GPU Buffers: Introduction to OpenGL 4.3 Shader Storage Buffers Objects
OpenGL 4.3
Shader Storage Buffers
GPU Buffers: Introduction to OpenGL 3.1 Uniform Buffers Objects
OpenGL 3.1
Uniform Buffers
Simple Introduction to OpenGL 4 Shader Subroutines
Intro to OpenGL 4
Shader Subroutines
Lorenz Attractor (Butterfly Effect)
Lorenz Attractor
Butterfly Effect
Mesh Exploder with Geometry Shaders
Mesh Exploder
Geometry Shaders
GLSL Hacker
GLSL Hacker
v0.7.0.0

Posts Tagged ‘heat’


[Geek3D-Test] FurMark, Catalyst 8.8 and HIS Radeon HD 4850: Torture Test PASSED!

9 comments - What do you think?  Posted by JeGX - 2008/08/29 at 17:54

Categories: Catalyst, FurMark, Graphics Cards, Test   Tags: , , , , , , , , , , , , ,

Ok now that I have in my hands a HIS’s Radeon HD 4850, I played with this card and of course did the FurMark renaming experience with Catalyst 8.8 (see ATI Optimizes Catalyst 8.8 to be FurMark-Proof!). The difference of score in simply… incredible, better it’s shocking! Here is the score when I launch FurMark.exe: 2234 points


And now, the score when FurMark.exe is renamed in… ati.exe (why not?): 4383 points


This score is almost twice the first one. That explains now the odd result I get with FurMark in this post: ATI Catalyst 8.8 vs 8.7: OpenGL Performance Drop.

I launched in the same time the Catalyst Control Center in the Overdrive panel to check how frequencies vary:


In both tests, the frequencies were the same. ATI Catalyst 8.8 does not downclock the GPU frequency but makes the GPU running slower (what does it means???). Anyway, it seems obvious the Radeon HD 4800 series have some serious thermic problems. Maybe I could add a kind of GPU temperature limitation for Radeon HD 4800 series. If temperature exceeds 90 degrees on Radeon HD 4850 and HD 4870, FurMark will stop the rendering or will render one frame over two or three… But actually I think I’m not going to do that. FurMark is a torture test but it’s most of all a standard OpenGL application. I don’t use low level code or different rendering path for Radeon or GeForce. The same code is injected in both rendering pipeline. And then any OpenGL 2.0 compliant GPU should process this code if the surrounding graphics hardware (I mean memory modules, power MOSFETs properly cooled, etc.) is well designed and implemented by graphics cards makers. The proof, HIS’s Radeon HD 4850 PASSED all FurMark tests I did, with stock clocks as well as with overclocked clocks.

Now dear readers, let’s burn HIS’s Radeon HD 4850.

- Stock clocks: GPU=625MHz and Memory=993MHz

  • 94 degrees after 3 minutes in 640×480 no AA windowed mode: TEST PASSED!
  • 96 degrees after 3 minutes in 1024×768 no AA windowed mode: TEST PASSED!

- Basic Overclocking: GPU=660MHz and Memory=1005MHz

  • 96 degrees after 3 minutes in 1024×768 no AA windowed mode: TEST PASSED!

- Overclocking: GPU=680MHz and Memory=993MHz

  • 96 degrees after 3 minutes in 1024×768 no AA windowed mode: : TEST PASSED!

- Overclocking: GPU=680MHz and Memory=1100MHz

  • 97 degrees in 1024×768 no AA windowed mode. After 140 seconds, VPU Recover:

Yes the latest overclocking was a little bit extreme, but it shows HIS’s Radeon HD 4850 is a very good product and is already FurMark-Proof. No need hidden tweak in Catalyst to run FurMark. I think graphics cards that do not resist to FurMark torture tests are either bad quality products or have some bad cooled parts like the VRM (voltage regulator modules)…

Related Links: