Rendering Benchmark (New Threadripper!)

[JonBDesigns]

Hi Everyone!

I just finished building a custom AMD Threadripper 1950x system. I would like to benchmark my system. I am super interested how my system will stack up compared to a dual Xeon system. I thought that there was a thread started which contained a common file that everyone was using to compare Raytrace times--however I was unable to find it after a quick search of the discussion boards. 

Please let me know if I need to move it somewhere else, or if there is already another thread started. 

 

Peace, 

Jon

 

Specs:

 

AMD Ryzen Threadripper 1950x (OC to 3.9 with plans to go further)

Gigabyte Aorus X399 motherboard

Asus ROG STRIX GeoForce GTX 1080 OC 

Corsair Vengeance 32 gb 2666 quad channel RAM

512 M.2 Samsung 960 Pro 

2x 256 M.2 Samsung 960 Evo

2TB Seagate Barracuda HD

Corsair HX750i Platinum Power supply

Corsair H115i Water Cooler

Corsair Graphite 780T case 

 

 

[Renerabbitt]

39 minutes ago, JonBDesigns said:

Hi Everyone!

I just finished building a custom AMD Threadripper 1950x system. I would like to benchmark my system. I am super interested how my system will stack up compared to a dual Xeon system. I thought that there was a thread started which contained a common file that everyone was using to compare Raytrace times--however I was unable to find it after a quick search of the discussion boards. 

Please let me know if I need to move it somewhere else, or if there is already another thread started. 

 

Peace, 

Jon

 

Specs:

 

AMD Ryzen Threadripper 1950x (OC to 3.9 with plans to go further)

Gigabyte Aorus X399 motherboard

Asus ROG STRIX GeoForce GTX 1080 OC 

Corsair Vengeance 32 gb 2666 quad channel RAM

512 M.2 Samsung 960 Pro 

2x 256 M.2 Samsung 960 Evo

2TB Seagate Barracuda HD

Corsair HX750i Platinum Power supply

Corsair H115i Water Cooler

Corsair Graphite 780T case 

 

 

If you would like to just pick out a sample plan from chief architects site and run the same settings I would sample it for you, I have a single xeon 14 core 

[JonBDesigns]

Hi Everyone!

 

Attached are two Raytrace renderings by me from sample plans provided by Chief. I have also attached screen shots from the end of the Raytrace to verify times. I just used the 1080 rendering profiles that were attached to the plans. My file size from the Raytrace exports are surprisingly low--my screenshots are larger than the renderings, which is not typical for me. I am used to giant file sizes--so I am currently trying to figure that out. One thing to note about he results is that I was performing tasks from web browsing to Photoshop while the Raytraces' were running---with zero errors or crashes. 

 

Results:

Grandview exterior- 32 passes in 1 hr 23 min = 2.59 min per pass

Fire Glass Kitchen interior- 26 passes in 30 min = 1.15 min per pass

 

I was eating during the Grandview Raytrace and is why it went so long. 

 

What times are you getting? 

 

Peace, 

 

Jon

[Renerabbitt]

On 9/4/2017 at 7:06 PM, JonBDesigns said:

 

Sorry, didn't subscribe to this thread...if you quote me then I get a notification.

Though you used the exterior 1080 defaults I think you may have had your resolution at something far bigger than the 1080 resolution...My exterior of grandview at 32 passes only took 23 minutes at 1080 resolution with aspect ratio locked, which makes me feel like your resolution was much higher. Try doing a saved camera with saved raytrace settings and upload the plan so we can compare apples to apples.

[AvoyeDesign]

To do a proper benchmark, I think you should save the camera you used with all it's settings, and upload the plan here.  We should all be using the same camera with the exact same settings, same raytrace setup, etc.  A slightly different angle of the camera can change what chief has to render, and different raytrace settings, well, I think that goes without saying.

[AvoyeDesign]

A faster way to compare systems though would be with cinebench.

[JonBDesigns]

11 hours ago, Renerabbitt said:

Sorry, didn't subscribe to this thread...if you quote me then I get a notification.

