Folding Forum

uncle fuzzy wrote:And they should benchmark a multi-core client on a single core how...?

Well for one it's actually not that difficult. You make all the threads run on one processor, it's the act of putting a single core client on multiple processors is where the trouble is and since they've solved that issue, the point is moot.

uncle_fungus wrote:

imzjustplayin wrote:How can Pande Group expect the PPD for the SMP projects to be an apples to apples comparison with the non SMP projects if they're using a different benchmark machine?

No-one has ever stated that it is an apples-apples comparison. The SMP benchmark machine, and the point values were set arbitrarily to provide a reasonable PPD due to the client being A) In beta testing and not stable, B) Running science faster than the regular clients and C) Having tight deadlines to enforce quick return of WUs.
The same factors affect the GPU client as well. They're called HPCs for a reason.

Well the closest I can make it an apples to apples comparison is to reduce it down the PPD per core and if you look at it that way, they have twice the advantage in terms of points accumulation. I see no reason why they don't allow its usage on a single processor system, the only loophole for this is to have it run on a P4 with HT which does in fact work quite nicely.

imzjustplayin wrote:...and if you look at it that way, they have twice the advantage in terms of points accumulation.

Which can be explained by factors A, B and C in my post above.

imzjustplayin wrote:I see no reason why they don't allow its usage on a single processor system, the only loophole for this is to have it run on a P4 with HT which does in fact work quite nicely.

Because Stanford is more interested in getting results returned as fast as possible than someone trying to earn as many points as possible yet slowing down science.
And, just to make this clear, if you are going past the preferred deadline on SMP WUs (or in fact any WUs), you are hindering the science being done.
The minimum CPU required to make the SMP preferred deadline is in the region of a 2 core Intel Core architecture CPU running at ~1.6GHz.

I've moved the thread into "General FAH" since we aren't discussing anything specific to the Windows SMP client.

uncle_fungus wrote:

imzjustplayin wrote:...and if you look at it that way, they have twice the advantage in terms of points accumulation.

Which can be explained by factors A, B and C in my post above.

imzjustplayin wrote:I see no reason why they don't allow its usage on a single processor system, the only loophole for this is to have it run on a P4 with HT which does in fact work quite nicely.

Because Stanford is more interested in getting results returned as fast as possible than someone trying to earn as many points as possible yet slowing down science.
And, just to make this clear, if you are going past the preferred deadline on SMP WUs (or in fact any WUs), you are hindering the science being done.
The minimum CPU required to make the SMP preferred deadline is in the region of a 2 core Intel Core architecture CPU running at ~1.6GHz.

I've moved the thread into "General FAH" since we aren't discussing anything specific to the Windows SMP client.

I'm missing the preferred deadline by one hour. Had they increased the preferred deadline to the same as the final deadline, then my computer would be making it. I think the point that I was trying to make that never made it across was that the SMP systems get an advantage in points because of the discrepancy in points between the SMP client and regular client. Running 10 computers with fast single processors should be helping "science" far more than 2 Core 2 Duos, however the way they've set it up, it's the reverse, and THAT isn't fair. Also increasing the perferred and regular deadline by 3 days wouldn't impact "science" too significantly and would allow many many more systems to run the SMP client.

If they're so hell bent on getting the WU as soon as possible, why don't they call it the "Conroe client" as anything else won't submit the WU as quickly as you're saying they want it. The SMP client should only differ from the regular client in that it can run on multiple processors and work on larger WU, that's it, it shouldn't be geared towards Conroe processors. I keep saying Conroe because AMD systems are having difficulty as it is with the projects the SMP client currently has.

AMDs handle the SMP just fine. My 6000+ has done 144 of them and even my 1.6GHz Turion laptop did quite a few before developing heat issues. The smaller L2 cache makes them slower than Intel, but they still work fine.

uncle fuzzy wrote:AMDs handle the SMP just fine. My 6000+ has done 144 of them and even my 1.6GHz Turion laptop did quite a few before developing heat issues. The smaller L2 cache makes them slower than Intel, but they still work fine.

And how quickly do they accomplish them? Unless they accomplish the WU in 1 day, I wouldn't consider it to be fine as the preferred deadline is 3 days and 2 days is 2/3 of that deadline, on a new processor no less.

6000+, 1d 7h 54m on a 2653, which IS fine within the limits of the hardware I chose to buy. If you want to run faster than that, buy different hardware.

