Yes, there are issues. They can be fixed.

[macfreek]

Yes, I have a few issues! No, not me personally. I mean with this script.

Most importantly, the Cybenetics website has changed, and the script no longer works. Furthermore, it misses newer PSU that have been released during 2024. This is important, because two new brands have uncrowned Corsair RM550x and Seasonic Prime Titanium. Corsair now barely makes the top 10 for efficiency at 20W loads.

Secondary, it reports the results of the 115V tests, and I'm more interested in the 230V test, and it does not correct all the errors on the Cybenetics website.

Of course, why not spend a few days fixing the script. And so I did.

But first, let's start with the results.

The Corsair RM550x (2021) barely made the top 10. Better performing (alphabetically):

• Cooler Master GX III 650 Gold
• Cooler Master GX III 750 Gold
• Cooler Master GX II 750W
• Cooler Master GX II 850 GOLD
• Seasonic Prime Titanium 650W
• Seasonic Prime Titanium 750W
• OMEN PS-4851-2 (850W)
• OMEN PS-4102-2 (1000W)

The OMEN and Cooler Master are from this year. These are all also very silent. Nice.

Whilst Cooler Master seems great here, note that not all modern Cooler Masters are that great. The Cooler Master M2000 Platinum (tested in 2022) only reached an abnominal 40% efficiency, rock bottom.

OMEN is not a brand I'm familiar with. My local shops don't seem to carry it. Thankfully, the Cybenetics website has a picture that shows this is a PSU manufacturerd by LiteOn, and branded for HP. The LiteOn website at https://power.liteon.com/ has a lot of info, but I can't find these two particular PSUs (which were the only ones tested by Cybenetics). We may have a case were the unavailable Corsair is replaced by an even more mysterious and unavailable OMEN akin HP akin LITEON PSU. Oh, Joy.

Cooler Master, thankfully, is more readily available brand, though it should be noted that the older Cooler Master GX II outperforms the newer Cooler Master GX III at 20W. So that also brings back memories how the RM550x has been replaced with less efficient counterparts.

Well, I see if I can upload the list, next to a pointer to the script. Link will follow in the comments.

A quick list what I fixed; the workarounds at the Cybenetics website; and a list of things that are nice to have, but I probably won't add. It was nice working on it, but also for me, it was a side-project.

First the list of things I fixed:

• Fixed your name. Maybe I should clarify. It seems that you copied the LICENSE file from some other project, without change the name of the person listed as copyright holder in that file. That, or I just reveiled your secret identity. I kind of doubt that.
• Fixed the URL of the Cybenetics overview page (params=2,1, no longer works. It's params=2,1,0 now)
• Properly escaped that regexp. My editor was complaining. "\d" is wrong. It is either "\\d" or r"\d". The last is more readable.
• Got rid of Selenium. Oof, that made things slow and complicated. No thanks.
• Get rid of PDF parsing. Thankfully, Cybenetics publishes it on HTML pages nowadays, and all the PDF links no longer work.
• Use functions instead of one long script. I know what is coming. Getting rid of Selenium removes some lines of code, but we will add more later.
• Spend way too long figuring out why my links index.php?option=database&params=2,1,0 end up as index.php?option=database¶ms=2,1,0. Well, it turns out Python's html.parser.HtmlParser always decodes entityrefs, even when it should not. It interprets &params as ¶ms and replaces it with ¶ms. Go figure. And it can not be turned off. I tried, trust me. So dumped HtmlParser and replaced it with the much more reliable html5lib parser.
• Use logging for output
• Lots of rewrite to get the code working again
• Even more rewrites to work around all the errors in the Cybenetics database or Cybenetics website.

Which errors you ask? I'm glad you asked.

• Light Load tests result disappeared:
  • Enermax Platimax D.F. 500W - tests can no longer be found (it was in the previous spreadsheet)
  • Aerocool Aero Bronze 650W - tests can no longer be found (it was in the previous spreadsheet)
• Missing test
  • Aerocool ACP-650FP7 - test are not present (neither in the previous spreadsheet nor now)
• Malformed table (data found):
  • SilverStone Decathlon DA1650 - Both 115V and 230V results are listed under 115V. The script has no issues, as it now verifies the actual voltage (e.g. 230.26V) and uses that to deduce which test this was.
• Malformed table (data manually found):
  • SilverStone Strider Platinum 750W - Table comes after the Light Load section, instead of in the section. The script only reads data within the section, and misses it. Checking all tables was too cumbersome, so I added these manually.
• Malformed table (some data found):
  • Super Flower Leadex III Gold 550W rev.1 - missing 40/60/80W 115V, and all 230V tests (the data in the previously spreadsheet were wrong)
  • Super Flower Leadex III Gold 650W rev.1 - missing 40/60/80W 115V, and all 230V tests (the data in the previously spreadsheet were
  • Super Flower Leadex III Gold 750W rev.1 - missing 40/60/80W 115V, and all 230V tests (the data in the previously spreadsheet were wrong)
  • Kolink Enclave 500W (#2) - HTML is seriously malformed, and 40/60/80W 230V tests are missing
  • Thermaltake Toughpower PF1 650W - HTML is seriously malformed, and 40/60/80W 230V tests are missing.
  • SeaSonic Prime Titanium 650W - 20/40/60W 115V tests report only zeroes. Previous spreadsheet had results here, but those seem completely different, so I wonder if it is a different test, or the previous spreadsheet was wrong.
  • Super Flower Leadex III Gold ARGB 550W (#2) - HTML is seriously malformed, only the 20W 230V test is there, whilst the previous spreadsheet reported all 115V tests.
  • - HTML is seriously malformed, and 40/60/80W 115V tests are missing
• Odd Datapoints:
  • SilverStone Strider Platinum 1200W (#2) - 115V 40W test is only 36.4W
  • ABKO STH-1000B-M - 230V tests run at 220V
  • Apexgaming AG-850M - 230V 20W test is run at 165V; 230V 60W test is run at 265V
  • Be quiet! Pure Power 10 500W - 230V 80W test is run at 250V
  • Be quiet! Straight Power 11 550W - 230V 40W test is run at 265V
  • EVGA SuperNOVA 650 G3 - 115V 40W test is run at 160V
  • SeaSonic Prime Titanium Ultra 650W - 115V 60W test is run at 310V
• Uncommon light load tests:
  • SeaSonic Focus SGX 500W - Tests ran at 40W/60W/80W/100W instead of 20W/40W/60W/80W
  • Montech Titan Gold 1000W - Tests ran at 10W/20W/30W/40W instead of 20W/40W/60W/80W (but were named 20W/40W/60W/80W)
• Wrong test name:
  • 1st Player Steampunk 850W - Tests were named 10W/20W/30W/40W instead of 20W/40W/60W/80W (but ran at 20W/40W/60W/80W)

