Build advice for 3D and video graphics machine (Please!)

[DanceswithUnix]

If the GPU is being used as a compute device, then frankly the output quality of the OpenGL drivers is irrelevant and so having a Quadro just doesn't seem important any more.

You want enough VRAM to make sure all the textures etc being rendered fit on the video card else the scene won't render.
You want something that is a monster at compute.

Don't think a Quadro 2000 will cut it tbh.

Edit to add: The Quadro 2000 is an underclocked GT450, a consumer card that retails new for £80. I would want 2GB of ram, and a lot more shaders.

[CAT-THE-FIFTH]

What about things like viewpoint speeds though?? If he is an architect,wouldn't most of his time be spent doing the modeling?? Unless doing it on the CPU only would work out faster??

[designosaur]

Most of my time is spent adjusting lights and textures and then of course waiting for renders to complete.

[Agent]

You want enough VRAM to make sure all the textures etc being rendered fit on the video card else the scene won't render.

Don't disagree with anything you've said, but just to add: This is only true when doing ray tracing calculations in the renderer. So things like VRay-RT and iRay

For none ray traced stuff, you don't need to fit it all in memory.

[Agent]

Most of my time is spent adjusting lights and textures and then of course waiting for renders to complete.

Quadro issues to one side - for the final render consider renting a render farm. It can be faster and in some cases, cheaper.

http://rentrender.com/all-render-farms-list/
http://www.renderrocket.com/

[CAT-THE-FIFTH]

Most of my time is spent adjusting lights and textures and then of course waiting for renders to complete.

That is why I am asking whether offloading it to the CPU would be faster or acceptable enough for the stages before rendering?? This is where,at least from what I gather from reviews,the professional cards tend to better overall. It would be nice if websites actually tried more consumer cards or even using software only too in their reviews.

Having said,it does make me wonder what Toms Hardware(even cgchannel when they polled their readers) were saying when they had image quality issues when using consumer cards. At least going by what is said here it is not the case. Websites trying to plug more expensive sales for companies?

[kalniel]

If the GPU is being used as a compute device, then frankly the output quality of the OpenGL drivers is irrelevant and so having a Quadro just doesn't seem important any more.

It's not about output quality in this case, I was recommending workstation cards/driver for their system stability. Desktop products graphics drivers are probably the weakest point of most modern systems.

[DanceswithUnix]

That is the tricky bit, there seems to be mixed messages out there. Using the graphics card in a traditional manner, you need a workstation card or you lose:

http://www.anandtech.com/show/5846/h...ion-heats-up/3

The GTX580 in those graphs gets mauled by a lowly re-badged GT450.

Whereas over here, you see a 560ti rendering in 35 seconds what it takes an i7-920 almost 3 minutes to do.

http://www.maxforums.org/threads/ira...test/0001.aspx

So the golden configuration would be to run a workstation graphics card for tweaking the design, and a pair of Tesla compute boards to do the final render.

We can't do that for £200, so I am left wondering what our sweet spot is.

Left field question: Just how good/bad is your current PC? Could you borrow a consumer nvidia card off someone just to try having it as a second video card in your current machine to try out GPU compute?

[CAT-THE-FIFTH]

Agent has made an interesting suggestion - perhaps the OP should look at that!

However, is it possible to run a professional and consumer in the system?? Have a suitable Quadro card,and a consumer card like a GTX560GB 2GB/GTX580 1.5GB,and switch between cards for different uses. Or would this be a recipe for unmitigated disaster??

[DanceswithUnix]

It's not about output quality in this case, I was recommending workstation cards/driver for their system stability. Desktop products graphics drivers are probably the weakest point of most modern systems.

Yet the strange thing is the Linux and Windows Nvidia drivers are supposed to be almost identical (hence you get it as a binary blob from Nvidia when you get the Linux driver), and yet I can't remember the last time my Nvidia driver crashed under Linux. The Linux driver for my Quadro card at work is the same as the one for my GTX460 at home.

The GT430 in my email reading work Win7 box blacks out the screen once every week or two and literally it just does web and email.

If you run Windows, your foundations are sand and anything you build on top will at some point sink.

[CAT-THE-FIFTH]

The GT430 in my email reading work Win7 box blacks out the screen once every week or two and literally it just does web and email.

Never had an issue with any Nvidia or AMD card I have run under Windows 7. Is the card overheating??

[DanceswithUnix]

Never had an issue with any Nvidia or AMD card I have run under Windows 7. Is the card overheating??

Nope, it is good spec Dell with plenty of ventilation in an air conditioned room. Can't find anything on the control panel that gives the core temperature though, perhaps GT430 doesn't expose that?

I see much the same at home on my dual boot machine, the screen blacks out sometimes under Windows but never under Linux. At least Nvidia are good at hiding graphics card reboots from the user these days.

[DanceswithUnix]

Just to say I think my reply to Kalniel didn't come over quite the way I intended (this is the Internet after all).

Basically I can't see any fundamental difference between a Quadro and a consumer graphics card. The difference seems to be in the BIOS that allows you to run the Quadro driver (under Windows). If the card is being used for compute, then I can't see that any of that matters.

[CAT-THE-FIFTH]

Nope, it is good spec Dell with plenty of ventilation in an air conditioned room. Can't find anything on the control panel that gives the core temperature though, perhaps GT430 doesn't expose that?

I see much the same at home on my dual boot machine, the screen blacks out sometimes under Windows but never under Linux. At least Nvidia are good at hiding graphics card reboots from the user these days.

GPU-Z,HWMonitor or MSI Afterburner should tell you the temperature. It could be also an artefact of the GPU changing clockspeeds and voltage - you can try fixing the clocks manually and see what happens,ie, stop it from changing between the various clockspeed states. Whenever I have the same issue, it has been down to me mucking around with the GPU core clockspeeds and voltage manually,ie, custom 2D and 3D clockspeeds.

[DanceswithUnix]

However, is it possible to run a professional and consumer in the system?? Have a suitable Quadro card,and a consumer card like a GTX560GB 2GB/GTX580 1.5GB,and switch between cards for different uses. Or would this be a recipe for unmitigated disaster??

This is why I asked if designosaur could try it before spending any money.
I know it wouldn't work in the big Dell I have at work, because the "workstation" bios seems to reject any video cards that haven't been approved by Dell and the machine won't boot. In a sane motherboard, I can't see that as a problem though.

[Agent]

However, is it possible to run a professional and consumer in the system?? Have a suitable Quadro card,and a consumer card like a GTX560GB 2GB/GTX580 1.5GB,and switch between cards for different uses. Or would this be a recipe for unmitigated disaster??

Since Vista Windows needs all GPUs to be using the same driver.

As the Quadro driver is actually different, I suspect it'll not work by default. No idea how far you'd get with ini / inf file altering and adding the device IDs.

This is getting a bit complex though for the OP I suspect