Quackmaster
Member
- Apr 19, 2003
- 68
- 0
- 0
Originally posted by: TerryMathews
Hate to pick nits, but here I go. The SMP switch in Q3 offloads geometry and world data to the second processor. With the advent of HW T&L graphics cards, only the world data gets offloaded to the second processor. The problem here lies with timedemos. In timedemos, the world data is static, leaving the second processor nothing to do and making the game look slower. In a real game, with bots or humans, running r_smp 1 would yield a higher overall performance, but possibly lower peaks. Like if you graphed your Q3 performance, the single proc might have higher peaks, but the dually would have more area under the graph, meaning higher overall frame rates.
You found a way to pick nits anyway tho, eh? /flamerwar on just kidding. That's interesting though, I did not know that about the Q3 codebase. The game is from '98 so I haven't tried SMP since the abit bp6 and pre T&L cards.
Also, Counterstrike is a Half-Life based game, not Q3. EDIT: And another point I wanted to make: HyperThreading is not SMP. It allows two seperate programs to take advantage of different parts of the CPU. For example, one process can use the FPU and the other the ALU. Games typically only make use of one piece of the core.
Well I think anybody who's got any sense knows who made Half Life (and thusly CS), but the code base originated with Carmack's Quake2 engine (althought they obviously did a lot of modifying the source code). As for the hyperthreading explanation, between Aces, Anands, xbit, Toms and everywhere else... I think anybody who's taken a microprocessor architecture college course or read a William Stallings book (including me) has no problems understanding how it works.
Now then Mr Terry Matthews (or is it Ms.? doh! ambiguous name ya got there ), I'll quit acting like an insecure little pr*ck and just drink this beer while I read more posts!