I’m sure this will appear as some evil plot by Autodesk but the original thread for this topic was mistakenly deleted earlier today when I had them delete a post made by a spammer. I’m truly sorry for this confusion and hope that those who submitted previously will submit their images and results again. Copied below is what I could recover of the thread from my browse cache.

Sr. members of the Autodesk and NVIDIA team are actively monitoring this thread so we appreciate your involvement in helping us narrow down the issues and confusion with the updated iray build included with Service Pack 2.

-----------------------------
-----------------------------
• kcpr raffaEl
• Posted: 2011-09-26 12:19 am

Hi guys!
Today I installed 3ds Max Design 2012 SP2.
Before it I did test renders with iRay, and then I did it after SP2 install.
According to my test, the new iRay is TWO or MORE times SLOWER than iRay shipped with 3ds Max Design 2012.
I did several tests, and I uninstalled SP2.
What a crap!
Did you noticed the same performance loss?

________________________________________
kcpr raffaEl

-----------------------------
-----------------------------
• Rawalanche
• Posted: 2011-09-26 12:58 am

Noticed the same. The incompetence of Autodesk staff that is taking care about renderer implementation never ceases to amaze me.
They already pretty much ruined Mental Ray, which is now lagging several years behind Vray, even when nVidia has already a lot of new solutions ready to go, but with Autodesk preventing them from getting into DCC packages, or F’ing up their implementation. They will be usable when it will be already wayyyy to late, or they will never even get implemented at all.
And this is just another drop of water in already a huge ocean of incompetence.
-----------------------------
-----------------------------
• kcpr raffaEl
• Posted: 2011-09-26 3:45 am

I’ve spent a LOT of money for CUDA cards, and now they want me to render everything twice longer???
I can’t believe it!!
OH! Maybe I should buy twice more nVidia cards??
If this is not accident I’ll be soon moving to VRay with my production and forget about mental ray forever.
iRay is THE ONLY reason that I stick to mental ray products.
-----------------------------
-----------------------------
• Jonathan de Blok
• Posted: 2011-09-26 5:23 am

Iray2.0 does more calculations per iteration, so you get a clear image sooner. So an iteration takes longer but you need fewer so the net result is faster.
-----------------------------
-----------------------------
• kcpr raffaEl
• Posted: 2011-09-26 5:31 am

So if I try to compare 10 minute renders, the iRay 2.0 version will have less noise?
Am I correct?
I’ll give it a try…
-----------------------------
-----------------------------
Yes from what I have been told that is the case. From looking at the posted logs iray 2 actually kicks in an “adaptive over-sampling” eventually which appears to be where the render time explosion comes from. If you test and iray 1 and iray 2 with iterations save them as a lossless format and do a difference in a compositing package. This should show you where the changes are happening.
-Eric
Author: PiXeL_MoNKeY
Replied: 2011-09-26 5:41 am
-----------------------------
-----------------------------
• a-black7
• Posted: 2011-09-26 6:45 am
well? any comments or results of tests? i tried to compare to sp2 and got worst image (using sp2) in 10 minutes of tests. and one time i got blue screen that’s never happened before (sp2). it’s very interesting situation
-----------------------------
-----------------------------
• Senorpablo
• Posted: 2011-09-26 6:45 am
If this is accurate, it’s pretty frustrating.
The 4X slower skylight shadows from Max 9 to Max 2012 costs me hours and hours.
-----------------------------
-----------------------------
• kcpr raffaEl
• Posted: 2011-09-26 7:07 am
Well my Friends,
Here are 10 minute renders.
So, it’s clearly visible that on iRay 2.0 images is MORE noised than on 1.2 version.
And I don’t see any noticeable quality advantage in 2.0.
-----------------------------
-----------------------------
• Jonathan de Blok
• Posted: 2011-09-26 7:29 am

Are you sure the max-bounce settings identical?
-----------------------------
-----------------------------
• a-black7
• Posted: 2011-09-26 7:39 am
yup. same thing. also iray2 uses more gpu memory. iray 2 not too much slower but nearly 1.3 times. that’s all big crap ( i think 1.2 have better quality(reflection,GI,shadows) even without noise (4 long-time renders)
-----------------------------
-----------------------------
• kcpr raffaEl
• Posted: 2011-09-26 7:51 am

