The top 20 (what it takes)

Patriot

[H]ard|DCer of the Month - March 2011/June 2013/De
Joined
Dec 15, 2010
Messages
2,502
The line to get into the top 20 varies a bit... currently its around 185k and that will bump someone off... if tons of people start trying... its gonna take a bit more.

so what you need is A G34 rig >= 4x 1.8ghz dodecas
or 2 or so SR-2 boxen >= 3.6ghz dual hexacores

As there is currently a trend starting towards quad G34 systems there should be a good many SR-2s popping up in the DC FS thread.... keep a look out.

the easy option is building a 4p system with ~1.7-1.9 12 core cpus... cheaper than a sr2 more ppd and less power usage....

this is my suggestions for the top 20... would love for others to post here...
or post your plottings :p
 
copied from milestone thread

Patience, money and a crap ton of PPD is what's needed. 10 months ago i stated that i would be in the top 20PPD within 18 months. At that point it was 53k to get into the top 20,- I now have a box that does 100k ppd and i'm no where near that goal. It doable but you have to have space and time and money to run a farm. To bet in top 20points wise - you just have to grind away at the science. To do it a couple more multi CPU boxen are probably needed, at the moment the jury is split as to whether to go SR-2 and over clock the snot out of 2 intel hex cores or just go for downright muscle and get a G34 board under linux. Given your current PPD you may get away with just a single addition but the bar is going up all the time
 
Currently, I'm sitting around 12-13 in the top ppd (averaging 225-230kppd). Currently I'm running 50 clients to maintain that number:

1 BIGADV Windows Box
8 Windows SMP Clients
1 Ubuntu PC w/ SMP Client
3 Ubuntu VMs w/ SMP
35 Not Fred VMs running SMP.
2 Not-fred VM's running Standard Clients

Due to the constraints at home (can't cool the room), I had to shut down 2 bigadv, and an SMP client there.

To stay safe in the top 20, you probably need an average of at least 200kppd, which would pop you in 19th; but there is a pretty large clump (9 of us) in the 200-250kppd range).
 
4 x 2600K machines running Linux should get you there as well, and is quite a bit more modular (i.e. you don't need to shell out the ~$2000 - $2500 up front.) These things are also pretty efficient - nothing like the big dog multi-processor boxen, but still very good for what they are.
 
I was in the top 20 for a brief period with 300K. What I had was a DL580 G5 and a DL580 G7. Both of those are quad socket Intel boxes. The G5 is 4x quad cores at 2.93 ghz and the G7 is 4 by 8 cores +ht at 2.0 ghz. The DL580 G7 by itself was over 200K PPD. If you find yourself with one of those at your disposal you will be in the top 20.

To get there with just SMP requires a bunch of boxes.
 
I'm very curious to see how the bigadv game will change once Interlagos hits the market in Q3 of this year, since current G34 owners will likely have a nice upgrade path once the extra special editions start hitting Fleabay.

For now, I just have to reminisce of the days when my 65-75k ppd got me as high as #14-16. This is truly amazing though, that the cost of entry into the top 20 is so high for this team.
 
Two months ago my electricity bill nearly killed me and I chose to rearrange/shut down all but 3 bigadv clients according to my own cost/benefit situation.
I am now maintaining >200K PPD using one dedicated SR-2 on Linux, one i7-920 HTPC w/ Win7 and one hex L5639 in an X58 board running Win7 as my everyday box.
My clients dropped from ~15 to 3, my PPD dropped 10-20% and my power bill is now manageable again, so I kept folding. Win-win in my book.
 
i do it with 2 2600k 1 980x and 6 Gpu's about 200k+ppd

im just now recovering from a house rearangement so farm was down couple days. should be back in top 20 in a few days.
 
What's the cheapest way to get 100K into one box these days (Linux preferred)?

I just got bad news from the doc so I think my backpacking days are over, may sell all my gear and invest in D.
 
What's the cheapest way to get 100K into one box these days (Linux preferred)?

