HD3000 Driver Performance Comparison

[MSchm21]

Comparing i5-2500k with HD3000 onboard graphics driver version 8.15.10.2266 with latest official driver 8.15.10.2361. Tested in 1280x1024 (except Mafia 2/Just Cause 2 1024x768, Final Fantasy 1280x720, Riddick 1280x960) with various game settings. Update: added 8.15.10.2401

Application 8.15.10.2266 8.15.10.2361 8.15.10.2401 3DMark VantageP1.734P1.751P1.7523DMark 06425842564264Crysis35,9838,8037,34Doom 336,746,947,6Far Cry64,8169,1070,17Hawx 2546771Prey38,136,236,2Serious Sam HD435353Dirt 244,545,345,9Half-Life 243,4842,9143,08Civilization 5505050Anno 140432,0044,8745,58Far Cry 243,7444,6944,57Stalker Pripyat81,7585,0085,25Mafia 219,719,719,7Lost Planet 236,536,737,1Battleforge41,844,644,3World in Conflict414344Riddick Butcher Bay36,5648,8348,97Farming Simulator145555Call of Duty 241,1545,8046,15Streetfighter67,2368,8368,91GTA 431,5331,8431,79Just Cause 228,9530,2630,68Final Fantasy XIV686709844Unreal Tournament 200462,4970,8874,01

[idata]

Hello,

Could you run 3Dmark06 at 1280x800 resolution and post what score you get?

I'm getting around 4255p with 1280x800 on MBP 13" 2011 i5

Cheers,

[MSchm21]

I cannot change the resolution on my Basic version of 3dmark06.

[idata]

Thanks for taking notice.

If you look at http://www.techyalert.com/2011/02/25/macbook-pro-2010-vs-macbook-pro-2011/ http://www.techyalert.com/2011/02/25/macbook-pro-2010-vs-macbook-pro-2011/ you can see he's getting around 4629p in 3Dmark06.

I'm really curious how he is getting that good score? Were there any better drivers prior to these drivers you compared?

I have identical MBP as his.

Cheers,

[MSchm21]

Basic resolution for 3dmark06= 1280x1024 - and that's what I have tested to get a score around 4260. To get a score of 4630 points with 1280x800 seems plausible. I can't see any anomaly in this score. Keep in mind some i5 notebook models does have a higher turbo frequency (1200-1300 Mhz) compared with my i5-2500k @1100 Mhz.

[idata]

Hi Yups, could you benchmark some of the games at the "low" settings? I mean verify the claims shown here:

http://downloadmirror.intel.com/20037/eng/Graphics%20driver%20release%20notes.pdf http://downloadmirror.intel.com/20037/eng/Graphics%20driver%20release%20notes.pdf

From what I am guessing, the gains are mostly limited to lower resolutions and settings.

[MSchm21]

I cannot verify it. Low game configuration does mean 640x480, 800x600 or which resolution? I tested some of the games in lowest settings as well, but with 1280x1024 excluding Mafia 2. Dirt 2, Crysis, Battleforge Hawx 2 for example. There are performance gains in this games, obviously not as much as mentioned in their release notes for me. Different CPU and benchmarks could make a difference though. Especially the benchmarked demo/save game can differ and bring different results. To verify it you have to run the same Benchmarks with the same settings and same hardware. The resolution shouldn't make a difference as long as the GPU limited the performance which is the case in almost every game with Sandy Bridge @IGP.

[idata]

Resolution and settings could likely make the difference. It looks like the 2361 build addresses deficiencies regarding OpenGL and hardware VS/T&L. You've got a big gain in Doom 3, which is an OpenGL game for example. I've seen numerous benchmarks regarding geometry processing, the gains are usually biggest in the lowest resolutions. Anyways, its not a simple "optimization patch", it enables a particular performance feature.

I have a 2600K system using HD Graphics 3000, which is closer to their config, but benchmarking is not simple. I will try running but I have limited disk space.

-Some games are easier to verify because they have a benchmark program. From there you should be able to check Batman: Arkhan Asylym/Hawx 2, Crysis Warhead/Resident Evil 5

-Crysis Warhead performs better and is more optimized than Crysis. Since its an updated version of the same game, might be worth it to test it instead.

By the way guys, performance is also somewhat dependent on CPU. It'll be mostly GPU limited, but if you have a 2410/2415M chip, you will perform less than those with 2520M or the quad core chips. The 2410/2415M Macbook Pro 13 also seems to have lower graphics frequency as a OEM specific part than the regular ones.

