fold.

In regards to hardware and Clients, these guys have got it spot on. Those 980x overclocked to 4GHz (or 4.2GHz) will be pulling some fantastic PPD (points per day) and if paired up with some graphics cards dedicated to Folding, as Stevehat said, you'll be looking at ~100K PPD per rig.

Cas latencies help a little but not enough to drive them riiiiight down. If you have access to lower latencies, via uber sticks or better rated kits, it wouldn't do any harm

Were I to spec and build those rigs I'd go for an Asus motherboard, something like the P6X58D, 6GB 1600MHz Cas 8 ram, those 980x @ 4(.2)GHz and as many 570/80 cards you could fit in.

Software wise I'd run one rig with the graphical GPU Client, showing the Protein Folding, along with one SMP (Multithreaded CPU) Console Client and another GPU Console Client (non graphical). (For a three GPU rig).

Once again, welcome to the Team. It's good to have you here.