The current owner of the Unofficial Blender Render Benchmark site cannot cannot maintain this site anymore and is open for others to step in.
Maybe with Blender 2.5 the Blend test suit could be expanded.
Jardo writes:
Hello Blender users and computer enthusiasts out there, I am very sad to announce that the Blender render benchmark is closed for an indefinite time.
I am awfully sorry, but in my rare spare time I am not able to maintain the benchmark as necessary.
The results will be stay online as long as possible, if some one is interested into the whole data, send me an email and I will transmit you all data (database dump) exclusive email addresses and IP addresses of course.
cheers Jarod
Website Link:
http://www.eofw.org/bench/
22 Comments
I fired off an email to him. I can certainly help and pick this up for him.
I think it should be dumped and started again anyway because people have entered their times wrong.
I have an almost identical system to the 3rd score on that list (my chip is the faster AMD Phenom™ II X6 1090T 3200MHz), and whereas it takes me 6 seconds to render, they have recorded 6 milliseconds. Any of those numbers that have less than a second have clearly been entered incorrectly.
Start again when 2.5x goes final...
It was quite unclear when I went to fill in my numbers (didn't bother when I realised that people had made mistakes) so it needs to be made more idiot proof (eg. screenshot from blender display and show how it translates to submitting the form).
But sorry, I don't have any hosting presence for this so I will shut up now.
I think the scene should also be updated and be made more "complicated" , thus raising the rendering times...
Normal fast machine nowadays render in around 6 seconds. Which is imho too fast for a benchmark.
It should be raised to about 30 seconds maybe even one minute on a 4-8 core machine, so the rendering time differences between machines setups get more obvious than a difference of 2 milliseconds.
o
It's a shame to see the faithful old benchmark disappear but I agree with Goosey that the data does need some cleansing.
It would also seem that the recent developments in blender and the onslaught of multi-core processors have supersceded the test file which could do with being more challenging for higher end rigs. This would of course render the previous dataset obsolete, however perhaps it is more relevant to archive this old data and start over. I'm sure it wouldn't take long to populate a new database with a quick post to blendernation and a few thousand users with a shiny new 2,57 build trying to get the top time for their setup.
Incidentally, the fields for categorizing this data (i.e number of processors) no longer apply (or have been poorly interpreted by users) in multi-chip, multi-core setups. So I would propose that whosoever accepts the challenge of maintaining this data makes provision for this by adding a cores per processor field or simply renames cpus to cores to avoid ambiguity.
I addition it may also be interesting to show results that have been acheived by overclocking a rig, memory timings etc.
Many thanks to Jarod for his efforts throughout, I fully sympathize that life can be too demanding.
It's a shame to see the faithful old benchmark disappear but I agree with Goosey that the data does need some cleansing.
It would also seem that the recent developments in blender and the onslaught of multi-core processors have supersceded the test file which could do with being more challenging for higher end rigs. This would of course render the previous dataset obsolete, however perhaps it is more relevant to archive this old data and start over. I'm sure it wouldn't take long to populate a new database with a quick post to blendernation and a few thousand users with a shiny new 2,57 build trying to get the top time for their setup.
Incidentally, the fields for categorizing this data (i.e number of processors) no longer apply (or have been poorly interpreted by users) in multi-chip, multi-core setups. So I would propose that whosoever accepts the challenge of maintaining this data makes provision for this by adding a cores per processor field or simply renames cpus to cores to avoid ambiguity.
I addition it may also be interesting to show results that have been acheived by overclocking a rig, memory timings etc.
Many thanks to Jarod for his efforts throughout, I fully sympathize that life can be too demanding ...
Why not making it an official test? It is something that BF doesn't have.
So, why not? Improve the scene & dump it again if it is necessary, and host the blend & time tables on the revamped Blender's site
Remove all possibilities for error as well. have a script that adds a benchmark property. once the benchmark is finished ask if it is ok to upload to a site. Then just post the data to a url. No typing and no chance for errors.
I agree with GREENie - that's a novel idea. I also agree that it should be made more complex, maybe a few scenes that render each via script, then complete a composite score. Even better yet, compile a version of Blender without access to the menus, etc - so users can't change a bunch of settings - it's not very efficient as a benchmark if people are changing things.
are we talking maybe of an addon? (hint! hint!)
this could a python script which would run the render and then create a text file with time and PC specifications to be uploaded to BF site.
PROs and CONs?
How do you crete a clay sphere on scult mode in blender 2.56, I can't figure it out
hmm..., i don't like this benchmark - it is too unserious.
this kind of benchmarktest is a big playground for 'swank' and 'poser' - like 'hey i have the bigger one'.
no, really - there are too many thinks that disort the results.
a benchmarktest should run multiple times on the same system with same setting. then you have to take the avarage time, the minimum, the maximum but better all values to build a correct 'gaussian normal distribution' to show the quality / qualtity of that system - and only for that system.
and the description of the system needs much more information. what about ram speed settings, what kind of ram is it using, how are the bios settings, what mainboard is used (e.g.: AMD Phenom X?, is it running on a AM2+ or AM3 board), which hardware revision, what background tasks are running, what is the IO transfer speed CPU to RAM to HDD, and so on...
but what you can do is to compair an older version of blender with the newest version on the same system - to see - oh, blender 2.5x is 'X'-times faster than blender 2.4x or 2.5 alpha x or like this. i think that makes a little bit more sence - if it do...