Jonathan de Blok 2011-09-26 7:29 am
Are you sure the max-bounce settings identical?
Yes, I’m absolutely sure, I’ve checked this again…

-----------------------------
-----------------------------
• JeffPatton
• Posted: 2011-09-26 8:26 am

For what it’s worth: I’m finding that Ver2 does indeed achieve similar results to Ver.1x but with fewer iterations however it does take longer to calculate those iterations in Ver2.
Data:
Ver1: - 1237 iterations in 15 minutes
Ver2: - 777 iterations in 15 minutes
I’m honestly still not sure how I feel about it yet.  To me Ver.2 certainly feels slower, but when I look at the raw data it seems to achieve similar results with fewer iterations. :/
-----------------------------
-----------------------------
Your scene looks like it uses a limited set of materials.  I wonder if the results would be the same with more textures, MSOMs etc?  Keep us posted!
Author: pfbreton

________________________________________Replied: 2011-09-27 4:53 am

Using the same CGtalk scene that kcpr used I get similar results.  More grain in the Ver. 2 than in Ver. 1.2.  Example:
5 minute render test between 1.2 vs. 2:
20 minute render test between 1.2 vs. 2:
There’s a 3 second delay between frames on those animated gif files.  I’m also aware that the gif format introduces artifacts itself because of it’s low quality, but the difference is still obvious.  I can supply the original renders (.tif format) if desired.
Also note the difference in chrome material that was used in this scene between versions.
Author: JeffPatton

________________________________________Replied: 2011-09-27 6:06 am

Since I didn’t mention the GPUs used in my tests: (1) Quadro 6000, (1) Tesla 2070, (1) Tesla 2050 with the 275.89 WHQL drivers.
Author: JeffPatton

________________________________________Replied: 2011-09-27 6:26 am

-----------------------------
-----------------------------
• Ken Pimentel
• Posted: 2011-09-26 9:16 am

We’re researching this with Nvidia. Clearly some scenes do what we expected - give better quality per iteration. However, there are other scenes that don’t. Iray is new for us too, and we haven’t had a chance to build out our automation performance testing to cover all cases. We do have such a suite and it did not report a serious issue when we tested iray 2.0. We now have a couple scenes that show the problem and it is in Nvidia’s hands.
-----------------------------
-----------------------------
Ken, please add skylight shadows to your automation performance testing. FYI it’s 4x slower in Max 2012.
For clarification, I’m referring to enabling shadows checkbox on a skylight light and using the scanline renderer.
Author: Senorpablo

________________________________________Replied: 2011-09-26 9:44 am

Senorpablo: Could you please specify what you mean by skylight shadows? You mean shadows generated by mental ray sun?
Author: Rawalanche

________________________________________Replied: 2011-09-26 10:06 am

Ken: How can YOU, Autodesk, assess the quality or performance of renderer when you have no competent staff to do it? The fact that 3ds Max 2012 was released with so significant Mental Ray problems only proves that you have no skilled rendering testers. I also really can not understand, how can you be eligible to make rendering related decisions. These decisions should be up to nVidia ARC, they are ones who know what they are doing, not you. They should make all the decisions about rendering portion of Max, and you should only implement their product according to their directions. If if continues in the current way, there will be no Mental Ray/iRay users left in 3 years.
Author: Rawalanche

________________________________________Replied: 2011-09-27 12:29 am

thread continued on next post....

Shane Griffith
Product Manager - 3ds Max & 3ds Max Design
Media and Entertainment Division
Autodesk, Inc.
AREA Blog
Twitter

Page 2 Continued.....

