PPD. Reliable metrics and performance comparison.

If you're new to FAH and need help getting started or you have very basic questions, start here.

Moderators: Site Moderators, FAHC Science Team

Post Reply
schernichkin
Posts: 4
Joined: Thu Feb 09, 2017 7:21 am

PPD. Reliable metrics and performance comparison.

Post by schernichkin »

Good afternoon, gentlemen!

I've joined folding@home few weeks ago and it's run just fine. Yesterday I've decided to compare my system with other gentlemen's hardware to ensure I've made proper setup and my hardware runs at optimal level. I've discovered that most commonly used metric (and the only metric I've found database for) is Parrots Per Day. The itself database is published here:

Code: Select all

https://docs.google.com/spreadsheets/d/1vcVoSVtamcoGj5sFfvKF_XlvuviWWveJIg_iZ8U2bf0/pub
But the results I've got was quite discouraging. According the database my average performance should be 146,814 PPD (Radeon HD 7970) per unit. I have two 7970 watercooled to 60°C, running 24x7 at stock frequencies and monitoring tool indicates constant 97-99% load. My results is about 23-24k PPD per unint (about 49k in total). This is more than 6 times slower than average performance reported by other users with same hardware. How could such thing be? Is PPD reliable enough to compare computing performance in F@H, or some other metrics should be used?

I'm not looking towards Parrots them self, but 6-times difference make me think that I've did something extremely wrong.
Joe_H
Site Admin
Posts: 7856
Joined: Tue Apr 21, 2009 4:41 pm
Hardware configuration: Mac Pro 2.8 quad 12 GB smp4
MacBook Pro 2.9 i7 8 GB smp2
Location: W. MA

Re: PPD. Reliable metrics and performance comparison.

Post by Joe_H »

Welcome to the folding support forum.

Usually such a large discrepancy in expected points versus what is received is due to not using a passkey. A passkey is required to get the Quick Return Bonus, for faster GPU's these bonus points are a significant part of the PPD being reported. Please see the Passkey FAQ. With a passkey it takes the successful return of 10 WU's before the QRB is awarded for future WU's as long as at least an 80% return rate is maintained.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: PPD. Reliable metrics and performance comparison.

Post by JimboPalmer »

While I also suspect no pass key, (and am amused by the Parrots Per Day theory, Jimmy Buffet would be proud) several items would help us help you.

1) the name you are folding under

2) the first 100 lines of your log showing your configuration.

3) at least one portion of the log showing a WU complete (may be in the above)
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
ComputerGenie
Posts: 236
Joined: Mon Dec 12, 2016 4:06 am

Re: PPD. Reliable metrics and performance comparison.

Post by ComputerGenie »

I agree with the passkey assessments; however, as a side note:
That "database" is only meant to compare 1 specific PRGC on 1 specific card at a time.
i.e.:
Project10494, Run5, Clone36, Generation303 performed on CardX compared to Project10494, Run5, Clone36, Generation303 performed on CardX
HagegeR
Posts: 5
Joined: Fri Jan 27, 2017 7:17 am

Re: PPD. Reliable metrics and performance comparison.

Post by HagegeR »

maybe you took fairly large WU: most of the point you will make with a GPU is based on how fast you complete the WU compared to the time it's expected to finish.
if a WU timeout in 10 days but you finish it in an hour, you will likely get 10 times the base credit, if your WU takes more time, you get a lot less bonus and thus a lot less PPD.
ComputerGenie
Posts: 236
Joined: Mon Dec 12, 2016 4:06 am

Re: PPD. Reliable metrics and performance comparison.

Post by ComputerGenie »

HagegeR wrote:maybe you took fairly large WU: most of the point you will make with a GPU is based on how fast you complete the WU compared to the time it's expected to finish.
if a WU timeout in 10 days but you finish it in an hour, you will likely get 10 times the base credit, if your WU takes more time, you get a lot less bonus and thus a lot less PPD.
That could be too. So much is wrong with most conversations that revolve around that damn sheet (especially since it rarely gets updated and all of the updates come from posts in a single forum); I want to scream every time someone mentions it. :x
schernichkin
Posts: 4
Joined: Thu Feb 09, 2017 7:21 am

Re: PPD. Reliable metrics and performance comparison.

