The much-anticipated Terragen 4 benchmark is here! Big thanks to Ulco Glimmerveen for creating the original scene and allowing us to use it as the basis for this benchmark. As with the Terragen 3 benchmark, this new version has been designed to test Terragen 4-specific features like the new v3 Cloud Shading (multiple scattering), Robust Adaptive Sampling, Lens Effects, and more. For the first time we are also including several bundled objects, among them two that were kindly provided by IDV Software, makers of Speedtree. You may use these objects only in non-commercial projects. As a result the benchmark download is much larger, but still a reasonable ~14MB in .zip format.
[attachimg=1]
Please note that due to the user of newer features like robust adaptive sampling, you will need Terragen 4.3 or newer to run the benchmark. Those of you whose maintenance is expired and who do not have access to 4.3 can still use the free version to run the benchmark. Just be sure to copy your older version install folder to a new location before installing 4.3 (or maintain the install file for later reinstallation).
Head on over to our Terragen 4 Benchmark page to download the scene and submit your render time, then check out the full list of results:
https://planetside.co.uk/terragen-4-benchmark/
We've also added a new results page that will allow for easier viewing and sorting than before (the full spreadsheet view is also still available):
https://planetside.co.uk/terragen-4-benchmark-results/
We can't wait to see results from some of the newer powerhouse CPUs from AMD and Intel!
- Oshyan
Excellent (and very Dune ;) )
I have been looking forward to testing this out :)
Do we need to turn off the logical cores for the test?
Sorry, I had to send the benchmark form twice, since I accidentally checked "overclocked" in the first form. It's not, so please delete the one with overclocked checked.
If I remember right (I don't "remember right" as often these days), we couldn't make more than one submission in the old benchmark database (for the same system.) Is this still true? I would think it beneficial to do that during the v4 time frame when new builds are released. [edit] It could show where a type of cpu/memory made a difference, positive or negative or just comparing different (TG) builds.
Quote from: archonforest on April 22, 2019, 05:39:24 AM
Do we need to turn off the logical cores for the test?
No need to change anything from your default configuration. If there are any issues with really high core counts, those should be shown in the benchmark and then we can see places to optimize further.
Also you can submit multiple results, so you could try with and without hyperthreading (I assume that's what you meant by "logical cores").
- Oshyan
Quote from: Hannes on April 22, 2019, 07:00:19 AM
Sorry, I had to send the benchmark form twice, since I accidentally checked "overclocked" in the first form. It's not, so please delete the one with overclocked checked.
No problem, taken care of. For future reference you can edit your response(s). The option is provided at the end of the result submission, and I believe if you revisit the form there should be a way as well.
- Oshyan
Quote from: jaf on April 22, 2019, 02:15:54 PM
If I remember right (I don't "remember right" as often these days), we couldn't make more than one submission in the old benchmark database (for the same system.) Is this still true? I would think it beneficial to do that during the v4 time frame when new builds are released. [edit] It could show where a type of cpu/memory made a difference, positive or negative or just comparing different (TG) builds.
With the old benchmark and with this one you can definitely submit multiple results. There is an option to submit a new result immediately after submission, and otherwise you can just revisit the form.
- Oshyan
4:46 here in build 4.4.18
And a great looking shot. Stuff to learn in there, been out of the TG loop for bit.
Holly cow a Threadripper kicked the butt of an i9!! WOW!
ops another i9 came to the party. 4minutes holly sh......
Wow, that's a surprise to me, that i9 result is crazy. I would venture to guess this benchmark ends up with just a few threads running toward the end and the i9 can really turbo boost at that point, up to its max of 4.2Ghz or perhaps even higher. That could explain why it's a bigger relative jump than you'd expect. Very interesting indeed.
- Oshyan
Should the 7980-XE say 16 cores or 18? It also says you have one CPU. Are you sure it's not dual, at that speed?
EDIT: Ah, of course it can't be dual because it's not Xeon. I'm amazed at that speed!
I edited it, it should be 18. This is why we have people supply the model # (and hopefully correctly). :D
- Oshyan
A new horizon is looming with those speeds, have to persuade the wife ::)
Quote from: Dune on April 24, 2019, 02:09:35 AM
A new horizon is looming with those speeds, have to persuade the wife ::)
Hehe, I know what u mean...not an easy task...lol
edited it, it should be 18. This is why we have people supply the model # (and hopefully correctly). :D
- Oshyan
It's only 30 seconds faster than the threadripper.
