OK, i wanted to start a new thread that will go through try to put real numbers to what a lot of what people claim is "bad" when setting up your F@H hardware. things like "you must have 1 physical core per GPU" vs "only 1 logical core per GPU is required". or "for best performance pcie 3.0 x8 or better is required" vs "on linux pcie 3.0 x1 is all that's needed", and many others. the point of this is to once and for all try and give people a solid understanding of just how far you can push the limits with F@H while still getting "good" performance. i do quote "good" as this will be measured in comparison with cost as well. so a 50% cheaper build with 15% reduced performance is logically a "good" route to take if money is tight.
 
Every test will be compared to my recommended build guide running the same GPUs at Stock speeds to eliminate variables outside the ones i am trying to compare to. i also will have each build run at the same time over a couple days to make sure that both builds have the same access to work units and that enough work units are completed to give solid averages to compare against.
 
i will try and continually add new data to this thread over the coming weeks and show off many different hardware setups with numbers attached to them and sometimes linking to a google spreadsheet for some more in depth look at the numbers behind the results. this thread will take a quite some time to populate as i am currently planning on running these test simulations for at least a couple days each so i can have plenty of work units completed for a good result. if you can think of any tests you would like to see please feel free to post them and i'll add them to my list, if possible.
 
Currently scheduled tests that need to be ran:

• 2, 4, & 6 GPUs on pcie 3.0 & pcie 2.0 x1 risers on dual core HT CPU
• 2, 4, 6, & 8 GPUs on pcie 3.0 & pcie 2.0 x1 risers on quad core HT CPU
• 8x GPUs on Threadripper using pcie ribbon cable extensions
• Logical Vs Physical CPU cores

Currently actively testing

• 8 GPUs on pcie 3.0 x1 risers on 6 core HT CPU - Vs - Recommended build guide

 
Results
 

• Control Data _-_ Experiment Data
  • Linux Result: Intel i7-8700K + 8GB RAM + ASUS Z370 + 8x 1080ti on Pcie 3.0 x1 Risers is 0.724% Slower
  • Windows Result: Intel i7-8700K + 8GB RAM + ASUS Z370 + 8x 1080ti on Pcie 3.0 x1 Risers is 78.16% Slower
• Control Data _-_ Experiment Data
  • Result: Linux + Intel G4620 + 8GB RAM + ASUS Z270-P + 8x 1080ti on Pcie 3.0 x1 Risers is 28.46% Slower

• Control Data _-_ Experiment Data
  • Result: Linux + Intel G4620 + 4GB RAM + ASUS Z270-P + 8x 1080ti on Pcie 3.0 x1 Risers is 34.26% Slower

 
Linux Driver Comparisons

• 384.69/384.98/384.111 Vs 390.25
  • Result: 390.25 ~6% faster

But what GPU's?  We already know a 1080Ti will be seriously hampered on a 1x riser, a 1080 and below should not.  Just inquiring :)

This might scare away new Members from Folding under the Team EVGA.
It is overwhelming for someone who only has one computer with one graphics card installed and only want to start Folding.
This adds a complexity that is not needed for a general Folder and makes to complicated to even try it out.

Chris21010
anyone can fold with their existing machines. this is simply a place for more advanced users to see real data on different hardware layouts who want to do more than let their machine fold during idle cycles.

I understand that, but this looks more like what the Folding Members Under Curecoin are doing to mine for coin.
Let's hope you are correct in your thoughts about this subject under the Folding@EVGA area of the Forum.

Ranmacanada
Actually bcavnaugh the top 20 members of team Curecoin are pretty much all rentals thanks to Vorsholk showing them where to get cheap VPS systems from, so they aren't doing this.  There's some pretty heated discussion on the reddit sub about it and most of the members seem to think that rentals are perfectly fine and that as long as the science gets done, who cares if they are fly by night operators who will just move their rigs to the next profitable coin.

