Bargain Build - BigAdv

S

Sunin

[H]ard|DCer of the Month - August 2008
Joined
Dec 27, 2005
Messages
3,421
1. http://www.newegg.com/Product/Product.aspx?Item=N82E16819115215 = Intel Core i5-750 Lynnfield 2.66GHz 8MB L3 Cache LGA 1156 95W Quad-Core Processor - Retail $199

2. http://www.newegg.com/Product/Product.aspx?Item=N82E16820144287 = GeIL 6GB (3 x 2GB) 240-Pin DDR3 SDRAM DDR3 1333 (PC3 10660) Triple Channel Kit Desktop Memory Model GV36GB1333C9TC - Retail $135.99

So my real question is go with a cheaper board 105 ish and build the entire thing for 450? or spring for a Triple channel board that will run about 250ish?

Is there a large difference between dual and tripple channel memory for the BIG ADV?

Thx!

 
Socket 1156 is dual channel, you should consider 1366:
1) will overclock easier,
2) HT helps a lot with bigadv,
3) you will get more pcie slots.
4) upgradability
Triple channel probably helps a bit too, I know ram speed does get you a bit more points.

i7 920 - $200 (someone on the forum could get you one if you're not near a MC store)
4 pcie board - $170
6gb ram 1600mhz - ~$110
 
Core i5 lacks hyperthreading so bigadv will not be possible. You'd have to try and overclock that way too high to meet the deadlines.

Edit: If you want to stick with the 1156 platform, the i7-860 makes for a nice bigadv folder however there will be no upgrades available for the 1156 platform.
 
Last edited:
I believe Nitrobass will get the 860 from micro center, pm him.
Otherwise, order the Xeon X3440.
 
I'm playing around with my i7, Pure Ghz vs ram speed.
For me it looks like ram speed makes as much difference as pure cpu ghz.
So I'm trying to find the fastest combo which is stable.

What I've found so far ...........
CPU @ 3.6 Ghz via 18x200, memory @1600 Mhz, ~35 mins per frame, unstable.
CPU @ 3.6 Ghz via 19x190, memory @1520 Mhz, Dont remember frame time but only semi stable.
CPU @ 3.6 Ghz via 20x180, memory @1440 Mhz, ~36 mins per frame, stable.
CPU @ 3.8 Ghz via 19x200, memory @1200 Mhz, ~36 mins per frame, stable.
CPU @ 4.0 Ghz via 20x200, memory @1200 Mhz, very unstable.
CPU @ 3.8 Ghz via 21x181, memory @~1460 Mhz, Its the next to try.

But 1 min difference in frame times is ~1000 PpD at these speeds.

So first indication are that ram speed is just as important as pure Ghz with bigadv.
Also getting bigadv stable is very depentant on having very stable memory.
Mine are rated for 1600 Mhz and test out fine at that speed, memtest and prime, but they don't like bigadv at that speed.

Luck .............. :D
 
Thanks I'll take those into consideration. I have true 120's is the socket that new that I'll need new HSF too?
 
Just an adapter for the TRUE so it fits the new socket.
Thats all I did for my Ultra 90's I'm useing to cool my i7's.
Its a lot cheaper than new good heatsinks.

Luck .......... :D
 
So I guess that leaves me wondering what the best combination is? I was looking at some of the Tyan boards that are 2 x 1366, which would be interesting for about 250, plus two processors, etc, but when you boil it down I think Its better to just get a good X58 mobo... question is which one.

http://www.newegg.com/Product/Product.aspx?Item=N82E16813151193

As I would think I would keep these for processor only folding something I can run off my 430 Earthwatts that I have spare of now.
 
Sunin said:
So I guess that leaves me wondering what the best combination is? I was looking at some of the Tyan boards that are 2 x 1366, which would be interesting for about 250, plus two processors, etc, but when you boil it down I think Its better to just get a good X58 mobo... question is which one.

http://www.newegg.com/Product/Product.aspx?Item=N82E16813151193

As I would think I would keep these for processor only folding something I can run off my 430 Earthwatts that I have spare of now.
Click to expand...
Don't bother with dual-LGA1366 motherboards. You can't run normal i7 CPUs on them (you need Xeons with the extra QPI link to run an MP setup), and you also can't overclock. Just about any X58 motherboard would be decent enough for a reasonable overclock and CPU-only folding. Take a look at lower-end offerings from ASUS, Gigabyte, MSI, etc. and see what piques your interest.
 
