X4 Ray Tracing - Optimal CPU

It looks as if I may need to do close to a 100 home sales brochures which will include ray traced images.

A new computer is needed... the question becomes:

Has anyone done a comparison of elapsed times for ray tracing with CA X4 on multi-core processors?

Note: this a CPU benchmark question. RAM / HD or SSD speeds are not a factor.

Since the computer will run Windows 7 Professional 64 bit, the other question becomes:

Is there any speed difference between CA 32 bit and CA 64 bit performing ray tracing?
(The only true way to test this would be on a dual boot (Win 32 bit / Win 64 bit) computer - ensuring exactly the same hardware).

At the bottome end: the ATI Radeon X1650 in ye olde Sempron 3400+ baseline computer died recently... have an Innovision card on order... maybe next week... - I will use that for single core CPU tests against my P4 dual core 3.6 Ghz box. In most single task applications the Sempron is almost as fast as the Intel. Both computers are 2006 vintage.

From what I have seen using a new Intel i7 running up to 3.4 Ghz, it's not much faster than the old P4 3.6 dual core (which was the fastest Intel CPU in 2006... all they seem to have done is reduce heat and add more cores since (?))

Results to be posted here when available

Ray tracing takes advantage of all of your cores about as much as anything can. Most of the time you will see all cores pegged at 100% while ray tracing.

This means that if you are looking for the fastest possible ray tracing machine more cores is best.

Hi Doug,

The "choice" is mostly between AMD or Intel....

New AMD FX8150 (8 core) seems good on paper.
Intel web selector recommends for CAD the i7-2600 with a separate video card (many Intel i series have varying inbuilt GPUs - but they are not specialised. DirectX rather than OpenGL targeted). The 2600 has minimal on chip GPU; then add a good OpenGL video card.

I found a web review of the FX8150 ("Bulldozer") which did a lot of single then multi-core benchmark comparisons with the i5-2500. Results generally indicated that the thread / core processing distribution was erratic with the AMD chip - Apparently Windows 7 doesn't know how to optimise for the new AMD chip. Some good results, some poor. Generally the Intel outperformed it until the task hit the sweet spot with the AMD, then it "Bulldozed" ahead.

CPU benchmarks have always been synthetic and not necessarily indicative of specific application performance.

What I do not know is: how the CA Ray Tracer will spectrum the load across threads and cores in the AMD vs Intel chip architectures. AMD claim 8 true cores vs Intel 4 core 2 threads per core. L1 and L2 caches may also be an issue. Large caches can slow down small fetches so a lot depends on the size and frequency of data calls. Ray Tracing will do a lot of data fetches? Will cache size and pipeline be significant factors??

An AMD system will probably be cheaper than an Intel. The computer will sit in the office working Chief Architect 90% of the time - so games, surround sound and Blu-ray movies are not required. Neither will it have several Microsoft Office apps open. AMD seem to be trying for the specialised users while Intel are after the mass market.

About 4 years ago, in a Revit users group we benchmarked several tasks (file opening, 3D view rotation, rendering, etc) and the humble AMD Sempron outperformed much more expensive Intel processors, against all expectations. The key seemed to be the small cache with an efficient data pipeline in the Sempron was quicker getting single, sequential tasks done than the Intel chips with dual cores and impressively large caches. With a 40 Mb Revit file, an 8Mb cache is a waste of time. Interesting.

Cheers.

I ended up with an AMD 8120 8 core cpu. It pegs all cores to 100% when ray tracing.

Doing some tests with the plan and same saved camera on my new and old machine were interesting. What would take an hour 20 on the old basic dual core machine takes about 10 minutes on the new 8 core.

No idea how it would compare all things being equal to an Intel fast chip, but, this works very well and is light years ahead of what I had. No issues on set up or anything. It worked out to be a very good fast budget build.

Thanks James,

That's want I want to hear... ray tracing down to 10 minutes.

It currently takes 2 to 4 hours on my old Intel. From what I've seen on a new Intel i-7 machine, it's not that much faster than the old box. I would have expected it to be at least 3 times faster.

My hopes are that AMD have produced a good workstation / CAD class processor range with the FX 8 series. I suspect Intel i7s to be more tuned to the Microsoft Office workstation than CAD / 3D modelling.

