2D versus 3D GPU performance

[SoNic]

Looking at my results with several modern GPUs I see that on the 3D performance I am well past 50% of the average scores, but on 2D I am well below average.
Below is a screenshot of what I mean. 2D performance is at the 2-1/2 stars (average), while 3D performance is at 4 stars (better than average).
Can anyone explain the discrepancy? Is the 2D performance bottlenecked by something else than the GPU cores, so that a lower performant 3D GPU has similar 2D performance like a better performance 3D one?

[David (PassMark)]

1) You don't need an expensive GPU for reasonable 2D performance.
2) Other factors influence the 2D score. Like CPU performance
3) Video card manufacturers aren't optimizing for 2D

[SoNic]

1) You don't need an expensive GPU for reasonable 2D performance.
2) Other factors influence the 2D score. Like CPU performance
3) Video card manufacturers aren't optimizing for 2D

I understand all of those. IMO the CPU differences should be averaged by numerous tests.

What's a mystery to me, is why my card seem to perform lower in 2D versus 3D - when compared to the average of those stars (cards)? What's the mechanism there of that averaging that's driving the discrepancy. My card is average in 2D but above average in 3D.

[David (PassMark)]

It is just that some card (or systems) are not equally good at 2D and 3D.
I agree it might be a reasonable expectation that all cards that are good at 3D are also good at 2D, but it isn't always the case.

[SoNic]

This is a RTX 3080Ti. Which cards are better at 2D than it?

Or is something else going on here?

[David (PassMark)]

I think you are assuming that expensive = fast.
But for 2D this isn't always the case.
1080 Ti is just as quick as 3080 for 2D, most of the time, all else being equal.
As mentioned above it also depends on what CPU the GPU is paired with.

We should do a graph with the best combos.

[SoNic]

The stats says that my 3080Ti, with a 2D score of 596 is just "average". That means that are cards that post a score better than that, doesn't? That's what an average value means.
So... I was wondering what cards are those? Or is something else.

Do we have a graph?

LE: I did use the search function in the website and ordered for 2D GPU performance, over 500. I see that there is a 3080Ti with score 1050, so I guess it's my CPU/memory performance that drags the number down.

[David (PassMark)]

We'll do a graph after the Christmas break.

But I am sure there are many factors at play here. Driver versions, maybe monitor resolutions and refresh rates, maybe G-Sync settings, memory and CPU speeds, etc...
We know already the meltdown and specter security patches had a really big impact (depending on CPU models) as they slowed down all O/S function calls and 2D does a lot of O/S function calls.

[SoNic]

Yeah, I have applied the BIOS patches for those two.
So maybe that's one of the variables.

[wonderwrench]

GPU's have very little to do with 2d performance as its mostly done in software by the CPU. Hardware accelerated 2d was for all intents and purposes killed off in Windows Vista and has been AFK ever since. In the good old days of Windows XP 2d flew at nearly the speed of light compared to Windows OS's of today. A great example is the windows interface test. Back in the day this test took about 5 seconds to complete on fast hardware. Time the test on your rig and you may have tears in your eyes and they are not tears of joy.

[David (PassMark)]

Story is slightly more complex. As there are various 2D programming interfaces in Windows. (GDI, GDI+, DirectDraw, Direct2D, DirectWrite, and the 3D APIs can also be used for 2D). Some of these have GPU acceleration for some tasks. So is hard to generalize as it has got pretty complex over the years.

[David (PassMark)]

So here it is.

The list of GPU/CPU combos that give the best 2D graphics performance

The surprise winning couple (for Jan 2022) is a GeForce GTX 1080 teamed up with a Intel Core i7-11700K.

2nd place is less of a surprise, a GeForce RTX 3090 with a Intel Core i9-11900K.

The chart more or less confirmed what we thought however. You get fast 2D from a CPU that has good single threaded performance plus a 'reasonable' GPU.
But there is very little difference between 1070, 1080, 2080, 3060, 3070 and 3080 video cards. So no point spending big on a GPU if you only use 2D, and not 3D.

Down at the bottom of the list we have the Intel HD graphics with a Celeron 847. Both known for being pretty rubbish.

[SoNic]

I think that result favoring single threaded CPU performance is unexpected. It's like MS didn't update that part of the Windows code for a decade or so...
Just repackaged it.

Thank you!

[David (PassMark)]

2D is stuff like line drawing, font rendering and BitBlts. It is hard to make this stuff multi-threaded.
But if you have multiple active 2D intensive apps running at the same time, then multiple CPU cores would be more important. This isn't such a typical use case however.