[wagurto]

where is the benchwell test scene?
thanks

[numerobis]

...providing other ways of displaying the info such as graphics...

yes, displaying bars for the benchmark or time would be nice
grouping the results also.

Is there a stand alone version planned? like cinebench? (maybe as portable without installation if this is possible)
I think it would be great if it could become a standart benchmark on hardware review sites!

[Bubbaloo]

where is the benchwell test scene?
thanks

Open Maxwell Render, Tools > Run Benchwell

[Fernando Tella]

Is the model of the teapot a 3D scan?

[RenderFred]

As a suggestion, I think there should be two tables: the main one for stock systems and one for overclocked systems. The point is that what is really interesting in such a test is the performance of systems at stock speed, that everyone can buy, and not overclocked ones. The vast majority of Maxwell users will never overclock their computer; it's also not that easy to do well. That is particularly true of Xeon systems since it requires specific boards and BIOS (eVGA comes to mind). I predict the Benchwell table will be quickly polluted by overclocking entries simply because some people build an overclocked system for bragging rights and will enter any public benchmark just to show off - Cinebench, Frybench, 3DMark, PCMark etc. you name it. It's human nature, but not that useful

I believe it would be more useful to have a table where only one entry per processor at stock speed is allowed (or better, an average value of the entries). There is enough CPUs on the market to have a big, meaningful table that people can quickly refer to before buying their hardware without having to browse through dozens of overclocked results which are not that interesting (and you can bet some won't have any 'info' attached so that will confuse things even more). One must also be aware of the fact that many overclocked machines are pushed to the limit just to get a good result on a quick benchmark, but would not run a 24 hours Maxwell Render at 100% CPU usage reliably... I assume people want their render to finish

Now don't think I hate overclocking, I did a fair share of that myself and it's quite interesting, but I believe overclocked results should be in a another table to avoid confusion and improve readability.

[def4d]

+1000, especially oc'ing for rendering is not a good idea!

[feynman]

Yes, agree - separating stock speeds and oc'ed speeds is a good idea.

However, oc'ing is very straightforward using the motherboard's vendor's software; I built several i7-3930K and i7-2600K in January for almost no money and they've been working in their oc'ed state 24/7 ever since (not just for rendering, also CFD/FEM analysis). If one or two should eventually fail, it's fast and cheap to replace failed parts.

[numerobis]

Now don't think I hate overclocking, I did a fair share of that myself and it's quite interesting, but I believe overclocked results should be in a another table to avoid confusion and improve readability.

Maybe there could be a second table with only one CPU at stock speed and no overclocked systems. Or better an option to filter the list to display only stock speeds and maybe only one entry of each model (like the average value of similar systems Juan suggested)
But a table with only oced systems is pretty useless... you need the stock speeds to compare.

[photomg1]

I know where you are coming from re overclocking. But I have run my 3930k at 4.3 for 24/7 since I purchased it I, I've let it render for a couple of days before and had no issues.I can push it higher to 4.7 but I don't find it stable enough for a 24/7 .The point that I'm trying to make , is that a stable overclock is relevant in that chart.Although you are right that the chart will have some pollution from unstable overclocks I still think its better for a potential cpu purchaser to see how far things can go.

[RenderFred]

Just to prove my point: there is an entry in place 6 (X5680 by O-VIZ) which is obviously overclocked, yet there is no info stating the real frequency. Now there is a stock speed Dual X5680 in place 15. So if a user without a sufficient technical knowledge to recognize which is which just takes a cursory glance at the beginning of the table, he/she will probably think that a Dual X5680 is 35% faster than it should be...

6 Intel R Xeon R CPU X5680 3.33GHz 4m43s 985.16 24 24 GB 2.7.0.0 Win64 O-VIZ
15 Intel R Xeon R CPU X5680 3.33GHz 6m22s 730.18 24 24 GB 2.7.0.0 Win64 John Layne

[numerobis]

and this is why i suggested to add a column to cleary display if a system is overclocked (first post)...

And i think the list has to be cleaned from obviously faulty, misleading and ancomplete entries from time to time. (...and from useless advertising like this info "Fedora 14 64-bit. We ONLY use Linux 64-bit for our rendering services." or comments like "Just testing on an HP Z820 - several background tasks running")
An entry like "Genuine Intel R CPU 2.90GHz" is also completely needless...

[photomg1]

What I really would have liked to have seen was a network benchwell scene . I would like to know what the real overhead is from running over a network.From my point of view if I want to do faster renders, looking at that chart the most cost effective/performance way is with me running another 3930k computer over a network. Or using a renderfarm such as your own : ) , but this chart is not showing the results from your renderfarm for me to assess.

[RenderFred]

What I really would have liked to have seen was a network benchwell scene . I would like to know what the real overhead is from running over a network.From my point of view if I want to do faster renders, looking at that chart the most cost effective/performance way is with me running another 3930k computer over a network. Or using a renderfarm such as your own : ) , but this chart is not showing the results from your renderfarm for me to assess.

To do that you can use our free RANCHecker utility which already has an integrated benchmark.

[photomg1]

I'm aware of that , but in the interest of this chart it would have been nice to see a result on the benchwell scene to give perspective. What would you say the overhead is on a network ? would two machines with a bench of 530 really put me in the ball park of a single machine running with a bench of 1000. How quick would the benchwell scene be done done on your network? less than 30 secs I guess.

[RenderFred]

Problem is, the benchwell scene in its current form is definitively not suited to test big renderfarms. Considering that the fastest workstation available today (E5-2187) takes only around 4 minutes to render the scene, the render time itself on the RANCH would be negligible (a few seconds), however the merging time is an important factor to take into account when you render on a farm, it varies with MXI complexity, and you don't have to care about it when you render on a single system. And of course people generally send much more complex scenes to a big farm; the benchwell was clearly calibrated to not take too long on a single system, although it would be interesting to render it in a much higher definition on a farm.