You can't cash the points in for anything, so ultimately they just foster scientific contributions and a little rivalry for us all to do more.

You can go down to Fry's or some other place and pick up a cheap dual core processor and motherboard, slap it together with some ram and maybe a salvaged power supply and old disk drive and install the SMP client on it. Let that be your point generator and take advantage of the point spread.

There's some logic in rewarding extra points to finickier less stable clients. I put more work into getting my SMP clients up than any others. I'm also running PS3's and single cpu clients. Together the homogeneous mix of clients I'm running cranks out science, FLOPs and points. I'm not running a GPU client yet. I may in the future when the 2xxx and 3xxx ATI cards get supported. I'll buy a card just for that purpose. I'll do it mostly because of the FLOPs, since my dual core SMPs do the most points per physical clients of all my folders. The PS3 is about the easiest client to set up.

You can analyze your folding clients in a number of ways and pick what works for you, or mix multiple client types. You may want to optimize for:

PPD/$
FLOP/$

PPD/Power (Watts)
FLOPs/Power (Watts)

PPD/physical client
FLOP/physical client

Ease of client installation
Reliability of client
minimum use of resources (memory, cores..)
maximum use of resources (memory, cores..)

Amount of time folding per day - 24x7 or a few hours a week.

I'm actually running the single CPU client on my dual core laptop so it will use 100% of one core which is 50% of the total processing power. This helps to reduce the heat produced, the fan noise and keeps my laptop very responsive, I don't even notice that FAH is even running except for the fans running a little more at low volumes.

I could get more points using the SMP client on my laptop but it would run hotter and the fan would run faster, more of the time. Since my laptop is often in meetings while I'm folding, it would be more distracting to me and others around me.

So do what works for you. If some clients get more points and you must have more points, take advantage and get or scrounge the equipment to run that client.

It's nice to have the choices so you can pick the client that best meets your needs. My needs are best met by running several clients depending on which device I'm running them on. I still think it's pretty cool that my Blu-Ray player (PS3) is doing protein research when I'm not watching movies. My DVD player did that! One piece of hardware that's very useful for multiple purposes. I have some 24x7 servers that are up all the time, the SMP is great for those but the SMP is not good for my laptop where I use the single CPU client. So having all the choices is the best choice for me!

Take advantage of the different clients and Points and Keep On Foldin'!

uncle fuzzy wrote:6000+, 1d 7h 54m on a 2653, which IS fine within the limits of the hardware I chose to buy. If you want to run faster than that, buy different hardware.

Anyways, they make an SMP client yet it can only run on a limit set of hardware, what kind of nonsense is this? The SMP client should be able to run on all SMP systems. Also there shouldn't be an increase the deadlines and there shouldn't be a per core point advantage to using the SMP client over the regular client. Not to mention that using 10 high performance single core systems provides much more to science than 2 Core 2 Duo systems yet receives and equal or lesser amount of points. This is wrong.

I neglected to mention that there is REAL scientific value in very fast clients because they can run the simulations in less real time. Complete simulations that would take dozens of years can now be done in months. Pande group obviously decides which projects to assign to which clients. I presume that when they have some promising research projects or maybe some projects that lay the groundwork for others, they may save those for clients that do the most scientific computation in the least real time so they can accelerate the scientific results from impossible or years down to months. Other important science that is not as time critical can be assigned to other clients.

Imagine you have 1000 work units to complete a whole simulation. If this is a very important project that benefits everyone by completing it as fast as possible, you assign it to the fastest clients that can complete it in a day or two running 24x7 instead of clients that might take 3, 4 days, maybe a week, and even longer if they're not running 24x7. Each of these work units is a single sequence in a simulation, so no-one can start he next one until the previous one is done, so getting each done as quickly as possible is very valuable. After all, even perpetual graduate students only stay in college for so long and they have to publish sometime! And there is plenty of science to go around that isn't as time critical for people to make important contirbutions using slower clients that may not be on 24x7.

Either way, your client is a node in the world's largest supercomputer which performs in excess of 1.3 PetaFLOPs!

Soooo.....I should be bouncing off the walls because I can't run the PS3 or GPU clients on my 386SX?

uncle fuzzy wrote:Soooo.....I should be bouncing off the walls because I can't run the PS3 or GPU clients on my 386SX?

Notice something there? They're called the PS3 and GPU clients, last time I checked, there was no PS3 with a 386SX and there was no GPU in a 386SX based machine so your point is void.