I just got bad news from the doc so I think my backpacking days are over, may sell all my gear and invest in D.

Probably 2x 2600k rigs.

It is 2x boxen, but probably the cheapest way to get that ppd, not too bad on power either.
 
Last edited:
4 x 2600K machines running Linux should get you there as well, and is quite a bit more modular (i.e. you don't need to shell out the ~$2000 - $2500 up front.) These things are also pretty efficient - nothing like the big dog multi-processor boxen, but still very good for what they are.

So just to add to this... I have a few of these things ranging from a E3-1230 to a few i7-2600K's. One of the biggest advantages is that you need under 200w to run them which is not that bad. The E3-1230 folds -bigadv on 107w and can use the stock HSF. You can literally put an entire motherboard + CPU + HSF + RAM together for <$450 and have ECC memory. 3-5 of these work fine.

One major note on the SB's is that units like the 2684's can slow the completion times to just over 2 days. Practically speaking, this means that with 4-5 systems you can end up like my graph and have a lot of peaks and valleys depending on the WU mix and if units are getting completed on the same day or different days.

SR-2 is probably $1400 for board + extra special CPUs + memory and then add another say $70 for the Supermicro 4U 19.5 dba coolers which are cheap, quiet, and work with Xeon mounts.

2P and 4P G34's are the next endeavor for me but I will say, the costs add up. For example, there is no cheap/ quiet cooler that I have seen. Noctua's cost in the $70 each range shipped + tax in California. Compared to the 19.5dba Supermicro Xeon heatsinks, that is more than double for example.

So basically, it takes $2000-3000. You can go a bit lower on this if you are willing to go used on the Sandy Bridge setups. Two dual-hex SR-2's are probably good for top 10-15 at this point so it gives some breathing room as the top 20 floor rises.

Not sure if I have standing to comment on this thread, but that's my two cents.
 
Probably 2x 2600k rigs.
2 2600K rigs in one box?! ;)

I wish I could jump on the 48 core bandwagon, but thats a ton of money. I think my next (hypothetical) dedicated folder would be a low power SB rig, I just can't afford another big capital investment for another folding rig and I'm bumping up against the limit of my power envelope as it is. The SB's seem to do pretty good PPD/W without actually consuming a lot like the SR-2's do (aka SR-2's putting out a ton of points to compensate for a high overall power consumption)
 
I dual MB case would be nice with a couple of SR-2s if it will hold them.
 
Thanks for the save tobit ;)
close one, but I think i've gotta let you slide on the technicality :D

So what does the breakdown of "good" folding platforms look like, from a ppd/w and ppd/$ standpoint?
aka:
SR-2 dual hex (3.6ghz ) ~$1600 150k ppd
Opteron dual dodeca ~$??? 100k ppd?
Sandybridge (4.5ghz) ~600$ 45k ppd?

you know, just in case people are trying to make a run for the top 20 ;)
 
cheapest way to get 100K into one box
The hardware for 2x 2600k builds is about $1300 so when you add on a $400 case I don't think it's the cheapest 100kppd rig anymore. I think the correct answer is still a nekid SR-2.
A 2600k at 5Ghz can put out 60kppd in Linux so I think 50kppd is more reasonable for 4.6ghz under linux.
 
The hardware for 2x 2600k builds is about $1300 so when you add on a $400 case I don't think it's the cheapest 100kppd rig anymore. I think the correct answer is still a nekid SR-2.
A 2600k at 5Ghz can put out 60kppd in Linux so I think 50kppd is more reasonable for 4.6ghz under linux.

You should be able to build a 2600k rig for less than $500 if it is a dedicated rig and only spend on what helps folding.
 
