VRAY slow on AMD FX 8350

[chaosmonger]

Hello guys, I'm fighting with this issue for long time without any solution...
(I'm italian forgive my bad english).

I'm rendering an animation with Vray 2 and 3DS Max 2011.
I'm asking a friend of mine some help, and I cannot understand why, but his renders are 250% faster than mine, even though I've a better computer.

Of course the scene is EXACTLY the same, with EXACTLY the same Vray Preset, texture, etc.

I've an AMD FX 8350, with 16GB RAM.
He has an I7-2600 with 8GB RAM.

My render time per frame (on same scene, same settings, etc.) is: 25mins.
His render time is: 10 mins.

How is it possible?!?! What's wrong!??!
I've already made a benchmark and everything seems to work perfectly. All the 8 cores are working at 100% at 4k clock. The RAM is working (9gb/16). Everything is good, but the render time... I've tried to install a newer version of Vray and also a brand new version of Max, but the problem remains.

So, my question is: how is it possible? Can some installed software create conflict with Vray?
Do I have to format? Or there's some glitch? I've to use BIOS settings? What can slow down my render time comparing to my friend's one?

I'm going crazy!
PLEASE HELP!

 

[chaosmonger]

what is your motherboard?

Mine is ASUS M5A78L-M/USB3
My friend's ASUS P8P67 LE

Basically he has an old chipset... so I guess it's not the motherboard

 

[random stalker]

As rmpumper said (and I said many times), we are rendering with CPU, not GPU (is not a VRay RT render, but a standard one, who use CPU).
So that's not the problem... thank you anyway stalker.

run this>
http://www.3d-passion.de/downloads/VrayRC3_Max8_Benchmark_01.rar
post results, and please don't mess with any of the render settings.
your friend should do the same...

 

[chaosmonger]

I made the test and I rendered that scene in 1m 12s (without changing anything). Don't know if it's a good result or not.

 

[random stalker]

I made the test and I rendered that scene in 1m 12s (without changing anything). Don't know if it's a good result or not.

Well, according to various sources, the results of a vanilla i7 2600K rendering that image should be around 62-65 seconds... So, your CPU should be fine, and should be performing perfectly.

As I stated before, the difference between the gen 3 i7 should be around 75-80/100, the older gen 2 should be within +-90:100 (lower is better)... Give or take... You can also add some statistical error...

But we'll see more if your friend runs that bench too

And then there are the following two scenarios:
1, he gets under 30 seconds and I give up.
2, he gets around 60-70 seconds and that proves I was right since the beginning, stating the GPU and especially the use of OpenCL /CUDA/ makes the huge difference...

 

[chaosmonger]

Well, he's offline now but probably he's gonna make the test later...
Anyway, if he gets same time of mine (60-70 seconds as you said), this doesn't proves anything! Because on my render animation he is going 2.5x faster! And anyway, he's using Radeon without CUDA, I'm the one who's using NVidia.... Don't know about the OpenCL stuff.

Thank you for your cooperation.

 

[DukeOvilla]

Hmm, the measly Human known as stalker may have a valid point to be proven.

 

[random stalker]

Well, he's offline now but probably he's gonna make the test later...
Anyway, if he gets same time of mine (60-70 seconds as you said), this doesn't proves anything! Because on my render animation he is going 2.5x faster! And anyway, he's using Radeon without CUDA, I'm the one who's using NVidia.... Don't know about the OpenCL stuff.

Thank you for your cooperation.

http://streamcomputing.eu/knowledge/what-is/opencl/

tl/dr: Well, it is a framework for writing programs that can be run at platforms consisting of various computing units.
Thus, you can, for example use it, to access a graphical computing unit (ergo the one worker inside your GPU) to do some non graphical related work (ergo, the stuff usually your CPU usually does).

Without it, you would render the image using only your 4x2 /or your friends 4+4/ core CPU (as it is set in the bench I'd gave you - also, that is the reason I asked you not to mess with the settings).
Using it (and last time I checked Vray can use OpenCl as well as CUDA) adds an thousand men strong army of small processors. And while one on one the are not as powerful as your CPU, they still do their job faster. "I am called Legion, for we are many" kinda thing.