-----------------------------
-----------------------------

Rawalanche, I suggest you watch what you say as there are some very experienced rendering and mental ray users as part of the beta. Just because you aren’t happy with the state of mental ray in 3ds Max doesn’t give you a right to attack those testers who do everything they can to ensure the product is capable. Unfortunately the testers can only provide their feedback, after that it is up to Autodesk to deliver based on that information and other information collected (CIP, bug reports, etc).
Keep your opinions about others to yourself as they are in violation of the forum rules:"1. Do not post messages that are inflammatory, nonconstructive, or at odds with the aims of the discussion groups.”
-Eric
Author: PiXeL_MoNKeY

________________________________________Replied: 2011-09-27 3:26 am

If at least one person over there did everything they can, situation would be totally different. I am loyal mental ray user for many years now, and i am extremely angry that situation came to the state, where even I, who defended MR for all these years and tried to push it to it’s limits, am realizing i am being forced to switch to other renderer. Not because it is bad renderer, but because it’s implementation is THAT bad.
Author: Rawalanche

________________________________________Replied: 2011-09-27 3:50 am

-----------------------------
-----------------------------
• kcpr raffaEl
• Posted: 2011-09-26 9:16 am
Since on Jeffs renderings noise is rather comparable, I’ve decided to maka another experiment:
I’ve made renders with 1500 iterations.
With iRay 1.2 it took 10:49 minutes, with iRay 2.0 it was 20:25 minutes.
According to the more-calculations-per-pass explanation, the quality of iRay 2.0 should be better.
And what is reality? Take a look - iRay 2.0 has MORE noise in my opinion - even with the same itarations number.
-----------------------------
-----------------------------
Noise looks the same, but its the contrast/brightness that is the major visible difference here in my eyes (most noticeable in the clock). I’m backing off on the SP2 update for now.
Author: PSStein

________________________________________Replied: 2011-09-27 5:26 am

What GPU hw did you use for this test?
Author: Ken Pimentel

________________________________________Replied: 2011-09-27 6:15 am

It’s GAINWARD GTX 460 GS 2 GB RAM.
Same performance difference I’ve noticed on GAINWARD GTX 580 PHANTOM, but the times are from GTX 460.
Author: kcpr raffaEl

________________________________________Replied: 2011-09-27 10:46 am

-----------------------------
-----------------------------
• hot chip
• Posted: 2011-09-26 9:40 am
Hi Guys,
iRay in 3dsmax is not iRay in RTT or Bunkspeed. I can´t believe why Autodesk or Nvidia don´t fix simple Problems.
- we can´t set the pregessive threshold. Why?
- Why is the Material Template very Fast and “every time without noise”.
- and when i render the same Sphere in iRay, can i wait hours of hours, it is every time grainy. This is de Demonstrate the Autodesk things with deliberate cuts.
- the iRay Version for RTT bring me very fast result without noise.
Autodesk say, they must learn, all is new. Bull Shit and a lie! This is politics of Autodesk or NVidia, and I do not let kidding me.
mfg
hot chip
-----------------------------
-----------------------------
It is same iRay… there is only on iRay renderer. But it is implemented differently. Only Autodesk has “artist programmers” that are capable of ******* implementation up so bad the final code runs twice as slow. Recently i learned that nVidia ARC (former mental images) are doing their very best, and have solutions that are way in front of concurrence (such as Vray) but their effort is getting constantly sabotaged by Autodesk
Author: Rawalanche

________________________________________Replied: 2011-09-26 10:12 am

That’s BS. You don’t know what you’re talkling about. Nvidia and Adsk work extremely closely together and they are initimately involved with building the integration with iray.
As to the integration with mental ray, that is a different story. Reality is that our small team has been focused on iray lately, not mental ray. We have to pick where we can make a difference. However, most of the work we’ve done on iray and Activeshade is directly usable with mental ray. However, we can’t just throw stuff like that out there.
If it helps, Zap has joined the max team and starts work next Monday. I can imagine having him dedicated to max will help things.
Author: Ken Pimentel

________________________________________Replied: 2011-09-27 6:13 am

Ken is 100% correct - we work very closely with Autodesk to deliver the best possible solution *together*. Autodesk is on the forefront here, adding a renderer update mid-stream to enable you to deliver better quality images to your current release - something that I believe is unprecedented.
We thank everyone here for pointing out their initial observations on this latest release and we’re deeply investigating issues being pointed out.
I don’t want to go into details right now, but iray 2 includes many quality improvements that sometimes had trade-offs in performance. We/NVIDIA already spent considerable effort to overcome slow downs these deeper calculations required, as we take this issue very seriously. We hope to have a comprehensive answer up here shortly.
- Phil
Author: Phil Miller