[MSchm21]

You are mostly GPU limited with a Sandy Bridge Quadcore and integrated graphics for gaming purpose. If possible I've used an integrated Benchmark. There are integrated Benchmarks or standalone Benchmarks used for Crysis, Lost Planet 2, World in Conflict, Battleforge, Dirt 2, Doom 3, Hawx 2, Stalker and Mafia 2. Far Cry 1/2 and Prey used with a Benchmark Tool and Half-Life 2 with a self-made timedemo. Yes there are big gains in OpenGL games.....not in all though. Prey for example was faster with the old driver. Quake 4 which is not on the list got a big boost similar to Doom 3.

[MSchm21]

To add another note, the driver 8.15.10.2372 includes different claims on a different system. To be specific with Core i7-2920XM this time.

See here:

http://www.abload.de/img/2372s7p1.png http://www.abload.de/img/2372s7p1.png

Low setting means 1024x768 for Intel. Far Cry 2 gained in DX10 mode a lot more than with DX9 mode. I tested Far Cry 2 in DX9 mode. All this affects the results.

[idata]

Yups, I tested 3 driver versions. The 2361 driver gives nice improvements in Quake 3 Arena and Crysis Pre-Beta, the only game benchmarks I tested. Little improvements elsewhere too.

15.21.10.2291

Quake 3 Arena

-640x480 Normal: 336.8/336.6/337.6, 337.6/337.2/337.1

-1280x1024 High: 209.2/208.9/209.8, 207.9/209.8/209.43DMark06 Vertex Shader: 175.440Mvertices/67.767Fillratetest 1.13: 12058MB/14924/17883/1537MTexels/3071 SiSoft Sandra Lite 2011.2.17.47 Graphics Memory Bandwidth:

7.72GB/15.75/3.78, 15.75GB/75.65%/5.54GB/2.58GB

Crysis 1024x768 Low GPU: 46.98FFXIV: 134

15.21.13.2342

Quake 3 Arena Demo

-640x480 Normal: 335.0/334.2/335.4, 336.9/337.7/336.5-1280x1024 High: 209.8/209.2/209.3, 208.3/208.6/208.93DMark06 Vertex Shader: 176.735MVertices/67.840Fillratetest 1.13: 12101MB/14922/17861/1541MTexels/3075SiSoft Sandra Lite 2011.2.17.47 Graphics Memory Bandwidth:

7.89GB/16/3.87, 16GB/77.24%/5.76GB/2.6GB

Crysis 1024x768 Low GPU: 47.10FFXIV: 138(+3%)

15.22.1.2361

Quake 3 Arena Demo

-640x480 Normal: 381.1/383.1/384.0 = 14.3% average gain for Demo 1, 385.3/385.0/380.6 = 13.0% average gain for Demo 2 -1280x1024 High: 249.6/250.6/250.7 = 19.5% average gain for Demo 1, 252.5/253.7/253.0 = 21.3% average gain for Demo 2 3DMark06 Vertex Shader: 176.770MVertices/67.867Fillratetest 1.13: 12126MB/15124/17893/1565MTexels/3067SiSoft Sandra Lite 2011.2.17.47 Graphics Memory Bandwidth:

8GB/16.15/4, 16GB/77.62%/5.81GB/2.62GB

Crysis 1024x768 Low GPU: 59.49 = 26.3% gain FFXIV: 142(+3%)

(-MB means MB/s)

The big gains in applications like Q3AD and Crysis Pre-Beta that doesn't have any additional patches or updates suggest that they enabled something in the hardware that wasn't there before.

[MSchm21]

Your Crysis gain is better than mine. Regarding OpenGL it's surely the activated OpenGL 3.1 Hardware Transform and Lighting. Riddick improved from 36.56 to 48.83 which is OpenGL based.

[idata]

http://www.anandtech.com/show/4084/intels-sandy-bridge-upheaval-in-the-mobile-landscape/7 http://www.anandtech.com/show/4084/intels-sandy-bridge-upheaval-in-the-mobile-landscape/7

You mean Chronicles of Riddick? From the 2820QM review from Anandtech, it didn't even run it properly on the early drivers. They sure have come a long way.

Edit: I ran another benchmark testing geometry performance for OpenGL. I haven't posted the results because it does 100 tests or so. I've took a look at the numbers and the gains are amazing. The geometry throughput increased by 3-4x.

[MSchm21]

I mean Chronicles of Riddick Escape from Butcher Bay, benched with the integrated timedemo. Dark Athena is a different version, don't know how it performs. Never tried it.

[idata]

If you have the time, run Crysis comparisons with 1024x768 this time. That's what I ran at.