 

[chaosmonger]

Ok, well, let's wait the benchmark of my friend, then we'll see...
Anyway, maybe I can try to change the settings of OpenCL / CUDA / whatever in Max, but I rememberd that I did it before.

 

[chaosmonger]

Ok, we made that benchmark again, restarting our machines, both me and my friend.
I'm little bit faster than my friend!!!

Here the final results:

ME:

MY FRIEND:

So, what could be the problem on a bigger scene?
Textures? Lights?
Of course the scene where he is 2.5X faster is quite complicated, 4 characters with bones, wires, links, high quality texture, resolution 1920x. But seems that my computer is faster...

 

[DukeOvilla]

Congrats!

 

[chaosmonger]

Congrats!

ahaha! Do I win something!?
Since I'm stil 0.4x slower than him on the renders that really matters!!!

 

[chaosmonger]

Still waiting the award!!! 😛

 

[chaosmonger]

Little update:
I guess that a lot is concerning the textures.
I'm not using a real texture for the materials, but a series of MASK textures (black and white) applied to different BLEND materials.
Sometime the texture is high resolution (4000x4000), but since it's just a JPEG in black and white, the size is maximum 400k.

How that can raise the render time? And why my friend is faster on calculating it?

 

[TenPc]

Have your friend do the exact same render as yourself and then see the rendering time differences, you might find that your render is faster than him. The reason that your render takes longer is that you are using masks which is much like layers and that each mask has to be "layered" to the image.

If you used 16gb of ram rather than 8gb, you will notice a remarkable difference. As you say, this type of work uses much more cpu grunt (and ram) than video card requirement.

You say the size of the jpeg is only 400kb, that is rather small, are you sure about that size?

 

[random stalker]

Have your friend do the exact same render as yourself and then see the rendering time differences, you might find that your render is faster than him. The reason that your render takes longer is that you are using masks which is much like layers and that each mask has to be "layered" to the image.

If you used 16gb of ram rather than 8gb, you will notice a remarkable difference. As you say, this type of work uses much more cpu grunt (and ram) than video card requirement.

You say the size of the jpeg is only 400kb, that is rather small, are you sure about that size?

Already had them done that - they did an old test suite, one that employs only CPU power. The score was FX's 52 s /72s on the first run/ vs i7's 62 s 😀
The second thing is, his FX has already twice the RAM compared to the i7 - which should help a bit.
Another difference is the Vray version 2.20 /i7/ vs 2.40 /FX/ 😱D
The last one is the GC - 6970 vs 660ti.

Also, I wanted them to run a bench, that employs all the resources /both GPU and CPU/ but haven't found a decent pic yet /mostly because it's that time of month for me 😀/. But I promise to look for a decent <15min bench picture, and we'll see more...

My current hypothesis is that the i7's HD6970 owns the FX's GTX660Ti in distributed computing.
So, that's why I need to find a bench, which will test both cards /OpenCL vs OpenCL and OpenCL vs CUDA/.

 

[TenPc]

Thanks, RandomStalker, I realise that they both had done the same test as shown in the images above, just not at the time I asked the Question, I'd scrolled past that entry, we're on Page 2 and it's a long read through it all. 🙁

 

[random stalker]

np, I'd just gave a quick tl/dr summary of what happened so far, since the thread is kinda long 😀

but, I wanted to ask you, do you have any good pic for bench? I've found something, but that takes about 1.5 - 2 hrs to complete, so it's kinda overkill for this situation 😀
Anyways, I'm off for a beer or two and resume my hunt later on

 

[chaosmonger]

Ahah, thanks random stalker. I'm waiting for another benchmark scene more "hard" (but maybe not 2hours 😀)
At the moment I've no solutions and I'm giving up... I've tried everything!

 

[TenPc]

Well, one thing you haven't considered is the amount of free space on your PC compared to your friend's PC.

Also, do you both start each project from a cold boot?

Also, too, as well, what is your OS as compared to his OS - both Windows 7 x64?

