Another New Computer Thread - Dual 6 Core Processors

[HumbleChief]

I just upgraded to a dual Xeon system, (specs in signature) as I was really bogging down with some larger models and RayTracing was just taking too long.

Even my 3D rendering was slowing down so I had a new system built with dual, 6 core Xeon L5640's, which is a low voltage older (2010) Xeon chip.

It was built using EVGA SR-2 Classified Mother Board because it's one of the few MB's that can be overclocked with socket 1366 Xeon chips.

Also threw in a Nvidia GTX 780 video card which,, IMO, might actually be overkill for the current state of Chief.

I still have my old system so I took a fairly simple room camera view, added about 18 lights and ran a RayTrace test. I never add so many lights but wanted to stress it a bit.

Old machine (specs still in signature) - 6 passes - 15 minutes 9 seconds

New machine - 6 passes - 5 minutes 44 seconds.

Haven't really used it much day to day but if can duplicate that performance I'll be happy.

For the super geek the pass mark score was over 18,000 which is just bit faster than the fastest single Xeon processor. http://www.cpubenchmark.net/high_end_cpus.html So it's pretty fast. Is it fast enough? A resounding no. I can't imagine a computer ever being fast enough but this should take some of the wait out of the design process.

Even though it's very fast a build like this is NOT for the faint of heart. The first MB got wrecked during shipping so even though I did not want to build another computer I ended up re-building this one any way.

The Mother Board can be a bit ornery with a few things to watch for. The MB is HUGE and the case is even bigger - really never thought a computer case could be so large. Not too bad on the power because they are low voltage CPU's - but there are 2 so it's still going to eat some electricity.

I know there are others with dual Xeons and was wondering if they are happy with performance?

Ask question if I missed anything and I'll update my sig with all the specs.

[perryh]

I just upgraded to a dual Xeon system, (specs in signature) as I was really bogging down with some larger models and RayTracing was just taking too long.

Even my 3D rendering was slowing down so I had a new system built with dual, 6 core Xeon L5640's, which is a low voltage older (2010) Xeon chip.

It was built using EVGA SR-2 Classified Mother Board because it's one of the few MB's that can be overclocked with socket 1366 Xeon chips.

Also threw in a Nvidia GTX 780 video card which,, IMO, might actually be overkill for the current state of Chief.

I still have my old system so I took a fairly simple room camera view, added about 18 lights and ran a RayTrace test. I never add so many lights but wanted to stress it a bit.

Old machine (specs still in signature) - 6 passes - 15 minutes 9 seconds

New machine - 6 passes - 5 minutes 44 seconds.

Haven't really used it much day to day but if can duplicate that performance I'll be happy.

For the super geek the pass mark score was over 18,000 which is just bit faster than the fastest single Xeon processor. http://www.cpubenchmark.net/high_end_cpus.html So it's pretty fast. Is it fast enough? A resounding no. I can't imagine a computer ever being fast enough but this should take some of the wait out of the design process.

Even though it's very fast a build like this is NOT for the faint of heart. The first MB got wrecked during shipping so even though I did not want to build another computer I ended up re-building this one any way.

The Mother Board can be a bit ornery with a few things to watch for. The MB is HUGE and the case is even bigger - really never thought a computer case could be so large. Not too bad on the power because they are low voltage CPU's - but there are 2 so it's still going to eat some electricity.

I know there are others with dual Xeons and was wondering if they are happy with performance?

Ask question if I missed anything and I'll update my sig with all the specs.

Nice set up Larry, you should be able to use it for along time.

[Joe Carrick]

Larry,
The Ray Trace difference may partially be due to the difference in Memory. 8GB vs 24GB Ram would improve performance by itself. There's also a question of whether Chief's RT is actually taking advantage of all the cores or not.

The GTX 780 performance boost will probably be more noticeable if you add some 3D Dentil Moldings and/or Complex Balusters (lots more faces) to the model. Naturally that is in Render Views with rotation.

[HumbleChief]

Yeah Joe the memory can't hurt but yes indeed all 12 (12 real and also 12 virtual) cores are pegged at 100% during RayTracing and 3 times faster I'll take all day long.

Accessing all cores was something the builder and I were curious about based on what CA has said about its Raytracing engine, its 100% CPU usage, and the advice that more cores the better. Always wondering where that limit might be and it seems to access 12 cores without problems.