There are some more questionable results. For example, the SilverStone HELA 2050R Platinum is tested twice: (#1) and (#2). There is a 9.4% difference in efficiency at 20 Watt and 11.7% at 40W. That's a big difference.

Finally, I was surprised that 50 of the results disappeared altogether. They are just no longer list or available in the Cybenetics database. This is particularly true for Corsair, where duplicate tests of the same PSU were removed. For example, the three RM550x (2018) tests disappeared, leaving only the RM550x (2021) test results. It is however, not always the last test that disappeared. For the Corsair AX1600i, the last test (Sample #3) remained; but for the Corsair HX750, Sample #2 disappeared, leaving only Sample #1. Not surprising, the samples that performed a little better remained. This is not true for all disappearing test, but is true for all Corsair tests. Usually it is just the last test that remains.

Anyway, the script was able to retrieve most results, and made sure it reported it at the correct place (Voltage and load). If no result could be found, the data from the previous spreadsheet was copied. It does not remove questionable results, because I can't judge that. Thankfully, this is only the case for very few results. Despite the above list, most results seems very reliable, and the issue is only in formatting.

While my modifications are freely available at https://github.com/macfreek/efficient_psu, you may notice that I have not made a pull request. The reason is that the script still has a few limitations:

• It uses a personal cached downloader script, that writes HTTP downloads to cache, and returns that when the URL was fetched again, even across multiple runs. This was invaluable when writing workarounds for the HTML errors, and made sure the script was both faster and not giving unnecessary load on the Cybenetics website. However, while the cached downloader script is publicly available, it is not a stand-alone pip package, and my attempt to use another package (like Pooch) were not successful. The current code tries to include the cached downloader, and if that fails, gracefully resorts to probing the fetching the URLs as usual. That said, this does not feel like production code. Of course, you are welcome to include that code here (I made sure to match the license), but be warned that I do not plan to maintain it.
• The script does not fetch any price data. I planned to get it from German geizhals.de, Dutch tweakers.net, and Amazon.com, but never did. It was so much quicker to simple go to those website, copy & paste the results in a text file, ran a couple of regexps, and just have a table ready.
• The script does not try to incorporate any 80PLUS tests. Those are publicly available at https://www.clearesult.com/80plus/manufacturers/230V-EU-Internal, and it would be interesting to see if the 10% load results are similar to Cybenetics. Would give a good indication of the quality
• As it stands, the script can retrieve 115V and 230V results, or both. I copied all to a big table (also ensuring to include the 50 tests that were removed). It also download the noise levels in dB(A) for various test, and the cable design (modular or not), since this was a purchasing criteria for me. It is all nice, but the table it rather big. Huge even. And perhaps not very readable for everyone because of that. I did make a trimmed down version for myself (removing duplicate tests and properties I did not care about), but don't plan to automate any of that.

If you want, I can still make a pull request. And otherwise, feel free to use my table in any way you see fit, and/or simple leave this issue open for other who are interested in the code.

Enjoy.

[macfreek]

And here is the overview (uploaded to a branded ownCloud service):

• Cybenetics Full Overview.xlsx (901 records, 30 columns, you may want to trim this a bit)
• Trimmed Overview 230V.xlsx (746 records, 21 columns, bit more manageable)

If you want to use the full version, please note:

• Columns M-U contain the results for 115V tests (or 230V if 115V was not available)
• Columns V-AD contain the results for 230V tests (or 115V if 230V was not available)
• Gray cells contain manually corrected data

I plan for these downloads to remain available for at least one year (oct 2025), but will not make any guarantees after that. Feel free to download and use or republish (at your own risk). While I obviously did my best to ensure everything is correct, there is no guarantee. I recommend to remain critical.