 

[chaosmonger]

We are both using Win7 64. I tried to render from a RAM virtual drive without any result. Also tried from my SSD driver (where I've the os), but nothing... In some pics above I also put our harddrive specs, I guess you can see the free space, if that really matters...

 

[TenPc]

Just things to consider, as the alternatives (above) don't seem to be the "fix" or solution you are wanting.

"Also, do you both start each project from a cold boot?" Do you start from a cold boot? Not awake from Sleep or hibernation or standby.

If he boots from a shutdown Pc but you "awake" from something then he is likely to have no data bits clogging up his memory.

If his free space is more than your free space , percentage wise, then he will have more virtual memory than you making his rendering faster.

 

[8350rocks]

1, why should he troll you? here is a 1st bench I've come across, which shows, that a i7 3770K takes about .76 time to render a scene under vray compared to 8350 (and yes, I'm too lazy to find an i7x bench)>
http://forums.anandtech.com/showpost.php?p=34471095&postcount=19

20% is not 250% faster...and, ironically, anandtech has some intel bias...but even so, the results aren't close to 250% faster.

2. about that "no intel part is 2.5 times better">
http://openbenchmarking.org/result/1210227-RA-AMDFX835085
Check Himeno bench - which means your hypothesis is false 😀
The Himeno benchmark is a linear solver of pressure Poisson using a point-Jacobi method.

Ok, so your point is what? If you use a PC for specifically one specific linear solving function the i7-3770k is fast at it? And...? Show me a real world benchmark that someone here would use on a regular basis.

The FX8350 is about 50% faster at encryption algorithms(40-60% depending on the application used), but I don't pretend it's 50% faster unilaterally as he did.

Also, as you see from the rest of the benchmarks, while AMD competes in some areas with the i7, there are many areas in which it has worse results as an generation old i5. So, please, stop behaving like a AMD sheep - this is a tech forum, we can at least use google to check your claims

I am not an AMD sheep...you quote an openbenchmarking.org benchmark for one specific thing, but don't quote the others where the 2 are neck and neck to make your "sheepish" example? This is a tech forum, provide accurate information. No intel product is 250% faster unilaterally. Period. I would even question the results of that benchmark as well, as there are conveniently no others with similar results.

 

[Cazalan]

"Another difference is the Vray version 2.20 /i7/ vs 2.40 /FX/"

Those need to be the same.

Since the test bench frame was faster on the FX, the issue is to do with the render job itself. Whatever code path it's using isn't optimal for THAT project.

 

[TenPc]

You said "Ok, we made that benchmark again, restarting our machines, both me and my friend. I'm little bit faster than my friend!!!"

What he does for his part of the project does not include masking whereas your part does involve masking so each mask has to be "compiled" prior to "layering" and the final "render" would take place, adding x amount of minutes to your outcome time frame.

If your taregt file is not located on the OS drive then you have to account for extra time used by the OS to create a temp storage on the Os to render the image and output to wherever the image is actually stored.

For example, copying a file from one drive to another does not actually stream directly to the other drive, a background compile is prepared on the OS drive in temp storage prior to the copying of data, packets of data are sent to the temp storage then those data bits are then sent to the target drive.
If you rendered to the OS drive, your time should be faster than the alternatives however, if your free space is less than 25% then that would impact on the rendering time. You should always save your image or project at the beginning prior to any work you actually do rather than working with a non-file.

Your ram, is it kits of 2 (x2) or a kit of 4 sticks. If a kit of 2 (x2) then it might be that you have one of one kit with one of the other kit as a pair.

 

[rmpumper]

As Cazalan said, the different Vray versions might have something to do with it. If the scene was prepared with vray 2.2 system, it might explain why vray 2.4 can't handle it as fast as some specific featured used are outdated therefore not optimised to run on 2.4 (the same issue is present in mentalray, i.e. if the scene is set up in 3dsmax 2009 or older, the render time on same machine is at least 50% slower when rendered in 3dsmax 2010 or newer as the code of mentalray 3.7 and up is not optimised to use mentalray 3.6 and older scenes).