________________________________________Replied: 2011-09-27 9:10 pm

-----------------------------
-----------------------------
• a-black7
• Posted: 2011-09-26 3:44 pm
is there any chance to set up sp2 without iray2?
-----------------------------
-----------------------------
• a3dcreator
• Posted: 2011-09-27 2:52 am
this is bull%#@*
i just got my tesla last night and was like damn this thing is not much faster even now that i have 3 video cards.
and i see the reason why now.
is there any way to revert back to Iray 1.2?

-----------------------------
-----------------------------
If you’re on Tesla, according to Nvidia, you should see the benefit of iray 2.0. Are you saying that after rendering for a fixed amount of time, you don’t?
Author: Ken Pimentel

________________________________________Replied: 2011-09-27 6:14 am

Just to clarify Ken: does your question also mean that according to Nvidia it would be normal to have performance drop/no benefit on Geforces?  As in are they now (Iray 2.0 - ) crippling the Geforces?
Author: vovov

________________________________________Replied: 2011-09-27 6:58 am

i get noisier renders than i ever did before.
and my renders take forever.
I have one quadro 5600
one tesla c2070
and one geforce gtx 580
also when i ran speed tests my geForce outperforms my new Tesla.... sigh
Author: a3dcreator

________________________________________Replied: 2011-09-27 7:11 am

Are they noisier? From what I see the iray 2.0 renders are sharper and the fireflies are brighter, but the noise is the same or less. Define the “noise” you see. Within a given pixel are the surrounding pixels blending better or worse? From what I see, at least IMHO, low fidelity noise is resolved by iray 2.0 better, but because of that the fireflies jump out more.
-Eric
EDIT: Sorry just saw Jeff’s new images. Can definitely see the noise difference there. There are more than a few differences in the results. So I do wonder what the calculation differences are doing. I also wonder what the results would be on a scene that actually converges. Right now none of the posted imagery is noise free.
Author: PiXeL_MoNKeY

________________________________________Replied: 2011-09-27 7:40 am

hey ken can you explain what the advantages of having a tesla are vs the gtx?
i have both and im just curious. cause like i had stated my gtx is faster than the tesla card i just got.
Author: a3dcreator

________________________________________Replied: 2011-09-27 1:09 pm

-----------------------------
-----------------------------
• raymarcher
• Posted: 2011-09-27 6:48 am
Thank god I ditched Mental Ray 7 years ago.
-----------------------------
-----------------------------

Continued in next post…..

Shane Griffith
Product Manager - 3ds Max & 3ds Max Design
Media and Entertainment Division
Autodesk, Inc.
AREA Blog
Twitter

Page 2 Continued.....

-----------------------------
-----------------------------
• Shane Griffith
• Posted: 2011-09-27 7:49 am

Hi everyone,
As mentioned we are actively running this down with Nvidia right now.
We can confirm the following:
• Render time per X iterations is not a valid measurement between the iray 1.2 vs 2.0 builds
This is explained by fundamental changes in the iray core, namely an improved material model and more complex (more accurate) sampling algorithms. The iray 2.0 build now does more in each iteration, but it also does it better, so the increased render time is typically outweighed by an improvement in the overall quality and a faster (time-wise) convergence speed. It’s therefore crucial to compare image quality versus render time, and never versus the number of iterations rendered. So far we’ve observed that, in most cases, iray 2.0 converges to a noise-free solution faster than iray 1.2, but we are still investigating this as results can very with HW and scene setups.
• Measuring rendertime against comparable error threshold settings is no longer valid between the iray 1.2 vs 2.0 builds
the error threshold was completely re-implemented in iray 2.0, and it can’t be compared to the one in 1.2. The new one should be more meaningful, but its value is essentially unrelated to the old one.
• Comparing image quality versus the same render time is the only accurate measurement.
• Results will vary with different generations of GPUs.