When I told the builder that CA uses 100% of the CPU for RayTracing this prompted the decision to go with dual Xeons instead of a single processor. They can actually work better than the sum of their parts which gets pretty technical but instead of getting say 190% of the 2 processor's combined power you can realize over 200% because of the way Xeons process data. Can't say I understand it but it is indeed really fast.

This is my second day and there's all kinds of speed benefits that go under the radar at first but it just snaps to as you might imagine. Didn't get an SSD drive but that will be next to speed it up even more.

I always wondered where the 3D bottle necks were and will keep the dentil molding etc. in mind as the models get more complex.

[HumbleChief]

Nice set up Larry, you should be able to use it for along time.

Yeah as long as CA doesn't implement the GPU for RayTracing which might render (pun intended) all that CPU power moot.

When I did the tests I had all the caustic photon and the other advanced items checked in the RayTrace dbx but never really use them. Just wanted to compare times. When I un-check those options the RayTracing is of course even quicker and makes generating RT's a real treat.

[jcaffee]

Yeah as long as CA doesn't implement the GPU for RayTracing which might render (pun intended) all that CPU power moot.

Who knows, maybe Chief goes hybrid ala Bunkspeed. There's always hope.

jon

[HumbleChief]

Who knows, maybe Chief goes hybrid ala Bunkspeed. There's always hope.

jon

Wow another great rendering engine I've never heard of. They always seem to be just out of reach or over qualified for my uses.

[CARMELHILL]

I had a dual Xeon E5405 2.0ghz with 30 gig ram and a GTX 580 card which Dell built for me in 2009. It came with a Nvidia Quadro FX4800 which I upgraded in 2012 to the GTX580. It was lightning fast when I got it for Vectorworks, but the original video card was really for Autocad work. When I swapped in the Nvida GTX 580 it's renders were cut 50%. Last month the front USB plugs died, and I use them a lot. So I built my own system from scratch,

see post #22
http://www.chieftalk.com/showthread....w-System/page2

I haven't switched over the to the new system quite yet because the office has been too busy to install all the new print drivers etc.., but that'll be next week. I've been toying with the idea of building a similar system using a single i7-4960 or going the dual Xeon route again. It was surprisingly easy to build myself so I might enjoy it during the winter. But I can't justify spending more than $1500 on just the CPU's.

[HumbleChief]

Thanks for the info Edward. I ended up going with the dual Xeons because the builder got a great deal on the chips and the entire system cost a little over $2500.00 and that was with paying the full $650.00 for the 780 which has dropped $150.00 since then. I was looking at around $2000.00 for a system similar to the one you linked to, which I just realized was at a similar $2000.00 price point, but couldn't match the CPU horsepower for only $500.00 more - it was a bank for buck decision.

The builder is interested in building and benchmarking Chief machines so if you're interested I can give you his name. As I mentioned above though you need some computer experience (which it sounds like you do) if you want to go this direction as I had a problem with shipping and had to replace the Mother Board. Not a daunting task as I've built many computers but something I was willing to pay to have done, again especially with the bang for buck.

[Doug Park]

The speed improvements in ray tracing are due to having more cores and generally faster throughput. More memory beyond 8GB is likely not a factor for the cases you tested, but is nice to have for disk caching. Adding an SSD would likely improve some things as well.

Chief is designed to use as many cores for ray tracing as are available.

[HumbleChief]

Thanks for that clarification Doug. I remember previous posts about RT being optimized for more CPU cores and also the need for not much more than 8GB RAM. I ended up with 24 GB RAM because of the triple channel Mother Board setup requiring 6 RAM sticks and 4 GB each stick was the natural choice.

[Doug Park]

More RAM is generally better, although it could lead to having models that are larger than practical for the performance of the machine.

[HumbleChief]

That's an intriguing comment Doug. If it's not too technical would you mind expanding on that thought a little?

[Doug Park]

It follows a similar "law" to "your stuff expands to fill all available shelf space".

In other words without a hard limit on how big a model can get one can get lazy about keeping the model small and before you know it the model becomes so large that it slows your system to a crawl. I don't want to imply you are lazy. Just that it is easy to ignore creeping complexity of a model until you hit some limit, such as performance, or memory to bring it back to your attention.

[HumbleChief]

So the larger and faster we make our systems the larger the model we might inadvertently create which will then require a larger more powerful system which will encourage larger models which will require larger....but isn't the idea with a fast system to create larger and more complex models? And how else would we do that without a faster system?