Folding Forum

Mod edit: moved topic to appropriate forum

Good afternoon all,

Not sure if this is the right place for this thread but here goes...

After much thinking and saying i would for ages, i have finally got around to spec'ing my office workstation... before i finally commit to it i wanted some advise on top of what i have found via searching the forum and the internet.

So folding boffins, i am looking at something alone the lines of HP z600 spec workstation, the z800's are not worth there 150% price tag for 10% in spec.

http://www8.hp.com/h20195/v2/GetDocumen ... =c04110961

Dual 55 or 56 xeons, currently looking at 5620's i believe, its a 2.66 hex core anyway, but i can go right up to hex core 3.06 all with HT and what not.
Depending on price and what i will get from the jump in cpu points wise?

They will take a max of 96gb of ram, folding isnt going to need as much as that is it? i was pricing somewhere around 24-36gb ish... again feedback, price and gain would decide what i do there, ssd drive for OS (windows 7 pro or ulti) and CAD/CAM software and a standard drive for storage. graphics card wise... am i better running a single 780 or two/three smaller cards that works out using less power, creating less heat (heat isn't an issue by the way) etc, as an example, 3x 760's over 1x780, 2x gtx 770's over a single 780 or 3 5 series cards over a single 780... again feed back and other users experience would help where.

Pc use wise, its for CAD/CAM between work hours, current software uses one core, new software will use 8 cores at most, quadro cards are useless to both software types i am told... (need to check for inventor however as i am considering upgrading to that for CAD) as a rule normally up to 30 tabs open in firefox and up to 20 in IE, Outlook, office as and when i use it, calculator and notepad open now and then, i plan to dual monitor at least. as soon as home time comes around the whole system will be dedicated to folding till i return to work and run folding through the day at a lower pace.

Any advise would be welcome

I wouldn't spend much extra money on the CPUs for folding purposes. Currently, having hexes over quads means that the machine would be capable of folding bigadv WUs, which need at least 24 threads. With HT, dual hexes just meet that 24 thread requirement, while dual quads don't. I'm not 100% positive if dual hexes are actually capable of completing bigadv on time, as that 24 thread minimum is a rough guideline; it does not guarantee that everything that meets it can run bigadv successfully. Throw in the other uses that machine will have, and it may be even less likely to consistently meet the bigadv deadlines. If you can make the deadlines, I'd expect a machine like that to be doing a couple hundred thousand PPD.

That being said, bigadv is going to end this coming January. Currently on regular SMP, the points are not particularly great for the money spent; I'd estimate dual 5650s would probably do 120k or so, but that's a rather rough estimate since I haven't seen any tangible reports (most people with that hardware don't run regular SMP). Obviously this machine is going to be used for other purposes, so you're going to spend a good deal of money anyways, but I wouldn't sink a lot more into the CPUs just for folding. If you think your CAD rendering would really benefit from an extra 400 MHz or so, base your decision on that.

Do not worry about RAM, as FAH uses comparatively little, even on bigadv WUs. I have a 32-core/64-thread rig doing BA, and it uses at most 6 GB. Regular SMP uses even less. While there are some minimal gains to be had by using faster RAM, it's not nearly enough to sink any decent money into.

I can offer some more concrete figures on GPUs. A single 780 will do around 130k PPD at stock clocks, a 770 will do around 90k, and a 760 will be around 60k. Don't bother with 5 series cards, as they don't run the new GPU cores very well.

I don't think running multiple 770s or 760s will actually work out to lower power/heat. The TDP of a 770 is not too much lower than that of a 780, and even the 760 is more than half of the 780. Folding doesn't remotely hit the TDP, but actual power draw numbers are harder to find.

The only situation I currently know of where this would work out would be if you ran a few 750 Ti's. The Maxwell architecture is super efficient, and this card can do around 40k PPD stock for ~40W draw (I believe that's actual power draw, not TDP). There are still some teething issues with Maxwell, however, as it crashes regularly on some of the projects right now.

One thing to keep in mind is that every Nvidia card needs a thread dedicated to feeding it with data due to the way Nvidia currently implements OpenCL, which is what the newer GPU cores run. This would probably disqualify you from bigadv anyway, as you'd want to use fewer than 24 threads, and the servers would then not give you BA work.