ElectricVehicle wrote:I neglected to mention that there is REAL scientific value in very fast clients because they can run the simulations in less real time. Complete simulations that would take dozens of years can now be done in months. Pande group obviously decides which projects to assign to which clients. I presume that when they have some promising research projects or maybe some projects that lay the groundwork for others, they may save those for clients that do the most scientific computation in the least real time so they can accelerate the scientific results from impossible or years down to months. Other important science that is not as time critical can be assigned to other clients.

Imagine you have 1000 work units to complete a whole simulation. If this is a very important project that benefits everyone by completing it as fast as possible, you assign it to the fastest clients that can complete it in a day or two running 24x7 instead of clients that might take 3, 4 days, maybe a week, and even longer if they're not running 24x7. Each of these work units is a single sequence in a simulation, so no-one can start he next one until the previous one is done, so getting each done as quickly as possible is very valuable. After all, even perpetual graduate students only stay in college for so long and they have to publish sometime! And there is plenty of science to go around that isn't as time critical for people to make important contirbutions using slower clients that may not be on 24x7.

Either way, your client is a node in the world's largest supercomputer which performs in excess of 1.3 PetaFLOPs!

Oh I see, so with that logic, how about we stop making Wu for systems that are older than 2 years old! Because you know, all of those systems submitting WU are no comparison to the powa of the Core 2 DUO!!!!! Yes that's right, the core 2 duo is so powerful that no matter how many computers you have crunching, anything that is older than 2 years old is a complete waste! Yeah let's just get rid of the regular client and make it a requirement to use the SMP, GPU and PS3 client. That's the spirit! If you want to contribute, you need to pony up or get lost!

@Everyone, is this guy actually that retarded to start this conversation?

@Whateverhisnameis(OP), stop trolling, you are annoying, if you dont like the folding system and points, then delete the client. If you cant do that, take your computer back to the store you bought it from and tell them you are to stupid to own a computer (because you are).

But seriously, you must have a really low IQ if you cant understand what Bruce and the Mod Team has said, i mean we have been doing this since before you got out of diapers. If you want to debate points im sure i can find you a team of lamers who do that on a forum of there choosing, not that i care, but at least you arent doing it here and disturbing the peace.

There is nothing wrong with the points system, if you arent scoring enough then buy better hardware, if you cant afford better hardware then i guess get a job and earn money to buy parts, or i guess your SOL. If it was up to me id kill the points system completely then we could export the points lamers to SETI, but alas this isnt a view shared by many others.

I leave you with a quote that even a monkey could understand, (given a little evolution).

Determining Credit

The amount of credits awarded for completed WU for each project are awarded based on a benchmark applied to all the projects.


The benchmarks are performed on a 2.8 GHz P4 running Linux with SSE2 disabled. The machine is set to receive 110 Points Per Day. This means that a project which is worth 110 points should be completed in 24 hours when running on an identical system. And a projects which took 3 days to complete on the benchmark system will be awarded with 330 credits.


BigWU Work Units get a 100% bonus on the credit due to their size.


The operating system (OS) on which the client is run shouldn't be an issue anymore these days. A 2.8 GHz P4 running Windows XP will perform almost exactly the same as if it were running Linux, because the Cores share the same source code and libraries. This used to be different, when each OS used libraries and code specific to that OS.

If you dont understand that, you may as well give up.

PEBKAC

Actually I've got two Fry's special AMD boxes I purchased almost three years ago folding using the single CPU client. $99 each, with low power AMD mobile processors. I've also got an old Dell (3+ years old?) Optiplex GX60 (Celeron processor) folding. Each of these boxes only does about about 50 PPD (well 60 PPD on a good day). They're not very fast, but I'd have them on anyway, so might as well let them fold. Individually they're pretty whimpy. Taken together, they add another 150 PPD to my team. So I run all types from sloooowww to faast, except for GPU at the moment. I might not have them folding if they weren't on anyway because the electricity bill from running them 24x7 for a year would pay for more efficient hardware in terms of money and speed. But as I said, they're on anyway, so they fold. having multiple clients also helps to ease the pain when one of my clients has a network, power or windows update issue! Even though I lose some points and folding time the other clients are still getting folding done. And I regularly nurse all the clients to make sure they're running. (Just keeping a windows computer running with windows, network and power issues takes some nursing )

Do whatever meets your particular needs best. It's nice that we have choices.