There are three sets of GPU generation for which iray is compiled: the 1.0 path (compute capabilities 1.0, 1.1, and 1.2); the 1.3 path (cc 1.3), and the 2.0/Fermi path (cc 2.0+). The compute capabilities differ in the features they support and the amount of memory (local, constant, registers, etc.) they contain. The ever increasing complexity of iray (materials, textures, motion blur, etc.) takes its toll on older generations: the 1.0 path can render up to 2.5x slower with iray 2.0 compared to iray 1.2 (this is the per-frame render time, not a time-wise comparison). The 1.3 path is much less affected, but still worse than the 2.0 path. For a list of GPUs by compute capability, see: http://en.wikipedia.org/wiki/CUDA#Supported_GPUs.  As mentioned above, the 2.0 path should generally render faster time-wise, while the 1.3 path should render about the same or a little slower, and the 1.0 path can be expected to render slower. When posting results please include this information.
Technicalities put aside, we are continuing to investigate this. Measurable results are difficult and do appear to change at various durations of render time. We are also unable to explain the differences in the metal shader on the alarm clock. This is important because a difference in brightness suggests a difference in material setup: a brighter/whiter material will be more reflective, cause longer paths, more fireflies, etc. It will take longer to render AND will cause more noise. This could be a new bug and the cause of the performance degradation on certain scenes.
-----------------------------
-----------------------------
According to the document you linked in this post, my GTX460 has compute capability 2.1. Yet, most of my scenes perform significantly slower.
By the way, difference in rendering of metallic shaders appeared also with new Mental Ray 3.9. It now behaves differently than 3.8.
And at last, Arch&Design material glossiness input is gammacorrected in different manner for Mental Ray and iRay, resulting in absolutely different output on same scenes.
iRay result
MentalRay result
Here is difference between MR and iRay result on exactly same scene. The glossiness map input is handled differently.
Edit: For some reason links got automatically displayed on the page, which i did not want, sorry for breaking your website apart… :X
Author: Rawalanche

________________________________________Replied: 2011-09-27 6:34 pm

thanks Rawalanche, this is the kind of test cases we are looking for. can you resize the images in half just so the forum police don’t come after us. ;)
Author: Shane Griffith

________________________________________Replied: 2011-09-28 1:52 am

-----------------------------
-----------------------------
• hot chip
• Posted: 2011-09-27 8:06 am

here a special link for the Autodesk Team + Nvidia.
On Filtering the Noise from the Random Parameters in Monte Carlo Rendering: http://agl.unm.edu/rpf/
-----------------------------
-----------------------------
• kcpr raffaEl
• Posted: 2011-09-27 11:36 am
I tested this scene with three options:
1. GTX 580 3 GB RAM
2. GTX 460 2 GB RAM
3. CPU Only
As you for sure know these two GPUs have Compute capability (version) 2.0 or 2.1 - THE SAME that newest TESLA and QUADRO have.
Dear Friends from nVidia / mental images,
I really really love iRay, and really I’m using it in my production (still images) from the time it was available through 3ds Max 2011 Subscription Advantage Pack. In my opinion it’s mental ray “last chance”, comparing to vray developement speed. It’s only iRay now can compete with vRay quality in my opinion.
But you have to be aware that if users only suspect that iRay is written only for QUADRO/TESLA cards - despite of fact that GeForces cards have the SAME GPU’s and Compute capability versions, this will be soon the end of iRay - not only because of economic reasons (for example: the price of GTX 580 PHANTOM card = vRay licence price - not to mention the price of QUADRO or TESLA card), but mainly because that gives bad-feeling that it’s only our money (to force us buying TESLA or QUADRO cards) is the goal.
I’m totally honest, and I want that’s scenario would never to happen.
-----------------------------
-----------------------------
>>But you have to be aware that if users only suspect that iRay is written only for QUADRO/TESLA cards ..... this will be soon the end of iRay
We primarily develop and test iray on Quadro/Tesla for both reliability and memory reasons, but that doesn’t mean we don’t support GeForce products. At this time, your experience on iray should be the same on our professional and consumer GPUs (cores and clock being equal). There are some programming possibilities that exist on our pro products that don’t exist on GeForce, but iray doesn’t exploit any of these at this time.
- Phil
Author: Phil Miller

