Why is there a bonus for SMP WU's

[Geoffric]

I am new, and was wondering why is there a bonus for SMP WU's but not GPU WU's?

I mean a slong as the work get's done, who cares HOW it was done?

The only thing I could think of was:

1. Perhaps the data returned from an SMP WU was in a different format, or was more refined or complete than a GPU WU and therefore required less processing on Dr. Pande's end.

2. Perhaps it serves as a very very good benchmark for the OC people, and so would guarantee many WU's being processed as geek to geek folding contests broke out into General Geek equipment escalation, which would in turn process more completed WU's, and so on, ad infinitum.

Geoffric of team DARKSIDE_RG

[Qinsp]

It appears that Pande Group is working in that direction for all jobs (QRB - quick return bonus), based on other posts.

F@H is a very, very "wide" computer, but it isn't always a "fast" computer.

SMP appears to be the guinea pig for a mechanism to encourage completing all jobs as fast as possible.

But GPU folding has a beauty all to itself. Any computer made in the last 10 years can make over 1000 PPD by adding a relatively inexpensive video card. The CPU in an old (but still useful) computer just isn't that fast by today's standards, and makes very few PPD in comparison.

An average desktop sold retail today will make more PPD by adding a video card (or even using the on-board graphics if supported by FAH) than by CPU folding by a wide margin. Typical desktop sold is a dual core in the 2.6ghz range and has a PCIe graphics slot available.

[Zagen30]

It mostly stems from the early days of SMP folding. Back then, the SMP core would always generate 4 threads no matter how many cores you had. Some people found that running 2 copies of the SMP client produced more PPD on their rigs than one copy. The thing was, running 2 copies meant that each copy was running at roughly half speed, and the Pande Group greatly prefers WUs to be returned ASAP since the generation of the next WU in a particular trajectory can't be done until the previous one comes back. Hence the bonus- they made it more profitable to return one WU in X amount of time over 2 WUs in 2X time. GPUs don't have this problem - you can't run multiple copies of the GPU client on a single-GPU card, so there's been less of a need to roll out bonuses for it.

[Grandpa_01]

The link below explains the reasoning behind the bonus points, PG gives bonus points to promote the quick return of projects they need faster and have a higher science value.

viewtopic.php?f=24&t=13039

[Geoffric]

Thanks!

I am running an Intel Q9550 slightly overclocked to 3.12 GHz and 1600 FSB and DDR3 1600 DRAM.

I have a high end video card, and the average GPU WU of 1200 points is about 2 hours, and the average SMP is 10 - 16 hours in the 400 -800 point range.

My Mom just came home yesterday after a sucessful operation to remove a tumor, and we are doing a new build without monitor, hard drive, or graphics card in order to have a 24/7/365 SMP dedicated AMDx6 1090t machine.

Every body keeps saying it won't be a -bigadv unit because it needs 8 cores, but I have seen a lot of posts saying you CAN do it.

[Nathan_P]

Thanks!

I am running an Intel Q9550 slightly overclocked to 3.12 GHz and 1600 FSB and DDR3 1600 DRAM.

I have a high end video card, and the average GPU WU of 1200 points is about 2 hours, and the average SMP is 10 - 16 hours in the 400 -800 point range.

My Mom just came home yesterday after a sucessful operation to remove a tumor, and we are doing a new build without monitor, hard drive, or graphics card in order to have a 24/7/365 SMP dedicated AMDx6 1090t machine.

Every body keeps saying it won't be a -bigadv unit because it needs 8 cores, but I have seen a lot of posts saying you CAN do it.

Have you registered for your passkey yet, PPD will go up once you have completed 10 WU successfully before the preferred deadline?
Re the x6 you are right on both counts, personally I would go with regular SMP as you need linux and a 4ghz overclock, I would look to add a graphics card to that new build at a later date or consider a different CPU if the funds allow. i7's allow bigadv units and are more power efficient than the AMD chip

[Geoffric]

Well, one of the reasons I am going AMD I am experimenting on being green as I can, and I have an 85% metal mesh Rosewill ARMOR gaming case. I am mounting a window fan on one side of the box at a very slow speed setting and see what I can do. Also I save $225 because I am not getting a new graphics card, but I do have a couple of GeForce 8600 and 8800 card. I also can do this build for like $425 total, and as a good i7 build would cost me $800-$1200 I can have 3 of these which should outproduce the single INTEL rig. Also I can pay as I go each time I add a build, and I would have to save for like 3 or 4 monthsd to get the money need for that single INTEL rig.

Still, if SPEED is important to the Pande Group, my 1400 point GPU WU's take 2 hours, and the 900 points SMPS take me 16 hours, so I complete a job MUCH faster on GPU rather than SMP, and so I would think the Pande group would want people to run the GPU clients specifically.

Also all my builds have been INTEL (My first was a E3110 XEON that I got from 3.0 to 3.6 by doing nothing but changing the FSB freq.)

