3 ms·
There is a lot of misinformation in replies to this post. The Tesla cards do NOT have 4x more DP performance (the whole Fermi generation, which includes Tesla a
by mrb 16y ago
There is a lot of misinformation in replies to this post. The Tesla cards do NOT have 4x more DP performance (the whole Fermi generation, which includes Tesla and gaming cards, have 4x the DP perf of the previous GT200 generation). Tesla cards NOT have higher accuracy. The differences between buying a Tesla card and a gaming card are:
- Tesla cards have ECC memory, video cards do not.
- When you buy a Tesla card, Nvidia supposedly gives you very good tech support (I've heard you can call their engineers directly)
- Tesla cards have more GDDR memory: 3GB for the C2050, 6GB for the C2070, compared to 1.5GB for the highest end gaming card GTX480.
Some people are willing to pay a premium for the above 3 advantages, that's why Nvidia can sell Tesla cards at such a high price. IMO, the first advantage is overrated: Nvidia's ECC implementation (secdec) is unable to detect many multi-bit errors. Gaming cards offer at 10x better perf/price ratio, so it makes it hard to ignore them.
- malkia 16y agoWe had NVIDIA folks coming to our game studio, and they've said almost the same but for another set of applications - Basically if you want full support for Autodesk or Adobe products (3DS Max, Maya, MotionBuilder, Photoshop) then you should get Quadra card. The Quadra has ECC, it's produced by NVIDIA itself (not 3rd party), and they have developed extensive driver support where each application that you run (Windows) would have specific settings that give you the best performace, quality or stability. There are tons of other manifacturers that produce NVIDIA cards, but they don't put ECC memory, boost the GPU frequency until they almost don't burn, etc. Also some specific features exist only for Quadra, that are not important for games - for example quick antialiased lines (just a simple example, there is more). So what MRB is saying is right. At the surface it looks Quadra/Tesla are worse, but their stability is much better, and customer support is guaranteed. NVIDIA can't simply support HW not built by them in large. It's not possible for them to know every kind of card there is.