[ivox3]

If I'm not asking too much, can you also provide SL for each?

Yes. ...but it'll have to wait a few hours, ... I'm being summoned to pay my bills.

[beatriz]

perhaps I should repeat the test but limit the threads to two?.

If you have the time... it might help at some point.

[numerobis]

i've just installed my new Q6600. it's running at 2,7ghz atm. so, compared to the old 1.1 values the bench should be around 180...

i get ~75 with 4 threads
~43 with 2 threads
and ~ 35 with 1 thread

i've stopped each test run after 1 minute... but i think it will hardly ever converge to the supposed 1.1 values...
(after 10 miutes at sl10 with 4 threads it was still at 75)

priority was set to normal

(winXP x64 + maxwell 64bit)

[JDHill]

Just a guess guys...Maxwell doesn't require 64bit precision internally...so, you're pushing around alot of unnecessary data in the form of pointers that take twice the room they do on 32bit. This overhead is probably not even measurable on most other apps, but Maxwell's whole deal is conquering the simulation through the sheer number of operations...so 64bit data may indeed be measurably detrimental. Fact is, 64bit computing was never about being faster - there's no universal reason why it would be...it can just be more precise at the same speed. So, if your app doesn't require greater precision, then you have nothing to gain.

Well...nothing but the ability to address completely insane amounts of memory, that is.

FWIW...

JD

[Left Side]

JD Hill,

Thanks for your input, very interesting.

So what are the advantages for Next Limit in producing a 64 bit version?. Is it just to remain compataible with 64 bit OS's?.

I re-ran my prevous test using just 2 cores.

The results were:-

Maxwell 1.1 rendered in 7h44m16s benchmark 92.46
Maxwell 1.5 32 bit rendered in 6h20m00s Benchmark 112.89
Maxwell 1.5 64 bit rendered in 7h03m54s Benchmark 101.28

So by limiting the processors to 2 both 1.5 versions are faster than 1.1.

I hope this helps

Rob

[beatriz]

JDHill is right, you don't expect speed from 64 bits versions (in general). The advantage is about being able to render and edit bigger scenes, bigger textures, more geometry, more output resolution... these kind of things.
Some 3d platforms do have speed improvements in 64 bits though.

In this case a loss of 40% of speed (first post in the thread) didn't seem reasonable/normal, and that's why we're making tests to see if there is a bug, and this last test of yours LeftSide might be a useful hint.
Thanks

[tom]

So what are the advantages for Next Limit in producing a 64 bit version?.

32-bit version would crash while loading/rendering a huge scene depending on the physical+virtual memory you have (this is an operating system limitation) while 64-bit version will be able to handle the scene without any crash. It's the benefit of memory addressing capabilities. 64-bit is 2^32 times superior to 32-bit about addressing and it's *slightly* slower.

[Tim Ellis]

So what are the advantages for Next Limit in producing a 64 bit version?.

32-bit version would crash while loading/rendering a huge scene depending on the physical+virtual memory you have (this is an operating system limitation) while 64-bit version will be able to handle the scene without any crash. It's the benefit of memory addressing capabilities. 64-bit is 2^32 times superior to 32-bit about addressing and it's *slightly* slower.

So the theory would be to use x64 MXST to open, modify and export a really complex mxs, then use x32 MXCL to render?

That would give the memory handling and the 'slight' speed increase, rather than one or the other.

Cheers,

Tim.

[simmsimaging]

Can you install both and choose which one to run with? I'm concerned because x64 version installs in the 32bit program folder no matter what I do and I don't want to wipe it or mess it up.

Thanks
b

[numerobis]

In this case a loss of 40% of speed (first post in the thread) didn't seem reasonable/normal, and that's why we're making tests to see if there is a bug...

for me the difference with 4 threads on a quadcore is more than 50%... there must be a bug

1.5 32bit bench ~180
1.5 64bit bench ~75

[beatriz]

Can you install both and choose which one to run with? I'm concerned because x64 version installs in the 32bit program folder no matter what I do and I don't want to wipe it or mess it up.

Thanks
b

Well you can, just rename the folder prior to install the second version.
But you will have to handle with environment variables all the time.

[tom]

So the theory would be to use x64 MXST to open, modify and export a really complex mxs, then use x32 MXCL to render?

What I've said is also valid for mxcl, not only mxst. mxcl would also crash with an mxs exceeding the limits of 32 bit memory handling.

[simmsimaging]

Thanks beatriz

[JDHill]

Hi Brett, in addition to what Beatriz said, you can also just rename the Maxwell folder, instead of changing env. variables...I do it all the time.

[simmsimaging]

Thanks JD !