Ooh, no, technology!

[Sarra]

I'm going to be getting a loan from my parental units for new computer equipment.

I'm going to have roughly $2,500 to play with, plus whatever I can save until November (don't count on anything more than $100 though).

With this, I will be building two computer systems. One will be my gaming rig, and will have mostly non-essential storage in it. This computer will be getting a small RAID 0 setup, plus whatever other drives I can stick in it.

The other PC will be a dual purpose server/workstation. This PC will be part server, basically a fileserver, and I'm going to get a very high-performance Dual Opteron mainboard for it.

This is my question. I would like to get a few SATA drives, and as a Server motherboard, the onboard SATA controller does multiple kinds of RAID, including RAID 5, RAID 1, RAID0, etc. I would like some kind of parity/mirroring. Should I do RAID0 with a backup drive, and backup from the RAID0 to other drive, or should I do an actual RAID0+1? Or would RAID5 be more effective? Everything that won't fit on the RAID0 drives in my gaming rig will need to access the fileserver over a Gigabit ethernet LAN, so access times need to be on par for streaming video/audio.

Suggestions? The boards (one board will be a socket 939 board, which wil be getting an X2 dual core processor, the other board will initially be getting a 140 Opteron until further funds can be located, at which point, the board will get a pair of Dual Core Opterons.

[Sarra]

I play some fairly processor intensive games, a laptop won't cut it for me. :p

[Cassandra]

Sarra,

I have a friend who is into real time aircombat simulation. It is highly processor intensive as it must react not only to his actions but those of his squadron and the squadron they are in combat with while maintaining the virtual reality scenerey. He is very knowledgable and could probably answer your question. If you don't mind I could copy your post and forward it to him. Might take a few days before he gets back to me though.

Cassie

[Jessica]

http://www.ageia.com/technology.html

Perhaps it won't be so processor intensive in the future.
The principle is the same as the Math Coprocessor.
Basically, the thought behind it is:
If Graphics Processors can take offloaded vector calculations from the processor and
Math Coprocessors can handle offloaded floating point calculations from the processor

Why can't something offload physics calculations from the processor?

Currently in games, vector processing for 4d (don't ask, but yes, it is 4d vector calculations due to rotations about a 2d axis plane (the monitor, 'player point of view')) is done by the GPU (Graphics Processing Unit) located on the graphics card.
Most of the rest is done by the processor.
It is estimated that something like 45% of the work done by the processor is simulated physics within the game environment. Collision Detection and Reaction is a HUGE amount of the 'physics' work. This physics chip is designed to offload that work.

Hey! I added something intelligent to a conversation 

I personally don't really game anymore, but I have studied 3d game programming (primarily in C++ with vector processing algorithms, collision detection algorithms as opposed to world design and graphics design... basically from a programmers stand-point, not a graphic design stand-point) and design fairly extensively and read a few very good books on the subject.  Basically, the only thing I see as a problem is communication between the processor and physics coprocessor.  It will add delay as it will have to communicate via a bus of some sort.  The PhysX chip will have to be compatible with existing architecture which means that regardless of how fast your processor is or how fast the PhysX chip is, they will still have to communicate across a fixed length bus (I believe current lines are 64 line bus's but I might be wrong) therefore, regardless of the speed of the processor and the physX chip, you are still limited by the speed of communication between the two.  Additionally there will have to be protocols in place specifically for the communication between the two which will be sort of a "interpret -> decode -> transmit -> decode -> interpret model which could result in additional overhead between the two.  Then the GPU will have to be able to communicate simultaneously with the processor AND the physics chip.  Since it has to be backward compatible with existing technologies, the only way to do that is to have the physX chip communicate to the GPU in the same language as the processor AND along the same bus as the processor, which means the GPU could very well get bogged down in the additional communication.

I have not seen how they have addressed these problems in communication and am leary of jumping in and saying yes, this is great.  I have not seen their communication model (which I am sure is proprietary).

Oh... well... sorry for rambling.... I guess I am saying, it is a really neat idea if it is implemented properly.

Sorry for all the technical talk, I'm a computer programmer, it's what I do bestest! <tigger>

Jessica

[Lisbeth]

http://www.ageia.com/technology.html
Why can't something offload physics calculations from the processor?

They've done that in the supercomputer world for years.  It's called an array processor, and does the same math calcuations to all (sort of, depending on how big the processor is) of the elements in a matrix at the same time.  If you have an 8 x 8 matrix to rotate, a 64-way array processor can do it in one machine cycle.

[Jessica]

From my understanding this isn't necessarily matrix rotations.  This PysX chip is designed specifically for collision detection and handling.  Matrix rotations, Matrix Calculations, and Vector Calculations & their corresponding draws have usually been the domain of the GPU in current PCs.  This chip isn't really changing any of that.  It is basically offloading the collision detection and has a lot of hardware level support for Physics Transformations and Calculations.

I am not sure about the mainframe world, but in the PC world, in theory, it will free up a lot of the processors time because collision detection is a HUGE amount of the processors time, and built in hardware support for physics functions will be really nice.

Jess

[Sarra]

Well, I play non combat sims fo rthe most part, and whatever ones I do play existed before the Pentium 4 was launched. I do play Flight Simulator 2004, which is more processor dependant than video card dependant. Upgrading my processor has helped tremendously, RAM upgrades helped as well. Video card upgrades help too, but to a lesser extent.

What I'm mostly concerned about is drive speed. I'm going to be running a 3 disc RAID in the gaming computer with a program that automatically backs up my data to another drive. It's what to do for the backup part that has me confused, as I've only ever worked with RAID0. Should I get 3 100 gig drives for RAID0 and a 300 gig drive for a RAID1, or do a RAID 5 (or even RAID 6)?

The server is going to be more complicated. Do I do RAID 5, RAID10, do I do it on SATA drives or should I use SATA and SCSI?

[Shelley]

I'm so out of my depth here sometimes. I remember when calculations were done with pencil and paper.

Shelley