Emulando Avx/Avx2 usando un CPU Xeon antiguo
Emulando Avx/Avx2 usando un CPU Xeon antiguo
Looking into optimizing AVX/AVX2 emulation via Intel's SDE.exe. The SDE program helped launch Star Citizen as a test for AVX emulation. Since the game runs at one frame per minute, you might explore alternative tools or methods to speed things up. Other programs could also be considered. Your interest is appreciated—this could be useful for older systems.
Following an instruction set emulation demands significant energy, which explains why your CPU struggles. What type of Xeon are we dealing with? Avx has been around for quite some time. These seem similar to Core 2 models. If they’re indeed that old, then yes—it will definitely not perform well. These are very basic processors, and running a heavy instruction set emulator on them won’t work.
I'm using two x5660 chips with six cores and twelve threads. Since I'm running them together, I wanted a way to divide the CPU load. I also considered if the software could run the entire game instead of just the AVX instructions.
In short, you're looking to understand how Intel's AVX state transitions affect moving SSE code to AVX. The article explains the process and implications for optimizing performance during this migration.
I completely understand your point—it's been a while since an update was needed! Honestly, most people see it as unnecessary, but I enjoy tackling these challenges and usually manage to solve them. This is my first request for help or a fresh perspective. Hopefully, someone can spot something I missed. Also, if you read that article I shared, it should give a clearer explanation than what I could provide. Intel actually built an emulator for similar projects. (Don’t worry about the heavy stuff, though—still fun!) P.S. Thanks, @jaslion, for taking the time to chat with me.
This emulator wasn't built for real-time high performance, which is what you're aiming for. This creates a conflict at its fundamental purpose. It was created to assist with the transition to AVX chips and succeeded in that role. Now you face the challenge of using software meant for more demanding work on a CPU that lacks sufficient power for real-time results. Unfortunately, that won't work.
If anyone in the future encounters this, I'd love to hear your thoughts. Perhaps using the emulator as a compiler for Avx to SSE conversion with GPU passes could help—so processing happens directly on the GPU rather than just through CPUs. Maybe packaging it in its own container or setting up a VM correctly would solve the issue altogether. This is just brainstorming, and I'm not a software engineer. For now, and probably always, your point stands: this hardware is likely too outdated for such tasks. Thanks to everyone who reads and shares any ideas.