folding on a budget

[Ali3n]

so, i recently have taking up folding. and i have two pretty decent rigs going and earning something like 750k-850k ppd on the two. 

 

noticed that the most recently posted requirements for hardware seems to be pretty low. anything from the roughly 2012 era GPUs

 

so i went and compiled a list of what i could find used on eBay for under $50 US. apart from some questionable deals on 9 and 10 series nvidia cards, the best deals are old amd workstation cards, the gt 650, 650ti and 660 and the 750/750ti 

note i culled anything over &50, which is almost every nvidia card. its not that i excluded incompatible cards just ones that are too expensive to be considered low budget. even $50 is more than i want to spend but these prices will certainly fluctuate with time. some of the cheapest are around $10 but are not cost effective.

the best value cards on the list are the 7,9, and 10 series nvidia low end gaming cards the 50s and 60s. there are listings for these cards in the 40-50 dollar range

next up is the amd workstation cards, notably hd 7850,7790,6950, firepro s7000 at similar prices to the gtx cards 

then there are the older amd gaming cards, r 260x, r7 360, r9 360 are good values as well

finally the cheap cards, the ones that will give much less ppd but are almost dirt cheap

the amd hd 6770, r5 240, and r7 240 are all under $25 and the only cheapo nvidia card i could find on ebay was the quadro k600 for about $18

 

excel sheet attached with my all my findings. i am assuming value by checking passmark's data bank of gpu benchmarks for 3d performance. also assuming that folding is comparable to 3d performance. and my test size of 2 gpus and what i understand it should be relatively linear

folding gpus.xlsx

[corrado33]

Nice work. I did the same for cpus a while back, but no one really cared.

[Ali3n]

Well with gpus you can throw a extra one in your system with much less effort than finding a cheap CPU, Mobo, ram etc. And likely still get better rewards than a cpu. Also I will probably update this either with more expensive cards or as I find more deals. If there is any interest.

[Gorgon]

9 hours ago, Ali3n said:

i am assuming value by checking passmark's data bank of gpu benchmarks for 3d performance. also assuming that folding is comparable to 3d performance. and my test size of 2 gpus and what i understand it should be relatively linear

folding gpus.xlsx

The best metric for comparison is Single Precision Compute Performance which, unfortunately, most benchmark sites do not report on with the exception of anadtech in their reviews. I started a similar spreadsheet mining Anadtech’s reviews but ... then ... got ... distracted.

The other problem is that the Single Precision results like those from FAHBench are not the whole story. The actual performance includes the Quick Return Bonus which applies a non-linear (inverse square root) function to provide more points for Work Units completed faster.

It would be nice if FAHBench could include an estimate in PPD by including the QRB as the “reward” in points stand is the incentive we competitive types care about most.

The spreadsheet maintained by a user at overclockers appears to to be the most comprehensive list of GPU performance figures for f@h but you need to dive into the By GPU tab to get the best use of it as some of the calculated averages are skewed by bad data.

[Ali3n]

5 hours ago, Gorgon said:

The best metric for comparison is Single Precision Compute Performance

good to know. i will do more research then. although i suspect that the gpus that i found that were some of the better perf per dollar, in the price range i was considering atleast, will be roughly the same.

[Gorgon]

On 2/6/2019 at 2:58 PM, Ali3n said:

good to know. i will do more research then. although i suspect that the gpus that i found that were some of the better perf per dollar, in the price range i was considering atleast, will be roughly the same.

You could also look at the P106 mining cards.  Should put out 500-600,000PPD

[Ali3n]

Found a site that has estimated gflops of pretty much every gpu. Not benchmarks but is probably the closest I will find to a list of single precision tests. I will also aggregate data for more card and repost the updated spreadsheet