My work machine is silly fast. It's also set up as a multiple GPU Octane box.
Early on last summer, Terragen would intermittently crash the hell out of it. I had to send 16K enviro sphere renders out to Pixel Plow to get them done...if the threadrippers were busy. They never crashed. Since 4.3 that hasn't occurred.
We have a threadripper here with Terragen on it as well, I should do the test there too.
Yes, the more the merrier! Also note that 4.4 will be faster than 4.3 due to some more optimization of render times.
- Oshyan
The Threadripper 2990 WX did the test in: 05:13.
The Threadripper 1950X did it in: 06:34sec.
But the price of the 2990WX is about 2x compare to the 1950X. Why would anyone pay double to save about 50seconds? :o
This is not what I want, but the result is very helpless. 2990 seems to have a memory delay problem?
Quote from: archonforest on April 25, 2019, 05:31:35 AM
The Threadripper 2990 WX did the test in: 05:13.
The Threadripper 1950X did it in: 06:34sec.
But the price of the 2990WX is about 2x compare to the 1950X. Why would anyone pay double to save about 50seconds? :o
But on a longer render, say 2 hours, it would save around 15 minutes. Then think of a long animation.....
Quote from: jaf on April 25, 2019, 10:48:54 AM
Quote from: archonforest on April 25, 2019, 05:31:35 AM
The Threadripper 2990 WX did the test in: 05:13.
The Threadripper 1950X did it in: 06:34sec.
But the price of the 2990WX is about 2x compare to the 1950X. Why would anyone pay double to save about 50seconds? :o
But on a longer render, say 2 hours, it would save around 15 minutes. Then think of a long animation.....
Yeah, true. Guess when you are making your living out of it then you count every minutes...
There are a couple of things to keep in mind about the 2990WX result vs. 1950x. First, 4.4 is going to be faster than 4.3 and the only 2990WX result we have so far is from 4.3. A 4.4 result from the same CPU would probably be 45 seconds faster, which doesn't sound like much, but is ultimately a good deal faster percentage-wise. Second, there are some scheduler and potential memory bandwidth/routing issues with the 2990WX, yes, but as far as I understand much of this is down to problems in how Windows handles thread management for Threadripper CPUs. It's something that can and should be improved over time. And if you happen to be able to render on Linux, those problems seem to mostly disappear (better thread management).
All that being said, of course a past generation CPU will be much cheaper, and unless the newer CPU really doubles performance (unlikely to happen in a single generation), then the *value* is usually with the past generation. This has been true of Intel for a long time, even more so in fact. I bought my dual Xeon workstation (currently #7 result on the list) on eBay for about $1800, and it nearly equaled some brand new Xeon workstations of the time for less than half the price. Since Intel was making only marginal improvements year to year with performance prior to AMD's Zen chips coming out (Ryzen and Threadripper), you had this glut of older but still close-to-current performance level hardware that was selling at much lower used prices. It made for a very interesting used market. :D
- Oshyan
Added my results, I had hoped to rate better but I guess my rigs are getting a little dated.
i7-5930K @3.5Ghz 32gb DDR4 13m44s
i7-980 @ 3.89Ghx (OC) 16gb DDR3 18m10s
i7-8750H @ 2.2Ghz - 3.89Ghz 16Gb DDR4 16m02s
I'm a little suprised that the most recent chip, the i7-8750H, is not faster. It would also appear that the core speed is bottlenecked in some way by the slower ram in the i7-890.
Interesting also that this new benchmark is not scaling the same as the old benchmark.
The new scene renders 2.22 times slower on the i7-5930K and 2.87 times slower on the i7-980.
As I've acquired a new PC made of the guts of an old one off Gumtree, I figured I'd bench this baby. Slightly faster than Cyphyr's slowest :), but then I'm focussing on deepening my GPU-rendering knowledge with this machine. Anyhow!
i7-5820k OC'd @4.4Ghz 64GB DDR4 12m30s
Thanks for the result Alex. All good fodder for the comparison. :)
- Oshyan