Basic Question Concerning PDF Rendering Test

[Götz]

Hi there

First of all congrats and thank you for making a super usable benchmark software. Been through a few and yours is the only one that I manage to use consistently to glean insights without a degree in hard- and software engineering.

I am also struck by the low PDF rendering score and have read everything closely related on this forum. All my othe scores are reasonably good to outstanding. My system is fresh out of the (individual) boxes, Win 11 Pro, PT 11.1

So before I dive even deeper into this conundrum, my first questions are:

What does the test "PDF Rendering" actually measure?
Is it relevant to anything else than rendering a PDF?
Can it point to a different problem?

[David (PassMark)]

The PDF rendering benchmark involves loading a 210 page PDF file (24MB in size) that contains a mixture of graphics, text and graphs and cycling through the pages displayed as fast as possible.

About 40 year ago a page description language PostScript was invented. It was used mostly for printing complex documents (really any document that wasn't just plain text). But there wasn't many options for displaying PostScript documents on a screen and the output could look different on different devices e.g. different fonts.
About 30 years ago a new version of PostScript arrived called. PDF, Portable Document Format. It supported embedded fonts and images, hyperlinks, and a bunch of other stuff to make it more useful as a way to create & share documents. Plus they looked the same on all devices (screens and printers). The internal format of a PDF file was very complex however.

Initially PDF was just a standard from Adobe, but in 2008, PDF became an open standard (ISO 32000). With Windows 10 Microsoft started to include built in support for the PDF format (it was part of the Edge browser), but any software could use the functionality to display a PDF, without needing to ship a huge amount of code their product.

Obviously PDFs are now hugely popular. So it makes a good performance benchmark.

This is what this benchmark test is doing. Opening this moderately large PDF, using the build in Window PDF render. So the test won't work on Win7 and below as a result. The PDF file's content is held in RAM, so the performance is not based on disk speed. Parsing the PDF file (a mix of binary code and some text) is a CPU intensive task. Rendering the text and embedded fonts and images should be a GPU intensive task. But the split of load between the CPU and GPU depends on the hardware & device drivers (and probably your version of Windows/Edge). Microsoft's PDF parser seems to be mostly single threaded. It is certainly the case that CPU's single threaded performance can be the bottle neck, meaning higher end GPUs don't get loaded up too much, and don't add much to PDF display performance. CPU cache size & RAM speed probably also has an impact, the run-time memory usage is around 60MB, which is at the top end of current CPUs Level 3 cache sizes.

So there are many hardware dependencies for rendering a PDF.

[Götz]

Thank you very much for this extensive reply which helped me quite a bit to give it some perspective! So it's complex...
The issue with Win7 I already gathered form other replies, which, finally, I will eventually move away from with this new build.

I wanted to edit my question because one part I immediately solved for myself after I sent it - but couldn't since it was my first post and needed approval.

It measuers OPs per second, as it says on one of the charts.
57/58 is awfully close to the 59.95 Hz of the monitor(s) I am using - could there be some connection?

My real issue was a persistent state of an abysmal result of 7.7, which kept me troubleshooting for 1.5 days including a parallel fresh install of Win11 with no internet connection to be able to control the drivers I want to install next. It turned out to be a complete waste of time though. I shall share my solve here because the problem did pop up a few times in the forum already - maybe somebody can take something away from my experience.

It all started with a single unexciting monitor (1920x1200 @ 60Hz) that I had plugged in with an adapter cable from DisplayPort to HDMI (for reasons). I set up the OP with nothing more than Windows Updates and most of the MB-software plus Firefox with 4 plugins. It all went swimmingly - or so I thought - until I noticed the 7.7 in the PerformanceTest PDF Rendering score. Some post of yours led me to experiment with a second monitor at the iGPU with a real HDMI cable and voila, it went up to 57/58. But nothing I did could improve the 7.7 which, ultimately, led to the fresh parallel install of Win11 on a different nvme. After many hours and much rebooting and benchmarking, everything stayed ok on the monitor with the adapter cable in the fresh install. It did not change in the initial install however, thick-headedly remaining at 7.7 - I started to suspect a misguided, over-enthusiastic user-intervention during the initial setup. Nothing came to mind though. In an act of desperation I began to swap out the monitors and cables and their ports. And now the big solution: after having the adapter cable removed from the first port I had plugged it in and chosing a different one, the bad result of 7.7 has never been seen again, even when I plugged the adapter cable back into the notorious port with the same monitor. From that moment onwards, it would be 57/58 rock solid on iGPU or my 4070, save for a few slightly slower scores which I attribute to background operations. No idea what might have happened, but it clearly must be related to monitor and/or cable and/or port PLUS windows settings (since it didn't happen on the same hardware connection on the fresh install). I did delete the monitors as devices in the settings in between switching them around, which might or might not have helped.

So there you go, I hope this might lead to something somewhere, somewhen...

BTW
Antec Flux Pro - Asrock Z890 Taichi Lite - Core Ultra 7 265k - Thermalright Phantom Spirit 120 Evo - Asus Proart 4070 Super OC - Corsair VENGEANCE DDR5 2x32GB 6600 CL32 - Samsung 990 PRO M.2 PCIe 4 (2x1 and 1x2 TB) - 1000 Watt be quiet! Pure Power 12 M - HP ZR2440w (2)

[David (PassMark)]

Your scenario sounds like some type of driver bug. Strange that the bad behavior didn't come back after switching the ports back. But if it is fixed, that's good.

It measuers OPs per second, as it says on one of the charts.
57/58 is awfully close to the 59.95 Hz of the monitor(s) I am using - could there be some connection?

Ops = Operations. Which in this case is the number of pages displayed per second.

And yes, as rendering in Edge and it's libraries use DirectX, (Direct 2D and maybe DirectWrite?) monitor refresh rates can also be a limiting factor. Especially on fast hardware connected to an older 60hz monitor. So turn on 120Hz+ if you have it !

[Götz]

Yes, the wish for a faster monitor did creep up on me during this process. Alas, it will stay at 60 Hz for the near future - energy consumption just ramps up everywhere and more than 60 is not strictly necessary for what I am doing.
That fits with an afterthought though that I wanted to add anyway: Like other users have mentioned, the displayed PDF rendering went fast for like ten pages and then paused for a second or so, which resulted it to only get to a certain page (the same very recognizable one), which means the result of 7.7 was quite stable. However, when I looked at the Test-PDF in Edge or Firefox, I was able to zip through the whole file in an instant and there was no visible lag whatsoever. BUT, there is still no preview available in the file explorer, which seems to be a whole different rabbit-hole.
So maybe, just maybe, the bug really only gets triggered by the way the benchmark is run? Because at least in my scenario it does not seem to have had much effect on real life performance.
That is also why I have decided to leave it alone for now. Thanks again for your great support!

[David (PassMark)]

Don't know why it would pause on a certain page for multiple runs. But then after a change in monitors / cables not pause.

There are reasons I could imagine for a pause that happened all the time for an arbitrary PDF (e.g. loading missing fonts in the PDF, or checking digital signatures, or rendering a very complex vector diagram, or running out of RAM and needing to swap to disk). But for it to come good and stay good seems to rule out all these explanations.