[Zagen30]

Speed is relative to the size of the protein being folded, not a constant across clients. Most GPU projects until recently have been simulating relatively small proteins, in the 600-800 atom range I believe. SMP proteins are larger and thus take longer to simulate the same timescale as a smaller protein, so the fact that an SMP WU runs longer before completion than a GPU WU doesn't really say much about how valuable it is. The Pande Group has never recommended one client over another because they all contribute different things to the project that are all valuable (they wouldn't have created each client if it weren't worth their time, money, and effort). The GPU cores can do certain types of simpler simulations very well, but can't do more complicated simulations, which the SMP (I think) and Uniprocessor cores can.

[mdk777]

Yes, currently -bigadv does not run on AMD X6 processors.(without some hacks)
The discussion you have read has been about whether it "should" be allowed.

I don't think you should base your build decision based on this insider debate.
Having been apart of these discussions, I don't want to mislead you. There is very little chance that -bigadv points will ever be optimized for the AMD platform.
We will have to wait awhile to see what BD brings to the table, but that is still pie in the sky for now.

Also I can pay as I go each time I add a build, and I would have to save for like 3 or 4 monthsd to get the money need for that single INTEL rig.

Similarly, I applaud your dedication and donation, but you need to start to figure in electricity costs once you get beyond one rig. Just Some thing to think about.
Those numbers add up fast.

[bruce]

Having been apart of these discussions, I don't want to mislead you. There is very little chance that -bigadv points will ever be optimized for the AMD platform.
We will have to wait awhile to see what BD brings to the table, but that is still pie in the sky for now.

I object to that characterization. The issue is NOT related to AMD platform, but rather the number of cores reported by the OS.

There is no chance that -bigadv WUs will ever be assigned to a machine that reports it has less than the required minimum of 8 cores. (You wouldn't be complaining if you had X8 to put in that machine, and as you point out, the jury is still out on what BD will report.)

[P5-133XL]

For the price the Sandy Bridge CPU is really hard to beat especially if you are concerned about electricity costs. That being said there is the chipset recall will make the MB's rather hard to find right now. If you can get a MB, the chipset problem is rather easy to avoid by simply using a SATA card or just use the two SATA III ports.

[Qinsp]

If you are willing to wade through the steps necessary to fold BigAdv using a six-core AMD, it is currently the cheapest system as far as sale price goes that has enough HP to fold all bigadv WU's in under the deadline. If you have an existing system that you are going to retire, a retrofit to AMD 1090T is under $300 today, or less than an i7 chip alone.

It can register as high as 22 gflops sustained average folding power at the factory clockspeed, and is fairly easy to boost up to 4ghz.

The steps necessary are complicated for someone unfamilar with Linux, so it pretty much guarantees that those who take that path are dedicated to the cause.

Another AMD path that is available is to run 2 AMD G32 quads on an ASUS ATX server board. This requires no "hacking" or overclocking, but the price to retrofit an existing ATX case computer for dual 4130's (2.6ghz quads) is just over $500, or back into Intel pricing. But it's saving grace is that it's understressed and highly expandable. It has 2 PCIe graphics slots and can support a large amount of RAM and HDD's should you want to make a powerful computer out of it.

[mdk777]

I object to that characterization. The issue is NOT related to AMD platform, but rather the number of cores reported by the OS.

There is no chance that -bigadv WUs will ever be assigned to a machine that reports it has less than the required minimum of 8 cores. (You wouldn't be complaining if you had X8 to put in that machine, and as you point out, the jury is still out on what BD will report.)

Really Bruce?
I was doing my best to inform someone who sounded very sincere and didn't have a great deal of money to waste.
You would have had me encourage him to try and get -bigadv to run on his AMD 1090T?
It will run SMP great as Qinsp says, but he asked about running -bigadv and I answered that he should not set his hopes on it....and you have a problem with this advice?

read the OP post again.

[bruce]

I agree with your recommendation, but I object to you calling it a support issue associated with a specific brand. The minimum requirement is 8 cores.

The forum policy is to be team agnostic and brand name agnostic.

[Geoffric]

I know I am the FNG, but c'mon guys, let's keep our eyes on the prize . . . .

I primarily went with the AMD because:

1. I have never done a non INTEL or XEON build

2. I have a E3110 XEON and a Q9550 on LGA 775 boards, and Oc'd they give me a great gaming rig with the hawk talon GPU.

3. I wanted to get one of the 58 series Intel board, but 20% of the feedback was DOA, or quit in the first 10 days.

3. I don't have the money for the INTEL 6 core, but I did have the $430 for the AMD3 board, and 1090t.

If I can't OC it enough to do the -big adv that's all good, I will move my GPU's onto this board once I have emough money to get an INTEL -big adv rig. Also it's not about the points so much, but 2 yeras ago I was afraid to open my case to read the serial number off the Mother Board, but I have done over 12 builds which, to me, is a lot. I have learned quite a bit as my compadres at DARKSIDE_RG were very kind by helping me with decisions on hardware that I did not understand at first, but now I do.

I can dso a simple OC to get my E3110 from 3.0 to 3.6 buy just changing the FSB MHz and multiplier. I am working on the voltage changes and the pci busd speed and draam latencies now, and have started making a few tweaks. I could never come close to the super rigs you guys have here because I can afford the new technology IF I stay 12 - 18 months behind the curve. Also, I just found out what trolling meant, and I am very serious about my questions, and seek to know and understand things as time goes by. Many thanks for the good advice.