SR-2 vs 2 systems

[twoodcc]

i have 2 systems for folding right now, a i7 920 and 930. first, could i put both those processors in an SR-2?

second, which would be better PPD wise? power consumption wise?

 

[Axdrenalin]

Standard i7 procs will not work with the SR-2 as a dual CPU configuration, because of the QPI link. The SR-2 is designed to use Xeon processors when using more than one CPU.

The SR-2 would give you a much better PPD/Watt ratio. Currently, I have two i7-920's crunching that comsume roughly 540 watts and give me no more than 40k PPD. My SR-2 on the other hand, consumes about 480 watts and produces roughly 136k PPD crunching the -bigadv units.

If you have the funds, the SR-2 is a much better way to maximize your points vs. powr consumtion ratio.

 

[twoodcc]

ahhh, ok. i didn't realize you had to have xeon processors. makes sense though.

Thanks for the info

 

[mekkanic]

@Ax - What kind of processors do you have on your SR-2?

 

[Kendrak]

Most of us that are running SR-2(s) are doing so to improve our ppd vs the amount of power we are using.

Also PG has said that for some projects they want the information back as quickly as possible (often new WU can't be made till the ones that are our are returned).

By making a SR-2 system(s) we can finish very large WU quickly and keep the whole project humming along.

 

[DarkStarr]

Wow..... That is insane. How much does a typical SR2 setup cost?

 

[Kendrak]

Wow..... That is insane. How much does a typical SR2 setup cost?

Depends on what kind of deal you can get on procs, or how all out you want to go.

Finding a deal in the forum or on Ebay is key if you don't want to go broke buying procs.

A quad box (16 threads) will probably run you $300/proc plus all the extras. So probably $1500 going bare bones.

A Hex box (24 threads) will run ya $450 to $600 a proc (you can go much much higher, but then what is the point of a dual Xeon mobo that can OC the chips?). So going bare bones again.... $1800 to $2100.

And by bare bones I'm talking no case, a basic HDD, and a $10 used vid card off Ebay.

 

[lehmann]

Standard i7 procs will not work with the SR-2 as a dual CPU configuration, because of the QPI link. The SR-2 is designed to use Xeon processors when using more than one CPU.

The SR-2 would give you a much better PPD/Watt ratio. Currently, I have two i7-920's crunching that comsume roughly 540 watts and give me no more than 40k PPD. My SR-2 on the other hand, consumes about 480 watts and produces roughly 136k PPD crunching the -bigadv units.

There no question the SR 2's are king but your comparison of i7's are missleading. 2 i7 systems will get close to 70,000 ppd running 4ghz for X58 chips. 65,000ppd for 2 P55 boards

 

[Kendrak]

There no question the SR 2's are king but your comparison of i7's are missleading. 2 i7 systems will get close to 70,000 ppd running 4ghz for X58 chips. 65,000ppd for 2 P55 boards

Very true, but they will use close to 2x the power.

 

[DarkStarr]

Wow that is some insane pricing but insane performance too. I just wish I had money to do stuff like that. Build a SR2 setup with as many GPUs as possible to fold, would get costly quick.

 

[Wyxe]

Wow that is some insane pricing but insane performance too. I just wish I had money to do stuff like that. Build a SR2 setup with as many GPUs as possible to fold, would get costly quick.

You'll find that most of the guys that run SR-2's don't run more than 1 if any GPU clients because it cuts into the CPU cycles, and the extra power to run the GPU client almost always cancels out any gain from using them.

 

[Filter]

i don't see how people say that's insane pricing.

i guess some of these people don't realize pc use to be about 3k.

 

[DarkStarr]

Really? My 275s are only using less than 25% of 2 cores on my 955 BE. I am saying insane pricing since my entire PC cost less than the basic setup.

 

[Axdrenalin]

@Ax - What kind of processors do you have on your SR-2?

I'm running the L5640 ES Xeon Chips that were scooped up by so many of us off eBay.

