Hello everyone. At last we have a new benchmark (candidate) that does a better job of showcasing Terragen 3 features and stressing modern systems without being as imbalanced as the older benchmark was on multi-core machines. The scene features several elements from the Terragen 3 presets pack, including Ulco's arch, Jack Marsh's green cliffs displacement and texturing, and Volker's marble texture and fake stones setup, along with Matt's Mackerel Sky. So it's got a decent amount going on.
[attachimg=1]
What I'd like you guys to do for the next day or two is run the scene on your machines and record the times (hours, minutes, seconds) and just post a reply here with your bench time and CPU type/speed (Ghz). Don't worry about OS or amount of RAM, we'll be collecting more complete system specs on the final benchmark, and there will be a form to fill out to submit results along with a public table of results for comparison. But for now I just want to make sure the benchmark doesn't take too long (and isn't *too* quick either) on a sampling of machines that you guys have here. On my i7 2600k here it takes about 10 minutes, and on my 2.0Ghz i7 mobile (also quad core) it's about 15 minutes. The Benchmark TGD is attached, just download, open, and hit Render.
I just don't want to see times that are longer than people are likely to bother with running a benchmark, so if we start to see 30, 45 minutes, or an hour or longer, I might adjust some settings or lower the resolution. Also feel free to let me know how long you think is reasonable for a benchmark to take, though keep in mind that a benchmark that's too quick will give you less accurate results.
Once I am able to get a preliminary idea of how performance is looking for average machines here I'll finalize the benchmark and, if settings haven't changed, you can just add your result to the final result list. Otherwise you'll need to re-run the benchmark before you can do so.
Thanks!
- Oshyan
Time: 13.02
CPU: I5-3570K-3.4Gz
Time 12:12
CPU I7-2600-3.4Gz
____________________________
Time 9:33
CPU I7-3770-3.4Gz
Time: 0.28.53
CPU: Intel Core 2 Extreme QX 6700 2.66 Ghz
System built: June 2007
Time: 0:32:58s.
CPU: Intel Xeon X5260 3.33Ghz (Dual Core)
Not too horrible so far. How do you guys feel about having to render for ~30 minutes for a benchmark on an older machine? It seems reasonable to me, but it's really more about how long people are willing to wait for the results and will see see fewer people with old machines submitting results because of it. I'm also hoping to see some of the newer 8 core CPUs or dual core machines here for some faster comparisons. Those machines ought to be able to get down to less than 5 minutes I think, and with a new batch of 8+ core CPUs from Intel now available in multi-CPU configurations, we might see even lower times, so I think this is a good target range for speed. But I definitely want your feedback too.
- Oshyan
I will redo the test when I get home tonight on my dual quad station. It should beat my office PC :D
0:04:02s
Dual Xeon E5-2650 v2 @ 2.60GHz (8-core CPUs, HT enabled)
I can run a test on my older dual CPU render machines in the next days. Nice scene!
Can u do one without HT pls? (Only 8 cores) Really curious about the time of that. ;)
Now we're talking! So pokoy, you've got a machine with 2 cpus of 8 cores each, so 16 threads for each CPU (with HT), and 32 threads total? The E5-2650 is 2Ghz by default as well, so yours are overclocked?
- Oshyan
QuoteCan u do one without HT pls? (Only 8 cores) Really curious about the time of that. ;)
This is my main workstation here and I can't take the time to restart with HT off for now, but will try to run a test in the next days. Too much work atm :-/
QuoteNow we're talking! So pokoy, you've got a machine with 2 cpus of 8 cores each, so 16 threads for each CPU (with HT), and 32 threads total? The E5-2650 is 2Ghz by default as well, so yours are overclocked?
Yes, 2 CPUs, 8 cores per CPU, 16 cores per CPU with HT, 32 threads total. Looks really neat in the task manager :)
Not sure about the 2GHz, system and CPU-Z report it to be at 2.6GHZ, so maybe it's faster because there's a 'v2' in the name...?
I don't think they're overclocked, they were purchased from a system builder shop and I bet they don't overclock systems, but I could be wrong.
13:31s
I7-860 @ 2.8 GHZ (4 core+ 4HT)
Oh I see, V2 was 2.6Ghz per core, yes. Interesting. Shame on Intel for such a confusing naming scheme though. *sigh*
So that's about the fastest time I'd expect to see, in terms of the currently available CPUs. There are now some 10 and 12 core Intel CPUs that could be even faster, though there are diminishing returns with more cores at a certain point, and the faster per-core speeds of CPUs with fewer cores might win out eventually.