The way i've always seen this test: a nice way to know approximately how good my pc was in terms of rendering. If the table says that my rendertimes are really far from the top i might think "ok it's time to look around for new hardware". Obviously it can be perfect, and (especially on windows systems) there are plenty of external-to-blender factor that might have influences on the results. But it's better than nothing, and it's better than installing some tons-of-MB-Cinebench-like test.
There is no coloumn for graphics card. Does the internal renderer not use it? (what a waste!)
How about turning this benchmark into a Wiki page that everybody can edit? That way burden of maintaining it can be shared by many users. I'd also like to see new, more demanding test scene, and less ambiguous data entry...
I second GREENie's idea. And what if we used a bit of mike's suggestion and have the script of GREENie's idea post the benchmark information to a free wiki site? Free hosting, low maintenance, reduced mistakes, and supported by contributors rather than just on one person alone?
Loads of ideas and critics, yet nobody who steps in to DO the actual work.
mike: Who is stopping you? Nobody.
Brian L. Go help mike.
Danielle: Did you actually read the post?
lsscpp: Go learn python and write it. Its a 2 day job, with the learning part.
Goosey: Great! When is your page online?
I would want to help!
I have plenty of time to maintain!
...rename the site to: http://www.blenchmark.com
:)
joeri67, your post is not fair play! you didn't even include yourself! ;)
and btw, since one of the things i'm waiting for is GI in BI, should I begin studying C++ and andvanced maths?
I am also interested in getting the benchmark more streamlined and updated. I have been in contact with Jarod a little about this (at least about getting the current dataset and permissions to modify the test.blend and redistribute it). I don't have a lot of time but would like to see a lot of the suggests above incorporated.
Such as:
- An automated reporting tool inside the benchmark
- Hardware/setting detection within the benchmark
- Maybe include a series of benchmark tests like SPECviewperf or 3DMark but run inside blender.
- Adding an information wiki about what the values of various tests mean ( some setups might be able to bake faster or run fluid simulation better than say ray tracing, etc..)
I am fluent with Linux and Java technology but only a novice with python and php so some help would be appreciated. I you are interested please contact me thajek at gmail dot com and maybe we can get something up a working.
Initially I wanted to maybe try something using google sites and their API that pull data from google spreadsheet (partly because of cost for hosting the site) but I also run my own web server at home running linux and mysql and using zoneedit for dynamic dns. It's not exactly fast but it has been up 24/7 for about 5 years and I do maintain it regularly.
I am also interested in getting the benchmark more streamlined and updated. I have been in contact with Jarod a little about this (at least about getting the current dataset and permissions to modify the test.blend and redistribute it). I don't have a lot of time but would like to see a lot of the suggests above incorporated.
Such as:
- An automated reporting tool inside the benchmark
- Hardware/setting detection within the benchmark
- Maybe include a series of benchmark tests like SPECviewperf or 3DMark but run inside blender.
- Adding an information wiki about what the values of various tests mean ( some setups might be able to bake faster or run fluid simulation better than say ray tracing, etc..)
I am fluent with Linux and Java technology but only a novice with python and php so some help would be appreciated. I you are interested please contact me thajek at gmail dot com and maybe we can get something up a working.
Initially I wanted to maybe try something using google sites and their API that pull data from google spreadsheet (partly because of cost for hosting the site) but I also run my own web server at home running linux and mysql and using zoneedit for dynamic dns. It's not exactly fast but it has been up 24/7 for about 5 years and I do maintain it regularly.
Also, I work in the IT department at a University and occasionally teach a course in Information Systems so I could possibly enlist the help of some students on the project. Our CS department does use blender in some of it's game development courses so I can probably find some support here for it as well.
Ok - looks like a lot of interested people, but I didn't spot someone willing to host.
I am willing to host the site as well as build out the "web end" of the data collection, reporting, etc.
I am a Joomla fan and site builder, so I would be leveraging that platform and I could build out web services to accept the postings if someone else can write the Blender scripts to send the data across.
I have already set up a new domain and site:
http://www.renderspeedtest.com/
Bear with me as I will need this week and weekend to start putting things in place.
Tomas - I'll capture your "feature list".
Others - I'll post back here as I progress to keep everyone in the loop. Has someone already been in contact with the original, "unofficial site" author? I will need to get looped into him as well.
As a side note, I would like to focus on the internal Blender Rendering Engine against various CPUs, but I am very keenly interested in the network rendering as well as GPU rendering, so I will be taking the site in that direction overall to begin benchmarking and communicating the best methods across not just Blender, but other products and engines.
Regards,
Dave Hudson
You can contact me directly via the site listed above......