There no question the SR 2's are king but your comparison of i7's are missleading. 2 i7 systems will get close to 70,000 ppd running 4ghz for X58 chips. 65,000ppd for 2 P55 boards

Well, one of my i7's are running at 4Ghz and the other at a meager 3.6 (I beleive, don't have much time to mess around with it much). Anyways, you'll note from this week old screen shot that I'm only hitting about 20k PPD there on a 6069, so that was what I was basing my estimates on. I guess I should have taken into consideration that I only process A3 units on my other boxen, and only run -bigadv on my SR-2. Your PPD on the i7's might be more if you've got them OC'ed good and run -bigadv units. I just figure that I'll maximize the number of WU's being processed by running A3's exclusively.

 

[APOLLO]

i don't see how people say that's insane pricing.

The pricing is not 'insane' unless you purchase two $1500 processors to go with the board or max out the memory configuration, etc. Nevertheless, it's far from cheap or affordable. Even a decade ago, when I could have justified the cost of assembling something like a SR-2 system, I would still have considered it an exotic build and seldom does anything exotic come with mainstream pricing.

i guess some of these people don't realize pc use to be about 3k.

Yeah, but how long ago was it when $3000 for a PC was the norm? In any case, PCs can still cost that much or more depending on the components selected. All is relative.

 

[SazanEyes]

Another way to look at it: one hex-core SR-2 box can replace three (not two) quad-core i7 boxen. Even without the F@H bigadv point advantage, that's a big win for those of us who run other DC projects like WCG.

I don't think an SR-2 is ever cheaper as far as build cost, but it's not much more (especially if you can find some Ebay deals as Kendrak said). The advantage is in long-term power savings and space savings. Plus they're just cool.

 

[Kendrak]

The advantage is in long-term power savings and space savings. Plus they're just cool.

I'm probably saving about $50 a month on my power bill now (more during the summer because there isn't as much heat for the A/C to get rid of. So on average probably $75.

With that I'm producing more than double the ppd I was.

And yes.... there is something very cool about having 120k come from a single box and only checking in on it once a month to clean out the fans

 

[DooKey]

I think you save space, but running a hex core in open box X58 motherboards like I'm doing is pretty cheap. I can buy 2 open box boards for $300 or less ( I did this), that's a savings of $300 from an SR-2.

RAM costs are the same, you have to buy an extra $10 vidcard, $40 PSU and $20 HD, still cheaper than an SR-2 rig going with two single proc systems with hexes. I'd say power usage will be a bit more, but not significant. Additionally, single proc boards are much easier to OC than the SR-2.

Either way it's a win for the pocket book and science.

 

[APOLLO]

I think you save space, but running a hex core in open box X58 motherboards like I'm doing is pretty cheap. I can buy 2 open box boards for $300 or less ( I did this), that's a savings of $300 from an SR-2.

RAM costs are the same, you have to buy an extra $10 vidcard, $40 PSU and $20 HD, still cheaper than an SR-2 rig going with two single proc systems with hexes. I'd say power usage will be a bit more, but not significant. Additionally, single proc boards are much easier to OC than the SR-2.

Agreed, and the cost savings are further reduced because one can find used components for single-processor systems easier. Look how rarely SR-2s show up in the FS forum if they even have yet.

Either way it's a win for the pocket book and science.

That's debatable. I am not convinced about the research being superior on one SR-2 vs a mixed farm. Actually, I believe it is less. OTOH, compared to two hex-core single processor system alternatives which is the topic of this thread, perhaps, because the comparison in that example is between the same client types (-bigadv) and completion speed is preferable to volume according to PG.

In the pre-bigadv era, people would have required to assemble a mixed SMP/GPU farm to attain equivalent production. Now, the allure is for one or few mega-core builds in lieu of mixed SMP/GPU setups. A boon for the power-conscious but whither the important research that comes from GPU and other clients? The increasing emphasis in points accumulation from fewer and fewer clients and client types (systems) is the current trend spawned by an exorbitantly remunerative bonus system.

 

[musky]

It all depends on your goals.

A 970/980x system is no joke in both performance and efficiency. Of course you are talking over $1000 for one. My 970 does almost exactly half the production of my SR-2s, and uses just a little over half the power. This is at 4.3+ GHz.