Regular SMP on a pair of X5670's comes in at 68-78k PPD depending on which core and project you are folding. I have not run them under the new 24 thread requirements but they used to complete all WU with about 10-12hrs to spare, If you get lower hex's 2.66 or 2.8 - you may not may the deadline, my 5670's turbo up to 3.2 from a base of 2.93.

The only situation I currently know of where this would work out would be if you ran a few 750 Ti's. The Maxwell architecture is super efficient, and this card can do around 40k PPD stock for ~40W draw (I believe that's actual power draw, not TDP). There are still some teething issues with Maxwell, however, as it crashes regularly on some of the projects right now.

Actually the numbers on the GTX 750 TI can be better than that. Right now I'm working on an x'17 unit and the estimated PPD is 50,564. I might be wrong, but I think something has changed recently since this card now seems to run the 17 units without crashing. I'm also noticing that the card is running almost 10 degrees C cooler when running 17's than other ones. 15 units give a maybe 27,000 PPD and 8018 (I think that's the number) only 10,000.

Zagen30 wrote:I wouldn't spend much extra money on the CPUs for folding purposes. Currently, having hexes over quads means that the machine would be capable of folding bigadv WUs, which need at least 24 threads. With HT, dual hexes just meet that 24 thread requirement, while dual quads don't. I'm not 100% positive if dual hexes are actually capable of completing bigadv on time, as that 24 thread minimum is a rough guideline; it does not guarantee that everything that meets it can run bigadv successfully. Throw in the other uses that machine will have, and it may be even less likely to consistently meet the bigadv deadlines. If you can make the deadlines, I'd expect a machine like that to be doing a couple hundred thousand PPD.

That being said, bigadv is going to end this coming January. Currently on regular SMP, the points are not particularly great for the money spent; I'd estimate dual 5650s would probably do 120k or so, but that's a rather rough estimate since I haven't seen any tangible reports (most people with that hardware don't run regular SMP). Obviously this machine is going to be used for other purposes, so you're going to spend a good deal of money anyways, but I wouldn't sink a lot more into the CPUs just for folding. If you think your CAD rendering would really benefit from an extra 400 MHz or so, base your decision on that.

Do not worry about RAM, as FAH uses comparatively little, even on bigadv WUs. I have a 32-core/64-thread rig doing BA, and it uses at most 6 GB. Regular SMP uses even less. While there are some minimal gains to be had by using faster RAM, it's not nearly enough to sink any decent money into.

I can offer some more concrete figures on GPUs. A single 780 will do around 130k PPD at stock clocks, a 770 will do around 90k, and a 760 will be around 60k. Don't bother with 5 series cards, as they don't run the new GPU cores very well.

I don't think running multiple 770s or 760s will actually work out to lower power/heat. The TDP of a 770 is not too much lower than that of a 780, and even the 760 is more than half of the 780. Folding doesn't remotely hit the TDP, but actual power draw numbers are harder to find.

The only situation I currently know of where this would work out would be if you ran a few 750 Ti's. The Maxwell architecture is super efficient, and this card can do around 40k PPD stock for ~40W draw (I believe that's actual power draw, not TDP). There are still some teething issues with Maxwell, however, as it crashes regularly on some of the projects right now.

One thing to keep in mind is that every Nvidia card needs a thread dedicated to feeding it with data due to the way Nvidia currently implements OpenCL, which is what the newer GPU cores run. This would probably disqualify you from bigadv anyway, as you'd want to use fewer than 24 threads, and the servers would then not give you BA work.

Hi Zagen, thanks for your massive reply, helped a lot!

I will have to Google TPD as i remember it form folding stats i believe but not sure what it is!

I think the plan is to hunt some gpus from ebay or the such like and see how we get on with those,i like the idea of the 750's due to the little power draw however i want as many points as a i can get but without spending a fortune! Part of me says use there old p6n diamond and the amd versions (same as atlas) and fill them full of 750's (this is my business's part to "charity")

I BA interests me but i dont think with the amount of use the pc will get that it will complete on time, what does interest me would be a link to your system as it sounds awesome! always wanted to be able to do a million PPD or something in the top 30 in 24hours-not looked at costs of what i would need for that tho and with BA dead as of 2015 that will be an even bigger cost now, imo that is a credit to do something like that in 24hours!

Nathan_P wrote:Regular SMP on a pair of X5670's comes in at 68-78k PPD depending on which core and project you are folding. I have not run them under the new 24 thread requirements but they used to complete all WU with about 10-12hrs to spare, If you get lower hex's 2.66 or 2.8 - you may not may the deadline, my 5670's turbo up to 3.2 from a base of 2.93.