________________________________________Replied: 2011-09-27 9:17 pm

It’s a matter of personal choice, there are pro’s and con’s for both sides of that argument. I use a GTX 480 in my personal system, but i don’t work with very serious production level scenes very often like you all. Jeff Patton wrote a good blog on this subject, http://jeffpatton.net/2010/11/gt...selections-for-rendering/ and I believe Jennifer O’Connor has written some thing also.
Author: Shane Griffith

________________________________________Replied: 2011-09-28 2:05 am

-----------------------------
-----------------------------
• Tom Mitchell
• Posted: 2011-09-27 2:05 pm

A little more kindness on all our parts will go a lot farther than venting on someone, just to vent.
Tom
-----------------------------
-----------------------------
• JeffPatton
• Posted: 2011-09-27 5:16 pm
>>kcpr raffaEl 2011-09-27 11:36 am
>>But you have to be aware that if users only suspect that iRay is written only for QUADRO/TESLA cards

I could be wrong but I don’t think that’s going to happen.  I don’t think the software is being (or will be) specifically coded/throttled for Quadro/Tesla GPUs to force sales. I would imagine it’s probably the other way around with the Quadro/Tesla drivers/hardware being more optimized for rendering.  At least I’d hope so due to their price tags. 
While I’m sure the “bean counters” come into play at times, these companies (Autodesk/nvidia) are filled with smart/passionate people, just like us.  I really can’t see them intentionally crippling their software with the hopes of selling a few more of the high end GPUs.
-----------------------------
-----------------------------
“I would imagine it’s probably the other way around with the Quadro/Tesla drivers/hardware being more optimized for rendering”
To put that in perspective:
A GeForce590:
1024 Cuda cores @1.2 Ghz (spread over 2 GPU’s)
327.7 Ghz mem bandwidth
3GB memory (1.5 per core so that’s the max you can use for a scene)
A TeslaC2070:
448 Cuda cores @ 1.15 Ghz
144Gb/s memory bandwidth
6GB
So the only advantage for the Tesla is the memory, if you can live with that the GeForce will render about 2.3 times faster.
Or pick up a Geforce 580/3GB for around $500.. you’ll get 512 Cuda cores @ 1.5Ghz, still faster then the Tesla.
The only thing is the Quadros/Teslas are tested for 24/7 100% use. On the other hand a E-VGA GeForce for example will come with a 10 year warranty. And you can always put a bigger/better cooling solution on the Geforce so it can handle the load better.
Author: Jonathan de Blok

Shane Griffith
Product Manager - 3ds Max & 3ds Max Design
Media and Entertainment Division
Autodesk, Inc.
AREA Blog
Twitter

Page 3 Continued.....

>>So the only advantage for the Tesla is the memory....
Not quite - as you mention, the Tesla is hardened, with far more burn-in time and higher quality memory as it’s intended for 24/7 usage in data centers and super computer clusters. This extends to board components, and heat control that are simply not on game cards. For example: examine a high end GTX - all the heat goes into the box, while the pro cards exit the heat out back (so if you use GTX, make sure your rig is prepared for that heat). The pro cards are also designed to fit into the power budgets of reliable workstation and server providers while high end GTX assumes you’re building a gaming rig.
>>On the other hand a E-VGA GeForce for example will come with a 10 year warranty.
I think you’re implying that you can replace them X number of times. While true, most professional customers would rather have reliability because swapping in replacements can mean downtime at the worst moments, and that delay could be at a considerable cost.
On your comparison, the GTX 590 is about 50% faster than the GTX 580 because of the power and resulting clock speeds. Memory bandwidth has no impact on ray tracing speed - only the graphics clock does. So compare cores and clock and you will have a good yardstick within a given GPU generation. The GTX 590 is the fastest single GPU card right now and this is a GF110 generation. The pro board to compare with here is the Tesla M2090, which is also GF110 of the same 512 core count.
Now that’s just the compute part that impacts iray performance. The graphics drivers deliver a different level of experience as well. GeForce drivers evolve and are tuned primarily for the ultimate games performance while Quadro drivers focus on the needs of professional applications like 3ds Max. 
- Phil
Author: Phil Miller