Z68 rigs would be nice, IGP + overclocking. It'd be great to have an OC'ing rig consisting exclusively of a mobo, cpu, ram, psu and hdd to completely minimize the required hardware (and thus power consumption)
 
2 x nekkid 2600Ks will be the cheapest route by a long-shot ($1000), especially if you get cute and split a PSU between the two of them. Anything else close requires some smokin' deals on used parts. #2 is probably going to be a pair of I7 970s if you are looking at new parts, in the $1750 range. #3 is probably going to be a dual 12-core AMD 1.9 at just over $2000, and #4 would be a dual hex SR-2 in the $2250 range. Of these 4 options, the SR-2 will be the most ppd and the dual AMD will be the most efficient.

(EDIT: Save the typing bashing my numbers here...they are blatant guesses...I have not memorizes Newegg's price list....yet... :))

The wild card is used/"extra spicy"/eBay special parts, if you can get your mitts on them. Patriot has the cheapest 100K+ box, and one of the fastest. He "cheated" in that he salvaged the procs, where most of us do not have that opportunity. If you caught the $300 1.8GHz AMD 12-cores on eBay and a Newegg open-box dually board, you could have one of those up and runniing for around $1000 and be just over the 100K ppd mark. Everyone had that opportunity. Used dual hex SR-2s have been hitting the For Sale section as Patriot pointed out, and I suspect several others will end up there shortly.

Keep your eyes open. You never know what you will find....:)
 
You should be able to build a 2600k rig for less than $500 if it is a dedicated rig and only spend on what helps folding.

Assuming a p67 mobo, 2600k, ram, psu combo folding naked you could pull that off for right around 500 probably. You'd have a box that would do 45-50k ppd an only consume around 200-220 watts at load. Pretty awesome if you ask me. For reference, I have an i7 920 @ 3.6 with a GTX 460 that only does about 33k ppd and uses about 380-390 watts. Sandy Bridge or multiproc boxes starts to look real attractive when you look at it like that.
 
Assuming a p67 mobo, 2600k, ram, psu combo folding naked you could pull that off for right around 500 probably. You'd have a box that would do 45-50k ppd an only consume around 200-220 watts at load. Pretty awesome if you ask me. For reference, I have an i7 920 @ 3.6 with a GTX 460 that only does about 33k ppd and uses about 380-390 watts. Sandy Bridge or multiproc boxes starts to look real attractive when you look at it like that.

Much less than 200w BTW. You also have to remember the i7-920 is 2008 technology and you are including a GPU in those figures. Have you tried bigadv with no GPU on that? I'd bet your PPD would stay the same with a lot less power consumption.
 
Much less than 200w BTW. You also have to remember the i7-920 is 2008 technology and you are including a GPU in those figures. Have you tried bigadv with no GPU on that? I'd bet your PPD would stay the same with a lot less power consumption.
You should see a ppd increase actually. I saw about 3k PPD increase on the bigadv, but power consumption dropped about 130W turning off the OC'd gtx460.
 
You should see a ppd increase actually. I saw about 3k PPD increase on the bigadv, but power consumption dropped about 130W turning off the OC'd gtx460.

Yep, thats inline with what I've seen so far with my rig as well.
 
You should see a ppd increase actually. I saw about 3k PPD increase on the bigadv, but power consumption dropped about 130W turning off the OC'd gtx460.

I basically see the same thing on the i7-2600K with a GTS 450. Probably should have said "at least stay the same."

Point being, someone looking in this thread might infer that there is a much bigger power consumption delta than there actually is.

I also noticed a big drop going headless FWIW.
 
looking at this thread i got to wondering, what solution would provide the best PPD value over a certain time line? like, a 2600K running bigadv today be able to still run with the bigadv dogs in 6 months or a year? or would money be better spent on a server / workstation setup? thanks
 
RAD_MAN: the problem is, bigadv isn't guaranteed.. Stanford has always stated that bigadv may not always be here. No one really knows.
 
looking at this thread i got to wondering, what solution would provide the best PPD value over a certain time line? like, a 2600K running bigadv today be able to still run with the bigadv dogs in 6 months or a year? or would money be better spent on a server / workstation setup? thanks

hard question... for a good solid ppd 2600k can do bigadv just fine...
and ivybridge should work in the 1155 boards...so it has an upgrade path.

now quad G34 @ even 1.8ghz competes and beats superclocked dual hexes on a SR-2 rig.... it also has an upgrade path to Interlagos promising a 50% performance increase.... so it unlike sr2s which give comparable performance levels SR-2s have no upgrade path...

now both of those types of setups will give about the same ppd/w close to ppd/$ quad setup just a bit better...

there will probably always be a need for larger work units... whether they get a bonus or not who knows... so ppd is not certain... we can't predict the future...
 