What Dookey is talking about is taking the great deal Xeon hex cores many of us got and putting them in single socket board. While that is still a great machine, you aren't going to see 4.3 GHz and 67K ppd out of it. You save $300 on boards over a single SR-2, and give back $100 of that for extra PSUs/hdd/memory/etc. Performance would be around 20K ppd less, and power draw would be about 50W greater. Still great machines, but not quite on the level of an SR-2 with two of the same processors.

For quad cores, single proc machines probably start making more sense due to the relatively cheap I7s available compared to what quad Xeons cost.

 

[mekkanic]

I think you save space, but running a hex core in open box X58 motherboards like I'm doing is pretty cheap. I can buy 2 open box boards for $300 or less ( I did this), that's a savings of $300 from an SR-2.

This is what I did as well. I did it cause I had spare parts just laying around.

But, I also needed 2 PSUs and 2 hard drives or USB sticks. That cost gets added in too.....probably not much more but...still...

 

[nimbyfaygo]

I could have went with the double board setup. Since I'm running quad core xeons and 6x 1gb ram sticks . I could easily split the system into 2, also I could see the PPD being nearly matched by 2 x58 systems.
I have 18x multipliers on my xeons, the SR-2 doesn't like very high bclk. So I'm sure that I could get higher clocks out of a good motherboard. I average about 85k PPD, so each single system would need 42.5k to break even.

However...there are several reasons why I went with an SR-2. It was my first upgrade in 3 years, I wanted it to be a very noticeable upgrade. The power draw is great for my PPD. I love the (wow) factor I get from my friends and family when I pop open my relativity tiny MAC G5 and the giant SR-2 pops into sight.

My whole system cost my around 1700$, including the custom work on that case. I consider that a very good bargain.

 

[Kendrak]

I could have went with the double board setup. Since I'm running quad core xeons and 6x 1gb ram sticks . I could easily split the system into 2, also I could see the PPD being nearly matched by 2 x58 systems.
I have 18x multipliers on my xeons, the SR-2 doesn't like very high bclk. So I'm sure that I could get higher clocks out of a good motherboard. I average about 85k PPD, so each single system would need 42.5k to break even.

However...there are several reasons why I went with an SR-2. It was my first upgrade in 3 years, I wanted it to be a very noticeable upgrade. The power draw is great for my PPD. I love the (wow) factor I get from my friends and family when I pop open my relativity tiny MAC G5 and the giant SR-2 pops into sight.

My whole system cost my around 1700$, including the custom work on that case. I consider that a very good bargain.

Not to mention, folding aside, you won't need to upgrade the mobo or CPUs for years.

 

[Nathan_P]

Not to mention, folding aside, you won't need to upgrade the mobo or CPUs for years.

Well not until Crysis 4 comes out

 

[nimbyfaygo]

Not to mention, folding aside, you won't need to upgrade the mobo or CPUs for years.

I may jump on the hex core bandwagon, that depends on my Q1 bonus and the prices of used xeons at the time.

At any rate the only real upgrade in the next 5 years I'd expect to get would be upgraded graphics, as I do game occasionally.

 

[may i be worthy]

I just replaced six Q6600 systems with 3 SR2s. I was planning to do one SR2 for my workstation, then single socket x 5 or so, but once I went SR2 I decided to stick with it and rationalise the whole farm to SR2 only.

Reasons in order:

• Reduced maintenance - 3 rigs easier to tune and keep running than 6. Time is money.
• Redundancy - if my godbox breaks, I have 2 near identical systems I can swap to and be up and running very quickly. This strategy worked a charm with my last fleet of 6 q6600s.
• Power saving. (Of course then I overlclock the bejeebers out of them, but I always have to option to back off 5%)

Folding - because of bigadv bonus curve, one fast dual hex SR2 = 2.8 times points of one single hex i7 at the same clock speed. 60kppd vs 170kppd. Not even close.

The bigadv curve rewards higher multiplier chips and higher overclocking.

(of course a 4.3Ghz i7 vs a dual hex SR2 at 3.6GHz looks a lot more competitive)

 

[HighYield]

.....Folding - because of bigadv bonus curve, one fast dual hex SR2 = 2.8 times points of one single hex i7 at the same clock speed. 60kppd vs 170kppd. Not even close.