Hi Nathan, thanks for your help too! 78k max PPD isnt really a great achievement considering i will have near £1000 in this build, i know it depends on work units and such but what is the AVE time to complete a smaller and a larger BA workunit on the CPU's i have said above? Do BA take days, or weeks?

Nert wrote:

The only situation I currently know of where this would work out would be if you ran a few 750 Ti's. The Maxwell architecture is super efficient, and this card can do around 40k PPD stock for ~40W draw (I believe that's actual power draw, not TDP). There are still some teething issues with Maxwell, however, as it crashes regularly on some of the projects right now.

Actually the numbers on the GTX 750 TI can be better than that. Right now I'm working on an x'17 unit and the estimated PPD is 50,564. I might be wrong, but I think something has changed recently since this card now seems to run the 17 units without crashing. I'm also noticing that the card is running almost 10 degrees C cooler when running 17's than other ones. 15 units give a maybe 27,000 PPD and 8018 (I think that's the number) only 10,000.

Cheers Nert,

So you too would give a thumb up for the 750 ti's, what is the Ti bit all about? Why is that better over a 750? What have you done to get just over 50k PPD out of them?

TDP is Thermal Design Power. Technically it's how much heat a chip produces under somewhat plausible real-world loads, so that people making the coolers for it know how much heat needs to be dissipated. In practice TDP is a rough shorthand for how much power a chip consumes, as a higher TDP means it uses more power on average. It is just a relative approximation for power draw, however, and folding usually draws well under the TDP since it doesn't use a number of GPU or CPU components very much (integer units on CPUs, RAM on GPUs, etc.)

The key components to my build are quad E5-4650's in a SuperMicro X9QRi-F+ mobo, with 4 4GB sticks of RAM for each CPU. I would have used less RAM, but the Sandy Bridge Xeons run best with quad-channel RAM for each CPU, and 4 GB sticks were the smallest I could find (and were pretty cheap anyway). I got most of the components used off of ebay.

You're right about BA not being a good fit with a machine that does other things, which I neglected to mention. This is especially true with one that is rather borderline for completing BA if running 24/7. The BA projects have an average preferred deadline of 2 days, and I think your hardware would take a little over a day to complete one. Missing the preferred deadline/timeout means no bonus points, and with BA the vast majority of the points come from the bonus.

Regarding the Ti, that's a naming convention that Nvidia resurrected starting with the 500 series. A Ti appended means it's more powerful than a card of that model number that lacks it. So, a "Geforce 750 Ti" is more powerful than a "Geforce 750."

BA takes about 48 hours on dual x5670's running nothing else, a standard SMP WU will finish in about 4 hours. I do know that any L56xx or E56xx will not make the current BA deadlines but are ample for SMP, my L5640's would get about 50k PPD.

£1,000 is not bad - I have £1,600 invested in my x5670 rig - now I am on e5 xeon's for folding the x5670's are in my daily use rig.

I am resigned to the upcoming changes now, whilst I will be getting a lot less in terms of points so will everyone else who currently run BA hardware, it make actually make things more interesting again. Also the qty of science MIGHT increase - I will be doing about 20wu a day instead of 2-3, although the WU will be smaller they might add up to more

HI Gooders,

Here's one other piece of info to consider regarding the GTX 750 TI. It requires one main processor core for each GPU when processing 17 units. I can confirm this behavior on my system. I don't know if this scales one for one for each GPU that you add or not (I only have one card). But, it's certainly something to check out before purchasing. Maybe someone else with more knowledge about how the software works internally can respond.

Regards

Zagen30 wrote:TDP is Thermal Design Power. Technically it's how much heat a chip produces under somewhat plausible real-world loads, so that people making the coolers for it know how much heat needs to be dissipated. In practice TDP is a rough shorthand for how much power a chip consumes, as a higher TDP means it uses more power on average. It is just a relative approximation for power draw, however, and folding usually draws well under the TDP since it doesn't use a number of GPU or CPU components very much (integer units on CPUs, RAM on GPUs, etc.)

The key components to my build are quad E5-4650's in a SuperMicro X9QRi-F+ mobo, with 4 4GB sticks of RAM for each CPU. I would have used less RAM, but the Sandy Bridge Xeons run best with quad-channel RAM for each CPU, and 4 GB sticks were the smallest I could find (and were pretty cheap anyway). I got most of the components used off of ebay.

