Introduction
Since I started the SQLFamily Folding@Home Team on March 8, 2020, I have gotten quite a few emails and Twitter DMs asking me questions. This is great, since it shows there is still curiosity and interest about this distributed computing project. As I have used Folding@Home on a number of machines, I have learned a few tips and tricks that I wanted to pass along.
Why Aren’t You Getting Work Units?
With the current pandemic, and the increased interest in Folding@Home, they have been overwhelmed with increased traffic and work unit requests. Because of this, whatever machines you have running Folding@Home may spend some or even most of their time idle.
Obviously, this is frustrating. The Folding@Home organization is fully aware of this, and they are working hard to generate more work units and add more infrastructure capacity. In the meantime, there are some things you can do to mitigate this issue:
- Periodically run the “Stop Folding”, and “Stop Now” commands with the FAH Web client
- Then run the “Start Folding” command
- This will reset the delay between work unit requests
- Periodically restart the machine where Folding@Home is running
- This is different than explicitly stopping and starting Folding
- It also seems to help in some cases
- Change your Windows Power Plan to Balanced
- This will reduce your power usage when it is idle
What Kind of Hardware Should I Use for Folding?
This depends on what your objectives and budget are. Unless, you are extremely dedicated or independently wealthy, I would not go out and spend a great deal of money on new hardware to run Folding@Home. Many people (especially people who read my blog) may have some existing computer hardware that could be suitable for Folding.
Here are my recommendations, in order of effectiveness.
Dedicated Folding Rig ($$$)
A well ventilated, desktop tower machine, with a modern AMD or Intel 6C/12T processor. A great choice would be an AMD Ryzen 5 3600. You want a machine that is power efficient and will support multiple modern NVidia graphics cards. I would use two 8GB sticks of RAM, one 250GB M.2 NVMe SSD, and an 80 PLUS Platinum or Titanium power supply.
I would use as many NVidia graphics cards as the system will support. The higher end the card the better! For example, an NVidia GeForce RTX 2080 Ti is the current performance king. The more graphics cards you use, the higher your power requirements will be. The model card(s) you are using also has a big effect on power usage. This will determine what size power supply you will need.
The FAHCore software runs much better on modern NVidia GPUs compared to AMD GPUs or any CPU. By default, the FAH client will grab one of your logical CPU cores to manage each GPU that it recognizes. That is why I think a 6C/12T CPU is a good choice.
If you want to really get in the weeds, you can scour motherboard, CPU, video card and fan reviews, to determine the performance/watt of each component. There can actually be pretty big differences in power usage across different brand and model components. You also want your system to run as cool as possible under a heavy load, so that it is not thermally throttled.
Building a System From Existing Parts
You may have old systems or old parts lurking in your basement or garage. Depending on their age and quality, they may or may not be worth using for Folding. For example, you might have an old laptop or desktop that has a quad-core or lower processor, with integrated graphics. A system like this might be perfectly adequate for general purpose light usage, but it will be terrible for folding.
Folding@Home will not use integrated graphics in most cases. A quad-core processor from several years ago is also going to struggle with FAH work units. Older systems will also be much less power efficient than newer systems.
Using an old gaming rig (or cast-off components from a gaming rig) would be a better choice. Your objective is to have the most GPU processing power at the lowest energy usage possible. Newer generation components are simply much more energy efficient. Anything more than 3-4 years old is probably not really worth using from a performance/watt perspective. If you decide to use old existing parts, I urge you to measure the total power usage of the system at idle and under a full load. This way, you will understand what you are doing from an efficiency perspective.
Buying Some New or Used Components
If you have an existing desktop system or some existing older parts, it might make sense to make some intelligent upgrades. Getting a newer generation NVidia graphics card or adding a second one would be a top priority. You may be able to find some bargains in the used market, perhaps from old cyber mining rigs. If you have an old, inefficient power supply, upgrading that to something with an 80 PLUS Gold rating would be helpful. Removing extra parts, like secondary storage devices will reduce power usage.
If you don’t want to spend any money, but you still want to do some Folding for a good cause, then you should carefully evaluate your systems and parts. You want as much GPU performance as you can get with the lowest power usage. With some research and by measuring the power usage from the wall, you can pick the system or components that you want to use.
Other Posts About Folding@Home
I have written quite a few posts about this recently. Probably too many! Here they are:
- Folding@Home News
- Building a Home Computer Lab
- Hardware for Folding@Home
- Folding@Home Power Usage
- Help Find a Cure With Folding@Home
Final Words
I spent much of last weekend “optimizing” my Folding@Home herd of machines. This included swapping 80 PLUS Gold power supplies for less efficient 80 PLUS Bronze power supplies, and swapping in several NVidia graphics cards that I had laying around.
The Folding@Home Stats Report has been stuck with old data for about three days now, which makes it harder for people to track their progress and stay motivated. You can still see your actual Credits from the Folding@Home Web Client though.
You can also reduce your power usage by disabling CPU work units, and only doing GPU work units. This will reduce your overall output, but will be more efficient.
