NVIDIA advises game developers to optimize their code for fewer CPU threads to achieve better performance. Some CPU-bound games perform worse when core count increases due to overhead. High-end desktop systems can see up to 15% performance gains by reducing thread count. Performance drop reasons vary across titles and systems. High-end CPUs with more than 8 cores can render games faster with fewer threads.

Reducing thread count can improve boost clocks on multi-core CPUs.

• Hardware performance: Higher-core-count CPUs sometimes have lower CPU speeds. Reducing the number of threads may enable the active cores to boost their frequency.
• Hardware resource contention: Reducing the thread count can often decrease the pressure on the memory subsystem, reducing latency and enabling the CPU caches to be more efficient. This is especially true for chiplet-based architectures that do not have a unified L3 cache. Threads executing on different chiplets can cause high cache thrashing.
  • Executing threads on both logical cores of a single physical core (hyperthreading or simultaneous multi-threading) can add latency as both threads must share the physical resource (caches, instruction pipelines, and so on). If a critical thread is sharing a physical core, then its performance may decrease. Targeting physical core counts instead of logical core counts can help to reduce this on larger core count systems.
• Software resource contention: Locks and atomics can have much higher latency when accessed by many threads concurrently, adding to the memory pressure. False sharing can exacerbate this.
• OS scheduling issues: An over-subscription of threads to active cores leads to a high number of context switches which can be expensive and can put extra pressure on the CPU memory subsystem.
  • On systems with P/E cores, work is scheduled first to physical P cores, then E cores, and then hyperthreaded logical P cores. Using fewer threads than total physical cores enables background threads, such as OS threads, to execute on the E cores without disrupting critical threads running on P cores by executing on their sibling logical cores.
• Power management: Reducing the number of threads can enable more cores to be parked, saving power and potentially allowing the remaining cores to run at a higher frequency.
  • Core parking has been seen to be sensitive to high thread counts, causing issues with short bursty threads failing to trigger the heuristic to unpark cores. Having longer running, and fewer threads helps the core parking algorithms.

Hyper-threading can degrade performance by competing for resources.

Hybrid cores add complexity; limiting thread usage maintains performance.

NVIDIA's guidance for better game performance.

Windows 11 will introduce an OS-level upscaler called "DirectSR" to rival AMD FSR and NVIDIA DLSS.

NVIDIA's DLSS technology is ahead with its fourth iteration, while AMD's FidelityFX Super Resolution has seen slower adoption.

Microsoft will showcase DirectSR at the Game Developers Conference, aiming to simplify super resolution support for game developers.

DirectSR may become a key feature on Xbox consoles in the future.

Having a Windows-based upscaler that works with NPUs will benefit lightweight notebooks without dedicated GPUs.

The quality of DirectSR remains to be seen, and its implementation could be challenging.

Several key industry players such as Google and Microsoft are investing significant resources towards attaining Artificial General Intelligence (AGI). Meta has also announced its intention to lean heavily into this department.

Meta plans to acquire 350,000 units of NVIDIA’s H100 AI GPUs by the end of 2024, having access to an equivalent of 600,000 NVIDIA H100 across all their compute units. This investment will cost Meta around $10.5 billion.

Meta is currently training their next-gen model Llama 3, and Mark Zuckerberg has expressed his interest in open-sourcing it “responsibly” in the future. Their previous models, Llama and Llama 2 were open-sourced as well. They also plan to merge their two major AI research efforts, the FAIR (Fundamental AI Research) group and the GenAI department.

Mark Zuckerberg took a dig at OpenAI without naming it for being less transparent over the years.

More companies are holding their cards close to their chest with OpenAI being less transparent over the years. AGI if ever achieved by any, will likely be a valuable commodity with the source never to be shared with the public, and in the case of Meta with Mark Zuckerberg having majority control over the company, maybe even resting in the palms of a single person.

Mark Zuckerberg to The Verge

Meta has suffered huge losses totalling billions in the Metaverse and VR departments together.

NVIDIA is the supplier of arms to the ongoing battle of the AI dominance.

Regardless, if there’s one company to benefit heavily from all this, it’s the supplier of arms to the ongoing battle of AI dominance: NVIDIA. Intel, AMD, and the other suppliers are still catching up and the gap is still wide. As of writing this, NVIDIA’s stocks have reached another all-time high of $594, and show no signs of stopping.