Sapphire Toxic 5850 or Fermi 470? For new AMD build?

[GoogleIntel]

GTX 470 is more closer to HD5870 than HD5850 in performance.

 

[Sean W.]

plus Nividia cards run hotter and use more power.

the only thing i like about Nividia is that you ca use it to increase your CPU score in 3dmark by using the physix(sp?) feature

 

[CalcProgrammer1]

I've gone back and forth a lot. Hated my X1600Pro due to poor compatibility so I went nVidia with my laptop, but had overheating issues with that and was convinced by people here on Tech Forums to go back to ATi instead of the GTX470. I got a 5870 and it's amazing, it isn't as fast as the GTX480 but it is still ridiculously powerful, supports 3 monitors (Eyefinity), and is more power efficient than nVidia's new cards. This means it runs cooler and quieter than a 470 would and you don't need as big of a power supply either.

As for the compatibility problems that I hated my X1600Pro for, they're pretty much gone now. Back then ATi had very VERY horrible Linux support and I couldn't get even the most basic of 3D rendering support on my card (Compiz, the 3d effects desktop that Linux has, worked on my old nVidia GeForce MX420 but not on the much better X1600Pro due to driver issues). However, after AMD bought ATi, they released a lot of previously restricted information about the Radeon hardware and open-source driver projects quickly brought solid support to all of ATi's cards. Today, ATi provides a solid proprietary Linux driver as well as a growing open-source, community driven driver. nVidia still only provides proprietary drivers but they have always been solid and powerful. On the Windows front, both companies have good driver setups.

The only thing nVidia has that ATi doesn't is CUDA. This in turn means PhysX but it really isn't a big deal. Both ATi and nVidia have implemented OpenCL as well as DirectCompute in their latest products. OpenCL is a high-performance GPU computing system similar to CUDA that complements OpenGL and is cross-platform. DirectCompute is another high-performance GPU computing system that is part of the DirectX 11 specification. nVidia also has CUDA (its existing GPU computing system that was introduced in their DX10 cards starting with the 8000 series) but most likely CUDA driven apps will begin to replace it with OpenCL or DirectCompute for better compatibility.

 

[Apokalipse]

I can say I am really happy with my 5870.
Not only is it really fast, but it also runs pretty cool and doesn't use a lot of power.

Though I would suggest you look for one with a reference cooler (often called the 'batmobile' cooler)

Anyway, CUDA/PhysX is pretty much an attempt to control the market standards. And I won't support that.
DirectCompute and OpenCL are used by Nvidia and ATI.

Directcompute is good because it's already part of the DirectX 11 specification.
OpenCL is good because it works well with OpenGL, and it can be run by the GPU or the CPU, or both a the same time.

I think I would like to see Nvidia either open the CUDA/PhysX standards, or just drop them in favor of DirectCompute and OpenCL, and change their strategy to try and achieve low power consumption and price while still having high performance.

 

[CalcProgrammer1]

Seconded on the reference cooler. I was originally looking at the Sapphire one but went with a Diamond 5870 because it had the reference cooler. The reference cooler is very solidly built, looks sleek and cool, and most importantly keeps the card cool without making too much noise.

With the reference cooler you get the reference PCB (circuit board). This original board design features the best voltage regulation of all the 5870 boards and thus can help you overclock if you want to do so. Many people complained about the Sapphire card not having good voltage control.