So I guess that leaves me wondering what the best combination is? I was looking at some of the Tyan boards that are 2 x 1366, which would be interesting for about 250, plus two processors, etc, but when you boil it down I think Its better to just get a good X58 mobo... question is which one.

http://www.newegg.com/Product/Produc...82E16813151193

As I would think I would keep these for processor only folding something I can run off my 430 Earthwatts that I have spare of now.
Click to expand...
It would probably be more cost efficient to just get two normal Core i7 boards, seeing as dual-socket boards don't allow you to overclock very well. With a good heatsink, the chances of getting at least 3.8GHz out of a 920 are very, very high.
 
yes, def. x58 motherboard! For a dedicated build Asrock 4 pcie is a great deal. 4 double-spaced pcie.
There's also an evga 3 slots for near that price, which physically looks nice if that matters.
For a workstation + folding I'd personally go with either asus or gigabyte, my favorite, instead. There's P6T from asus or cheap EX58-UD4P from gigabyte floating on ebay sometimes.
 
ASRock x58 Extreme open box $118
MC i7 920 B&M $199.99
OCZ DDR3 3x2 1600 kit $149-$30 MIR
Cogage TRUE Spirit $34.99

Get a cheap HD or USB drive for *nix and a decent psu if you don't already have one.

Reason I'd go with the TRUE Spirit over the cheaper CM Hyper 212+ is because the TRUE Spirit will have lower load temps than the Hyper 212 especially after you OC.

That's similar to what I'm ending up with...bought the open box ASRock, bought the U12P SE2 used (no fans $29), bought 1 6gb kit of ram used for $100 and bought the other at ZZF, and just got back from picking up my 920 @ MC in Brentwood. :)
 
Those ASRock boards are very nice for the money. There are many big-time folders using the 4xPCI-E slot versions and running them with 4xGTX-295's and they are very stable. I've recommended that same config as zero2dash before. ASRock has come a long way with these new X58 boards.
 
Close to pulling the trigger, got a combo together for 565 plus 30 MIR, .... and with a decent OC the 920 should pull 80k PPD aprox?
 
Hmm if that is the case then the PpDp$ still leans in the favor of the gpu?!?!
 
Sunin said:
Hmm if that is the case then the PpDp$ still leans in the favor of the gpu?!?!
Click to expand...
Not really, you can put together an i7 rig for around/less than the price of two GTX285s and it will put out slightly more PPD. Plus, you'll have room to add more GPUs, since with a GPU-centric system you still need a mobo and CPU to run the cards. Right now it makes more sense to build a -bigadv box with room for future upgrades than it does to build a box with a cheap CPU and mobo combo that has a bunch of PCI-E slots and fill it up with some GPUs.
 
Agreed, but if you happen to have 5 multi gpu boards it makes sense to just wait to the Fermi's hit and stack 4 of them, and probably pull 100-120k from a single setup, no?
 
Sunin said:
Close to pulling the trigger, got a combo together for 565 plus 30 MIR, .... and with a decent OC the 920 should pull 80k PPD aprox?
Click to expand...
I don't believe I have seen production approaching 80k except possibly for 32-core servers, but we'll see what Gulftown brings to the table. BTW, very glad to see you ramp up. We need all the help we can get!! :cool:
 
Zero82z said:
Not really, you can put together an i7 rig for around/less than the price of two GTX285s and it will put out slightly more PPD. Plus, you'll have room to add more GPUs, since with a GPU-centric system you still need a mobo and CPU to run the cards. Right now it makes more sense to build a -bigadv box with room for future upgrades than it does to build a box with a cheap CPU and mobo combo that has a bunch of PCI-E slots and fill it up with some GPUs.
Click to expand...
I totally agree with this current assessment of the folding landscape. However, things change with the winds at Stanford so what holds true in this quarter can radically change the next. In 6 months we can very easily find our investments spawning the cobwebs of folding obsolescence... :eek:
 