________________________________________Replied: 2011-09-27 9:49 pm

>>So the only advantage for the Tesla is the memory....
Not quite - as you mention, the Tesla is hardened, with far more burn-in time and higher quality memory as it’s intended for 24/7 usage in data centers and super computer clusters. This extends to board components, and heat control that are simply not on game cards. For example: examine a high end GTX - all the heat goes into the box, while the pro cards exit the heat out back (so if you use GTX, make sure your rig is prepared for that heat). The pro cards are also designed to fit into the power budgets of reliable workstation and server providers while high end GTX assumes you’re building a gaming rig.
>>On the other hand a E-VGA GeForce for example will come with a 10 year warranty.
I think you’re implying that you can replace them X number of times. While true, most professional customers would rather have reliability because swapping in replacements can mean downtime at the worst moments, and that delay could be at a considerable cost.
On your comparison, the GTX 590 is about 50% faster than the GTX 580 because of the power and resulting clock speeds. Memory bandwidth has no impact on ray tracing speed - only the graphics clock does. So compare cores and clock and you will have a good yardstick within a given GPU generation. The GTX 590 is the fastest single GPU card right now and this is a GF110 generation. The pro board to compare with here is the Tesla M2090, which is also GF110 of the same 512 core count.
Now that’s just the compute part that impacts iray performance. The graphics drivers deliver a different level of experience as well. GeForce drivers evolve and are tuned primarily for the ultimate games performance while Quadro drivers focus on the needs of professional applications like 3ds Max. 
- Phil
Author: Phil Miller
________________________________________Replied: 2011-09-27 10:04 pm

-----------------------------
-----------------------------
• Jonathan de Blok
• Posted: 2011-09-27 10:49 pm
“Not quite - as you mention, the Tesla is hardened, with far more burn-in time and higher quality memory as it’s intended for 24/7 usage in data centers and super computer clusters. This extends to board components, and heat control that are simply not on game cards. For example: examine a high end GTX - all the heat goes into the box, while the pro cards exit the heat out back (so if you use GTX, make sure your rig is prepared for that heat). The pro cards are also designed to fit into the power budgets of reliable workstation and server providers while high end GTX assumes you’re building a gaming rig.”
Sure you have to make sure it will have good airflow, a decent workstation case and some case fans will take care of that.
About ‘better’ memory, do you have some datasheets on the chip components to back this up? If it’s about ECC error checking, that’s something you want to turn off for iRay to boost speed.
from the iRay faq:
“ECC should be OFF (for Quadro and Tesla GPUs based on “Fermi”)”
“Now that’s just the compute part that impacts iray performance. The graphics drivers deliver a different level of experience as well. GeForce drivers evolve and are tuned primarily for the ultimate games performance while Quadro drivers focus on the needs of professional applications like 3ds Max.”
Quite the opposite is true nowadays, with the introduction of Nitrous viewport the special performance drivers were discontinued. Max’s viewport is more similar to game technology then to an OpenGL CAD application.
So for everyday 3dsMax usage a GeForce will definitely out perform a Quadro/Tesla on both iRay as in the viewport, the only limitation is the amount of memory.
Like I said before, the thermal issues can be fixed by adding a better cooler and/or underclocking it a little, that will reduce the heat output as well.
But to each his own, if you want out-of-the-box stability and have a high budget go for a Quadro/Tesla.
If you want max performance, have a small budget and are willing to tweak things a little go for a GeForce.
If I’d run a company that has a 100 workstations I’d might put Quadros in them because it might be less work and for support.
But I’m a one man show. If it somehow breaks I can just walk to the hardware shop get a new card and install it all within an hour or so.
-----------------------------
-----------------------------
>About ‘better’ memory, do you have some datasheets on the chip components to back this up? If it’s about ECC error checking, that’s something you want to turn off for iRay to boost speed.
Jonathan, Phil is the Director of Software Product Management at NVIDIA. We’ve asked NVIDIA to help reply here so that you have accurate information.
Author: Shane Griffith

