2D, Acceleration, And Windows: Aren't All Graphics Cards Equal?

[ljbade]

Incase you want to know what GDI operations are currently accelerated look here:
http://msdn.microsoft.com/en-us/library/dd445349.aspx

Currently covers:
FillRect,
BitBlt,
StretchBlt,
AlphaBlend,
TransparentBlt,
ClearType

and all the ROP2/ROP3 combinations of the above.

Note that lines\outlines, bezier curzes, round parts of ellpises\rectangles are not accelerated.

This makes sense as CPU faster at 2D drawing than GPU (as GPU only does 3D).

But the GPU can perform very fast memory operations on rectangular regions.

 

[rubberjohnson]

[citation][nom]tsnor[/nom]Outstanding article. Bet the author was freaking at the first measurements and grinning at the AMD/ATI "We now have our driver team engaged to optimize this path and will release a new driver to address this workload as soon as possible."[/citation]
Yeah cause only AMD have these 2D issues, do you Nvidia fanboys even read the articles or just jump to the conclusion then bash in the forums?

 

[nforce4max]

The Cyrix cpu in one of the pics I found one working with original stock cooler. amazing find it was.

 

[matt87_50]

so really... what the hell is the difference between 3D and 2D save for a few matrix operations on the vertices? maybe have the depth buffer switched off?

the ONLY conceptual difference between 3D and 2D is that all reference points (such as vertices) get converted from 3D to 2D with a simple Vector times Matrix operation. and that is negligible compared to all the operations done at a 2D level.

for instance, a graphics card only processes triangles, maybe dx11 has added some of the more complex GDI curved shapes back with "native" support by using the tessellator?

 

[hixbot]

great article, it's about time some of these problems are getting attention. stick it them toms!

PS. It sure would like to see these benchies with many, many more cards.
integrated cards, workstation cards, more gaming cards etc.

 

[hixbot]

no edit button, dang it! oh well, you all get my meaning.

 

[ufo_warviper]

Wow, Iv

 

[ljbade]

@matt87_50:
You are basically describing what Direct2D does.

Normal GDI cannot be easily converted to 3D as there is no way to cache\batch the vertex buffers. Also Direct3D would not render pixel-for-pixel identical to GDI.

Direct2D forces you to write your code around vertex buffers which is significantly faster.

 

[edlivian]

W7 32bit Intel Core 2 Duo P8600 2.4ghz 4gb ram
Intel GMA x4500HD

BENCHMARK: DIRECT DRAWING TO VISIBLE DEVICE

Text: 25853 chars/sec
Line: 26558 lines/sec
Polygon: 2534 polygons/sec
Rectangle: 1940 rects/sec
Arc/Ellipse: 5934 ellipses/sec
Blitting: 5451 operations/sec
Stretching: 380 operations/sec
Splines/Bézier: 17229 splines/sec
Score: 930

 

[Guest]

W7 64bit Core 2 Quad Q9300 2.5GHz 6GB DDR2 667MHz
NVidia Quadro NVS 290 PCI-e x16

BENCHMARK: DIRECT DRAWING TO VISIBLE DEVICE

Text: 26483 chars/sec
Line: 39401 lines/sec
Polygon: 10147 polygons/sec
Rectangle: 2484 rects/sec
Arc/Ellipse: 7203 ellipses/sec
Blitting: 2908 operations/sec
Stretching: 443 operations/sec
Splines/Bézier: 24564 splines/sec
Score: 1201

It sure beats my gaming Q9550 + HD 5770 (Score 740)

 

[TheGreatGrapeApe]

[citation][nom]cangelini[/nom]Benchmark was included in Part 2, which is still in translation in the US. I'll pass along the feedback re: those legacy cards to see if the German team has them available![/citation]

Actually I grabbed it from the news article about the first one.

Would be nice to see S3 and Matrox, since otherwise it's really a 2 pony show with intel's (primarily desktop-only) graphic solution thrown in.

I think S3 would be another nice 3D + Media centric IHV and Matrox a desktop + Media IHV, especially since Matrox's focus is primarily 2D now but also not big enough that M$ changes their WDDM/Vista implementation for them (much to their chagrin), so a good comparison.

Anywhoo, just my pet peeve about the impression that there was still a possibility of suggestions for the next test, but part 2 had already been done for theirs, as you can see by my first comment about inclusion, didn't know it initially.

Anywhoo just picky here, overall a very interesting investigation, and like that it's getting everyone thinking 2D too and hopefully working on improvements to match XP performance. It would be interesting to see the affect on apps like CS4, GIMP, Vegas, etc.

And perhaps a follow-up investigation into the default colour accuracy on a known quality LCD (which should have much less varaiability than a CRT which needs alot of tuning and also needs a good 10mins warm-up).

 

[cangelini]

Ah, but there will be a Part 3 with the fix in place, as well That's where I'm hoping to get the folks overseas to broaden their look with some of the suggestions you guys are making here! Unfortunately, I haven't had the chance to look at Part 2 yet.