Even more of a reason this thread would be best under another area of the Forum, like maybe the Cryptocurrency area.
 

ok, if you want this thread moved which is 100% on topic with F@H just delete it instead. i'll simply post to the main F@H forums my results as to how their program works with different hardware setup. this has nothing to do with crypto and 100% to F@H unless you read between the lines and jump to unintended conclusions. its not my fault that F@H has become synonymous with Curecoin but Curecoin is the reason why F@H has grown so much over the last few years.

nikkocortez
I would say keep it here as there are some F@H EVGA team members who could find the results valuable.  Lets keep this going, Chris!

sorry for the delays. real life has been crazy and i have been having one heck of a time setting up the 2nd test rig and getting it functional. at this point in time i will share with you my first set of test data comparing my suggested build guide vs a 4 thread 8x GPU system on x1 risers with 4GB and 8GB of system RAM.
 
 
 
Control Data
Experiment Data
Result: Intel G4620 + 8GB RAM + ASUS Z270-P + 8x 1080ti on Pcie 3.0 x1 Risers is 28.46% Slower
 
Control Data
Experiment Data
Result: Intel G4620 + 4GB RAM + ASUS Z270-P + 8x 1080ti on Pcie 3.0 x1 Risers is 34.26% Slower

Definitely interesting. Thank you Chris for doing this. Would 16Gb RAM make any bit if difference? I'm thinking not after a certain point. 

Thanks Chris, for sharing the information in this thread - BR awarded
 
looking forward to your updates

i honestly will never run windows simply because i can not justify the additional cost of windows keys, performance isnt even an issue here. though if anyone were to be building rigs with this many GPUs one would think that they can go to linux and get the free performance upgrade.
 
though now that i think about it doing one test with windows wouldn't be a bad idea... though i think i will be forced into buying an SSD for the system as i do not think windows will run well, if at all, on a flash drive. and if i remember correctly i can instal windows 10 with the creation tool and say i do not have the key at the time of install and it will run perfectly normally for a few weeks, long enough for testing.

yea, thats what i meant. i do not need a key as long as i get the testing done before that trial period is up.

Control Data _-_ Experiment Data

• Result: Intel i7-8700K + 8GB RAM + ASUS Z370 + 8x 1080ti on Pcie 3.0 x1 Risers is 0.724% Slower

 
the latest test data and results. Pcie lanes do not seem to be as important as your CPU/RAM. the only test left to confirm this is trying pcie 2.0 again and see what happens. because having a poor CPU hurts and having too little RAM hurts but there is no difference between pcie 3.0 x8 and x1 from the data i see thus far.
 
Next test though will be a windows vs linux test, though to be honest the early estimates are looking grim... 

 
EDIT: oh and i also threw in an i7-7700K into the Z270-P board to replace the G4620. i assume it will be doing much better now that is has 8 logical threads vs the old 4 threads. we will see: https://folding.extremeov...ry.php?s=&u=745999

i mainly pushed the rig to the limits so i can really find the true limiting factors and not have to guess that it may be a couple of things. for example those 3 tests clearly show that a 2 core CPU can handle 8x 1080ti's worth of information at about 25% loss. and that not enough RAM really hurts performance too. and pcie 3.0 x1 works perfectly fine (on linux) even with 8x 1080ti's. knowing that bandwidth is not an when compared to the CPU was a complete flip on how i thought these tests would work out. without these tests many would still think that pcie 3.0 x1 will have performance losses, when they do not. and that you must have 1 physical core per GPU, when again you do not even really even need a logical core per GPU. (in linux anyway)
 
i just like how the data shows exactly what is good and what is bad when trying to build a F@H rig and what kind of rules to follow when trying to build a system optimized for F@H.

CRAP! i forgot to grab the log file before moving back to linux.... No matter though this result wasnt even worth it to begin with and i do have a couple screenshots.
 
Result: Windows + Intel i7-8700K + 8GB RAM + ASUS Z370 + 8x 1080ti on Pcie 3.0 x1 Risers is 78.16% Slower

I have had that machine back on linux again for a while now and production is back up to what it was before, with only ~1% loss in performance.