View Single Post
  #6  
Old 03-14-2011, 03:04 AM
TUSA/TX-Gunslinger's Avatar
TUSA/TX-Gunslinger TUSA/TX-Gunslinger is offline
Approved Member
 
Join Date: Oct 2007
Location: Austin Texas
Posts: 195
Default

Great question Buzzsaw. I read those sections several times myself. Of all the posts that have come to this board in it's existance - this is the thread I hope one or more of the developers posts in.

If they don't respond to this, I believe it's because - "They don't really know, exactly". Why? Because at some point, the number of various systems and software becomes an counter-productive systems-administration burden on developers, so that they are paying almost as much money keeping the varied CPU/GPU/OS's current and documented - as they are on programmers and graphics developers.

With 15 years of working with and managing development of threaded, advanced graphics and signal processing software, I can safely say: "I certainly don't have a clue"

As a simmer, I spent 18 months on the ROF beta team testing various configurations of hardware (will never do that again, too much work and too little play) against many beta builds, and comparing my results to other testers performance, in reasonably controlled conditions. What did I learn? or relearn?

That testing on 3D Mark or other benchmark software and/or examining hardware performance on non-simulation software (i.e. FPS type games) - tells you next to nothing about how applications like FSX, ROF, Il2, DCS will behave.

This gets even more complicated with threaded and partially threaded applications, of which ROF is/was.

Without knowing how the routines are structured and how the software scales under load, we can't know.

Between the Intel and AMD offerings - how significant is real cores vs hyper-cores in performance of this application?

The only "wild speculation" I would even hazard a guess on is that in my development group at work, where we do have access and choice between Intel and AMD solutions - we've tended to stick with AMD (Opteron mainly) due to cost/performance and inability to obtain significant threaded performance from hyperthreading in our applications.

Our code is fully threaded. Period. On the other hand I know of no entertainment software which is.

Hehe - so with all that I don't know crap

What I would recommend we do - in lieu of explict recommendations from 1C - is form our own voluntary testing team early on. Find a benchmark, set of standards and go for it. Like on day 2 of the release. With the purpose of informing the community of hardware/OS performance. All we have to do is get organinzed a little.....

It would be at least as productive as much of what's currently being discussed in every forum concerning CoD, IMHO.

S!

Gunny

Last edited by TUSA/TX-Gunslinger; 03-14-2011 at 04:15 AM.
Reply With Quote