 

[overdrive5k]

I'd like to see a S3 Virge thrown into the mix to teach these new cards a lesson about 2D acceleration and a S3 Chrome 540 GTX.

Cyrix-VIA-S3 fan

 

[brendonmc]

Could this be the 'Watergate' of the graphics card industry????
At any rate, proffesional software packages mentioned in this article like Autocad require workstation class graphics cards to run optimally. Lets hope that part 2 analyses the performance of Firegl and Quadro cards with respect to 2D performance.

 

[Guest]

I agree that it will be very interesting to know if the Quadro/FireGL variants have full ( or better) 2D acceleration. We'd be hearing of differences in their firmware (and drivers) as the main technical advantage. Professional 3D is clearly better on those chips, but lot of pro apps run in 2D.

Testing IGP is also of note, maybe those awful GMAs aren't that bad after all.

On a Win7 note, that's probaly why I am force to use that awful Aero Theme on Win7, as soon I disable it and go back to good old Win2000 look, graphics refreshes slow down to snail speed.

Maybe is all a marketing strategy to force people to use DWM and have the new "cool" look of Win7 on their desktop.

 

[Chetou]

C2D E7300 @ 3.6 GHz
ATI Radeon 4670
Win7 x64 Aero ON
Catalyst 9.11

BENCHMARK: DIRECT DRAWING TO VISIBLE DEVICE

Text: 54585 chars/sec
Line: 43116 lines/sec
Polygon: 4575 polygons/sec
Rectangle: 6329 rects/sec
Arc/Ellipse: 10165 ellipses/sec
Blitting: 9136 operations/sec
Stretching: 773 operations/sec
Splines/Bézier: 34176 splines/sec
Score: 1845

How come my scores are substantially higher than 4870's in the article?

 

[brisingamen]

whats next? benchmarking at 640 x 480?

 

[lostcomma]

Finally after asking this question for so many years.
In fact every time they did their monthly gaming card
comparo for value and said if you dont game don't waste
your money on these cards.(Which is even more truer now)
So what should I spend my money on for good 2d. Which really
is by far a larger segment than gamers.
Great article.
I always wondered just where a matrox millennium would compare
to todays cards. I may not find out but it nice to have some
reference.

 

[brisingamen]

2d performance barely means squat,

nvidia fans/writers will do anything to take a shot at the other company while they are on top eh.

the funny thing is because amd is so great they are going to optimize its paths anyway and when they do it will yeild a 20x performance increase.

ive never been so bored reading an article in my life partially due to the fact i dont believe half of it.

 

[Narg]

Wow, what a waste of an article. Unless you are married to 20 year old software, you would never and should never turn off Aero. Even then, how would this affect you? It won't. This whole idea is a waste of time and resources.

 

[Narg]

[citation][nom]zachsaw[/nom]I think what's clear about ATI is that they skim on driver development until someone starts complaining.[/citation]

Same goes for nVidia. Their DX9 implementations were aweful. They only fixed them when a developer screamed about it.

 

[jaybus]

portlandtom: "And all this time I thought it was a bug in my software."

Same here. Well, I was leaning toward a bug in GDI relating to the 2.5D / z-buffer re-drawing. The author remarked that it was sad that developers were redrawing entire windows rather that just the affected portions. Well, this is why I did exactly that. Otherwise it didn't freaking work. I just had no idea that it was the GPU driver and/or hardware and instead blamed Microsoft. Who would have thought the GPU manufacturers would have screwed up 2D acceleration after all these years? Kudos to Tom's.

And yes, I said screwed up 2D, because I will be very surprised if the hardware primitives aren't there. My guess is they've done something really stupid in the driver. That goes for the Linux driver too, and would explain why the FOSS driver is so much faster than AMD's.

Not that nVidia is off the hook. Their $400 card can't best a cheap IGP either, and again, my guess is that they too botched the driver. Both companies get a D- in test and QA.

 

[edlivian]

W7 64bit Intel Core 2 Duo E6600 2.4ghz 4gb RAM
Nvidia GT 240 512MB GDDR5 (Stock)

BENCHMARK: DIRECT DRAWING TO VISIBLE DEVICE

Text: 27886 chars/sec
Line: 38324 lines/sec
Polygon: 12376 polygons/sec
Rectangle: 4201 rects/sec
Arc/Ellipse: 7592 ellipses/sec
Blitting: 6396 operations/sec
Stretching: 497 operations/sec
Splines/Bézier: 23878 splines/sec
Score: 1333

 

[jmmazuch]

ATI 5870 core i7@3200 win7 x64 10.1beta + stream 2.0 SDK

PAssmark 7

all 2D test ok, except those
FONTS and TEXT
258 - 176/aero on/
Windows interface
130 - 61 /aeron on -(/


2D with 5870 is real shame

 

[Guest]

Other than synthetic benchmarks, very few (if any) modern apps use things like lines and arcs. They are basically API vestiges from an earlier time. All that most "2D" apps use these days are blits (copies), solid fills, and alpha blends for AA fonts and translucency effects. Most users won't perceive any performance issues with a lack of acceleration for this older operations.