Folding@home Points FAQ

Table of Contents

Introduction

Much of what drives distributed computing is the sense of collegiate competition to compute the most for the project. One way to quantitatively assess this is through the points that Folding@home (FAH) keeps track of (in our statistics or Stats pages). Here we detail the nature of how our points are determined and why that method is used. There are lots of methods one could use, but we've found that over time (and with extensive discussions with FAH donors) that our current method is a reasonable compromise, given all the complexity of awarding points for FAH work units (WUs).

When will my points, user name, or team # appear in the Folding Statistics?

When you first start to fold, your user name will not appear in the statistics until your FAH client has completed its first work unit. And because the stats pages are updated every few hours, your stats may not appear immediately. Please be patient. If you created a new team, it will appear right away.

How do you decide how much credit a work unit is worth?

General scheme (relevant for all types of clients)

Points are determined by the performance of a each contributor's folding hardware (CPU, GPU, etc.) relative to a reference benchmark machine. Before sending out any work units from a new Project #, we benchmark one or or more work units from that Project on a dedicated machine. We plug the results of this benchmark test into the following formula:

Points = PPD * (DaysPerWU)

where DaysPerWU is the number of days it took to complete the work unit.

We summarize the details of the machine and the PPD below. Please note the very concept of a reference machine will mean that some WU performance will vary from the performance on your machine. Even between various Xeon processors, there are significant differences in architectures. Moreover, there are variations between WUs within a given project which can lead to speed differences.

CPU client

Benchmark machine: 2.8 GHz Pentium 4 computer with SSE2 disabled. (more specifically, as reported by /proc/cpuinfo on linux: vendor_id : GenuineIntel, cpu family : 15, model : 2, model name : Intel(R) Pentium(R) 4 CPU 2.80GHz, stepping : 9, cpu MHz : 2806.438, cache size : 512 KB). This machine runs linux, so all WUs are benchmarked with the linux core. Note: Currently the linux and Windows fah_cores run at exactly the same speed (same code base), so this does not impact points performance.

PPD: 110. This PPD was chosen to match the points on Gromacs WUs from the previous point system. The upshot is that Tinker WUs will be worth more than before we set up the new point system (i.e. before April 2004, which previously ran on a Celeron 500 MHz CPU (which did not have SSE2, so it was disabled above to maintain similar points performance before and after the benchmark upgrade)). The benchmarking and points for each client are detailed on the FAQ page specific to that client.

Note that this P4 machine is fairly old and we are working to transition to the new i5 machine (see SMP2 below). As we phase out non-A3/A4 Gromacs cores, we will phase out the P4 benchmark machine as well.

PS3

Benchmark machine: PS3

PPD: 900

See the PS3 FAQ for more details.

GPU2 client

Benchmark machine: ATI Radeon 3850 GPU. Details:

• GPU: ATI 3850 GPU 512 MB, 320 Stream Processors
• Host: Dell Inspiron 531, with a 2.16 GHz dual core AMD 64 X2 4000+.

PPD: 1500. Note that the GPU client still relies on a fast CPU, so the CPU is an important part of this. The Points Per Day (PPD) given here assumes that a CPU is heavily needed, with a larger PPD to compensate for the use of that CPU.

see the GPU2 FAQ for details.

GPU3 client

Benchmark machine: GeForce GTX460 . Details:

• GPU: VGA GeForce GTX460 768MB DDR5 PCI-Express 2.0 Graphics Card 768-P3-1360-TR
• Processor: Dual Core AMD Opteron(tm) Processor 275 (2210 MHz)
• Operating System: Windows 7 Professional, 32-bit
• DirectX version: 11.0
• GPU processor: GeForce GTX 460
• Driver version: 258.96
• CUDA Cores: 336
• Core clock: 720 MHz
• Shader clock: 1440 MHz
• Memory clock: 1800 MHz (3600 MHz data rate)
• Memory interface: 192-bit
• Total available graphics memory: 2047 MB
• Dedicated video memory: 768 MB GDDR5
• System video memory: 0 MB
• Shared system memory: 1279 MB
• Video BIOS version: 70.04.1B.00.60
• Bus: PCI Express x16

PPD: 9787

See the GPU3 FAQ for details.

