(PC) Desktop specs for optimized rendering?

Replies to This Discussion

Permalink Reply by Alan Crighton on October 30, 2011 at 9:20am

Michael,

Building, or more accurately, assembling, a machine will be the best route to take, and save a bundle vs. buying a hi-end PC from one of the major manufacturers. I built my own system for rendering purposes, as my workstation grade laptop couldn’t keep up. I’m not using VRay, but Brazil and Rhino, so issues shouldn’t be much different. Rendering is generally a processor intensive task, so the core count is the primary feature that drives (or reduces) the rendering time. Personally, I’m using a quad core AMD 3.2GHz system with 8GB memory, substantially faster than my old dual core 2.0GHz system. I’ve found 8GB to be more than adequate, even for large scale renderings with numerous detailed materials. I opted for AMD as the cost point was better, and not a lot of performance difference between Intel & AMD that could justify the additional costs. A 6 core AMD processor can be purchased for $170-$180, motherboard $100-$250. For specifically rendering purposes, it won’t make much difference if you load up with DDR2 or DDR3 memory.

The graphics card has no effect on your final rendering. If you’re viewing large complex models in shaded/rendered view then a higher end video card will provide some benefit, but again, no effect on your actual rendering. I use a $100 graphics card in my system. My ideal spec for a rendering PC:

AMD 6 core processor, Asus/Biostar motherboard, 8-16GB of DDR2/DDR3 memory, Windows 7 Professional, 64 bit. I wouldn’t waste a lot of money on a video card, but rather spend it on additional hi-capacity hard drives. Hope this helps…

• ▶ Reply

Permalink Reply by Alan Crighton on October 30, 2011 at 9:54am

Forgot to add, additional processor cooling would be a great addition, running all cores at 100% increases processor temp quite a bit. An after market liquid cooler can be had for $75 - $100.

• ▶ Reply

Permalink Reply by Charles Fee on November 2, 2011 at 12:10pm

Alan-

I find your reply to be SPOT-ON ! GOOD coverage of the main-processor vs Vid card for rendering. CLEARLY some work lies ahead for developers in this area, but for now I'd say you pretty well nailed it !

 

Just a couple of clarifying questions, if I may-

1. Since RH is a single thread process, help me understand the effect that multiple cores has on the rendering process ? I can see possibly multiple processor cores coming into play with animated renders , especially those using a plug-in like VRay or Brazil, but I remain unclear where the break-out lies and whether or how these make use of the additional cores. I have a dual-core iCore-2:2.5, and I'm considering a 4-Core i-7:3.3.

 

2. A mid - to higher end Video Card may offer SOME assistance, however during the modeling and setup for animation processes, tho, as the OpenGL processes and an RH Viewport set to RENDER display mode happens at the video card, and not the main processor. I've seen some VERY nice mid-range options in the $100 - $250 range (VERY reasonable) from both manufacturers, making this an option worth considering?

 

Now here's something I find fascinating- Up to a certain point I was relying on a basic renderer (Flamingo) as the results exceeded our needs, and the price was reasonable. The downside was that as a renderer, JUST AS YOU'VE DESCRIBED, it was main processor based -  took NO advantage of the MULTIPLE rendering cores available on even our lowest end cards - and as such was S-L-O-W ! I'd come in the next morning and discover multiple HOURS left on even the most MODEST animation renders !!! Ok, and here's where it gets interesting - I had set the render to run on Flamingo, but it was just a check run and I just wanted to see how it would look fully rendered, too ! I aborted the run, discarded the frames, set the animation to run using NOTHING but the Viewport RENDER mode, and the results were ready for review in just over an hour. I came back from lunch, and was SHOCKED to find it sitting there waiting on ME ! THEN the boss came by, saw the check-run playing, and said PUBLISH ! The results were so NICE that I've completely abandoned Flamingo, and now go straight to VIEWPORT RENDER as my go-to setting ! Since the VIEWPORT RENDER mode DOES happen on the video card, I get at least SOME up-tick in performance, and the results are genuinely excellent ! Ok, I have to do a few "tricks" for things like shadows, etc, but the result are almost completely indistiguishable from the Flamingo Renders, are much more PREDICTABLE based on in-viewp0ort content, and the render times are STUNNINGLY better ! Ok, for most this is just good enuff for check runs, but for those who find this trick useful, I say GOOD HUNTING !

 

In conclusion, I find no fault with Alan's analysis, and based on my experience, I'd say his input is right on target.

 

Good luck and post here with results !

 

C.

• ▶ Reply

Permalink Reply by Alan Crighton on November 3, 2011 at 4:20pm

Charles,

Brazil for Rhino (and I think Vray and Maxwell also) utilize all possible cores to the rendering process. Rendering is performed in small regions called buckets, and each bucket is rendered by an individual core. Therefore, a 4 core will simultaniously render 4 buckets, a 6 core 6 and so on… Based on my experience, going from a dual core to quad core isn’t perfectly linear, but very close to it, so render time is just about cut in half.