For the last 6 years, my preferred video cards have been ATI FireGL for Revit and Chief. They are CAD specialist OpenGL power cards. I have yet to use an NVIDIA card that comes near the FirePros. Besides render speed, the FirePro cards have very good text and vector line display that present a sharp image when zooming in and out. NVIDIA cards have generally been fuzzy with text in Revit. Also ATI Hydravision is very good at multi monitor display handling... NVIDIA just hasn't bothered with niceties like fill child windows to monitor. Both Chief and Revit are MDI (Multi Document Interface) applications so a good software tool like Hydravision is really handy. (ATI Radeon cards are very poor cousins to the FirePros... so please don't think Radeon. Radeons chase the gamers market.)

... money saved on the base system and CPU can go into a more upscale FirePro video card.

I'll put a simple test Plan together for benchmarking if anyone is interested.

Cheers.

I wouldn't go with AMD over Intel when it comes to CPU's. I recently built my rig for raytracing, and all my research pointed to Core i7 CPU's (that was about a year ago). AMD are generally less expensive, but they do not benchmark as well as Intel CPU's when it comes to raytracing. Check out www microcenter com for great deals on Intel CPU's if you are building your computer yourself - they sell them under cost.
I have an old Dell with a Pentium dual core processor (around 2Ghz or so). The Pentium i7 processors have 4 cores and hyper-threading, so the PC thinks it has 8 processors. Renders that take 5 minutes on my old computer are done in the 47 second range on the i7 rig.

rongobong,
Thanks for the reply.

I'm back in Australia... it's not worth buying computer parts from overseas... darn hard to enforce any warranty; or deal with shipping (a fortune by air or 2 months by sea).

The AMD FX-8xx series have 8 full, true cores... not 4 with dual threading as do the Intels.

I have created a test Plan that is currently running POV on the other computer.
Chief X4 ray tracer completes it in just under 20 minutes. Exported to POV; POV Ray 3.6 took 1 hour, 52 minutes to ray trace it.

Conclusion so far: the CA X4 ray tracer is about 5.5 times faster than POV Ray on an Intel P4 660 3.6 GHz. Unfortunately, the POV Ray image is brighter and cleaner than the CA X4 image.... there is a bug/feature in the X4 Export that doesn't quite send the same lighting data. It literally would take hours to try all the settings to get the same render result.

I will post results and the test plan file later.

Cheers.

Attached is a test plan for ray trace benchmarking.

It's an exterior frontal view of a simple house with plants to test textures, reflectivity and shadows.

In the plan file is a camera named: "Benchmark Camera" - Open the view (don't move within it) then hit the Ray Trace button and Ray Trace with the set "Benchmark" configuration.

Also in the file is the ray traced image.

When finished, Chief will display the ray trace elapsed time in the view title bar.

My results have been intriguing:
Intel P4 660 3.60 GHz - timed: 18:09 minutes
AMD Sempron 3400+ - timed: 25:19 minutes

The Sempron only has 1GB RAM and it hit the pagefile a little... slowing it a mite. All synthetic CPU benchmarks have always rated the Sempron as being a lot slower than the Intel than it really is.

Atempts to obtain a meaningful comparison by exporting to POV Ray were foiled by Chief not exporting the configuration settings. Dismissed the results (after re-running at 640x480 NoAA) of 24:00 minutes and 29:48 minutes respectively... but the relative times seem consistent.

These are 2006 vintage computers.... how much faster are the current breed?

Take the challenge and gain bragging rights for your computer.

Cheers.

These are 2006 vintage computers.... how much faster are the current breed?


Ian:

seriously ???

I suspect they are "light-years" faster :eek:

I don't do raytraces - nor benchmarking
so I have no personal experience

but I can say that a 2006 PC is rapidly approaching "antique" status

Lew

Lew:

I think Ian is just having fun with us, as POV is 1990s tech, and nobody in their right mind uses this.

I'll go back to my broken record: buy a Boxx Technologies renderPro w/12 cores and call it a day. Boxx was in the Intel booth @ NAB and their 128 core rig was truly a thing of beauty, expensive as all heck, but really, truly stunning.

jon

Good deal posting that plan. I was wondering what my new one would really do.

This is a result from my 1 year old $400 HP laptop. It is an AMD a6 quad core at 3400 ghz.

7:04, just opening the plan, clicking the camera, and hitting raytrace. I tweaked the image properties a bit from yours after the fact.

I will do this with my work PC on Monday.

1 min. 8 sec. with my system. The image was pretty dark though.

Does 43 seconds seem too slow? ;)

I got 1minute 26 seconds. Thought I would check it out.

Holy Cow George, your hardrives cost more than my whole computer...I'm sure it's fast though...very nice.