Last edited:
I have my personal doubts about this, TTYTT. It might take a long time, but there has to be a definable limit becasue the human body is not a technological device that can continue to scale in complexity indefinitely.
That may be the case, but the actual complexity of the human body is still well beyond the computational abilities of current computers. It will be a long time before we reach the point where a single computer is capable of simulating the behaviour of an entire human body within a reasonable timeframe.
 
That may be the case, but the actual complexity of the human body is still well beyond the computational abilities of current computers. It will be a long time before we reach the point where a single computer is capable of simulating the behaviour of an entire human body within a reasonable timeframe.
We'll see. Perhaps I'm a bit overly optimistic, perhaps not. We're currently up to how many calculations a second with the highest performance machines working on the project? I'm sure the coders and engineers on the project would know far better than anyone on this forum. It would be very interesting to have a rough idea how close we are to seeing a -bigadv WU complete within an hour, for example. It's not only the performance levels for each core that is improving at a fast pace. Unlike in the past, the number of cores and virtual cores is increasing in tandem with architectural developments within the processor cores themselves. The compounding effect is accelerating performance levels at an even quicker pace as software catches up with the hardware. Once quantum computers become available (which can happen sooner than anyone thinks) assuming the project is still around, well, we'll be talking in terms of WUs per time frame instead of TPF.
 
Last edited:
I'm standing outside the cool group, wondering if I can join back in with the cool kids. Can I come in?
 
1 Eagle07 1,034,166 41,004,308
2 musky 773,393 114,649,284
3 dropper 434,340 92,191,760
4 Mtnduey 401,842 68,274,832
5 Kanakis[H] 392,330 65,489,782
6 lassiterb 390,274 75,115,981
7 Rick5127 347,417 4,957,376
8 dwdawg 293,900 19,052,375
9 TheMTZ 286,343 18,841,662
10 pjkenned 285,505 6,231,006
11 beekeeper([H]ardOCP) 280,113 108,707,419
12 Tigerbiten 273,049 134,382,811
13 mplee73 270,102 53,412,844
14 Kendrak 258,877 66,788,886
15 Syran 233,381 26,908,136
16 KMac 208,656 25,184,806
17 capreppy 207,966 80,665,746
18 bilbois2k 206,986 12,076,533
19 0ldways 199,252 25,161,094
20 Jaech 196,732 59,988,923

Almost 197k for top 20 now... and Damn... 1mil for number one! funny how people clump up, 12k ppd range between 20 and 16, 23k ppd range between 13 and 8, 11k ppd between 6 and 4.
 
Last edited:
The top 20 is an ever moving target... if you manage to muster up 200k you will be in...until 2 ppl muster up a little more... and anyone in the top 20 won't just give it up...
musky for instance... I managed to bump him out of 1st... he is amassing a 4p g34 army to fight back with...



 
The top 20 is an ever moving target... if you manage to muster up 200k you will be in...until 2 ppl muster up a little more... and anyone in the top 20 won't just give it up...
musky for instance... I managed to bump him out of 1st... he is amassing a 4p g34 army to fight back with...




It is a brutal multi-way folding cold-war arms race.
 
arg. I should be surpassing 200k ppd. HFM says about 280k. It hasn't showed up in my stats yet though. I might have to check into that. OH. I remember I lost a bigadv last night. ugh.

In any case, I'm hoping the top 20 holds together so we can be all over 200k
 
Back
Top