Intel 14th/13th gen, latency/stability issues?

[Scotty]

Interesting use case scenario. I’m a VCV Rack user but I don’t go higher than 48kkz . Have you found any bench marks apart from core to core latency to be a good relative indication of what kind of performance you’ll get with such high sample rates? Does memory tuning help? Do you have to avoid performance cores or certain cpu series altogether? Are you using pcie hosted interface like an RME Raydat for lowest latency?. What is the best solution you’ve found so far both CPU and interface? I'm really interested in your observations.

Look at the latest DAWbench
http://www.scanproaudio.info/2024/09/12/5075/

Looks like the testing was done prior to AGESA 1.2.0.2 (which brings in the core latency fix) - it should probably be done again.

For my interests (which are somewhat niche, admittedly), DAWBench DSP / VI are not overly informative measurements. Whether a platform can execute 300 or 500 instances of a single-threaded plugin doesn't tell me much, as 300 or 500 instances is far more than I'll ever need.

I'm interested in multithreaded modular synthesis (in my case, using VCV Rack), running at extreme sample rates (e.g. 384KHz, 768KHz) and the lowest possible buffer. As the modules, individually, are often very simple, low level compute tasks, how quickly those modules communicate between their hosted cores, via on-die cache and physical barriers like CCDs, is critical in determining how well a patch can scale across cores on any given platform - and core-core latency gives me the clearest picture of how much of a CPU I can effectively utilise in an environment like this.

[Pictus]

I'm interested in multithreaded modular synthesis (in my case, using VCV Rack), running at extreme sample rates (e.g. 384KHz, 768KHz) and the lowest possible buffer. As the modules, individually, are often very simple, low level compute tasks, how quickly those modules communicate between their hosted cores, via on-die cache and physical barriers like CCDs, is critical in determining how well a patch can scale across cores on any given platform - and core-core latency gives me the clearest picture of how much of a CPU I can effectively utilise in an environment like this.

I like VCV Rack, very interesting!
You may like this. viewtopic.php?p=8820694#p8820694
I use ALL, but keep MPO(GPU) enabled.

[Scotty]

I recently built a single socket X99 Xeon system specifically for VCV Rack and haven’t yet done those advanced tweaks. Just the basics for optimizing windows 10 for audio. I am using the Expert Sleepers Es 9 interface. When I get home in 10 days I’ll do some comparison and I’ll take screenshots so I can review any performance gains or losses along the way.

I did many of these tweaks (all of the purple) for my dual socketx99 Chinese motherboard build and my Cubase performance did improve when benchmarked before and after. There we’re fewer spikes and the asio performance meter didn’t dance around as much when running low latencies.Using an RME Raydat on that machine with all 34inputs connected and monitored and 16 or so outputs connected.

Both systems are on Windows 10 as they lack the TPM security modules. The single socket build does have a Third party TPM module which is working and I’ll be ordering that soon for windows 11 on that machine,

Both of these rigs are using Xeon E5 2690 v4 processors on Chinese motherboards.

[NER]

- server boards would be especially prone to destroy CPUs, as they would use "safe" settings, meaning high loadline > high voltage

It's primarily because the servers run heavy loads 24/7

[Zombie Queen]

It's primarily because the servers run heavy loads 24/7

My understanding is that 24/7 all core loads would be OK. Crazy voltages would appear in idle or single core boost.

My server remark was basically that enthusiasts would mostly undervolt their systems, while servers would run "safe" settings, where safe means not crashing means high voltage.

Anyway, there's yet another microcode update, still in beta for ASUS bios.