Though you used the exterior 1080 defaults I think you may have had your resolution at something far bigger than the 1080 resolution...My exterior of grandview at 32 passes only took 23 minutes at 1080 resolution with aspect ratio locked, which makes me feel like your resolution was much higher. Try doing a saved camera with saved raytrace settings and upload the plan so we can compare apples to apples.

I think that I may have figured out the problem:

1. I was writing to 2 drives (I should know better)

2. I had every light in the house on. Without any lights on I am at .32 min a pass. With some lights on I am at .625 a pass. 

I will include the plan that I used with saved camera and the lights on that I use, in the following post.

 

How many lights did you have on originally? 

 

Peace, 

Jon

[JonBDesigns]

Here are my Cinebench scores. I am unsure if the leader board on the left of the screen is correct, but if it is I am leading in both categories. I ran two CPU's and one was considerably higher after I shut down almost every running service. 

 

Peace, 

 

Jon

[JonBDesigns]

11 hours ago, KervinHomeDesign said:

A faster way to compare systems though would be with cinebench.

Here are my Cinebench scores. I am unsure if the leader board on the left of the screen is correct, but if it is I am leading in both categories. I ran two CPU tests and one was considerably higher after I shut down almost every running service. 

 

Peace, 

 

Jon

[BrownTiger]

Are you trying to benchmark real world app like ChiefArchitect and or your system synthetics/perceived benchmarks (Cinebench) - measures Cinema 4D performace.

 

Because Cinebench  rendering is different from ChiefArchitect. 

 

If later, take a plan adjust it as you would render your own project, place a camera, run your passes, post your plan and results with the saved cameras.

 

>1. I was writing to 2 drives (I should know better) 

THis depends, having to drive may improve or cause adverse effect. Using busy system drive could be detrimental. SSD are definitely faster.

[JonBDesigns]

11 hours ago, KervinHomeDesign said:

A faster way to compare systems though would be with cinebench.

I personally like to use PassMark--it is very extensive and very fun to watch. Attached are my results.

[JonBDesigns]

12 hours ago, Renerabbitt said:

Sorry, didn't subscribe to this thread...if you quote me then I get a notification.

Though you used the exterior 1080 defaults I think you may have had your resolution at something far bigger than the 1080 resolution...My exterior of grandview at 32 passes only took 23 minutes at 1080 resolution with aspect ratio locked, which makes me feel like your resolution was much higher. Try doing a saved camera with saved raytrace settings and upload the plan so we can compare apples to apples.

Here is the plan that I used. 

 

Peace, 

 

 

Jon

Grandview-savedbyjon.zip

[JonBDesigns]

11 minutes ago, BrownTiger said:

Are you trying to benchmark real world app like ChiefArchitect and or your system synthetics/perceived benchmarks (Cinebench) - measures Cinema 4D performace.

 

Because Cinebench  rendering is different from ChiefArchitect. 

 

If later, take a plan adjust it as you would render your own project, place a camera, run your passes, post your plan and results with the saved cameras.

 

>1. I was writing to 2 drives (I should know better) 

THis depends, having to drive may improve or cause adverse effect. Using busy system drive could be detrimental. SSD are definitely faster.

I had accidentally saved Chief to an M.2 drive that was not my C drive, then I was also writing to two different m.2's. Also, I re-posted Passmark scores and a new Grandview plan with camera and lights save. 

 

Peace, 

Jon

[JonBDesigns]

14 minutes ago, BrownTiger said:

Are you trying to benchmark real world app like ChiefArchitect and or your system synthetics/perceived benchmarks (Cinebench) - measures Cinema 4D performace.

 

Because Cinebench  rendering is different from ChiefArchitect. 

 

If later, take a plan adjust it as you would render your own project, place a camera, run your passes, post your plan and results with the saved cameras.

 

>1. I was writing to 2 drives (I should know better) 

THis depends, having to drive may improve or cause adverse effect. Using busy system drive could be detrimental. SSD are definitely faster.

I am just trying to compare my system on Chief as I built this system by hand almost entirely for Chief, and some Revit. 

Peace, 

Jon

[TheKitchenAbode]

2 hours ago, JonBDesigns said:

I am just trying to compare my system on Chief as I built this system by hand almost entirely for Chief, and some Revit. 