________________________________________Replied: 2011-09-28 2:21 am

Shane Griffith
Product Manager - 3ds Max & 3ds Max Design
Media and Entertainment Division
Autodesk, Inc.
AREA Blog
Twitter

-----------------------------
-----------------------------
• Daniel Levesque
• Posted: 2011-09-28 1:33 am

We’d like to chime in with our perspective from the iray development team. iray is not a simple piece of code, it’s grown to be a complex, feature rich renderer; it’s quite possibly one of the most complex pieces of software to ever run on a GPU. We’re constantly breaking barriers and pushing the limits of what the GPU can do, but those limits are much narrower on older generations of GPUs. Our team works very hard to continue supporting all existing GPUs (and CPUs) and combinations, be it Quadros, Geforces, Teslas, from the earliest CUDA generation to the latest. The performance degradation we see on older GPUs is a compromise we had to make in order to continue improving iray and make it more useful for all users. Newer GPUs (Fermi-class, compute capability 2.0+) shouldn’t see a performance drop, in fact they should usually be faster (time-wise, not iteration-wise). The reduction of performance on older generations is not intentional, it is the reality of rapidly evolving technology, those older boards can’t cope with the improvements we put into iray.
The full benefits or iray 2.0 are not immediately visible after installing the service pack, but they are there. If used properly, they can make your work easier and your renders much faster.
Here’s a glimpse of the things you get form iray 2.0, as bug fixes and improvements:
* Better/fixed A&D material support
** “no diffuse bump” now supported
** better glossy refractions (improved frosted glass)
** better translucency
** improved IOR behavior with nested volumes (e.g. ice in a whiskey glass)
* Daylight portals are now properly supported, these can vastly improve the performance of interior scenes
* Improved environment lighting, reduces noise faster in exterior scenes
* Large resolution rendering: your render resolution is no longer limited by the amount of memory available on the GPU (e.g. 6k images with 1GB of GPU memory)
* Automatic adaptive sampling to finish off high noise areas We hope you’ll agree that it’s not just about the strict per-frame render time. We understand that some of you are reporting results that don’t agree with the performance expectations stated above, and rest assured that we will investigate these cases.

Daniel Levesque & Carsten Waechter
iray developers
NVIDIA ARC

Shane Griffith
Product Manager - 3ds Max & 3ds Max Design
Media and Entertainment Division
Autodesk, Inc.
AREA Blog
Twitter

Hey guys!

Did you try another tests, scenes?

I’m testing iRay 2.0 in many conditions and I didn’t find situation with speed advantage.

Could you show any example where iRay 2.0 performs better?

Thanks!

kcpr raffaEl

http://www.wizualizacje-3d.eu/en/portfolio

hi guys a a quik question for anyone who might now.

is Iray faster if you limit the bounces? or not?
cause i see little squares on my render that dont go away.

around the eyeballs as you can see in this image.

i have limited the bounces to 4.

visit http://www.wickedresin.com/shop/pc/viewcontent.asp?idpage=5
to see what 3d printing can do for you.

my personal website: http://www.stevenguevara.com

Core i7
Windows 7
24 gigs of RAM
Quadro 4000
Tesla c2070
3dsMax2013
Mudbox 2013

	Attachment

Maybe Mr. mental ray can explain the “exact” problem? Why not?

http://www.youtube.com/watch?v=SWT_9lSoCoc&feature=youtube_gdata_player

Intel i7 3820 3.6Ghz 8 CORE HT / CoolerMaster Hyper 212 EVO / OCZ Vertex4 256GB SSD v1.5
Nvidia EVGA GEFORCE GTX 660 3GB v314.22 (Nitrous in max) / Dual Dell Professional P2311H 23" Displays
ASUS P9X79 PRO / G.SKILL ARES 64GB 1866 DDR3 / Antec…

We made a related post on the iray dev blog recently:

http://blog.irayrender.com/

See the November 10 post about portals.

cheers
daniel

daniel levesque
iray dev
nvidia arc