hi all,

I have a rendering system running with SLI QuadroFX 4400’s, 8GB DDR2 and a Quad-Core 6600 with XP64.

Problem I seem to have is bottlenecks.

My 3DMark05 score is ~5500 and SpecViewPerf10 scores are consistently 40-50% lower than the system baseline scores shown here.http://spec.unipv.it/gwpg/gpc.data/vp10/summary.html.

i upgraded the PSU to an OCZ 600W and checked the systems power draw was well under max and stable.

i know 3DMark is DirectX orientated and SpecView is OpenGL but i would expect more on both scores given the setup and specs.

All drivers are current from Nvidia and Microsoft and i also have the Maxtreme performance driver (for 3DS Max2009).

I need to get the most out of this stoopidly expensive system.

any help or ideas would be much appreciated.

Have you investigated whether the 4 cores are splitting up the tasks properly?  I know there are numerous issues with this...for instance a few years ago when dual cores came out, I was stunned to discover that a computer with a Win XP Home rendered twice as fast as an identical one with XP Pro.  I can only assume this was the XP Pro system splitting up the rendering process(es) in a way that conflicted with the way Max was trying to split it up?

--3ds Max Design 2009 64bit, Direct3D 9, Win Vista Premium 64, 3ghz quadcore Xeon, Quadro FX1700--

Max is - as far as i know - a single threaded app. so it doesnt push work onto more than 1 core at a time. it does split the work among them but one instruction at a time. so you only get 25% total output. ive monitored the usage patterns and heat output to assess the workload on the CPU and it is thus. edit At least when SpecViewPerf is running, it is running as a single thread test anyway. Ensuring Max is optimized for quad-core may come later /edit

its the cards i am more concerned about. For the money and given the RAM available to the CPU, i’d have thought the system would fly.

But its scores are sometimes half that of lesser machines.

keep in mind a lot of those benchmark apps from futuremark are designed for gaming cards. your card is a pro DCC card and might therefore not have on-board certain “effects” that the benchmarking app needs to run. at least this was the case a few years ago with the old quadros. things might have changed now, but back in the day a crappy old £50 geforce 2 mx would outperform a £800 quadro in certain benchmark apps.

also what HDD are you using? i assume at least 7200rpm or perhaps raptor 10ks? slow HDDs will affect all sorts of things.

alex york
for and on behalf of Atelier York | Bespoke Architectural Visualization
http://www.atelieryork.co.uk
MentalRayTips Twitter Feed

yes the Futuremark benchmark picks up on Direct3D performance while the Quadro’s are OpenGL optimized for 3D.

but it gives me an idea of comparisons for other cards we have here (FireGL and Radeon/GeForce).

harddrive is standard 7200RPM Sata.

its the SpecView results that alarms me.

Whilst the performance the designers are getting is ‘ok’, i’m just curious why the scores are low and whether I can somehow optimize the system to get much better (i mean in all fairness, i dont care what the scores are, i am just using them as pointers)

has anyone worked with or played with hardware and driver optimizations using max09?

Is the question here about the viewport or the rendering? Becos pl4yd34d is interested in the graphic card
and alexyork is talking about Harddisk and CPU which is the rendering side of things… confused..

during modeling stage you use mainly the graphics card. It updates the viewport. While when the work is
done , and rendering takes place ... cpu, ram and harddisk takes over. Graphics card don’t do any rendering only
refreshes realtime on your viewport. Gaming Graphic card is design to do dx or openGL realtime , imagine
a first person shooter that does cpu rendering ... all too slow. :P

Studio Max 2009 x64
X5000 Chipset | Dual Core Intel 5140 | 4G RAM | Nvidia FX3450 | WindowsXP 64

victor.nsy 05 September 2008 01:36 AM

Is the question here about the viewport or the rendering? Becos pl4yd34d is interested in the graphic card
and alexyork is talking about Harddisk and CPU which is the rendering side of things… confused..

during modeling stage you use mainly the graphics card. It updates the viewport. While when the work is
done , and rendering takes place ... cpu, ram and harddisk takes over. Graphics card do any rendering only
refreshes realtime on your viewport. Gaming Graphic card is design to do dx or openGL realtime , imagine
a first person shooter that does cpu rendering ... all too slow. :P

Fantastic - this is a good lesson. i didnt know that the render work and Viewport work was split like this but it makes sense now you say it.

So what was said earlier by Gary Hasler about splitting the work into threads to make best use of the 4 cores is all the more important.

Perhaps this is a potential source of performance loss.

BUT - the SpecViewPerf benchmark allows you to run in Single Thread, Dual or Quad mode. And I have run the Single Thread mode and used the Online Results table for Single Thread. And I monitored (using Task Manager and Speedfan) the core usage during the benchmark run so I can see that only one core at a time is receiving instructions and total usage is 25% (1 core).

So I suppose it comes down to: the system just isnt as powerful as we think it should be. (sour grapes)

Success!!!

People - I have found the problem.

the reason the scores are so low is explained by this piece of info hidden deep on Nvidia’s website:

What NVIDIA Quadro graphics cards can work together in an SLI Frame Rendering configuration?
The supported models are NVIDIA Quadro FX 5600, 5500, 4600, 4500, 3500, 3450 or 3400 graphics cards.

What NVIDIA Quadro graphics cards can work together in an SLI Multi View configuration?
The supported models are NVIDIA Quadro FX 5600, 5500, 4600, 4500, 4400, 3500, 3450, 3400, 1500, 1400, 560, 550, 540 and 350.

The cards we have do NOT support Frame Rendering in SLI configuration.

THEY DONT BLOODY WORK!

so all the time, we have only been getting output from one card.

I just disabled the second card and ran a benchmark and got exactly the same result as before.

Oh well…

the upshot of this is I have a spare QuadroFX card…

always look on the brightside of life - de-dum- - de-dum-de-dum-de-dum…

thanks for your time guys.

interesting. i had no idea quadros differed that significantly. usually you just expect differences in gfx cards to be purely in power/speed, not functionality.

alex york
for and on behalf of Atelier York | Bespoke Architectural Visualization
http://www.atelieryork.co.uk
MentalRayTips Twitter Feed

agreed - i would have thought cards were SLI capable or not - end of story. The rest is numbers. But low and behold - there is a fundamental difference with Quadro’s and their Geforce cousins.

What got me sniffing in the right direction was a lack of control functionality in the Nvidia Control Panel (which was the main way other forums said to monitor the cards workload). With that missing, I started asking - well is it even working?

What may well happen next is i will start concentrating on that quad core CPU and see what i can get out of that.

watch this space…

glad you found the answer ... well your research will benefit someone on their next buy.

It good to check 3 times before you commit on a product though. :)

Studio Max 2009 x64
X5000 Chipset | Dual Core Intel 5140 | 4G RAM | Nvidia FX3450 | WindowsXP 64