Last edited:
Sunin said:
Agreed, but if you happen to have 5 multi gpu boards it makes sense to just wait to the Fermi's hit and stack 4 of them, and probably pull 100-120k from a single setup, no?
Click to expand...
Not as far as power consumption or efficiency is concerned, especially since we don't know how well Fermi cards will actually perform. I think it makes a lot more sense to replace as many systems as you can with i7 rigs that have 3 or more PCI-E slots if you can afford it. Tigerbiten can attest to how good the results from that would be.
APOLLO said:
I totally agree with this current assessment of the folding landscape. However, things change with the winds at Stanford so what holds true in this quarter can radically change the next. In 6 months we can very easily find out investments spawning the cobwebs of folding obsolescence... :eek:
Click to expand...
Things have been going that way ever since Stanford introduced the first high-performance client back in the SMP days. You never know what's going to happen next, so all you can do is plan things based on the current landscape. In any case, I doubt the -bigadv program is going anywhere considering how much work Stanford is getting done with it. -bigadv is basically replacing the need for the F@H guys to rent out server farms, which is invaluable to the project.
 
I happen to think there is going to be more CPU-based developments at Stanford.. bigadv is just the start.
 
Tobit said:
I happen to think there is going to be more CPU-based developments at Stanford.. bigadv is just the start.
Click to expand...
I haven't heard anything else besides Stanford R&Ding a cluster client, but that has long been relegated to low priority status, IIRC.
 
I think in terms of what the future holds, building an x58/i7 is a better investment than buying several gpus. (That's what sold me at least)

i7's pull very good PPD at surely less power demand than a couple of new gpus. Fermi is still up in the air, no one knows what'll happen there. We know what PPD you'll get off an i7. We also know hexacores are coming for 1366 which should be a drop in cpu replacement on any x58 board (with or without a bios update). If quads (8 with HT) i7's get 20-25k PPD (mathematically that's 2,500-3,125 PPD per core) I would estimate a hexacore will yield (conservatively using those figures) at least 10k more PPD.

This is all of course in a perfect world assuming the hexas perform core to core and clock to clock what the i7's currently do. :)
 
If quads (8 with HT) i7's get 20-25k PPD (mathematically that's 2,500-3,125 PPD per core) I would estimate a hexacore will yield (conservatively using those figures) at least 10k more PPD.
Click to expand...
Trouble is, the new hexa-core nehalems are apparently going to retail for $1000 each. Is the extra money spent on that worth it over a few extra quad-core i7s?
 
Tobit said:
Core i5 lacks hyperthreading so bigadv will not be possible. You'd have to try and overclock that way too high to meet the deadlines.

Edit: If you want to stick with the 1156 platform, the i7-860 makes for a nice bigadv folder however there will be no upgrades available for the 1156 platform.
Click to expand...

I know I'm late to this thread, but I'm quoting this one. If you choose the 1156 socket, the i7-860 is a much better choice than the i5. I'm running this CPU here at 3.5 Ghz under an H50 with 8 gigs of RAM. Frame times are at 35 minutes 15 seconds with little fluctuation.

Don't have a 1136 so I can't compare the two, but I'm very happy with the i7-860 series.
 
Mr. Pedantic said:
Trouble is, the new hexa-core nehalems are apparently going to retail for $1000 each. Is the extra money spent on that worth it over a few extra quad-core i7s?
Click to expand...
The i7s will, but the non-EE Xeon variants will be cheaper.
 
Mr. Pedantic said:
Trouble is, the new hexa-core nehalems are apparently going to retail for $1000 each. Is the extra money spent on that worth it over a few extra quad-core i7s?
Click to expand...

If you are looking to spend $1000.00 on a CPU, it would be better to buy 2 x E5530 and build a gainestown system like i have.

Ebay can also be your friend ... *wink*

That would give you 16 logicial cores ( 8 physical + 8 HT) and get much better performance.

The -bigadv units have a bonus based on how fast you complete the units.
If Stanford uses this bonus on future units, current dual cpu systems will get more PPD then a future Core i7 980x for around the same price range.

