3D V-Cache and CPU Performance

[rigoschr]

Hi all, 

while @Minh Nguyen has compiled truly great guides and comparisons for GPUs, I cannot find detailed information on CPU performance in Archicad.

My parents have been struggling with slow 2D navigation (redraw?) in large projects, hence I'll upgrade them from an 9700k to a 13700k, as the latest intel CPUs have made a huge leap in single-core performance.

However, with the new Ryzen 7000 CPUs with 3D V-Cache coming out, I was wondering if the additional cache can help some Archicad workflows. I personally know that racing simulators (and Microsoft Flight Simulator) benefit hugely by the additional cache. On the other hand, CPU rendering is not helped much by large amounts of cache, but what about 2D redraw? Without background knowledge on the code the only other option is to test it, but I lack the resources and time.

Anyone have experience on that? A sensitivity analysis for RAM bandwidth and latency would also be a great resource for everyone. @GRAPHISOFT let's go, please help us clear-up some performance questions!

My wishlist:
- Does thread scheduling work properly with 12th and 13th gen Intel CPUs? Both on W10 and W11?
- Does 3D V-Cache have a significant benefit for 2D redraw or other operations?
- Intel 12th vs 13th gen comparison would also be interesting, due to the increase in L2 cache.
- Is there a benefit of tighter RAM timings and higher frequency?
- Does Archicad favor Intel or AMD at the moment?

[Haris Matthaiou]

Hi rigoschr,

Thanks for initiating this discussion as I am sure other users are having similar questions given how many new technologies are available and how it all glues together when it comes to performance.

• I believe you're referring to Intel's Thread Director technology that was introduced with 12th Gen Intel CPUs. Thread Director is utilizing a microcontroller embedded in the CPU that works with the Windows scheduler to determine whether to assign a thread to performance (P) cores or efficiency (E) cores. Although the technology is compatible with Windows 10, it is primarily optimized for Windows 11 Task Scheduler. This is a software-agnostic feature, the implementation occurs at a low level between the CPU and OS.
  
  The main idea behind separating P cores and E cores is to detect whether a computation is I/O bound or low priority background tasks (E cores) or CPU bound (P cores). By utilizing this technology, performance can be optimized based on workload characteristics. However, the real-world effects of Thread Director will depend on the specific tasks being performed and the characteristics of the workload.
  
  The general concept behind Intel's Thread Director technology is to improve thread scheduling and power management on the CPU. By utilizing Thread Director, the CPU can achieve a more responsive multitasking environment, where foreground and background applications are competing for CPU resources. This is accomplished through a smarter scheduler that leverages machine learning to identify and profile the computation requirements of each application.
  While Thread Director can provide better power management and thread scheduling on the CPU, it may not have a noticeable impact on high-priority tasks like Archicad, as the software will utilize P cores anyhow.  However, it's worth noting that Archicad relies heavily on single-threaded performance, and while efforts are being made to extend multithreaded operations in each version, the single-thread performance of a CPU is still the most important factor to consider when selecting a CPU for Archicad and no matter how smart scheduling can be, it won't help in boosting performance. 
• Regarding AMD's 3D V-Cache, although it will surely improve things on gaming performance, I personally believe that is unlikely to have a significant impact on 2D redraws as the technology behind it is focusing on gaming industry. Practically it's just an extended L4 cache size and it does not act as a dedicated memory but instead caches L1/L2 memory data. Because of Archicad's complexity and significant RAM allocation requirements, the expected overall gains in performance from AMD's 3D V-Cache technology may not be significantly different from another CPU with the same single-thread performance characteristic.
  
  In general, single-threaded applications can benefit from a larger L3 cache, but the extent of the performance boost can vary widely depending on the type of computation. For example, games may see a significant improvement with a larger L3 cache because of the need to perform different operations, such as continuous UI refreshing, object tracking, simulating weather effects and nature simulations, intersection bounces, audio, etc. As these operations fill up the L3 cache, data flows constantly, resulting in better performance. While BIM software like Archicad are extremely complex in nature, apart from the geometry, calculations are mainly matrix based computations. That's why my take is that L2/3 cache size is not a priority when choosing a CPU for Archicad. I am not arguing that there are benefits, but being a broken record again, single thread performance is the number one buying factor.
• About Intel 12th vs 13th gen CPUs,  Archicad's performance can benefit a lot from multi-cores but mostly from single core performance. Based on the mix of operations, a rough estimate of 10%-20% performance gain on 13th gen may be expected. However, this figure can vary depending on the nature of the computations being performed inside Archicad and the project size/type. Therefore, it's important to keep in mind that this estimate is a very rough estimation that cannot be measured  and should be taken with a grain of salt.
• Lower RAM timings and higher frequency can increase memory bandwidth and reduce CPU/RAM latency, which can theoretically result in better performance in applications. While it is common in the gaming community to apply tighter timings and higher RAM clock rates to try and improve FPS, the actual percentage increase in bandwidth is often disproportionate to the perceivable performance improvement on most applications.
  Given the potential risks involved, it is generally advised to keep RAM settings at their stock levels when working in Archicad. Mild overclocking won't give any notable performance boost , while aggressive settings will make the system prone to instability. You really don't want to have those notorious CLOCK_WATCHDOG_TIMEOUT, IRQL_NOT_LESS_OR_EQUAL BSOD when working on a project with a tight deadline!
• Archicad does not favor Intel over AMD because we do not implement any vendor-specific optimization. Our instruction sets when compiling Archicad, are utilized by both AMD and Intel processors. Thus, it is up to the user to decide which CPU to choose based on their specific needs, such as identifying their project's type and size, considering the price/performance ratio in their local market, and their particular budget.
  For instance the i7-13700K mentioned will be equally fast with Ryzen 9 7900X in Archicad.

Let me know your thoughts on these remarks.