SMP client

Benchmark machine: Macintosh Pro with 2 x 2.33 GHz Dual Core Xeon processors. Details:

• 2 x Woodcrest 5140 processors with 4 MB cache (each)
• 5 GB FBDIMM Memory (667 MHz DDR2), 1.33 GHz Bus)

PPD: 1760

See the SMP2 (pre-A3 core) FAQ for details.

New Gromacs cores: SMP2 (A3)

Benchmark machine: Quad core i5-750 ("Lynnfield" (45 nm)) class with Turbo Mode off. Details:

• Intel SR1630HGP Xeon 3400 1U Rackmount Server Platform
• Primary CPU Intel Xeon UP Quad Core X3450 2.66Ghz 8Mb Cache 95w
• 4 x DIMM 2048mb 1333 MHz Registered ECC DDR3 for SR1630HGP
• Seagate Barracuda ES.2 500gb Disk Drive 7200rpm 32mb Cache SATA 3.0Gb/s
• Intel Slim SATA DVD Drive

PPD: See the SMP2 (A3 core) new points FAQ for details.

Big adv

Bigadv is an experimental new FAH configuration, with a FAQ for the experiment posted in the FCF Forum.

Benchmark machine: Same as SMP2 (see description above)

PPD: normal SMP points * 1.5

Note that bigadv deadlines are also much shorter than normal deadlines. Also, in this experiment, we are seeing points variation as the FAH cores get upgraded (which can change the values based on a faster or slower calculation involved). We stress that many bigadv details are subject to change, as this is an experiment in a larger type of WU. The restrictions may change and the bigadv may end.

Frequently asked questions

Why use benchmarking to set the point values?

The original concept for benchmarking is very simple; give equal pay for equal work. Everyone who completes the same task will get the same reward. Fair and equitable! For example, everyone who completed a work unit in Project 2653 would earn the same amount of total points for that WU. This reference benchmark system of assigning points to projects works very well with consideration for the large variations in work unit size and complexity, and the nearly infinite combinations of computer hardware contributors use to process work units.

Please note that the very concept of a reference machine will mean that some WU performance will vary from the performance on your machine. Even between P4s, there are significant differences in architectures over the years. Moreover, variations between FAH WUs can also lead to small differences in benchmarking points.

Our goal is consistency within a given definition of a reference machine setup (described above for the CPU client). However, differences in memory speed, CPU cache size or speed, and even the typical data drift from one generation of a work unit to the next will all influence points performance up or down slightly. These and other variations will never allow any point system to perfectly reflect what you get on your machine. But the typical contributor will receive points within a very small deviation relative to the benchmark.

Our benchmark scheme is pretty simple: we would like to standardize benchmarks to a single machine and standardize and simplify the bonus schemes now employed. Bonuses have played a key role in aligning points with science and we will continue to use them. For example, returning work units (WUs) promptly can be very important for the science we’re doing, so we provide bonuses for this, especially with the high performance clients.

Why are some projects given significantly more points than others?

Certain projects require substantially more computer resources than others, either in terms of more disk space, more network transfer, or more RAM used. By default, these work units are given out to clients that opt in to request them. To reward those contributors for donating resources beyond the typical client, we currently give bonus points for these larger work units.

How much higher are the bonus points?

Currently the bonus points are a 50% increase over the standard benchmark point determination (described above). Please note that this value is subject to change at any time, and may be increased or decreased to better align points with specific scientific goals.

How do I configure my client to get bonus point WUs?

Please see our Configuration FAQ for these details.

Is there any risk of taking on bonus point work units?

These work units are larger and more experimental. For example, they often involve new cores, such as the SHARPEN core. Thus, one should not run big work unit clients on non-dedicated machines. The bonus points are a reward for contributing more resources to FAH, and so you should not be surprised if these work units impact system performance and use the full resources of the computer.

For More Information, Please See:

• FAH FAQ
• Console configuration documentation for information about console switches
• GUI documentation for Windows or Macintosh
• Folding Support Forum

Last Updated on February 16, 2011, at 12:20 PM by