SU Podium Forum
Register  |   |   |  Calendar  |  Latest Topics
 
 
 


Reply
  Author   Comment   Page 1 of 2      1   2   Next
Janusz_Biela

Avatar / Picture

Registered:
Posts: 33
Reply with quote  #1 
Hi!

We just released official K3 Benchmark:


web score:
http://www.kraytracing.com/kraybench/

download:
http://www.kraytracing.com/pub/kray3bench/

Facts about benchmark:
- full and pure 10 bounce lights without fake solution
- Progressive Photon Mapping
- QMC GI sampling (brute force)
- reflection blur , refraction, transparency, translucency, instances, LEM Lights, Physical Light, Physical Sky and Sun, bump.
- 16 passes of Anti-Aliasing (Grid 4)

Greetings from K3 Lightwave Render Engine!

__________________
http://www.januszbiela.com
bigstick

Avatar / Picture

Moderator
Registered:
Posts: 10,249
Reply with quote  #2 
Thanks Janusz! This will be a good test for Dave's new 72 thread machine!

For the benefit of people who don't know, Kray is the render engine that powers Podium. The latest version, Kray 3 has just been released, and it is what Podium 3 will be based on.

You can download and run the benchmark file, which creates this render:

Screen Shot 2017-09-18 at 22.24.09.png 

This is a 19 minute render on my 2013 Core i7 MacBook Air, which would place me in last place on the Kray benchmark table!

As you can see, it's an unbiased QMC algorithm. The application simply runs a pre-packaged Kray script. You can download and run the benchmark application in a couple of clicks. If you want to add your results to the table, I think you have to render on the Kray forum. There is lots of cool and interesting stuff there, and lots of information on the render engine.

Janusz is a professional visualiser, Kray's main tester and he is a key part of the Kray engine development process. He's one of the men I ask when I have difficult questions!

Check out his website - and try not to swear too much [wink]

Edit: Dave has been getting questions about Podium 3. It is not imminent.
We have started work but we do not even have an alpha version yet.

We do not have a feature list, we do not have any upgrade costs and we do not have a timescale.
If anyone is wondering whether to buy now or wait for V3 - buy now.



__________________

That which does not kill us makes us stronger
-Friedrich Nietzsche

Janusz_Biela

Avatar / Picture

Registered:
Posts: 33
Reply with quote  #3 
Also we released Android version.
__________________
http://www.januszbiela.com
arqcova

Avatar / Picture

Registered:
Posts: 287
Reply with quote  #4 
Cool! going to run benchmark right now!
__________________
https://www.behance.net/CuevaRoja
arqcova

Avatar / Picture

Registered:
Posts: 287
Reply with quote  #5 

instances!? yeah!

 

kray.jpg 

 

Thanks!



__________________
https://www.behance.net/CuevaRoja
Nick00

Avatar / Picture

Maître érudit
Registered:
Posts: 3,294
Reply with quote  #6 
Well 2558- nowhere near Janusz's score [smile]
__________________
Podium Tech. Support
JustinSlick

Avatar / Picture

Moderator
Registered:
Posts: 306
Reply with quote  #7 
Wow I just ran it on Dave's 72 thread machine and it was disgustingly fast. (Still short of Janusz though).

So the 36-thread limit that was a problem in V2/V2.5 is obviously solved!

kray3-benchmark.png


-----------------------------------------
And.....

Laptop Test: Core i7 7700HQ
5:01 - Score 1990


__________________
What's New in SU Podium?
What's New in Podium Browser?
arqcova

Avatar / Picture

Registered:
Posts: 287
Reply with quote  #8 
WOW less almost a full minute! very nice!
__________________
https://www.behance.net/CuevaRoja
Janusz_Biela

Avatar / Picture

Registered:
Posts: 33
Reply with quote  #9 
This is our near future:
- almost Path Tracing quality in less than 3-5 minutes for half HD with normal complex scenes
- neat images (K3 has the best noise balance ever) - there is no situation when one region of render is clear and second one has noise
- in future real time tone mapping with all control aspects (clip dark/bright, contrast, burn  out, LUT, glow, etc)
- shader system for realistic  surfaces
- VPR render with full real time interaction (no more static renders)
- more optimization for PPM and also VCM technology for more accuracy

__________________
http://www.januszbiela.com
jlo

Registered:
Posts: 56
Reply with quote  #10 
Can you provide info on that limit?

Both my machines rendered into 123rd place...
xeon w3565| 7:28 | 1338 score
xeon w3520| 7:49 | 1278 (now 124th place)
both 4core 8thread
Janusz_Biela

Avatar / Picture

Registered:
Posts: 33
Reply with quote  #11 
Quote:
Originally Posted by jlo
Can you provide info on that limit?