 

The video card will only provide faster viewing in shaded/rendered view, zero effect on the final render. If you frequently work on large detailed models, the screen refresh can become noticeably slower if you change view in these modes and may prove to be a nuisance, for me it hasn’t been a concern.

 

Flamingo is an old rendering package, I haven’t used it since I switched to Brazil a few years ago, however Flamingo NxT will run on all available cores.

• ▶ Reply

Permalink Reply by Charles Fee on November 4, 2011 at 9:20am

Alan-

This agrees with my experience, too. Your input that the "buckets" get shared across cores really clears up that piece of the puzzle for me, but makes me want those extra 2 cores available in a 6-core processor !!! How do you think the AMD 6-Core compares performance wise to the i-7 4-Core? That seems to be the $ match point, but those extra 2 cores are suddenly VERY tempting !!!

 

I haven't been able to get any input as to how to tell Fl-nXt when to stop a render so I can use it in my animations, so if it renders at the video card, there may still be no net gain if I have to wait for multiple passes. Its interesting -  I recall back in the mid 90's when I was using ACAD-12/Win-3.11W-G and doing some amazing modeling and rendering even by today's standards, that the tool I was using (the direct precursor to Flamingo) used multiple passes, and I'd just do a CTRL-C to cancel the render at the point I was satisfied ... Its nice to see that technology has progressed so far !

 

On the Shade/Render Viewports thing- I regularly work in RENDER mode in my viewport to anticipate the final render results. I'll swap to GHOSTed mode when needed, but find a significant up-tick in performance when I swap back to RENDER mode, so a capable video card is a real help where it counts - during the work process itself.

Interestingly enough, because I've pretty well mastered Scene Lighting, and am fairly well set on Environment creation, my "Scenes" look almost IDENTICAL in viewport RENDER mode to the final rendered output, with the ONLY discernable difference being shadows provided by my renderer. Since shadows are a nice but often distracting element of a scene, I often turned them off. I stumbled on the fact that this led to my renders being virtually indistinguishable from the scene provided by the native viewport set to RENDERed operating mode, which led DIRECTLY to my discovery that if I set my output to VIEWPORT RENDER, and my operating antialias setting at 4x, I could get a per-frame time of about 3 -5 SECONDS per frame instead of Minutes to portions of an HOUR per frame ! My render times dropped from DAYS to MINUTES with STUNNING results ! Ok, I'd love to be able to get special effects like multi-colored paint (depending on view angle) - oh wait I get that by the use of specular control and lighting in my scene - oh, never mind ! Ok, tongue in cheek, but you get my point ! AND all of this is available using RH-4 right out of the box .

 

Its looking like an associate I do some of my work for is looking into some of the things we've been talking about, so if things go as I hope, it'll lead to a significant software and hardware purchase, and I'm looking at Vray as a possible contender based on your input here and others who have used it here with success. If I can get a real-time in-viewport scene renderer, tho, I'll probably go that route, as the results will be very superior, and except for shadows (see above) the render frame rate should be exceptional, as well ! Anyway, more on THAT as it develops !

Thanks for your input - its been a huge help !

C.

• ▶ Reply

Permalink Reply by Alan Crighton on November 4, 2011 at 11:55am

Charles,

 

Currently, the hi-end Intel core I7's are outperforming AMD. But since there's a myriad of versions to choose from, it makes a direct comparison difficult. Intel may score higher in some benchmarks and AMD in others, depends on the tasks. I haven't seen a direct raw processing comparison between comparable models of I7 (4 core) and Phenom 6 core to see what difference there is, though I would have to think x6 vs. x4 rendering would have to be faster?

• ▶ Reply

Permalink Reply by Jon on November 9, 2011 at 7:47am

Most of the information posted thus far has been correct.  The intels have been doing much better than the AMDs in most situations.  I have also seen some reports that the 2700 is outperforming new bulldozer chips when using vray for 3ds.  How this will translate to VfR I couldnt tell you, but I would still take it into account.  Most people using a single box for rendering will go with 2700 chipsets.

 

That being said, have you considered going to a farm with servers?  You can pick up some inexpensive 4x3.6 ghz boxes for a couple hundred bucks on ebay.  All you need to load is the DR spawner onto each system and you are ready to go.  I went with servers because all I really needed was processor and nothing else.  I dont even have a monitor on the server rack because it isnt needed.  

 

One last thing to consider is where render engines are heading in the future.  Currently VfR is processor based, hoewever I think it is going to go to GPU in the next iteration.  When ASGvis owned it, they were co-developing it with a Sketchup version.  I believe the sketchup version 1.49.xx was released as GPU based.  They never got around to releaseing the 1.49xx for rhino before Chaos purchased it.  If you look at Chaos group, they are starting to build vray for 3ds as GPU based.  When that happens, there are going to be a lot of farms out there out of luck.  