Another fact that many users forget, adding GPUs to a -bigadv system will drop your bonus by a large % (somewhere in the 20%-35% range) which can really add up!
 
EvilAlchemist said:
Another fact that many users forget, adding GPUs to a -bigadv system will drop your bonus by a large % (somewhere in the 20%-35% range) which can really add up!
Click to expand...

What is the cut off line? Adding a 2nd gpu? 3rd? 4th?
If having the 2nd 260 cuts up my -bigadv bonus then I may relegate it to the remaining systems.
 
EvilAlchemist said:
If Stanford uses this bonus on future units, current dual cpu systems will get more PPD then a future Core i7 980x for around the same price range.
Click to expand...
Stanford will definitely be using a bonus system on new A3 SMP2 cores.. Vijay has already confirmed this:

The SMP2 cores are currently in testing; watch for future announcements regarding their release. The first SMP2 core to be released will be a Gromacs-based core with core number A3. In conjunction with this release, we are implementing some updates to our points benchmarking system. In particular, we will use early-completion bonuses, which we have been testing with the bigadv work unit program, more extensively for SMP2 cores.
Click to expand...
 
zero2dash said:
What is the cut off line? Adding a 2nd gpu? 3rd? 4th?
If having the 2nd 260 cuts up my -bigadv bonus then I may relegate it to the remaining systems.
Click to expand...

Any GPU cuts into the PPD of the -bigadv units.
That many not be the case with the newer SMP client.
Time will tell.

I would suggest shutting down your gpu and test you PPD over a few units.
That way you can see how it is affecting your points and can make a decision based on your results.

Tobit said:
Stanford will definitely be using a bonus system on new A3 SMP2 cores.. Vijay has already confirmed this:
Click to expand...

Awesome. I have to remember what is public and what is in testing.
Hard to keep track sometimes.
 
EvilAlchemist said:
Any GPU cuts into the PPD of the -bigadv units.
That many not be the case with the newer SMP client.
Time will tell.
Click to expand...
Any GPU client will cut into any CPU client period, simply because the GPU cores still require some CPU time, although not very much. When I was running three GPUs in my secondary PC (with your old E2140 ;)) I saw my SMP PPD drop down to about 1350 whereas now with one GPU it's at about 1950. Someone I know who was running -bigadv on an i7 rig with two GTX260s folding saw an increase of about two minutes per frame with the second GPU folding.
 
well I'm also folding bigadv -smp 7 along 3 gpus but my TPF is within what tigerbiten posted above. 36min30sec TPF, 3.76ghz 1200mhz. Doesn't look like it has a big effect.
 
404 - File not found said:
well I'm also folding bigadv -smp 7 along 3 gpus but my TPF is within what tigerbiten posted above. 36min30sec TPF, 3.76ghz 1200mhz. Doesn't look like it has a big effect.
Click to expand...

Right but your are only running your SMP across seven of eight cores.

If you were running across all 8, and then turned on GPUs you would notice it more :)
 
404 - File not found said:
Doesn't look like it has a big effect.
Click to expand...
This is what I have always assumed as well. 36m 30s frames are still very acceptable and grant a little over 55K points/unit with the bonus for an effective PpD of 21.8K. The points gained from the 3 additional GPUs more than covers the few minutes of lost CPU time per frame.
 
nitrobass24 said:
Right but your are only running your SMP across seven of eight cores.

If you were running across all 8, and then turned on GPUs you would notice it more :)
Click to expand...
Precisely. After all, the whole reason for running -smp 7 instead of -smp 8 is to give the GPU clients more breathing room.
 
My biggest hope is that the new SMP client can run the -bigadv WU in a windows based platform.

That will get rid of the VM overhead and make things much easier.
 
Kendrak said:
My biggest hope is that the new SMP client can run the -bigadv WU in a windows based platform.

That will get rid of the VM overhead and make things much easier.
Click to expand...

Yea i seriously cannot wait for the new client.

I would pay Pande Group $ to let me have the beta client:eek:
 
You must log in or register to reply here.
Back
Top