Both my machines rendered into 123rd place...
xeon w3565| 7:28 | 1338 score
xeon w3520| 7:49 | 1278 (now 124th place)
both 4core 8thread


This is what I explained on our forum:
- complex render system is NOT only about measurement speed of CPU
- when you have very synthetic CPU test which needs only CPU the score will be different when you compare it to test which use a lot RAM/Cache/CPU swapping data - in this case Global Illumination
- K3 uses Full Photon Mapping solution and QMC sampling so data for calculation (RAM octree build, segmentation of RAM, CPU Cache, CPU - RAM swapping information) is really big.....it means, two same CPU will have different scores when MOTHERBOARD and RAM are different. 
For example low motherboard bus speed and RAM frequency.
- benchmark K3 test whole computer system: CPU, CPU cache, RAM, BUS speed, RAM speed

We have here in office exactly the same hardware (almost 100% identical) but one of them is 3% faster.
This faster one has newer version of motherboard with newer chip (Asus).....

__________________
http://www.januszbiela.com
jlo

Registered:
Posts: 56
Reply with quote  #12 
Thank you Janusz for that info!
That sounds spot on in terms of reality on how minor differences output big resulting differences. It's about the entire chain.

Having said that, I was looking for more info in the cpu limit in podium v2.5.
Here is where I'm coming from, my renders are taking way too long so I'm considering a monster used (old) server of something like 40 cores/ 80 threads as opposed to getting a new something like ryzen/ threadripper server that will actually cost more to purchase. I'm curious of what walls and issues I'm going to hit before I actually make a purchase end up running into things (maybe literally).
Janusz_Biela

Avatar / Picture

Registered:
Posts: 33
Reply with quote  #13 
I think situation for Ryzen vs Intel must be a bit develop on market. We must wait (soon) for Intel move with i9 CPU (already i7 and i9 got cheaper after Ryzen release).
Suggest wait up to end of this year.
We have here 3 machines with Dual Xeon 72HT...I know this is expensive solution but this hardware works under stress almost non stop without any problems.
To clarify situation watch benchmark scores with Ryzen and Intel. Compare speed and prices and think about future Intel moves. Intel for me means reliability.
If solution will be 10-20 % more expensive for Intel I would choose Intel.

__________________
http://www.januszbiela.com
JustinSlick

Avatar / Picture

Moderator
Registered:
Posts: 306
Reply with quote  #14 
jlo, I'll add that the limit I mentioned above does still affect V2.5.

I don't have comprehensive knowledge on this, but above 32 cores Windows divides the load into two discreet processing groups.  Most software wasn't equipped to handle it right when the 18 core Xeons were released so people running dual Xeon E5-2660s ran in to problems.

https://area.autodesk.com/blogs/maxstation/n268-how-many-cores-does-3ds-max-support/#50873
http://www.tomshardware.com/answers/id-2507034/applications-cpu-power-cores-threads.html

We only discovered it in relation to Podium this year when Dave bought a dual Xeon (36 core) workstation.... and found that Podium V2.5 would only use 18 cores and 36 threads instead of the full 36/72.

It has now been solved in Kray 3 (as seen above in the Kray 3 benchmark image above which used all 72 threads).

So this will be remedied in Podium V3, but I don't have any time-frame for you; it is pre-alpha essentially.


__________________
What's New in SU Podium?
What's New in Podium Browser?
jlo

Registered:
Posts: 56
Reply with quote  #15 
There is an old 1 message thread on here that provided a msft link. The link said windows has a group limitation/segregation at 64 threads in the os that unless software is made to account for this grouping then that software will be limited to 1 group and max 64 thread usage. 64 may be the number you are thinking or that thread may be incorrect.
That forum post got around the limitation by changing the bios so the OS only knew it had 64 threads and he was able to render using all 64 threads (but never could use the extra threads for anything while in that configuration).

ebay has cheap old servers less than $1000 that have around 32 to 40 cores (64-80 threads) with tons of ram. These are typically intel based and should be monster machines but the cpu speeds will be lower somewhere near 2Ghz. I assume the thread count would still overcome any speed limitation and should be more than amazing for render purposes. Do you guys see any downside to this approach especially any downside against a new amd/intel cpu of a consumer workstation cost type of ratio? 

Another related issue is that these servers have 4 cpus per machine and sketchup, podium require a consumer os -and- any consumer OS like win10 only allows 2 cpus per OS so a windows server version would need to be the OS. How have you overcome this limitation? Does sketchup and podium install on winserver?

on the kray benchmarks- all the top numbers are around these 72 thread count numbers so im assuming the old slower server but crazy high thread count is still the best way forward for (probably long) time being - especially on a budget. 

your thoughts?
Previous Topic | Next Topic
Print
Reply

Quick Navigation:

Easily create a Forum Website with Website Toolbox.