Running on modern machine settled by -CPUCount:1 option

[A Nonny Moose 8,261]

I've been running for some time with the -CPUCount:1 option set. This seems to have almost completely killed the tendency of the game to crash on modern hardware.

Yesterday, I ran for over 12 hours without a single crash, and today for over two hours before I needed to stop.

Over the last few weeks, I have upgraded to a new version of the Operation system (UBUNTU 12.04) and to a beta version of wine 1.5.3 when, with the exception that wine can't handle the multi-buffer requests for hardware graphics, I have had smooth sailing. I run in software rendering quite satisfactorily.

My machine summary:

Athlon II X2 model 245

2.9 GHz

L2 cache 1024 KB

4GB Ram 25 GB Swap partition

SimCity 4 Deluxe 1,1,641,0

Ubuntu 12.04 LTS (GNU/Linux 3.2.0-24-generic x86_64)

[Piteri 17]

I find most of my crashes to be GFX related. That is; they all seem to happen whilst scrolling or switching between subway-ground view.

[jdenm8 362]

Eh, most graphics cards don't support several of SC4's graphical components properly now anyway. They generally hand processing off to the CPU resulting in modern GPUs performing about as well as a GeForce 2 Series card when running SC4. It's one of the issues that Maxis identified with the engine only in the last few years. I run SC4 in Hardware mode on both my PCs (AMD Radeon HD6870 and ATI Mobility Radeon HD3200) and it generally runs terribly. That's a lot of data to pipe across a PCIe 2.0 x16 bus (Or in the case of my laptop's GPU, its insipid integrated solution meaning it has to come from system memory, to the GPU, back to System memory, to the CPU, to System Memory, back to the GPU, back to System memory, back to the GPU and finally on my screen) then back again, the whole time fighting over the attention of my i5 2500K.

The crashing when scrolling and zooming is probably caused by your fancy graphics card's poor swapping performance when doing things that SC4 uses. SC4 just can't wait for the sluggish GPU to respond.

[A Nonny Moose 8,261]

You make a very good point. The memory being on the same bus as everything else is a real drawback in modern machines. Other than some complications, solved in the 1950s, there is no reason for single bus machines to exist any longer except for the price point. An asynchronous, interrupt request driven, memory oriented architecture would work much better especially with GPUs that share memory space with the CPUs. It would also increase the overall performance of the multi-core chips if they had a private bus to memory. This amounts to putting the DMA in charge of all cycles, operating via memory requests from the active modules. Examples of the past: CDC6600, GE635, IBM7090. And by the way, the really big mainframes still work this way.