Search found 58 matches

by mpeterson
Wed Mar 20, 2013 2:19 pm
Forum: Links & Papers
Topic: Haswell benchmarks show up
Replies: 20
Views: 17971

Re: Haswell benchmarks show up

Dade wrote:the speed up is quite small: something in the 10%-25% range.
right, the same holds for gpus when we compare the last two generations.
by mpeterson
Fri Feb 15, 2013 5:10 pm
Forum: General Development
Topic: Can't understand Timo Aila's GPU traversal paper
Replies: 16
Views: 9219

Re: Can't understand Timo Aila's GPU traversal paper

jbikker wrote:SOA is good for SIMD processing, because it allows you to read 4 coordinates at once in an 128bit register.
wakeup call ! we have already 512bit in place.
by mpeterson
Tue Feb 12, 2013 5:01 pm
Forum: Scratchpad
Topic: tbp
Replies: 5
Views: 8860

tbp

someone ever heard something from tbp in the last month/years. he seems to be fully disappeared.
by mpeterson
Mon Dec 17, 2012 9:31 am
Forum: Visuals, Tools, Demos & Sources
Topic: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?
Replies: 29
Views: 30074

Re: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?


these are results from the intel labs with really poor scaling. we have seen much better rt results with phi comparable to up2date gpus.

mp
by mpeterson
Tue Nov 27, 2012 9:09 am
Forum: General Development
Topic: Ward tone mapping
Replies: 8
Views: 8237

Re: Ward tone mapping

in practice (having a rt, interactive etc. renderer) use the avg. luminance from the last frame.
all local tm operators need some kind of "global" luminance values to adjust too. for me the avg.
is the most stable and the workload can be easily distributed over all threads.

mp
by mpeterson
Thu Nov 22, 2012 11:59 am
Forum: Visuals, Tools, Demos & Sources
Topic: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?
Replies: 29
Views: 30074

Re: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?

$400 ? no its more like $2400. who will give it a try ?
by mpeterson
Thu Nov 15, 2012 8:57 am
Forum: Visuals, Tools, Demos & Sources
Topic: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?
Replies: 29
Views: 30074

Re: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?

Only time will tell of course, but I suspect that the 4 threads per core will only serve to partially fix the fact that the cores are in-order. A 4x improvement factor is definitely way too optimistic :) right, each core is 4x ht (to compensate for non-vectorized codes). when running with all threa...
by mpeterson
Wed Nov 14, 2012 4:55 pm
Forum: Visuals, Tools, Demos & Sources
Topic: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?
Replies: 29
Views: 30074

Re: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?

My take is that it will be easy to get core-wise parallelism out of the chip. If your code scales well on CPU cores already, it will scale to the cores on Phi. However, without making good use of the SIMD lanes, I would guess your performance will not be any better than an 8-core Xeon chip. This re...
by mpeterson
Tue Nov 13, 2012 3:36 pm
Forum: Visuals, Tools, Demos & Sources
Topic: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?
Replies: 29
Views: 30074

Re: Intel Xeon Phi Preferred Pricing - Only $400 per Card ?

I don't know if i would want to redo a several 10000 line of code project via intrinsics (even if hidden within C++ wizardry).. ;) i am working with that crap since two years now. no complex algo can fill the __m512 registers over time. so you have to go with one of the > 200 threads and trust the ...
by mpeterson
Sat Oct 27, 2012 9:27 am
Forum: General Development
Topic: Csillámtrace - my new graphics engine
Replies: 153
Views: 100341

Re: Csillámtrace - my new graphics engine

shit, i like this thread...