Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - Mattschia

Pages: [1]
1
Regarding the new light solver in v3 i can see somebody already reported issues, I wanted to add my recent experience as well while working on a project. Please check the images attached.

I can see opacity map issues on both examples. I thought this bug was fixed long time ago.

Actually I didn't mention that because I remember it was a known issue and the focus of my post was on the light solver.. but now that you mention it yes, alpha still doesn't work in all conditions, that error was fixed in our scene making a different material

2
[Max] General Discussion / Re: Corona 1.3 Benchmark
« on: 2018-12-04, 01:44:13 »
I've used corona benchmark a lot recently to test some workstations with OC that would be used as rendernode, I've noticed a big difference in watt consumption and temperature of cores between the benchmark and the actual render with v2/3, to the point that I dropped the benchmark completely. I assume it's because the latter uses the new AVX instrunction and a more recent version of embree, among other changes. Are there any plans to release an updated version of the benchmark? I believe it would be helpful also to show the difference between the newer CPUs.

3
Regarding the new light solver in v3 i can see somebody already reported issues, I wanted to add my recent experience as well while working on a project. Please check the images attached.

4
I would not worry about this. There are already some reviews testing the performance degradation after the patch and everyone agrees the difference is around 1-2% or none. Only in storage benchmark we can reach 5% but in real life scenario it will be unnoticeable. The issue will affect the big players, not us, luckly.

5
Eheh I've done worse. It's actually quite simple with the right tools, look for some video online.

Talking about frequencies 4.4 it's actually averge for a delidded chip, most people are in the 4.6/4.7 range but they probabilly don't have the pc rendering 24/24 eheh.. anyway, real frequency in Corona is lower as I've mentioned because AVX instructions run hotter. This chip has 2 AVX offset, one for the classic 256 and a new one for AVX-512 just like xeons. But it's up to you to configure if you want to run at lower multiplier or not and how much offset.. i wish xeons were so configurable as they used to be.

You can expect a reduction of 12-15deg from delidding anyway.. if you don't break it of course, in that case it will be much much colder lol


6
Hello Juraj , I'm glad that this test can be useful.

I get your point regarding price or Ram, it's a shame really. However that is common for all platforms nowdays so when i considered the value of the CPU I left them out.

My thermals are very good but I delidded the CPU and I ran a custom made waterlooling loop with an EK Fullcover..  temps never go above 65deg in overclock and 40degs at stock and here it's pretty warm. I downclocked a little for the test because Ryzen and Threadripper rarely go above 3.9Ghz, but normally I run the 7980XE at 4.4 GHz.

Mind that Corona triggers AVX frequency that has a negative offset, in my case it's limited to 4.0 Ghz on all cores.

My MB is the Asus x299-A, it's the cheapest that Asus makes for this socket.

Honestly it's a very good platform, as long as you delid the CPU or ask the shop to do it.

7
[Max] General Discussion / Re: new iMac Pro 18 cores
« on: 2018-01-01, 17:22:14 »
I've just posted a similar test here with a 18 core consumer cpu. Anyway, the new 18 core MAC should be pretty expensive and if your'e going to use it with windows it kinda defeat the point of having a MAC in my opinion. Of course if you don't really care about budget is a good option but you can have similar or better performance with much less nowdays.

8
[Max] General Discussion / Interior Benchmark - AMD vs Intel
« on: 2018-01-01, 17:12:03 »
Hello Everyone, I've done some testing during the holidays and I tought to share the results bacause it could be useful for people who want to build a workstation.



Settings: 4k resolution | 25 Pass | GI 16 | Light Samples 4

INTEL 7980XE 18 Cores (36 Threads) @3800 Mhz (AVX):

Rays/s total 12.850.000
Time: 00:15:45


AMD Ryzen 1700 8 Cores (16 Threads) @3800 Mhz (AVX/2):

Rays/s total 4.200.000
Time: 00:45:20

(Combined Time in DR: 00:12:50)

Being both CPUs at the same frequency, if we consider a core to core comparison Intel CPU has 2,25x the amount of performance on paper, however in real life with Corona the performance is almost exactly 3x.

I think this is a mixture of slightly better IPC and full AVX implementation by Intel but maybe I'm missing something else.

Of course if we consider price/performance/power ratio AMD is the clear winner here (500,00 vs 2.500,00 EUR now for MB+CPU and 170W vs 450W) but I believe that in our field hardware cost for this type of consumer grade systems is quite manageable and what is left over time is a huge performance difference.

I thought this could be an interesting comparison as those 2 are the reference CPUs at the moment for a cost-oriented consumer workstation and a performance-without-compromise one, I believe the above numbers can easily transfer to AMD Threadripper and lower Intel i9 processors as well since it's actually the same architecture.

I expect that AMD 1950x at the same frequency would complete the above render in exactly half time with 00:22:00 vs 00:15:45 of the 7980XE.

This comparison is only good for Corona and in particular 1.7 version, the official benchmark uses version 1.3 core and after some testing i've noticed a smaller performance gap between the 2 architectures, not sure what happen under the hood but something is changed and most review websites are not telling the whole story in real life application.

9
[Max] General Discussion / AVX-512
« on: 2017-10-25, 02:43:42 »
Hi eveyone,

I have a question for the developers and I believe it would be interesting for other users as well to know the answer.

Since I'm moving to Corona and I'm about to update my main workstation, I've been wondering if the AVX-512 implementation in some processors will be used by Corona in the near future and how much difference could it make. Considering the price of hardware it would be useful to know if a performance gain is to be expected choosing one brand over the other. Thanks.

Pages: [1]