Dedicated PhysX Card Experiment - How Powerful Does it Have to Be? Linus Tech Tips
1 - 20 of 83 Posts
Joined
·
1,183 Posts
Quote:
to be clear, that was a 8600
![]()
Originally Posted by alchemik  Wow the results actually surprised me, guess people will stop saying all you need is a 8800 card for physx. |
Joined
·
9,337 Posts
I'm sorry, but linus is a straight up noob. He was using the bottom pci-e slot for the 8600gts, and the middle pci-e slot for the gtx 550 ti and likely the other cards as well. This creates a problem because the bottom pci-e slot is running off of the chipset's pci-e lanes, where as the middle pci-e slot is running off of the cpu. I bet you this is why his results are so extreme.
Joined
·
3,742 Posts
Quote:
A G92 8800 is a whole lot more powerful than a 8600. Considering a 8600 is 53% as powerful as needed to run it optimally, a card roughly twice as powerful will do the trick. A GTS250 or 260 would probably be overkill.
Quote:
Why would this be an issue? Even PCI-e 2.0 x4 would probably not bottleneck a 8600.
Originally Posted by alchemik Wow the results actually surprised me, guess people will stop saying all you need is a 8800 card for physx. |
Quote:
Originally Posted by Behemoth777 I'm sorry, but linus is a straight up noob. He was using the bottom pci-e slot for the 8600gts, and the middle pci-e slot for the gtx 550 ti and likely the other cards as well. This creates a problem because the bottom pci-e slot is running off of the chipset's pci-e lanes, where as the middle pci-e slot is running off of the cpu. I bet you this is why his results are so extreme. |
Joined
·
9,337 Posts
Quote:
Because there is a decent difference between pci-e bandwidth for the chipset and for the cpu. There is much higher latency going through the chipset, and we don't even know if the chipset is creating issues for physx because of driver issues or what not.
All i'm saying is that I don't really believe these results unless the test method is exactly the same across the board, which it isn't.
Originally Posted by angrysasquatch Why would this be an issue? Even PCI-e 2.0 x4 would probably not bottleneck a 8600. |
All i'm saying is that I don't really believe these results unless the test method is exactly the same across the board, which it isn't.
Joined
·
9,337 Posts
Quote:
Yeah, it amazes me how many people get paid to review hardware when half the time they don't know what the hell they are doing. Linus is a noob, I don't know why he has this job.
Originally Posted by mbudden Dude is a prick. His "testing" is BS. |
Joined
·
688 Posts
Quote:
I wish I was paid to review hardware - I at least have some idea of what I am doing.
![]()
Originally Posted by Behemoth777 Yeah, it amazes me how many people get paid to review hardware when half the time they don't know what the hell they are doing. Linus is a noob, I don't know why he has this job. |
Joined
·
84,195 Posts
Anyway, this thread is an excellent addition to OCN because it proves that what we have been saying all along is correct: make sure that whatever card you are using for PhysX processing is not so slow that it bottlenecks the main card.
Joined
·
1,247 Posts
You can't run a ATI card like a 6950 with a dedicated Phys X card without hax right?
Joined
·
2,241 Posts
HDR-Sucks.
Bloom-Sucks.
Morphing mesh textures- Waving grass and stuff suck.
Internal and external shadows- Suck.
All my shaders ever have to do is simple texture and light transforms and havok.
So all I have to say is
*cough* havok *cough*
Oh I do like wavy water so I do let shaders do those.
Bloom-Sucks.
Morphing mesh textures- Waving grass and stuff suck.
Internal and external shadows- Suck.
All my shaders ever have to do is simple texture and light transforms and havok.
So all I have to say is
*cough* havok *cough*
Oh I do like wavy water so I do let shaders do those.
Joined
·
84,195 Posts
Quote:
??
Originally Posted by Hephasteus HDR-Sucks. Bloom-Sucks. Morphing mesh textures- Waving grass and stuff suck. Internal and external shadows- Suck. All my shaders ever have to do is simple texture and light transforms and havok. So all I have to say is *cough* havok *cough* Oh I do like wavy water so I do let shaders do those. |
Joined
·
1,247 Posts
Quote:
Sounds like someone isn't happy that they can't run phys X
Originally Posted by Hephasteus -Ranting- |
Joined
·
3,742 Posts
Quote:
So then sort of like a southbridge, which was used on all 775 stuff..?
I do agree that using different slots is a bit odd, and probably not the best testing procedure, but I don't think there is any actual performance differences between using a bridge chip and going direct to CPU.
Oh, and PhysX can die in a volumetric fire calculated with havok physics.
Originally Posted by Behemoth777 Because there is a decent difference between pci-e bandwidth for the chipset and for the cpu. There is much higher latency going through the chipset, and we don't even know if the chipset is creating issues for physx because of driver issues or what not. All i'm saying is that I don't really believe these results unless the test method is exactly the same across the board, which it isn't. |
I do agree that using different slots is a bit odd, and probably not the best testing procedure, but I don't think there is any actual performance differences between using a bridge chip and going direct to CPU.
Oh, and PhysX can die in a volumetric fire calculated with havok physics.
1 - 20 of 83 Posts