I would love to hear more details about the system(s) you are using for Folding@Home. I also want to thank everyone who has pitched in on this effort!
Good Morning,
I stumbled across this page while I was looking for ways to optimize F@H application. It is a few years too late but I have been folding on and off since 2014 but I have been heavy on Boinc a few years earlier and never left that platform until recently. I believed I knew enough about Boinc and their projects to get myself to the top score in Thailand for a few years now. So, let see what i can do for F@H. It took 8 years for me to climb to the top for Boinc and I am assuming that it will required a similar time span for F@H as well. This is my hobby that I have invested a lot of resources in throughout the years and I am willing to continue the tradition.
I am running 2 dedicated machines. One at work and one at home. They are 24/7 number crunching stations. Every month or two, I’ll do some software updates and give them one day off. Then they will be on full load all cores and all the GPUs working again until a repair or replacement is needed. I used my stuff to the last penny and i dont have to pay for the power bill.
As technology progressed over the years, I had several setups. Odd setups too. I have access to various hardware and I experimented with them. A single all the way to quad socket Xeons (not recommened due to cost and setup complexity that provide so little PPD). Once, I tried using mining rigs with a dozen of R9 290xs and another dozen of Nvidia 980TIs for distribute computing. PPD were huge but that didn’t last long. Finally, I have concluded that running 2 machines for distribute computing seems to be a sweet spot for a hobby.
My current setup since 2021 are AMD 3700x at home and 3900X at work. As for them GPUs, I am running very old units both at work and at home. A pair of 1080Tis at work and 780Ti + Fury Nano + Rx 480 at home. Yes, i like having assortment of GPUs in one machine. regardless of architectures or manufactures, they rarely give me any issues. I do believed that software were written well enough to support this varieties of hardware. The oldest GPU is the 780Ti that still pump out 6-700k a day on stock setting is still not ugly to me even if it is consuming 230W by itself. The Nano is doing 7-800k a day again not too bad. I am pushing 1.5 million+ PPD from home and i am happy with it. At work however, around 3.5million PPD is excellent.
Regardless, I have a feeling that i can get more by optimizing something but so far i could not find anything that would give me that edge yet. In Boinc, each project have their own way to optimize and the differences in PPD is HUGE. I am hoping it is the same for F@H. so, if you have a trick or two, pease let me know. thank you very much and nice to meet a fellow cruncher. have a good day
Happy to say I just picked up a Geforce RTX 4090. I am getting about 15-30M PPD from it. It is an upgrade from 2 Vega 64s which each did about 1.2M PPD. I am running this on my main system which is a 3900x with 64GB RAM and a 2TB NVMe SSD. It also houses 5x 8TB HDDs and 2x 240GB SATA SSDs for caching. The system runs Windows 11. Right off the bat, I could improve performance by running linux on this system. However at current, this system is my workstation and gaming PC so it serves as an all around general purpose system. My 2 Vega 64 GPUs run at about 300 watts each under full load, so this setup pumps out minimum 8x the PPD for 1/2 the power. From a pure number crunching perspective, this is outrageous and should afford me the ability to quickly overtake competitors here in the US for a minimum of power cost. While the 4090 cost a bundle, the savings in electricity will pay for the card easily when considering the costs, plus I now have one of the best GPUs for gaming and for doing other GPU computing work. I am very happy. Now, I have 2 spare Vega 64 cards which I can use in another system build. I also have another system with 2 Vega 64 GPUs (for a total of 4 Vega 64 cards), so I could easily do 20M PPD.
It would be interesting to play around with projects to find those which might yield more PPD from F@H, however right now I am just content to let my 4090 crank out some work units and build my score. With a conservative estimate, I can do 1B points in just about a month, which should easily bump my score quickly. I estimate that with just the 4090 running for about 3 years, I can reach top 10 position throughout the world, assuming things remain static (which they won’t I’m sure), but it’s a goal worth trying for! Very excited.
That sounds like a good system. You might consider dropping a Ryzen 9 5900X or 5950X in there since they are very affordable now, and they use a little less power. Then you could use the 3900X for another system with your older Vega 64s.
Thanks, really useful insight. I think the energy efficiency argument depends a bit on your local climate, your property, and how it is heated/cooled and insulated. If where you live is hot, I can see energy efficiency is important. Spending money to further heat up the room folding is probably not a great idea, especially if you than require AC to cool it back to comfort. But if you live somewhere where heating is required in winter, spring and autumn, I reckon any folding, inc quite old machines , is probably good during these seasons. the energy efficiency of most computers is close to that of electric heating. Almost all energy ends up as heat. As long as your property , or better room, has thermostat that adapts to the folding output, I think it’s probably fine. Better to spend money on heating which also benefits medical science, than just spending money on heating, would be my view.
Having multiple computers running Folding@Home can definitely serve as an alternative or addition to heating a room or small house during colder weather. Picking more energy efficient components or components that deliver more performance/watt is still important in my view though.