Still, 4 minutes is a pretty good lower limit I feel like. I wouldn't want it to be too much faster than that on current machines if this benchmark is going to last a year or two (the last one had to go for quite a few years!).
- Oshyan
0:28:13
Intel Core 2 Quad Q9550 @ 2833.3 MHz
Quote from: Klas on September 18, 2014, 05:41:11 AM
0:28:13
Intel Core 2 Quad Q9550 @ 2833.3 MHz
How is that possible that a quad core 2.8Ghz takes 28 min to render it and my office dual core 3.3Ghz took 32? Your quad should probably 40-50 percent faster than this dual...hmmm
QuoteStill, 4 minutes is a pretty good lower limit I feel like. I wouldn't want it to be too much faster than that on current machines if this benchmark is going to last a year or two (the last one had to go for quite a few years!).
I guess the brand new Intels will get to around 2 mins on dual CPU setups.
Seeing the times posted here I feel like I'd never want to go back to anything else than dual Xeons, but they come at quite a high price tag unfortunately.
11:57
i7 2600K @3.4Ghz
16GB ram
Quote from: archonforest on September 18, 2014, 05:52:27 AM
Quote from: Klas on September 18, 2014, 05:41:11 AM
0:28:13
Intel Core 2 Quad Q9550 @ 2833.3 MHz
How is that possible that a quad core 2.8Ghz takes 28 min to render it and my office dual core 3.3Ghz took 32? Your quad should probably 40-50 percent faster than this dual...hmmm
Maybe because it's not the fastest chipset?
Computer Brand Name: FUJITSU D3041-A1
Motherboard Chipset: Intel G41 (Eaglelake) + ICH7
DDR3 Frequency Support: 533 MHz (DDR3-1066)
(hwinfo (http://www.hwinfo.com/))
7:59 Mac Pro Dual, Quad Core 3.0GHz Xeon Mac OS 10.9.4
8:26 MacBook Pro Retina, i7 2.6 GHz Mac OS 10.9.4
7:25 Dell Precision T7500 Dual Xeon E5645 (6 Core) Win 7 Pro 64.
8:44 Intel® Core™ i7-4770 3.4 GHz
6:38 :) 16 core VM (2.6Ghz) I hope they let me keep it.
23:44 at home i7, 3Ghz
I think anything under an hour is fine for a benchmark
Intel 4930 6 core 3.4 Ghz 6:27
interestingly the Subdivision Cache warning popped up advising increasing the cache size to prevent performance loss, so I bumped the cache from the default 1 gig to 3.2 gigs and it rendered in 6:12.
The sytem is built for OC so I might try another run in a few days.
07:43
i7 2700k @ 4.6GHz
16 GB Ram
0:08:55
Macbook Pro 2.7 GHz Intel Core i7
16 GB 1600 MHz DDR3
OSX 10.9.4
Might be even a little bit faster as I was surfing during the render
Something is wrong here:
Ben used 16 cores and got 6:38s.
Zaxxon used 6 cores and got 6:27s. :o
This is not possible that 6 cores beats the s.. out of 16 cores ??? ??? ??? ???
Time: 09:44s
CPU: AMD FX-8350, 4 GHz, 8 Cores, 16GB
Time: 18:40s
CPU: AMD A10-7850k, 3.7 GHz, 4 Cores, 16GB
Time: 20:37s
CPU: AMD Phenom II-955, 3.2GHz, 4 Cores, 16GB
Time: 01:57s (average per frame)
CPUs: Four FX-8550's, One A10-7850k and One Phenom II-955 working together.
Archon, I don't think there is anything wrong. One "core" or "thread" is not inherently the same as another, and having the same Ghz CPU speed does not always mean the same actual processing speeds.
Core 2 is a previous generation CPU technology. Newer Core i5/i7 architecture is a good deal faster *at the same clock speed*. So what you're seeing is the combination of a faster per-core speed (3.3Ghz vs. 2.8Ghz), and faster overall execution speed. It makes good sense to me.
Zaxxon is running a higher clocked, 6 core CPU, while BigBen is running a *virtual machine* with 16 "cores" (or threads, presumably). So you're seeing several effects here that create an interesting but not impossible result. First, on a pure Ghz equivalency, Zaxxon's got 20.4 while BigBen's got 41.6, so about double, and indeed it looks surprising to get a similar result. BUT, BigBen is running a virtual machine and that has some overhead, sometimes significantly so. There is also some amount of efficiency lost with greater numbers of render threads, if you could have 1 20Ghz CPU core, and 16 2.6Ghz cores (40Ghz theoretical in total), the 20Ghz single core would actually beat the 16 threads by a lot, despite being in theory half the speed. Lastly, we don't know what generation of CPU BigBen is working with, it could be older, meaning less performance per-Ghz compared to Zaxxon's. So, a surprising and interesting result, but not without explanation. :) Frankly I'm a bit wary of including results for virtualized machines in the results table for this reason, but as long as they're clearly labeled I'm OK with it.
