[macray]

For every SL update it gives me a warning that the picture cannot be saved because of an invalid path...

[ivox3]

macray, .... it doesn't affect the benchmark.

* you would have to adjust the path, ...but in this case, ..we're not interested in that.

Adrian: what is the current required SL? ...wouldn't a simple clean SL of 10 be good ? ..or is that too fast for the octa-set ??

btw: ....very attractive site.

[macray]

SL is set to 17 if I'm not mistaken.

[ivox3]

...strange number. ...is that when the scene begins to clear, ..maybe ?

[mgroeteke]

I had to edit 3dworks' entry, he manually added a "Dual-Core Xeon" which applies to pretty much any Xeon in the database. Since he ran OSX, i blindly changed it to a MacPro 2.66 hoping that's what he has.

I agree with nachob, bejack's entry is the one that's off. Maxwell scales near linear so the 33m of SmellyPenguin seems realistic. Also, can you equip a Mac with 9.0GB ram?

...the times are very high for a "quick" benchmark (~90min for a 3ghz core2duo). wouldn't it be better to lower the requirements a little bit to make the benchmark more usable?

Who said it was a quick benchmark?
I really wanted to make sure that current octa-cores (did't have one at my disposal) and the soon to come 45nm chips don't finish this in 5 minutes, there could be some overhead involved when the scene voxelizes and threads are launched. If everyone agrees, i can lower the target SL and we start over of course.

hi thanks for correcting, it is a macpro 2.66 quad. i wasn't sure about the Xeon specs

btw. yes, a mac can be equipped with 9 G... (4x2048 + 2x512 for example).

about bejack's result: i was wondering too - but he repeated the test - making sure it was set to 8 threads - and the result was the same! what can be the cause for this? anyone else with an octo-mac can please test again?

cheers and thanks for providing this great test,

m.

[-Adrian]

Ah ok, good to know that it was right. I feel a little odd to temper with entries, and i had to do quite a bit of research myself to figure out the elusive specs of a MacPro
I think i'll remove the Intel names from the Macs, so there's just the clockspeed listed, like when you purchase a Mac.

ivox: The number is pretty random, i actually tweaked the SL to get about 1.5h rendertime on my system, but maybe that was over the top. Hilarious new avatar btw.

[nachob]

about bejack's result: i was wondering too - but he repeated the test - making sure it was set to 8 threads - and the result was the same! what can be the cause for this? anyone else with an octo-mac can please test again?

There must be something bad in that machine, the differences in an 8 core with the 1.1 test scene versus a 4 core mac were much bigger, almost doubled also on mac. The only explanation I can imagine for it is that bejack is running maxwell under Rosetta in his machine (that means he is running under the ppc emulation). The performance penalty of doing it is similar to what we are seeing in his result.

nachob

[nachob]

An beside last point. It would be also very good to add wich version of maxwell is running in the machine in the case of windows: maxwell 32 bits or windows 64 bits, and also the exact version of the engine (if there is some patch released)

nachob

[-Adrian]

Yes, due to the differences reported between 32 and 64bit i thought it would add chaos if we mixed them. Thus 2 separate lists for each architecture. However i found out i only had like 2m difference between the two Maxwell versions, maybe we can merge them after all and add a version column. But to include further patches would certainly make the list unreadable, it has to be simple enough to quickly get an idea what hardware is a good investment and what's not.

[big K]

hi adrian,

thanks for all your efforts on this. very informative !
(as i have plans to buy new machine soon)

cheers
michael

[KurtS]

...maybe we can merge them after all and add a version column.

great idea!

[sandykoufax]

I added my result.

[-Adrian]

Thanks m8, i'll propose to Intel that their next codename be Koufax.

So now that many results have been entered already, i'm wondering when or if i should change the benchmark scene (shorten the time) and remove the separateness of the lists. We have to get it right at once for the sake of the testers donating their time

• Shorten time, roughly cut in half
• Add column for Maxwell Render version (32/64bit)
• Remove Cinebench Column? Seems very rarely used.

[numerobis]

Shorten time, roughly cut in half

- yes, i vote for sl16... should be enough and more handy atm.

Add column for Maxwell Render version (32/64bit)

- i think a combined list for 32 and 64bit would be more clear...

Remove Cinebench Column? Seems very rarely used.

- no, i would keep the cinebench results, even if they are not complete - any additional information could be useful. and cinebench is a very common test and many results available.

btw: what number do i have to select for the threads for a quadcore? 0 or 4 or a higher number - i'm a little bit confused looking at the "thread test thread" http://www.maxwellrender.com/forum/view ... hp?t=23715

[sandykoufax]

O-VIZ's quad core system is faster than bejack's octa core system.

What's the problem? maybe wrong thread assignment?

and Opteron 885 result is interesting.

We've seen that opteron 285's result is much slower than qx6700 in the 1.1 benchmark.

but now it's faster than qx6700 even though allowing that overclock.