Folding Forum

foldinghomealone has been building a PPD database with help from the FAH community viewtopic.php?f=38&t=34510
Using that database, I created a tool to compare processors in Google Sheets. Simply select up to 4 processors using the dropdown menus and the chart will automatically display your results.

Download: https://docs.google.com/spreadsheets/d/ ... vDUs4/copy

Changelog:
V1
Initial release

V2
Adds 2 series to one graph

V2.1
Series labels
Y axis now scales with data
Hid Data & Scratchpad sheets

V2.2
Added Changelog sheet
Added version
Notifies if newer version available

V2.2.1
Linked to this post

V2.3
Added blank option
Linked version to changelog URL
Added 4 series to one graph
Created About sheet
Color coded series

V2.3.1
Fixed issue where graph displays NA if no results were found
Fixed issue where data points, which were cast as text, were not displaying

V2.3.2
New results went beyond set search range. Made search range unlimited.

V2.4
Points within 1000 atoms of each other are now averaged
Switched to linear trendlines

Wishlist/TODO:
Optimize to reduce resource usage and lag
Add table below chart that fills with information (min, max, average, standard dev, etc.) regarding current selection
"Data" sheet that allow filtering & sorting

Let me know if you have issues or feature requests.

Placehold

Nice one!
Thank you for the work!

NormalDiffusion wrote:Nice one!
Thank you for the work!

My pleasure! I just like graphs

FACT: The Radeon's performance is known to degrade when the atom-count goes down. This graph shows that to be true. I'm surprised that it's not also true for nVidia but to a lesser degree. The benchmarking team is succeeding.

bruce wrote:FACT: The Radeon's performance is known to degrade when the atom-count goes down. This graph shows that to be true. I'm surprised that it's not also true for nVidia but to a lesser degree.

All of the the Turing GPUs I looked at do increase performance with atom count, but with a much flatter curve. Interestingly, several of the older Pascal GPUs do the opposite and perform slightly worse with higher atom counts.

bruce wrote:The benchmarking team is succeeding.

That's what I like to hear

NoMoreQuarantine wrote: My pleasure! I just like graphs

Makes life easier

Thanks

Hopefully people continue to upload data for you

NoMoreQuarantine wrote:

bruce wrote:FACT: The Radeon's performance is known to degrade when the atom-count goes down. This graph shows that to be true. I'm surprised that it's not also true for nVidia but to a lesser degree.

All of the the Turing GPUs I looked at do increase performance with atom count, but with a much flatter curve. Interestingly, several of the older Pascal GPUs do the opposite and perform slightly worse with higher atom counts.

bruce wrote:The benchmarking team is succeeding.

That's what I like to hear

I think there are a lot of factors at play here. I lurk the Beta forum as I'm considering signing up, and it appears there are times when either camps modern cards return low PPD while my little onboard graphics blaze through them. Really small GPU WU's in particular seem to make some of the "power" cards lack, but one with an atom count of only 5000 returned almost twice my average PPD on my Vega 11.

If someone really dug deep into the entire GPU architecture, they might be able to figure it out. But until then, your charts will hopefully allow us to see the "sweet spots" in performance for various hardware. Then we can all see the WU's we hope to get when the servers roll the dice for us.