The bigadv curve rewards higher multiplier chips and higher overclocking.

This is exactly what I worry about with folding now. We certainly aren't doing 2.8 times the science since these are the same work units, and getting the work unit in a day or two earlier really won't make that much difference in the big picture (of finding cures or actionable biology in a medicinal chemistry sense). What we risk is that folks still folding on capable quad cores and gpu clients are going to feel left behind and not valued because of the kfactor bonus multiplier. Don't get me wrong here. I love folding, but someone has to bring balance to the force (I mean point system). Hear that Xilikon....we need some balance Go get your light saber and shake it up at the next advisory board meeting.

 

[osrk]

I have two hex core systems and I prefer it over one SR-2 box. The main thing being redundancy. If one goes down I don't have to rush to bring it back up. I learned from my past mistake of building a dual opteron system /w 4 video cards. It was a pain in the ass to keep it up 100%. Plus my machines are dual purpose as file servers. Overall I average 110k ppd, which is close to what an SR-2 produces with a slightly increased power draw (power supply efficiency is big). There is a temptation to go SR-2 but the board alone is $500+, naw.

 

[Axdrenalin]

This is exactly what I worry about with folding now. We certainly aren't doing 2.8 times the science since these are the same work units, and getting the work unit in a day or two earlier really won't make that much difference in the big picture (of finding cures or actionable biology in a medicinal chemistry sense). What we risk is that folks still folding on capable quad cores and gpu clients are going to feel left behind and not valued because of the kfactor bonus multiplier. Don't get me wrong here. I love folding, but someone has to bring balance to the force (I mean point system). Hear that Xilikon....we need some balance Go get your light saber and shake it up at the next advisory board meeting.

Hence the reason I run a variety of different WU's here. I've got a single SR-2 running -bigadv units, 6 video cards running GPU, 3 i-7's running A3 units and even a standard client or two running the old original units. I believe it's good to run all sorts of stuff thus helping the overall cause.

 

[APOLLO]

This is exactly what I worry about with folding now. We certainly aren't doing 2.8 times the science since these are the same work units, and getting the work unit in a day or two earlier really won't make that much difference in the big picture (of finding cures or actionable biology in a medicinal chemistry sense).

You are so correct. You even mentioned the temporal factor of the research I have deliberately withheld from posting because it would just heat up the controversy even more. Suffice to say that when I had alluded to other reasons why this is becoming untenable, that was one of them. Kudos for apprehending this.

What we risk is that folks still folding on capable quad cores and gpu clients are going to feel left behind and not valued because of the kfactor bonus multiplier. Don't get me wrong here. I love folding, but someone has to bring balance to the force (I mean point system)..

We are essentially conversing amongst ourselves. It is plainly evident 99% of the community is enamored with the system in its existing state. I seriously doubt there will be amendments if the status quo and consensus weight is against reform. We'll see the adoption of this system across all other clients and that will lessen the discrepancy between clients yet worsen the disproportion within. Best case scenario we'll see is probably a slight change to assuage the near negligible dissent, but that's about it (if history is any indication).

 

[SazanEyes]

I have two hex core systems and I prefer it over one SR-2 box. The main thing being redundancy.

This is why you should buy two SR-2s (or three like FTFDU).

 

[may i be worthy]

This is why you should buy two SR-2s (or three like FTFDU).

Beat me to it. Yes, one should have several spare... as one does, when one can.

I will have to amend my sig - to save all the "why do they call you FTFDU?" PMs

 

[Kendrak]

I'm runnning a few other clients other than the 2x SR-2 rigs. My borg is a big part of that, but it has been hobbled, ill have a chance to get it back to 100% over the holidays.
I should run the classic client on my WHS box.... maybe ill get that going when it makes it into the rack.

The thing with SMP folding, when you get a stable OC.... you don't have to touch the box for months on end. They just hum along.

 

[Parja]

...and getting the work unit in a day or two earlier really won't make that much difference in the big picture (of finding cures or actionable biology in a medicinal chemistry sense).

You don't know that.

Clearly time is of the essence on the bigadv WUs or they wouldn't have instituted a bonus points system.

 

[APOLLO]

You don't know that.