You're right about BA not being a good fit with a machine that does other things, which I neglected to mention. This is especially true with one that is rather borderline for completing BA if running 24/7. The BA projects have an average preferred deadline of 2 days, and I think your hardware would take a little over a day to complete one. Missing the preferred deadline/timeout means no bonus points, and with BA the vast majority of the points come from the bonus.

Regarding the Ti, that's a naming convention that Nvidia resurrected starting with the 500 series. A Ti appended means it's more powerful than a card of that model number that lacks it. So, a "Geforce 750 Ti" is more powerful than a "Geforce 750."

Thanks for answering my questions Zagen, Monster build btw my builder has said that i can go to 2.8g cores with a turbo boost of 3.2 for an extra £170... for that price i am not sure to say yes or no, part of me says yes but i could buy a second hand 760 in the uk for that from ebay or a brand new 760 ti.

I think with the amount of use and hours i spend here, BA is going to be out of my path witch does is a pain bur thats life... quick ish SMP it is

if i go single or twin card what config would be better for me Zagen dual 750's or dual 760's one of each etc etc, what is your advise?

Nathan_P wrote:BA takes about 48 hours on dual x5670's running nothing else, a standard SMP WU will finish in about 4 hours. I do know that any L56xx or E56xx will not make the current BA deadlines but are ample for SMP, my L5640's would get about 50k PPD.

£1,000 is not bad - I have £1,600 invested in my x5670 rig - now I am on e5 xeon's for folding the x5670's are in my daily use rig.

I am resigned to the upcoming changes now, whilst I will be getting a lot less in terms of points so will everyone else who currently run BA hardware, it make actually make things more interesting again. Also the qty of science MIGHT increase - I will be doing about 20wu a day instead of 2-3, although the WU will be smaller they might add up to more

Thanks Nathan for your help, im going to avoid BA as i wish to use the GPU for folding too. so far the price is at 740 without GPU witch i didnt think was too bad really.

What is your opinion about what GPU's to run for this build?

Nert wrote:HI Gooders,

Here's one other piece of info to consider regarding the GTX 750 TI. It requires one main processor core for each GPU when processing 17 units. I can confirm this behavior on my system. I don't know if this scales one for one for each GPU that you add or not (I only have one card). But, it's certainly something to check out before purchasing. Maybe someone else with more knowledge about how the software works internally can respond.

Regards

Hi Nert, cheers for your help, i have agreed BA is out of the question so just SMP and GPU folding for me.

How many PPD are you getting from your 750's? witch model are they?

Thanks guys!

Summery:

780 is out of the question i believe due to costs and power use, so really its going to come down to 750,760 or 770's and if i run a pair of both or one and another, ideas on a postcard please guys

Power wise the new Maxwell based 750Ti's are supposed to be really good but I don't currently GPU fold so I am not the best person to ask, Last time I fold with a gpu the best bang per £ was the GTX 460!!

Let me support Nathan_P's suggestion.

Measured only in terms of FAH production, the Maxwell is a huge step over older GPUs, and today, only the 750Ti is a Maxwell. (Be careful: most of the 700 series GPUs are rebranded Keplers, though that may change soon.) Look for the GM107 chip.

bruce wrote:Let me support Nathan_P's suggestion.

Measured only in terms of FAH production, the Maxwell is a huge step over older GPUs, and today, only the 750Ti is a Maxwell. (Be careful: most of the 700 series GPUs are rebranded [Keplers], though that may change soon.) Look for the GM107 chip.

That's good to know Bruce, thank you!

Gooders wrote:
Thanks for answering my questions Zagen, Monster build btw my builder has said that i can go to 2.8g cores with a turbo boost of 3.2 for an extra £170... for that price i am not sure to say yes or no, part of me says yes but i could buy a second hand 760 in the uk for that from ebay or a brand new 760 ti.

I think with the amount of use and hours i spend here, BA is going to be out of my path witch does is a pain bur thats life... quick ish SMP it is

if i go single or twin card what config would be better for me Zagen dual 750's or dual 760's one of each etc etc, what is your advise?
...
Summery:

780 is out of the question i believe due to costs and power use, so really its going to come down to 750,760 or 770's and if i run a pair of both or one and another, ideas on a postcard please guys

I wouldn't put the money into the CPUs. You'll get a far better PPD return on GPU(s).