[MSchm21]

Crysis Pre-Beta Benchmark_GPU

8.15.10.2266

1024x768= 43.04 fps

1280x1024= 35.98 fps

8.15.10.2361

1024x768= 51.99 +20,8%

1280x1024= 38.80 +7,8%

I'm surprised there is a much bigger gain in the lower resolution.

[idata]

Interesting! The 2361 build fixes things related to geometry performance and like I said, the gains are greatest in the lower resolutions. Now I just need to find a synthetic benchmark that shows a difference in DirectX so we can see in detail what they improved on.

[idata]

I finally found a synthetic benchmark that shows differences with DX.

3DMark06

2342:

-Pixel Shader:127.620

-Vertex Shader: 185.827MVertices/67.763-Shader Particles(SM3.0): 35.891-Perlin Noise(SM3.0): 45.031

Batch Size Tests

8 Tri/32/128/512/2048/32768: 1.971MTri/7.910/31.864/64.946/66.918/67.316

2361:

-Pixel Shader: 129.686-Vertex Shader: 176.971MVertices/67.875-Shader Particles(SM3.0): 35.960-Perlin Noise(SM3.0): 45.033Batch Size Tests

8 Tri/32/128/512/2048/32768: 2.450MTri/9.664/37.673/62.622/64.111/64.459

In the batch size test, the 8 Triangles/32 Triangles/128 Triangles show a gain of 24.3%, 22.2%, and 18.2% respectively.

More in depth Crysis tests, also Unigine Heaven:

Crysis GPU benchmark

2342:

-800x600 Low: 48.91

-1024x768: 47.02-1280x1024: 42.42

-800x600 Med: 24.30

-1024x768 Med: 22.90-1280x1024 Med: 18.41

-800x600 High: 17.44

-1024x768 High: 14.31-1280x1024 High: 10.17

2361:

-800x600 Low: 64.46(+31.8%)

-1024x768: 59.47(+26.5%)-1280x1024: 46.79(+10.3%)

-800x600 Med: 31.71(+30.5%)

-1024x768 Med: 25.72(+12.3%)-1280x1024 Med: 18.70(+1.6%)

-800x600 High: 18.98(+8.83)

-1024x768 High: 14.42(No gain)-1280x1024 High: 10.21(No gain)

Unigine Heaven(Default settings, except renderer and resolution)

2342

DX10:

800x600 - FPS: 18.1, Score: 456, Min FPS: 12.9, Max FPS: 30.8

1280x1024 - FPS: 7.9, Score: 200, Min FPS: 5.0, Max FPS: 13.7

OpenGL:

800x600 - FPS: 11.8, Score: 298, Min FPS: 5.8, Max FPS: 22.0

1280x1024 - FPS: 5.4 , Score: 137, Min FPS: 2.6, Max FPS: 9.9

2361:

DX10:

800x600 - FPS: 17.9, Score: 452, Min FPS: 12.5, Max FPS: 30.3

1280x1024 - FPS: 8.0, Score: 202, Min FPS: 6.1, Max FPS: 13.8

OpenGL:

800x600 - FPS: 13.5, Score: 340, Min FPS: 9.4, Max FPS: 21.7 +14.1%

1280x1024 - FPS: 6.8 , Score: 170, Min FPS: 4.9, Max FPS: 10.3 +24.1%

Note on OpenGL: With 2342 driver, you won't notice the sun shining on some parts of the screen. Its like it doesn't exist at all, and the whole thing looks more dull than on DirectX. There are also frequent rendering errors, which are shown as white geometry shape in the center of the screen.

[MSchm21]

Regarding Unigine OpenGL, with driver 2345 they fixed this matter.

Unigine: Heaven Benchmark 2.1* application now runs and does not hang during loading screen in OpenGL mode

Root Cause/Resolution:  

Application uses pairs of vertex + fragment shaders that are not valid (according to our driver). In some cases, a gl_Texcood[x] varying is used in fragment shader, but it is not written in vertex shader. OGL Spec says, that this causes undefined behavior. Fix: immediate variable of zero value is used instead of gl_Texcoord[x].

[MSchm21]

HT4U with another driver test: http://ht4u.net/reviews/2011/treibervergleich_intel_2361/ http://ht4u.net/reviews/2011/treibervergleich_intel_2361/

[MSchm21]

Added driver 8.15.10.2401 with some new games. Basically slightly faster here and there in most games except Crysis. Three anomalies noticed performance wise.

- Prey runs faster with previous driver 8.15.10.2261 (36,2 vs 38,1 fps or ~5% loss)

- Crysis runs faster with previous driver 8.15.10.2361 (37,34 vs 38,80 fps or ~5% loss)