Clearly time is of the essence on the bigadv WUs or they wouldn't have instituted a bonus points system.

The math behind the bonus scaling speaks for itself, that's all we're saying. When you have a line that shoots to the bloody moon beyond a certain performance level, it simply doesn't translate into a true representation of research gains for the additional performance, i.e. work completion time reduction. There are diminishing returns as there exists for every field of endeavor, yet the way the bonus operates one would believe the reverse is true. If that were the case, PG would have implemented a bonus structure to the incentives system when F@H2 had commenced nearly 10 years ago, or way sooner than late 2009. No way they would have waited this long. Necessity is the mother of invention.

My understanding of -bigadv and the rationale for the bonus incentive had been implemented was directly due to the high hardware requirements at the time of its inception. This has been stated many times. The initial hardware requirements described dual-processor servers running Linux with gobs of dedicated RAM to complete these WUs within a prescribed period. The high hardware requirement was reduced after the release of Windows A3 -bigadv, at least in the form of memory use. Moreover, I suspect PG had not accurately anticipated the potential performance level of this year's hardware when they plotted the bonus graph. Otherwise, things may have been implemented differently. Next year, we may very well see single processor systems performing close to what a typical SR-2 performs presently. Higher performance dual-processor systems one year from now will be off the scale instantly rendering anything prior 'obsolete' in terms recently established by F@H...

You can believe the extreme gradient of the bonus line represents an accurate picture of the research actually produced and that is entirely your choice, but one is hard pressed to defy the mathematical scenario as it is presented to us. If I supported the POV that states bonus is equivalent to research, then my entire farm is instantly rendered archaic, and I should be swiftly rid of my 5 last generation Xeon systems with their dozen GPU clients in order to procure a single ~$4000 SR-2 system that would produce roughly similar production in the form of points...correct? I would not be able to do that, and taking hypothetically that your assertion is indeed correct, I would be a fool to continue in this project with a vain belief my 'antiquated' farm with its 20 trivial clients is achieving anything truly worthwhile.

 

[may i be worthy]

. If that were the case, PG would have implemented a bonus structure to the incentives system when F@H2 had commenced nearly 10 years ago, or way sooner than late 2009. No way they would have waited this long. Necessity is the mother of invention.

I agree with you in general - you know you have convinced me of your arguments, but not on this particular point... I thought the bonus point system was conceived to grab a bit of the future early.

Bigadv tackles molecules far bigger than anything attempted before 2009, thus needing much faster production of fewer trajectories. bigadv is trying to get an early look at molecules that average 2011/12 hardware is suitable for - but in 2009/10. It had the intended effect - it was one of the reasons why I went with 3 SR2s rather than one SR2 and 5 single cores.

They just did not think through the implications of what happens when you create incentives towards a special type of hardware, by definition you are removing incentives for the alternatives. They overdid the curve, and created a monster that directly attacks the traditional model of running FAH on any old CPU you can find. But perhaps the advent of GPU compute made this model a dinosaur anyway. It looks to me like PG have a mindset that GPU is 90% of the future, and therefore most of the remaining value in CPU folding is really in the stuff that GPUs are weakest in - for now.

PS - you have probably seen this, but being a visual person I needed to see pictures before it finally got why bigadv is so much harder a task than smp. Jmol viewer comparing a SMP project to a bigadv. (6701 vs 2683)

 

[HighYield]

You don't know that.

Clearly time is of the essence on the bigadv WUs or they wouldn't have instituted a bonus points system.

How do you know I don't? I qualified my statement. They wanted a quick turnaround time; that is for sure based on the bonus curve. However I talked about actionable biology and cures in a medicinal chemistry sense, and an extra day or 2 for that work unit won't make that much difference. It would take years from the point that the simulation is complete, interpreted, presented at the appropriate forum, and then picked up in some fashion by actual biology/medchem teams until you might see tangible progress towards a small molecule therapeutic. I guess it just depends on what you want to accomplish. The conclusion of the simulation is just the beginning of the process as far as I'm concerned not the end. APOLLO has articulated very well issues surrounding the math of the bonus scaling and I don't think there is much more I can add in that regard.