Post by schernichkin »

Thank your for reply, gentelmen. I've added my passkey and my PPD jumped close to the database values almost instantly.

Now, knowing that Parrots Per Day not just raw folding performance metric but fancy combination of different factors such as "Quick Return Bonus", second question comes into play. How good is it for comparing performance and how does it scaled? For example, if my graphic card runs at 146K PPD at average and GTX Titan X Pascal runs at 1442K PPD does than mean that GTX Titan X Pascal will complete same folding workunit 10 times faster?
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: PPD. Reliable metrics and performance comparison.

Post by JimboPalmer »

I do not think the QRB is linear, if you can return WUs 10 times quicker, I think you get more than 10 times the points.

I also believe each WU returns how long it took, so you get more complex or simpler WUs in the future, so it is possible those two cards would never get similar WUs to 'race' with
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
7im
Posts: 10189
Joined: Thu Nov 29, 2007 4:30 pm
Hardware configuration: Intel i7-4770K @ 4.5 GHz, 16 GB DDR3-2133 Corsair Vengence (black/red), EVGA GTX 760 @ 1200 MHz, on an Asus Maximus VI Hero MB (black/red), in a blacked out Antec P280 Tower, with a Xigmatek Night Hawk (black) HSF, Seasonic 760w Platinum (black case, sleeves, wires), 4 SilenX 120mm Case fans with silicon fan gaskets and silicon mounts (all black), a 512GB Samsung SSD (black), and a 2TB Black Western Digital HD (silver/black).
Location: Arizona
Contact:

Re: PPD. Reliable metrics and performance comparison.

Post by 7im »

QRB is exponential. With a fast enough GPU, you could earn Infinity Points™.
How to provide enough information to get helpful support
Tell me and I forget. Teach me and I remember. Involve me and I learn.
Joe_H
Site Admin
Posts: 7856
Joined: Tue Apr 21, 2009 4:41 pm
Hardware configuration: Mac Pro 2.8 quad 12 GB smp4
MacBook Pro 2.9 i7 8 GB smp2
Location: W. MA

Re: PPD. Reliable metrics and performance comparison.

Post by Joe_H »

Well with an infinitely fast GPU you could, but otherwise as they say normally in math it approaches infinite.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
schernichkin
Posts: 4
Joined: Thu Feb 09, 2017 7:21 am

Re: PPD. Reliable metrics and performance comparison.

Post by schernichkin »

So, I'd like to write resume. Correct me if I'm wrong.

If I understood correctly, PPD is not even linear. Practically this could mean that dual GTX 1070 could do more job, than one GTX Titan X Pascal, despite the GTX Titan X Pascal will produce more PPD.

On the other hand, PPD formula was picked by project authors to reflect project's actual needs. They should be aware of impact of the PPD formula when new user deciedes which hardware to use. Current formula stimulates to use fewer but faster GPU than more but less fast, despite the fact that raw throughout could be lower. Probably, project authors have good reason to do it in this way, which I not aware though.
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: PPD. Reliable metrics and performance comparison.

Post by bruce »

schernichkin wrote:If I understood correctly, PPD is not even linear. Practically this could mean that dual GTX 1070 could do more job, than one GTX Titan X Pascal, despite the GTX Titan X Pascal will produce more PPD.
That's true. It's easiest to explain if I talk about WUs for the CPU, but the same answer applies to GPUs.

Suppose your system has multiple CPUs and you complete two similar WUs. You have a choice between allocating all of your CPUs to one project at a time or splitting the resources so that both WUs are being processed concurrently. Using all of your CPUs, let's say the WU is completed in a day but using half of your resources, each WU will take two days. Completing a WU in one day and a different WU on the next day moves science along faster than delaying each of them by an extra day -- and the Quick Return Bonus (QRB) reflects that fact.
Leonardo
Posts: 261
Joined: Tue Dec 04, 2007 5:09 am
Hardware configuration: GPU slots on home-built, purpose-built PCs.
Location: Eagle River, Alaska

Re: PPD. Reliable metrics and performance comparison.

Post by Leonardo »

Schernichkin, here is an online Quick Return Bonus calculator to estimate F@H points per day. It is very accurate. I think you'll find it very interesting. http://www.linuxforge.net/bonuscalc2.php
Image
Post Reply