THE PERFECT VRED RENDER PC

[Anonymous]

Hello everybody,

do you guys know, what would be the perfect render pc for VRED?

What PC hardware components are really important for VRED to render pictures and animations super fast? 🙂

And what is necessary for a jerk-free-workflow?

Regards Matthias

[michael_nikelsky]

Hi,

I assume you want to use Raytracing, so in that case the CPUs should be as fast as possible. The best you can get at the moment (ignoring the E7-Processors since these are server only) are two E5-2699 V3 with 18 Cores/2.3GHz Baseclock each but these are very pricy ( about 4750$ each at the moment). You can get a slightly better price/performance ratio going with two E5-2697 V3 with 14 Cores/2.6GHz which should be around 3000$ each at the moment.

GPU however is less important in this case, any recent Nvidia Quadro or AMD Firepro will do (and Geforce GPUs will usually work as well, even though not officially supported).

Kind regards

Michael

[Anonymous]

Oh Thank you very much! Do you know how I could configure the cheapest PC for using VRED? I don't have so much money. I'm a student!

[michael_nikelsky]

That is hard to answer without knowing your actual budget and whether or not you are willing to overclock your system. In general: the more cores and the higher the clock speed the better. You can pretty much just multiply to clockspeed with the number of cores to get a rough comparision of the performance you will get from processers and an i7 will give you about 10 to 20% more performance compared to an i5.

[Anonymous]

ok I understand. I want to spend around 1500€.
is there a possibility to buy cluster. so only CPU Power? I wok on a Macbook and my idea is to make a setup in VRED and render it out on the cluster System. (like a Renderfarm)

[michael_nikelsky]

Please change your user name so I can sent you a private message with some recommendations.

[Anonymous]

can't find where i can change my user name?

[michael_nikelsky]

When logged in you should be able to edit your account at the top and change your name to something unique.

Not sure why this forum system allows hundreds of user to have the same name in the first place but it makes it impossible to find the correct recipient.

[Anonymous]

sorry I can not find where I can change my Username?

[Bob.Bon2000]

Dual 20 Core xeon

128gb+ ram

2x 512gb M.2 SSD 2gb read/write + raid 0

Everything else dont matter.

That will be fastest u can get for RT renderings. 

[richardlevene]

Hi Michael,

Does Vred fully utilise all the cores of a dual E5-2699 cpu setup? I know other renderers have some issues.

Would be great to know.

Best,

Richard

[michael_nikelsky]

Hi,

yes, it does, provided that the scene is complex enough and the resolution is high enough to allow for good load balancing (at the moment a single unit of work on an AVX2 system contains 32x16 pixels when rendering locally, so the higher the resolution, the better the chances to balance out the differences between the various tiles). We just recently tested a Quad-CPU Setup of the E7-Variant with 72Cores/144Threads and even this still worked quite good.

Sadly I don´t have first hand results for interactive performance since the systems I tested were servers and therefore had very slow GPUs that required us to use software emulation for uploading the image which is quite slow. My personal workstation however has a dual 14 Core CPU setup, so 56Threads in total, and this works flawlessly for pretty much everything you throw at it.

Of course, it always depends on your scene setup and complexity. For systems with so many cores I strongly advice against using too many material overrides since this can make the workbalancing a lot harder and often results in worse performance and longer overall rendertimes. Especially setting the glass material to a precomputed mode while rendering in Full GI is a very bad idea since this will drastically increase the number of rays being traced in a headlight for example. If have seen up to 4 times longer render times for such a setup compared to using no overrides at all without any noticable differences in the final image.

Kind regards

Michael

[Bob.Bon2000]

Hey

So are you saying that under Windows 7 x64 you did not hit any thred per application limit that other render engines are hitting and you are able to in fact run vred at 100% using anything anything north of 80 threads/buckets.

Souds like great news.

[michael_nikelsky]

Hi,

yes, I can´t remember seeing that limit anymore, we had this about 5 years ago with an 80 Core system but that is long gone. The test I did recently with the 72Core/144Thread system scaled by a factor of 68x compared to a single core with hyperthreading and turbo disabled (so 72x being the theroretical maximum but that is nearly impossible to achieve since your memory bandwidth will hit a limit eventually and you always have some sort of non-parallel code) and about 84x with hyperthreading enabled (keep in mind that hyperthreading gives you at most about 20-30% performance increase over your physicial cores, so these numbers were pretty good). Of course this was computational heavy scene and I disabled some things like adaptive antialiasing and motion blur since I wanted to test pure raytrace performance and keep the workload consistent for each sample. This gave me pretty much 98-100% usage on all 144 Threads.

So except for a rather unexpected behaviour when artifically limiting the number of cores to use in VRED when Hyperthreading was activated everything was fine (shortnote: Just never limit the number of cores VRED is allowed to use. Windows will make no difference between hyperthreading cores and real physical cores so you might be a bit surprised when telling VRED to use 72 Threads of the 144 Threads with hyperthreading enabled and end up with a performance roughly equal to running 40threads with hyperthreading disabled. It makes sense if you think about it since that basically is 36 Cores + the 20% from the 36 Hyperthreading cores, but it was a bit surprising to me).

Kind regards

Michael

[ajewitt]

I am doing research on specifications for my new machine and was hoping we could update this forum with 2016 hardware. While I don't really have a budget, I'd like to keep it reasonable (under $15k).

It seems like the Intel Xeon E5-2600 series CPUs are the way to go and I'm leaning toward a NVIDIA Quadro M6000 24 GB GPU. My current workflow (in VRED) consists of raytracing large images (over 4k) and some short animations (under 10 seconds). I also work with some very large model assemblies in other software so my machine must handle that as well. With the new machine, I'd like to be able to shorten my raytracing render times and have the ability to render out longer animations efficiently.

Does anyone currently use a Xeon E5-2600 series CPU or the Quadro M6000 GPU? If so, how is your experience? If not, what do you run?

Any recommendation, suggestion and/or other applicable information is more than welcome!