GroovyMAME on enterprise hardware: Can 16 cores (Xeon Gold) compensate for a 2.0GHz clock?

[rojop64664]

Hey everyone,
I’ve been lurking here for a while, mostly soaking up info on pixel-perfect setups and the eternal struggle for sub-frame input latency. Recently, I’ve decided to go down a bit of a hardware rabbit hole that I haven't seen discussed much in the shmup community, and I'm hoping some of the technical experts here can chime in.
I’ve recently repurposed an HP ProLiant DL380 Gen10 https://serverorbit.com/pc-and-servers/ ... -xeon-gold that I salvaged from a local office decommissioning. It’s a beast of a machine, currently sitting in my garage rack with a 16-core 2.0GHz Xeon Gold processor. Normally, for a dedicated GroovyMAME or ShmupArch build, the standard advice is to grab a high-clocked i5 or i7 because single-core performance is king for arcade emulation. However, I’m trying to see if I can leverage this enterprise iron to build a "multi-instance" emulation server for my game room.
The specific point I’m struggling with is frame-time consistency when under heavy multi-threaded loads. On this 16-core Xeon Gold, the 2.0GHz base clock is admittedly low. I’ve noticed that while the server handles modern tasks effortlessly, certain demanding shmup titles—especially Cave’s CV1000 stuff like Mushihimesama Futari or Dodonpachi Saidaioujou—seem to exhibit tiny, almost imperceptible stutters that I don't get on my 4.0GHz desktop.
My personal insight is that I might be fighting against the server's power-management states or perhaps the way the Xeon Gold handles its Turbo Boost. In an enterprise environment, these CPUs are designed for steady, parallel throughput rather than the "bursty" single-thread performance we need for high-speed bullet hells. I’ve tried pinning the emulator process to a single core and disabling C-states in the ProLiant BIOS, but I’m still seeing some frame delivery variance that just feels "off" when the screen gets busy.
I’m also curious if the memory architecture on these ProLiants—dealing with ECC and much higher latencies—is introducing a bottleneck that we don't usually talk about in shmup circles. We’re so focused on the GPU and the display lag that maybe the internal bus latency of server-grade Xeon platforms is a hidden enemy for high-level play.
Has anyone else here tried to go the "server route" for a dedicated emulation box? I'm wondering if I should keep tweaking the CPU affinity settings on these 16 cores, or if a 2.0GHz base is simply an insurmountable wall for the more demanding arcade drivers regardless of how much "Gold" is in the name.
Do you think we are finally reaching a point where high core counts can mitigate lower clock speeds in emulation, or is the architecture of these server CPUs fundamentally ill-suited for the 1/60th-of-a-second precision we need?

[Josh128]

Didnt you post a similar topic here over a year ago? If you didnt, someone certainly did. Getting super deja vu reading this post.

**EDIT - you did lol. I found the post below. Its an intriguing idea, but I think you got your answer last time. The cold hard truth is that even the cheapest AMD Zen 3 or higher gen or 4 core Intel 12th or higher gen CPU will run circles around these higher core count, low clocked server systems for any and every kind of video game hardware emulation.

viewtopic.php?t=77131

Do you think we are finally reaching a point where high core counts can mitigate lower clock speeds in emulation, or is the architecture of these server CPUs fundamentally ill-suited for the 1/60th-of-a-second precision we need?

This will never happen as newer systems (to be emulated) have faster CPUs and GPUs, which require ever increasing clock speeds (on the CPU used to perform the emulation) per core to emulate. Some older MAME drivers such as Atari Seattle & Vegas use multiple cores/threads to emulate the boards CPU and the Voodoo Graphics subsystem separately, but at 2.0GHz, Im not sure that Skylake era x86 architecture Xeon you are looking at is even fast enough to achieve 100% emulation speed. It probably is, but just barely. I would be very interested to see what emulation % you get in game during an actual fight in Mace:The Dark Age, or San Francisco Rush.

You will never see a fast CPU being emulated by multiple slow cores, etc. Its just not feasible. Clock speed and IPC is and always has been king for emulation.

[orange808]

Timing gotchas and scheduling delays associated with multi-threading have been a staple ongoing subject of spirited development and debate for decades.

The only realistic solution for your machine is to implement less accurate emulation and reduce the load. Sounds like you need to fork some existing or legacy emulators to me. I'm not interested in this at all. Maybe try reddit. Good luck.