- Oshyan
Quote from: Oshyan on September 18, 2014, 02:38:30 PM
Archon, Core 2 is a previous generation CPU technology. Newer Core i5/i7 architecture is a good deal faster *at the same clock speed*. So what you're seeing is the combination of a faster per-core speed (3.3Ghz vs. 2.8Ghz), and faster overall execution speed. It makes good sense to me.
- Oshyna
Hmmm...the difference between the core speeds are not that big. But still 16 render line vs 6?? and 6 wins :o :o It just blowing my mind...
I updated my explanation above.
- Oshyan
Virtual machine can be slower...sure...
I would like to see that table at the end if possible. Pls do not close the benchmark yet as I still want to include my home workstation timing in there. :D
Don't worry, this is all just feedback! I may tweak the final benchmark a little (or not), but in the end you'll need to fill out a form to submit your result(s), and then you'll be able to see all the results in a nice, sortable table. :)
- Oshyan
I think the guys in Research Services will still be happy with the performance. It's not necessarily designed to be the absolute fastest processing around, but it does provide much better performance in a platform that can be shared to researchers for projects they have without requiring them to invest a lot of money in computers. Given that, I wouldn't be surprised if there was some compromise on quantity vs quality. I'll have to get them interested in letting me try their other flavour of VM which is for mostly CPU processing and high RAM usage. The one I'm using is designed more for providing GPU processors as well (They gave me 8 :)) .... or I could ask them for the other 50% of the system resources for my VM ;)
Just ran the benchmark on a render node:
0:08:23s
Dual Xeon E56320 @2.53GHz (2 x 4-core, 8 threads with HT per CPU, 16 threads total).
0.08:20 s
Intel(R) Core(TM) i7-4770 CPU @ 3.40 GHz
Here is my homie workstation:
Time: 00:09:02s
CPU: Dual quad Xeon E5530-2.4Ghz
1:09:39
Intel Dual Core T4400 @ 2.20 GHz
8 GB Ram
22:45
Intel Quad Core i7-3537U @ 2.0 2.5 GHz
16 GB Ram
7:12s
Intel i7-4790k @4.00GHz, 16GB Ram
10:12s
Windows Vista Ultimate 64
(2) Xeon 5420 2.5 Ghz 16 GB Ram
9.55 with my i7 2600 (3.4GHz) and 16GB set at medium performance, but interestingly: 7.56 (~20% faster) when I change performance settings in bios to the highest level. Good to know for myself as well. Doesn't even make more noise ;)
6.35 on my i7 3930K (3.20GHz) with 32Gb ram
8:53 intel i7-4790 3.2GHZ
8:45 with subdivision cache moved up to 3600 and preallocate checked - woo-hoo !
Quote from: Dune on September 21, 2014, 03:01:19 AM
9.55 with my i7 2600 (3.4GHz) and 16GB set at medium performance, but interestingly: 7.56 (~20% faster) when I change performance settings in bios to the highest level. Good to know for myself as well. Doesn't even make more noise ;)
Interesting! I wonder what that actually changes. Do you know? A performance difference like that would seem to me to only be possible from higher CPU speed, or perhaps less throttling due to heat or noise (i.e. prioritize performance over noise or power user concerns). If you could check CPU speeds in Windows with something like CPU-Z or Coretemp, it would be interesting to know that info.
- Oshyan
My shiny new MacBook Pro:
0:07:25s
2.8 GHz quad core i7 (I7-4980HQ)
Oshyan, I've noticed that the benchmark has settings which exceed the Free version maximums. I wonder if it would be better if it could be run with the Free version.
Regards,
Jo
0:11:17
AMD FX 8-core cpu (8320, 3.5 GHz) 7 cores for Terragen
8 GB RAM
Win 7 prof., 64 bit
Oh good point Jo. I had intended that it would work on the free version but forgot to adjust final settings for that. Hmm. Well, looks like you'll all have to re-run the benchmark coming up soon then. *sigh*
Thanks for the input everyone! I'm closing this thread as the benchmark will now be finalized and published later this week.
- Oshyan