Peace, 

Jon

Finding the best CPU for Chief is somewhat challenging as there are two aspects that must be considered. When just using Chief for developing a plan and standard camera views then the CPU's single threaded performance is likely the most critical factor and when Ray Tracing it's multi-threaded capability is what you are looking for. For the cost the AMD Threadrippers offer a lot of cores/threads which will definitely prove beneficial when Ray Tracing but from the test reports I have read these CPU's do not deliver similar single threaded performance improvements when compared to Intel's latest offerings.

[Renerabbitt]

5 hours ago, JonBDesigns said:

I personally like to use PassMark--it is very extensive and very fun to watch. Attached are my results.

Here are mine, about what I thought:

 

Thought I would add for anyone interested:

motherboard $249

Ram $100

Case $50

M.2 Drive $220 

CPU $500, ebay refurbished server

GFX used on ebay $350

Power Supply $80

Water cooler-$100

All in~1650 before taxes

passmark per dollar ~2.3 not bad

[AvoyeDesign]

 

6 hours ago, JonBDesigns said:

Here are my Cinebench scores. I am unsure if the leader board on the left of the screen is correct, but if it is I am leading in both categories. I ran two CPU tests and one was considerably higher after I shut down almost every running service. 

 

Peace, 

 

Jon

Wow, that threadripper is ripped!  My cinebench CPU score for an i5 is 560.  Open GL is 118 on a GTX 1060 6ghz.  I wonder if you could get a little more out of that 1080?

 

I think that the scores on the left are from other systems within your range.  Although I would guess that you are pretty high up there.  Out of curiosity, if you don't mind me asking what did you pay for your system?

[AvoyeDesign]

On 9/4/2017 at 1:31 PM, Renerabbitt said:

If you would like to just pick out a sample plan from chief architects site and run the same settings I would sample it for you, I have a single xeon 14 core 

How does that Xeon 14 core stack up against your i7?  I assume you get 14 cores, 28 threads?  Xeon is more of a server/workstation CPU if I'm not mistaken?  What is your MOBO?  I've been curious for a while about building a system around one of those xeon multi core beasts, perhaps even on a dual cpu motherboard.  

[AvoyeDesign]

57 minutes ago, Renerabbitt said:

Here are mine, about what I thought:

 

 

6 hours ago, JonBDesigns said:

I personally like to use PassMark--it is very extensive and very fun to watch. Attached are my results.

 

Here is mine.  I don't think this puts the CPU through a rendering task like cinebench does, but I don't think cinebench puts the cpu through a more comprehensive set of tasks like this does.  So while your CPU's may be rated 5 times as well as mine on a cinebnech rendering, they are only about 1.5 to 2.5 times on this test.

[Renerabbitt]

1 hour ago, KervinHomeDesign said:

How does that Xeon 14 core stack up against your i7?  I assume you get 14 cores, 28 threads?  Xeon is more of a server/workstation CPU if I'm not mistaken?  What is your MOBO?  I've been curious for a while about building a system around one of those xeon multi core beasts, perhaps even on a dual cpu motherboard.  

I'll do the i7 a bit later, and post...these Xeons are a steal on ebay, they pull them out of servers and sell them for 1/5th the market price. Next step was going to be buying another and putting it on a multi mobo like you said. I run it in a mini itx, its an ASRock X99E-ITX/ac LGA 2011-v3 Intel X99 SATA 6Gb/s USB 3.1 USB 3.0 Mini ITX Intel Motherboard.

Curious as to why my 1070 8gb was so much slower than your 1060 6gb...I am pushing 4k resolutions, maybe that has something to do with it?

[Barton_Brown]

10 hours ago, TheKitchenAbode said:

Finding the best CPU for Chief is somewhat challenging as there are two aspects that must be considered. When just using Chief for developing a plan and standard camera views then the CPU's single threaded performance is likely the most critical factor and when Ray Tracing it's multi-threaded capability is what you are looking for. For the cost the AMD Threadrippers offer a lot of cores/threads which will definitely prove beneficial when Ray Tracing but from the test reports I have read these CPU's do not deliver similar single threaded performance improvements when compared to Intel's latest offerings.

