 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Quite impressive. I just bought an Athlon II x4 (thank you AMD for maintaining
backwards compatibility!) and wanted to see how fast thing thing was. One
interesting test I did was to render a particular scene using 3.6 with 4
instances vs. 3.7. I was expecting the render times to be about the same. Yet,
3.7 took half the time that 4 instances of 3.6 did. What's allowing this speed
up?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Also, it might be time to assemble a 16-core computer with
2 of these:
http://www.newegg.com/Product/Product.aspx?Item=N82E16819105266&cm_re=magny-cours-_-19-105-266-_-Product
1 of these:
http://www.newegg.com/Product/Product.aspx?Item=N82E16813151212&cm_re=socket_g34-_-13-151-212-_-Product
and gobs of RAM...
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"jhu" <nomail@nomail> wrote:
> Quite impressive. I just bought an Athlon II x4 (thank you AMD for maintaining
> backwards compatibility!) and wanted to see how fast thing thing was. One
> interesting test I did was to render a particular scene using 3.6 with 4
> instances vs. 3.7. I was expecting the render times to be about the same. Yet,
> 3.7 took half the time that 4 instances of 3.6 did. What's allowing this speed
> up?
if it's a scene with radiosity, clipka contributed a lot of high-caliber
optimizations and cleanups for this feature.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
> Quite impressive. I just bought an Athlon II x4 (thank you AMD for maintaining
> backwards compatibility!) and wanted to see how fast thing thing was. One
> interesting test I did was to render a particular scene using 3.6 with 4
> instances vs. 3.7. I was expecting the render times to be about the same. Yet,
> 3.7 took half the time that 4 instances of 3.6 did. What's allowing this speed
> up?
>
>
With 4 instances of 3.6, it takes 4 times as much memory. You may start
swaping. You need to parse 4 times, and only once with 3.7.
With 3.7, data collected by a core can be reused by the others, that's
impossible in 3.6 and multiple instances.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Alain <aze### [at] qwerty org> wrote:
> > Quite impressive. I just bought an Athlon II x4 (thank you AMD for maintaining
> > backwards compatibility!) and wanted to see how fast thing thing was. One
> > interesting test I did was to render a particular scene using 3.6 with 4
> > instances vs. 3.7. I was expecting the render times to be about the same. Yet,
> > 3.7 took half the time that 4 instances of 3.6 did. What's allowing this speed
> > up?
> >
> >
>
> With 4 instances of 3.6, it takes 4 times as much memory. You may start
> swaping. You need to parse 4 times, and only once with 3.7.
> With 3.7, data collected by a core can be reused by the others, that's
> impossible in 3.6 and multiple instances.
No radiosity in the scene. Also, definitely no swapping since I have 4GB and the
scene only uses 200MB max. I thought it might be memory issues, but I then
compared the test scene on 3.6 running one instance vs. running 4 instances and
they all finished at the same time.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"jhu" <nomail@nomail> wrote:
> Alain <aze### [at] qwertyorg> wrote:
> > > Quite impressive. I just bought an Athlon II x4 (thank you AMD for maintaining
> > > backwards compatibility!) and wanted to see how fast thing thing was. One
> > > interesting test I did was to render a particular scene using 3.6 with 4
> > > instances vs. 3.7. I was expecting the render times to be about the same. Yet,
> > > 3.7 took half the time that 4 instances of 3.6 did. What's allowing this speed
> > > up?
> > >
> > >
> >
> > With 4 instances of 3.6, it takes 4 times as much memory. You may start
> > swaping. You need to parse 4 times, and only once with 3.7.
> > With 3.7, data collected by a core can be reused by the others, that's
> > impossible in 3.6 and multiple instances.
>
> No radiosity in the scene. Also, definitely no swapping since I have 4GB and the
> scene only uses 200MB max. I thought it might be memory issues, but I then
> compared the test scene on 3.6 running one instance vs. running 4 instances and
> they all finished at the same time.
....
for a fair comparison, each instance of 3.6 should be rendering 1 corner of the
image divided by 4. It's roughly what 3.7 does automatically (to as many cores
as you have).
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
jhu wrote:
> instances vs. 3.7. I was expecting the render times to be about the same. Yet,
> 3.7 took half the time that 4 instances of 3.6 did. What's allowing this speed
> up?
Did you try with the non-SSE2 version of 3.7?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Christian Froeschlin <chr### [at] chrfrde> wrote:
> jhu wrote:
>
> > instances vs. 3.7. I was expecting the render times to be about the same. Yet,
> > 3.7 took half the time that 4 instances of 3.6 did. What's allowing this speed
> > up?
>
> Did you try with the non-SSE2 version of 3.7?
Custom compiles for 3.6 and 3.7 on Debian 5.0 x86-64 (gcc 4.3), so I would
assume they're both SSE2 enabled.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
I think I found the problem. Back when I was using the old processor, I compiled
3.6 with -march=k8. But when I switched processors I didn't think to recompile.
After recompiling with -march=barcelona, I got a 40% speed increase with 3.6!
The configure script for 3.7 uses -march=native so I think that might account
for why it seemed to be so much faster.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