• ▶ Reply

Permalink Reply by Charles Fee on November 14, 2011 at 1:15pm

Alan -

" ... it makes a direct comparison difficult. Intel may score higher in some benchmarks and AMD in others, depends on the tasks."

Spot-on CORRECT ! (in my humble opinion ...!)

 

Jon -

Yeah, I'm seriously considering the option of "farming" out the rendering process as you've described. It seems that in the near-term there may be some significant advantage to duplicating the render task across multiple base platforms as you mentioned. It SEEMS that someone interested in render-farm platform power as you describe could gain a significant near term advantage for relatively small $. AND I'm not CERTAIN that a shift to GPU based rendering would really negate that advantage ! I could EASILY see plugging in a dual or triple video card solution into EACH farm "member" as the budget would allow, and for multiples of $100 to $300 per farm node . Taking advantage of the re-write to the GPU you mention, someone could inexpensively populate an impressive scalable rendering solution !

Thus the advantage MIGHT shift from multiple moderately paced main processors back to fewer much faster main processors. The key would STILL reside in the strength of raw processing power. intel offers fewer processors per core, but they are clocked higher and offer internal methods for breaking out processor streams or "threads" to provide effective MULTIPLE internal cores and the resurrection of an OLD idea - "TURBO" mode temporarily boosting clock performance ONLY when needed from ~3.4 Ghz to well over 4 Ghz, dropping back when the draw load backs off. Of course multiple cores necessitate maximum RAM to take full advantage.

SO - the question remains - 4 faster and more powerful cores per processor, or 2additional (to 6-core) but less powerful cores in the base CAD unit ? That's where you'd have to start, and where you'd jump off from as you then expanded on your core system with the additional nodes as you could afford them.

Please note - its not TASK BENCHMARK but THROUGHPUT-to-PRODUCT we're discussing here.

Maybe when we've been running CAD as an application of interest long enough, someone will stumble on a market interest in replacing a general best-guess approach to component selection with a more pertinent approach to understanding what the CAD market is really looking for - EFFECTIVE and specific performance comparisons matched at various budget levels, each meaningful in it own sphere of operation ? After all, CAD as a standalone PC application has only been around since the early 80's - apparently not nearly long enough for the technical geniuses among us to really master the issues involved in arriving at genuinely masterful clear, detailed and meaningful conclusions to the kinds of questions raised here - after all, the debate goes on !

No matter which evaluation you see, it always devolves to a typical disclaimer - " ... of course results vary and you're really on your own - take your own best guess ... !" We see lots of PC laboratory evaluations of some favorite pet construction of some sort submitted for evaluation, each claiming to be a fastest or bestest PC/MAC system around !!! And of course so much MORE clarity is brought to the issue when you factor in the MAC/PC debate ! A note to those reading here who provide us with these wonderful insights - Any substantive response to the questions we've raised ?

In the mean time - Jon & Alan's input agree completely with my experience, and the RENDER FARM idea is an excellent one I'm really looking into !

Tks-

C.

• ▶ Reply

Permalink Reply by Jon on November 14, 2011 at 1:47pm

Some news from Chaos about Vray for Rhino popped up on their page today.  Looks like they are going RT like 3ds, so there will be more of a load on the GPU.  Looks like Im going to have to start filling in the vid slots of my machines in the near future...

Chaos Group Brings Exciting V-Ray Capabilities for Product Designers and Architects at Shape to Fabrication

Our team is getting ready for a productive conference and exhibition at this year’s Shape to Fabrication event organized by our partner Simply Rhino. Our presenter Konstantin will be showcasing key features of V-Ray for Rhino and V-Ray for 3ds Max. Highlights of his presentation include:

Interactive rendering in Rhinoceros with V-Ray RT
- Setting up a scene with the RT used for preview
- Changing materials lights and camera settings interactively
- Using the RT for production purposes

Improvements of the V-Ray material
- Using the dispersion parameters
- Controlling the anisotropic reflections with textures
- Interpolating glossy effects

Global illumination and light improvements
- Using the GI ambient occlusion effect
- Improving the quality of the light cache with the retrace threshold options
- Using dome light for image based lighting

Improvements of the V-Ray displacement
- Using the water level value

Using V-Ray proxies in Rhino
- Creating and importing proxies
- Rendering extremely high detailed geometries in the 32bit version of Rhino with the help of the V-Ray proxy
- Using the proxy for architectural purposes

V-Ray Lens effects
- Setting up the bloom and glare lens effects

 

• ▶ Reply

Permalink Reply by Charles Fee on November 14, 2011 at 5:35pm

WOW ! This is GREAT News ! THANKS ! Seriously. Since in-viewport rendering (as far s I'm aware) happens on the GPU with only final renders taking place back at the main processor, this could be a real game changer. SO- It ssems that the right answer is a fast main processor with multiple cores AND a good Video Card with CUDA Cores ! Ok- looks like we're getting very close now to an answer. Thanks for the update !

-

C.

• ▶ Reply