I believe that Chief, especially X9 is no longer single threaded when developing plans and standard camera views. I recall reading a statement to this effect a number of years ago here on the forum but can not locate the thread at this time. For what it is worth, a quick demonstration confirms, at least to me, that Chief is multi-threaded in plan preparation. I use a windows gadget (CPU Usage) that displays the % usage of each CPU core/thread. Just the act of opening an inactive camera will cause all the threads to bump to 100% until the view is displayed.

[TheKitchenAbode]

15 minutes ago, Barton_Brown said:

I believe that Chief, especially X9 is no longer single threaded when developing plans and standard camera views. I recall reading a statement to this effect a number of years ago here on the forum but can not locate the thread at this time. For what it is worth, a quick demonstration confirms, at least to me, that Chief is multi-threaded in plan preparation. I use a windows gadget (CPU Usage) that displays the % usage of each CPU core/thread. Just the act of opening an inactive camera will cause all the threads to bump to 100% until the view is displayed.

I believe I also saw something about this and that CA was attempting to optimize Chief to take better advantage of multi-threaded operations. However, I think you will find that this is not applicable to all operations Also, watching core activity can be a bit misleading as even single threaded operations are spread out over the available cores. For example, I'm watching my cores right now and there are 168 processes and 1965 active threads, there is activity on all cores as these are being distributed out over my 4 available cores. This does not necessarily mean that these processes are optimized for multi-core processors in the same manner as say Ray Trace is. I don't proclaim to know exactly what processes in Chief are true single threaded or multi-threaded but I believe you will find it to be a mix and that single threaded throughput is highly relavent.

[TheKitchenAbode]

Here is an interesting article that should provide some insight into single versus multi-threaded performance. It also demonstrates that some operations that have been optimized for multi-core processors do not necessarily result in improved performance and in some cases performance is actually degraded.

 

https://www.pugetsystems.com/labs/articles/Solidworks-2016-Multi-Core-Performance-741/

 

The overall take on this is that it is a complex issue and in this case for Solidworks single thread performance is the most important performance factor other than when rendering.

 

You can also draw some further conclusions about this and Chief through the many comments concerning performance. The trend is that those with the best single thread performance CPU's seem to be the most satisfied with their system performance for most general Chief functions. For rendering the most satisfied users are those with the highest core count.

 

To be as satisfied as possible you want the fastest single threaded performance CPU with the most cores you can afford. If to have one is at the expense of the other then you need to think carefully about what you will be sacrificing and whether or not the gain on one side is worth the lose on the other side.

 

[Barton_Brown]

Interesting article. Thanks for posting.

[JonBDesigns]

On 9/7/2017 at 9:15 AM, KervinHomeDesign said:

 

Wow, that threadripper is ripped!  My cinebench CPU score for an i5 is 560.  Open GL is 118 on a GTX 1060 6ghz.  I wonder if you could get a little more out of that 1080?

 

I think that the scores on the left are from other systems within your range.  Although I would guess that you are pretty high up there.  Out of curiosity, if you don't mind me asking what did you pay for your system?

 

Thanks, I think I am going to nickname it "da Chief Ripper". When I ran those tests I had the GPU running stock and the CPU was actually only OC to 3.6. I just got detailed instructions last night on how to safely OC to 4.1 with my current equipment. Also, I am going today to replace the memory with a 3600 quad kit instead of the 2666 that I currently have which is throttling me back a little bit. Currently it is only scoring in the 60th percentile of other threadripper systems. People have been cryo cooling the threadripper and getting 5.1! 

The huge huge upside that I am noticing is that I can continue to work on my computer, even inside Chief, while the Raytrace is running as the Threadripper allocates at least one core for system processes and you can allocate more. I could never do this on a quad core or even the 1800x Ryzen that I tried for a little bit. I am sure the Xeons can do the same however, the Threadripper is considerably cheaper than dual Xeons that are benchmarked at the same level of the Threadripper.

I built the system myself with part from Microcenter after much research. I spent days dealing with "phantom" motherboard issues with unknown solutions as the motherboard and processor are on the bleeding edge as they were just released at the beginning of August. 

 

I have attached my shopping cart from Microcenter--they no longer have my motherboard in stock so I had to substitute it with something else. 

 

Peace, 

 

Jon