Hi Burr, it looks like a killer system!

I have not been really following every single detail of the latest CPUs to be able to give much specific advice...

I think it's a pretty good idea to just let DELL worry about having all compatible stuff put together for you, definitely saves a lot of potential problems!

I guess the only thing that is maybe not quite optimal is pricing - the Xeon should work great but they may be pretty expensive compared to the brand new Core i7. But I'm not sure if there is any way right now to set up a 2-socket Core i7 system to get 8 cores instead of only 4, that's why I was interested in what you were doing for that.

But actually I'm surprised that the DELL 8-core system is not really as expensive as I would have thought neither. It actually looks pretty good in that department overall as well,

- Michael

> How do you think it will handle with 2 quad cores??
> :) (of the Xeon type!)


I only have two cores :-(


But still, two is better than one :-)

Seems your more up than I . Didnt even know about a "Core i7" :) Maybe 4 core i7 is faster than my 8 core xeon?? (Thats usually what happens) still stuck in old core technology.

The I7 has four cores but they are hyper threaded so you get eight virtual cores. It's slower than actual double quad xeons but I am waiting for them to make dual cpu I7 boards so I can get 16 cores running in one box for under $4000. I too have a Dell with dual quad xeons and I can't wait to run mesh comparison times.

Hey. That means I will make models and renders "JUST LIKE GRENDEL!" :o

Wait just a biiiit longer and get the core 8! (Actual 8, not virtual)

Hi Grendel,

> I too have a Dell with dual quad xeons and I can't wait
> to run mesh comparison times.

I'll also add in a moi.ini switch that can be set to limit the mesher to only one thread, that would be good to get a benchmark of the new mesher with that limited as well to see how well it is scaling out to a higher number of cores.

There are also quite a lot of optimizations that improve just single-core meshing speed by a significant factor as well, so without a thread-limiter setting it won't be easy to know what factor is coming from multi-core and what is coming from the other optimizations.


It is possible for certain kinds of scalability problems to get magnified when the number of cores gets fairly large. I don't think that I should have much of that but it is hard to know for sure without actual testing on a higher number of cores, I have been testing mostly on quad core.

Are you using XP or Vista? There appears to be some tweaks in Vista to make a certain kind of multi-core scaling problem kind of go away ("lock convoys"), but I'm not quite sure right now how much that affects MoI in particular.

- Michael

I'm using XP 32-bit.

I'll run a comparison with just a straight v.1 mesh and a beta mesh with the multi-core mesher if that helps remove the intangibles.

Hi Grendel,

> I'll run a comparison with just a straight v.1 mesh
> and a beta mesh with the multi-core mesher if that
> helps remove the intangibles.

That would certainly be good to know too!

If you could do 3 tests it would help to get the full picture though.

One with version 1.0, another with the new mesher running full out, and a last one with the new mesher limited to 1 thread (by a special moi.ini setting enabled).

Almost ready, I think I've got the meshing all wrapped up now and I just need to clear a list of some other bugs first before releasing. Maybe a couple of days.

- Michael

No problem Michael three tests

Hi Michael....

> Maybe a couple of days.
I think you will release the next beta at Jan 21, 2009.
Base on the history, you like to release at 21 :)

Hi Anis, yes it is funny how it has been working out to be the 21st several times now! :)

This time I think probably before that though.

- Michael

i just bought a dual-quad Xeon about 5 months ago. So i'll stick with it for at least a few years. I think my next move is to build a stripped down render slave unit...

jonah

Thats great JB, I remember when I used to wait for several hours for a render to come out of the oven. ;P)

I've looked at getting a renderboxx(four quad xeon rack). The price is around $6k which is roughly the same as getting two more workstations but the space factor is a lot smaller. Plus your not using up as many resources.

When i spoke to Boxx about the Renderboxx ("Pizzabox" as they call it), they told me a price of around $2500. There is no GPU or anything, just chews the data and returns it (excrements it?) back to the master unit... So one of must be wrong about the price... Hopefully it's you! :P

I hope it is me because I'll have one sooner than I thought....must check again.

I am also holding on to a hope that 3d soft companies will be able to harness the CUDA processors that nvidia introduced. 240 processors on a unit, how sweet would that be to have stuffed into an excrement box.

Yeah CUDA is pretty crazy. What they don't tell you is you need a cooling unit (the CULA) which is the size of a small condo! :P

Heat factor. What drives your speed! Imagine your PC powering your refrigerator!???

Hi Grendel,

> I am also holding on to a hope that 3d soft companies will
> be able to harness the CUDA processors that nvidia introduced.

Some certainly will! But it is not really a great fit with every kind of software problem. Yes you've got 240 processing units in there, but each one of them is not as powerful as a regular main CPU...

It tends to be something that works well for speeding up things where you have a million pretty simple calculations that you want to get done. Rendering can often fit into this category well, and some things like particle or voxel processing is a good fit too.

But if the problem at hand does not easily split up into a million simple sub-tasks then it doesn't really fit into that area.

- Michael

The rendering is what I am hoping will fit.

Hi Grendel, have you tried nVidia Gelato yet? I think it is the main existing renderer that has been tuned to take advantage of this kind of stuff so far.

- Michael