xeon 8 core slowdown

[pinpan71]

i am working on a highway scenen with a lot of proxy trees, vray sun but no fancy stuff.
i precalculated the irradiance map for a flytrough animation and started rendering the final
images (full hd, rpf, lots of channels, irrmap size about 1gb).
now whats strange is that my new workstation (dual quadcore xeon 2,4ghz, max 64bit) takes lets say 2 hours for a frame whereas my render slave (core2 quad 2,4ghz, max64 bit) does the same frame in 50 minutes. both machines have 4gb of ram and the both seem to be doing fine with the amount of ram.
what might be the cause? i was wondering if the preloaded irrmap causes that behaviour.

[stevesideas]

Hi Pinpan,

What motherboard are you using in your xeon system ?

I have sent back my equipment to be tested. No solution found yet. I did everything I could have done..most up to date drivers, most up to date bios.

Still couldnt work out what was wrong. I did a number of benchmark tests using Sisandra...mainly cpu tests that showed the xeon system was twice as fast as my quad system, but when it came to rendering any scenes using max it was a lot slower.

Really not sure what is wrong ?

Keep in touch, maybe we can resolve this problem together..

[pinpan71]

hi steviouk,

its an asus but i have to look up the exact type (w 4gb of kingston fb-ram).

the ram was not cooled in the beginning and the board was extremly unstable when i started rendering. in the bios i found a lot of logged ram failures and without a fan the ram temperature was at 98°C! once cooling was setup it went down to 50 and all the hickups were gone...but the speed at rendering the scene was the same.

also latest bios. driveres i don't know - have to ask my hardware guy.
i thought the high temperature might have fucked the dimms.

i did not make any comparisons of the render time in general. calculating the irrmap seems bo be going really fast on that machine. i will have to do more testing...
problem is that for testing is no time at the moment.

[jacksc02]

pinpan- what are you vray settings?
Do you have the "bitmap pager" turned on or off?
What are you virtual memory settings?
What is you network setup?

Sounds like there is an issue somewhere.

[stevesideas]

From previous experience, Im thinking that if you system is stable...the memory should be fine. I think if the high temps had buggered up the memory sticks..or anything to that matter, the component would not work at all or at least behave erratically, causing BSOD or other problems.

Try the benchmarking programme from here:

http://www.sisoftware.co.uk/index.ht...uy&langx=en&a=

Test it on both computers and see if the results come out as expected. Basically your dual quad core system should be roughly twice as fast. Based on what other people were saying.

My cpu scores seemed to work fine for this benchmark, as it did with others. But when it came to max..the results were just odd. The dual xeon system would always under perform.

[pinpan71]

so i just check whats going on in the office and i checked the type of mobo.
its an asus dsgc-dw and the information about processor speed was wrong:
the xeons are xeon 2 x L5335 @ 2ghz
the core2 quad is a Q6600 @ 2.4ghz

so in theorie the speed of the xeon should be 16/9.6=1.67 of the quad.
here for this particular animation its for example xeon: 49min; quad: 21min.
which means the single core2 quad is 2.33 times faster than the dual quad xeon!

to jacksc02: virtual memory pagefile size both set to fixed 6gb
bitmap pager i neverer touched; so i assume its off
my network is a gigabit lan - nothing fancy but i don't think its a networking issue.

[stevesideas]

Im pretty sure the network isnt the issue. If you do the tests independently without the use of the network you can cross that off the list.

Try rendering the same scene using max, but only using the scanline renderer. The dual xeons should be faster than the quad core...surely.

Try downloading this max benchmark scene:

http://spec.unipv.it/gwpg/downloadindex.html

Goto the SPECapcSM for 3ds Max 9 download. Install it

Load the arch scene (a building with some water and some boats).
My dual xeon 5355 system managed to do that scene (standard scanline max render) rendered out at 1900x1080 in 16 mins. Way too much.

My quad core did the same scene in 220 seconds i think (around that).

Would be interesting to see how your 2 systems handle the same scene.

[pinpan71]

i am downloading now. will give it a shot on monday...
i will keep you updated on the times.

i checked my backburner list for other scenes to compare and found:
accidentally sent the irr pass to both computers and there everything is relatively normal:
4x-core2 17:55min, 8x-xeon 13:28min (render final image off) -> xeon 1.33 faster
another city scene renderpass for testing:
4x-core2 12:50min, 8x-xeon 9:21min (preloaded irrmap, 600mb) -> xeon 1.37 faster

so i think it must be related to this particular scene.
i use a matte background terrain (.bmp image sequence from terragen) with opacity mapped grass to be mixed with the matte terrain, not that many trees (all vray proxies, 6 types, total maybe 1000, incl opacity mapped leaves w. filtering off), a very simple mapped road texture and some railings, signs, displays and such stuff. there must be something the xeons do not like at all.

will post an image tomorrow.

