Cut Crystal Tests

WytRaven · Post by **WytRaven** » Wed Oct 10, 2007 3:09 pm

I am currently saving ($10,000AUD budget) to build a dual quad-core machine. Estimated build time around February 2008. I'll see what the CPU situation is at that time before commiting to AMD Barcelona Opterons or Xeon's. I'm hoping the Barcelonas will be up to the 3GHz mark by Feb. Fingers crossed

CoolColJ · Post by **CoolColJ** » Wed Oct 10, 2007 3:17 pm

Well the new Intel Penryn Xeon quads are at 3ghz for sure already, and can clock higher if needed

I would go Intel, I don't think the AMD will be as fast on pure grunt

WytRaven · Post by **WytRaven** » Wed Oct 10, 2007 3:23 pm

They are very different beasts, those two CPUs, you can't directly compare them clock for clock. Bareclonas are true quad cores, not 2 x Dual Core attached via a huge L2 cache...

A 3GHz Barcelona should, due to better architecture, easily out perform a 3GHz Xeon at the same task. It will also do it with lower power consumption which starts to become an issue as you get more cores in the same machine.

CoolColJ · Post by **CoolColJ** » Wed Oct 10, 2007 3:37 pm

I haven't seen too much of a disadvantage even though Penryn are supposedly inferior arhcitecture wise - as they say proof is in the pudding

They're also running cooler, lower power consumption and clock up higher due to the smaller die size

zsouthboy · Post by **zsouthboy** » Wed Oct 10, 2007 4:27 pm

Scaling for Barcelona vs. Peryn for 4+ cores is amazingly different. The Peryn quickly falls behind.

WytRaven · Post by **WytRaven** » Wed Oct 10, 2007 4:31 pm

Uh-oh... we've started something here

As I said I'm leaning towards the Barcs as they are an obviously better architecture but if they haven't got high enough in speed by the time I build I may have to settle for Xeons

CoolColJ · Post by **CoolColJ** » Wed Oct 10, 2007 4:45 pm

or wait for Intel Nehalem and get the best of both worlds and with up to 8 cores per CPU

or just getting something to hold you over and upgrade as you go along

Stur · Post by **Stur** » Wed Oct 10, 2007 8:01 pm

If you're always waiting for the next better processor to come, you'll never buy anything

WytRaven · Post by **WytRaven** » Wed Oct 10, 2007 8:40 pm

Stur wrote:...you'll never buy anything

I think my wife would be very pleased if I adopted that particular way of thinkging

Render Update:
~122h, ~8300 samples/pixel

The render is obviously much closer to convergence than it was this morning. I still think 200h is probably the likely end time.

Schott Glass data update:
I figured out on the bus ride home from work today one of te issues that has been plaguing my Cauchy calcs. Much closer to get it right now.

Stur · Post by **Stur** » Wed Oct 10, 2007 11:09 pm

WytRaven wrote:
Stur wrote:...you'll never buy anything
I think my wife would be very pleased if I adopted that particular way of thinkging

ROFL !
Ok, here is a good method : start arguing with your wife AND saving money in the same time. When she'll be fed up and finally say yes, you should have a nice amount of dollars to buy the very high end processor of the moment.

WytRaven · Post by **WytRaven** » Thu Oct 11, 2007 1:38 am

@Ono:

I have been sitting here for days trying to figure out what I'm doing wrong with these cauchy calculations when I finally realised why I can't get it to work. It's not because I'm mad (thank God) it's because below about 430nm Cauchy's equation simply cannot continue accurately.

I finally discovered why ryjo chose the nF and nC from which to derive his B coefficient. The difference between these two points is know as the Principal Dispersion. I Used ryjo's method to calculate B and used a rearranged Cauchy equation plugging the median of nD and nd to solve for A. The results are as close as you will ever get to correct using cauchy.

If you look at the spreadsheet the first sheet is now a great big graphing station

You can select the glass you want to look at from the box up the top and it will graph meassured vs calculated values for n, a, and t. I put the transmission in as a sanity check for my calculated absorption values.

So after days of trying to fine tune Cauchy I realise that given n/a pairs already calculated with great precision via Sellmeiers equation, for IoRs, and spline interpolation, for Absorption Coefficients, you shouldn't have any need for the Cauchy stuff at all. This spreadsheet can be altered very quickly to produce n/a pairs at whatever level of detail you desire and you can then interpolate between the values in indigo if required.

If I'm wrong about the above and you still do need the Cauchy stuff for whatever reason, then it's still there, but you can see how crap Cauchy really is

There is a little issue with the transmission data, mainly missing info at some wavelengths and zero values at others which I got around by filling in myself with zeros (marked red in source data) and then replacing in absorption generating functions with a fairly small non zero number. You could just set a threshold in indigo, anything over absorption coefficient value x = kill path; it died a lonely death right here. RIP.

The spline function I used is a part of a handy little free addon for Excel which I have included along with the spreadsheet.

So anyway let me know how you want these pairs of numbers supplied and at what interval; a string of values in a text file per glass or whatever and I will do that for you.

EDIT: Download removed

alex22 · Post by **alex22** » Thu Oct 11, 2007 7:14 am

Lol it took my parents five years to choose and buy a new TV

You shouldn't wait to long...

WytRaven · Post by **WytRaven** » Thu Oct 11, 2007 3:37 pm

mmmmm....Tyan Thunder n6650W (S2915-E)

Yes please!

CoolColJ · Post by **CoolColJ** » Thu Oct 11, 2007 3:57 pm

can't the regular AMD/Phenom desktop chips run in dual CPU?

That's the way I would go myself - save some bucks
and there is a 4 CPU MB from Tyan

Supermicro make the better MB for Xeon, dunno about AMD though

WytRaven · Post by **WytRaven** » Thu Oct 11, 2007 4:50 pm

4CPU goes beyond my $10K budget I'm afraid.

Cut Crystal Tests

Who is online