I would use two identical cards over mixing and matching, especially if you're mixing architectures in a situation such as a 750 Ti (Maxwell) and 760 (Kepler). There's fewer opportunities for mix-ups when both cards are the same.

I'm assuming that you'd only be looking at a single 770, as dual 770s would normally cost more than a single 780, and you ruled that out. Points-wise, dual 760s (~120k PPD) should outproduce dual 750 Ti's (~100k), with a single 770 bringing up the rear (~90k, possibly closer to 100k). Power-wise, the 750's should use the least (by a lot, to boot), followed by the single 770, and then the dual 760s using the most.

Nathan_P wrote:Power wise the new Maxwell based 750Ti's are supposed to be really good but I don't currently GPU fold so I am not the best person to ask, Last time I fold with a gpu the best bang per £ was the GTX 460!!

Last and only time i GPU folded Nathan was gtx 290's!

however thanks for your help

bruce wrote:Let me support Nathan_P's suggestion.

Measured only in terms of FAH production, the Maxwell is a huge step over older GPUs, and today, only the 750Ti is a Maxwell. (Be careful: most of the 700 series GPUs are rebranded Keplers, though that may change soon.) Look for the GM107 chip.

Bruce the legend, thanks for your help too, i thought kelpers where the muts nuts?? :S ok so if i go down the 750 route ONLY buy the TI version and look for the GM107 chip within those?

Due to my lack of pc knowledge and being a stupid englishman most of your post baffles me bruce, but from what i see if i use 750's buy TI's and make sure there that chip you listed above? guessing most spec lists will tell me that?

Jesse_V wrote:

bruce wrote:Let me support Nathan_P's suggestion.

Measured only in terms of FAH production, the Maxwell is a huge step over older GPUs, and today, only the 750TI is a Maxwell. (Be careful: most of the 700 series GPUs are rebranded [Keplers], though that may change soon.) Look for the GM107 chip.

That's good to know Bruce, thank you!

Glad this thread helped more than me

Zagen30 wrote:

Gooders wrote:
Thanks for answering my questions Zagen, Monster build btw my builder has said that i can go to 2.8g cores with a turbo boost of 3.2 for an extra £170... for that price i am not sure to say yes or no, part of me says yes but i could buy a second hand 760 in the uk for that from ebay or a brand new 760 ti.

I think with the amount of use and hours i spend here, BA is going to be out of my path witch does is a pain bur thats life... quick ish SMP it is

if i go single or twin card what config would be better for me Zagen dual 750's or dual 760's one of each etc etc, what is your advise?
...
Summery:

780 is out of the question i believe due to costs and power use, so really its going to come down to 750,760 or 770's and if i run a pair of both or one and another, ideas on a postcard please guys

I wouldn't put the money into the CPUs. You'll get a far better PPD return on GPU(s).

I would use two identical cards over mixing and matching, especially if you're mixing architectures in a situation such as a 750 Ti (Maxwell) and 760 (Kepler). There's fewer opportunities for mix-ups when both cards are the same.

I'm assuming that you'd only be looking at a single 770, as dual 770s would normally cost more than a single 780, and you ruled that out. Points-wise, dual 760s (~120k PPD) should outproduce dual 750 Ti's (~100k), with a single 770 bringing up the rear (~90k, possibly closer to 100k). Power-wise, the 750's should use the least (by a lot, to boot), followed by the single 770, and then the dual 760s using the most.

Cheers Zagen,

so from a points per watt basis go for 750 TI's, for raw out and out points keep going up the spec scale... dual 750's would get me the 100k a day i would like as a minimum,

ok, so the £170 i would be spending on the next level up CPU's based on what you say, i think spend on a gpu so that's at least one 750 minimum, Time to go ebaying going to see what deals i can find on there and then post on here what i find...

ok next really stupid question... can i for an example use 2x 750 ti's for folding and that alone... nothing else, and have my monitors/software run on there own card e.g a 770 and that folds when i remember to turn it on, always on, but at a reduced rate, weekends or not at all!?!?

Next question is going to be power requirements... but lets get the gpu's out the way before i ask that.

Yes, cards can fold even when they're not driving a monitor. My main desktop has dual 780s, but I normally drive the display using the integrated Intel GPU since I rarely game on it. The cards fold just fine in that configuration.