960sec to 220 sounds terrible. thats worse than my scene. hope we can both resolve this problem.

[jacksc02]

pinpan- i have the EXACT system.
And my machine renders much faster than any single quad we have at work (qx6600)

It does sound like a network problem.
Also with your page file, its better to let windows allocate the memory it needs. Also it is better to set the page file to use a separate partition to avoid fragmented disk space.

[gilpo]

We built a couple dual quadcore machines about 6 months ago and had similar findings as you guys have. The primary factor in this was the use of proxies and dynamic memory to render memory heavy scenes. Comparing a dual, dual-core machine with a dual, quad-core machine, both with 4gb of ram equals different amounts of ram per core, which affects rendering performance in high-memory scenes.

The dual-core is thus running at 1gb ram per core while the quad-core is running at 512 mb per core. This causes the dynamic memory to have to page geometry in and out more frequently on the quad-cores and negatively affects performance. Also, the FB-Dimms required by the Xeons are a little slower than DDR2, so this adds to the problems.

To verify this, notice that the Xeons will render circles around everything else as long as the entire scene fits into memory. As soon as it needs to start paging the scene in and out, it slows down.

To get around this, we have a very diverse render farm and have identified the strengths and weaknesses of each platform and allocate jobs appropriately. It seems, in our case, that the Xeons are faster at calculating IM/LC passes so we do our pre-calculations on those machines and leave the frame rendering to our Opteron systems (where they are faster).

Hope this helps, but I'm sure its not what you wanted to hear.

[stevesideas]

So comparing a dual xeon system to a quad core would you say the quad core is faster.. doesnt really make sense to me. When it comes to pure mathamatical calculations the xeons should be a lot quicker.

[pinpan71]

that sound very reasonable what you say gilpo...
with some scenes an older opteron system is performing same/better than the faster core2quad...
also ir-calc seemd speedy on the xeon....so adding memory might help...i will do that anyway...

steveiouk: no time yet to do the benchmark. i ran into other problems ;(

other problem solved: jagged/no aa shadow on mattes. hope that will be fixed soon, bacause i forget quickly

[gilpo]

steveiouk:
The quads are overall faster, but they are only as fast as data can be given to them. If your scene can't fit all in memory at once, your system has to unload some geometry, re-load other geometry back in, then send it on to the processor. By having less memory per core, you have a smaller buffer in which information can be stored prior to rendering and geometry has to be swapped in and out more often.

Hopefully I'm making some sense.

pinpan71:
I didn't see you say, but you'll need to make sure you're using 64-bit windows to fully take advantage of 4+ GB. Also, crank up your dynamic memory limit. Watch task manager while you render. If your processors aren't running at 100%, but are going up and down, you are hitting your memory limit and need to raise it.

[stevesideas]

Ok, im not hijacking the thread, but, if i have a very simple scene that has one cube in a room with global illumination on. I have a xeon system 5355 with 4gb memory and a quad core6700 system with 8gb memory, you are saying the quad core will be a lot faster. What about scenes that dont ever use the full memory, the xeon system should be faster then. I dont see the point in even purchasing a system like that if it wont be faster.

The scene with the cube is something i setup to test the 2 systems and the quad core thrashed the dual xeons. The xeons are both rated at the same speed as the quad core, but they have 8mb level 2 cache unlike the 6700.

[gilpo]

steveiouk:

Yeah, the Xeons should be dominating that scene.

In our tests, when we benchmarked the Xeons with the benchmark scene from this forum (the one with the teapots and the lasers) it benchmarked correctly with what others had shown. However, once we threw some real-world, memory-heavy scenes at them, they choked until we added more ram.

But if you can't even get the standard benchmark or another 'light' scene to work right, then there's a problem somewhere. If that's the case, it's not memory related but either motherboard or processor related. Something is definitely wrong. I would check your chipset drivers. I don't know what you've tried, but try both the motherboard manufacturer's chipset drivers or the Intel reference drivers from Intel's web site.

I recall having a system one time that refused to benchmark properly until I re-formatted totally and tried different drivers. It was somehow hanging onto the old drivers.

Another thing that could be happening is either thermal or load throttling. Check the bios. Make sure the system isn't slowing the processors down when they reach a certain temperature. Vray pushes processors more than most programs and it might be triggering something.

Good luck.