Quadro 2000 vs. GTX580

(Mixen) #1


For my new pc I’m looking at either a Nvidia Quadro 2000 or a GTX 580. I’m primarily using Blender with cycles and several Adobe products - and almost no gaming.

From what I hear the Quadro is a workstation card optimized for pro apps like Blender? But from benchmarks I find that the GTX is far more powerful for GPU rendering like cycles?

So which card should I choose? The price difference is not an issue.

(arexma) #2

Never too lazy to answer for the 100th time :slight_smile:
Never too tired to talk about graphics hardware.

With the Quadro2000 you´re good for 3dsmax, maya and others, but Blender has no real quadro support, so there isn´t a real benefit. And without the Quadros features used to it´s fullest, I say as an assumption, a GTX580 is propably the faster card.
I don´t find the screenshot anymore so I am not certain, but I think one of our machines in the studio has a Quadro2000 to be able to cut movies in Stereo3D, a feature that does not work with a GeForce driver (thanks nvidia) and the Blender performance was worse than my GTX470 at home.

Your problem is, you most likely need both:

GeForce: Fast DX, crippled OpenGL, crippled CUDA
Quadro: Fast OpenGL, crippled CUDA
Tesla: NO graphic connector, fulll CUDA performance.

I have no idea if cycles runs with single or double precision and currently I don´t follow cycles that closely, I am waiting for OpenCL to be fully included and for the Radeon HD7000 series which is still expected this year, and the GF600 series expected early next year.

Personally I´d say get a 560 or a cheap 470/480 on ebay if you need more power right now and wait how it develops.
My bet is that Blenders future is in Radeon and OpenCL if nvidia follows it´s current company strategy:
Generate one GPU that excels at DX, OpenGL and GPGPU, cripple the features not needed for a particular type of customer.

Another path would be to buy a Radeon for viewport, and ram a GeForce in a free slot to use it for CUDA. It works, had it in my system, there´s a thread in… dunnow i think Offtopic where I posted it.

Basically it boils down to this:

  • The Quadro would give you no real benefit in Blender, it would just circumvent the deliberate bad performance in OpenGL a GeForce has. Might change in the near future as nvidia sponsored hardware to the Blenderfoundation, and has shown interest in Blender, so there might be Quadro drivers for Blender on the way sooner or later.

  • The Quadro card has it´s double precision capped to 1/8 of the single precision, while it would be capable of 1/2 the performance of single precision. Unfortunately I have no relieable source on how much the GeForce is crippled in it´s CUDA performance, but it´s certain from various blogs/forums that it is. With the Quadro it´s an open secret - nvidia released the info.

It really depends on your usage as well. Genreally I´d say a 580 is plenty for gaming, Cycles and Blender´s OpenGL.
A Quadro or Radeon would be faster for the OpenGL viewport.
A Tesla would be the choise for hardcore rendering because of the speed, let alone because the card is made to run at 100% for years - I actually doubt in this case a GeForce could handle that without having kickass watercooling - also the Tesla has EEC registered RAM.
If you need a lot of video memory, there´s now way around a Quadro, as a GeForce comes with puny 1.5GB, or you throw in a kidney, your wife and a dog for a 3GB version :wink:

I am loosing my hair already about this whole issue. I am a long time nvidia customer, deliberately not saying fan, because I don´t give a flying fork. I buy what satisfies my needs and meets my requirents best for my money.
Currently it´s nothing. You can just choose the lesser evil.

(Mixen) #3

Arexma, thanks for the great answer! Really helpful!

(holyenigma74) #4

Here is something new read it…


(s12a) #5

(sorry fur bumping this relatively old thread, I got here through Google while searching information about Nvidia’s policy on crippling OpenGL performance on newer Geforce cards)

So, a few months later, after the Radeon HD7xxxx series got releasedand given that Cycles will slowly improve its performance and compatibility with AMD cards and the OpenCL path, If I had some money - let’s say 300-350 euro - to spend just on a single card (due to limited budget and hardware compatibility because of power supply and motherboard constraints), would you still suggest buying a Radeon? I don’t play 3D games - same as user Mixen.

By the way, I suggest that a suite of performance “regression tests” should be created to carefully monitor changes in OpenGL, OpenCL and CUDA performance with Blender as new consumer GPU get introduced to the market. I can’t say I’m extremely happy after recently finding out that on Geforce cards OpenGL performance is crippled on purpose.

(Herbert123) #6

That is why I decided to test the 7970 and the older 5870 (did you read my test?) And it seems that, at this point, you will get good/great opengl performance with a 5870 (except in textured mode) - most of the time the cards were limited by the way Blender operates (and slows down), not the other way around.

The problem with nVidia still persists, with people having more or less success increasing opengl performance in Blender by turning off double-sided faces - yet not for everyone, it seems.

(reC) #7

(sorry fur bumping this relatively old thread, I got here through Google while searching information about Nvidia’s policy on crippling OpenGL performance on newer Geforce cards)

Yeah same here ! But I thought I should bump it some more because my problem is the same as Mixen’s

Anything new going on around the GPU sector ? It’s been several months from the last post. The new 680 4gb vRam is out. The 685 is rumored to be released.

I’m quite unsure on what to buy. I would buy the 680 if it would be faster in cycles than the 580.

From the results here it seems that the single precision for the 680 is better than the 580. Is that relevant to rendering with cycles ?