Hi Dirk,

Ich bin auch immer noch staending an herumsuchen und spekulieren. Ich bin kein Techniker daher sind Technik vergleiche nicht so einfach fuer mich.

Angenommen das der SenseRobot etwa gleichstark ist mit Stockfish. Obwohl ich finde die beide spielen voellig anders from stil her.

Ich machte neulich diesen search (auf Englisch:

1) how much faster is a busy server than an i7 4 cpu processor

A modern "busy" server processor is fundamentally designed for massive multitasking and sustained loads, making it roughly 5 to 10 times faster (or more) in multi-threaded tasks than a 4th Generation i7 (e.g., i7-4770K). While the i7-4th gen is still a capable desktop CPU for light tasks, it struggles with modern, high-load, multi-threaded workloads, whereas server processors can offer 32, 64, or more cores compared to the 4 cores on that i7.
Key Factors Driving the Speed Difference:
Core Count & Threads: The 4th Gen i7 has 4 cores and 8 threads. Modern servers (like AMD EPYC) can have up to 192 cores (384 threads), allowing them to manage dozens of virtual machines simultaneously, unlike the single-user focus of a desktop i7.

Performance Under Load: A 4th Gen i7 is a consumer processor optimized for snappy, single-threaded performance (like gaming), but it can overheat and throttle under 24/7 heavy workloads. Server CPUs (e.g., Xeon) are engineered for 24/7 continuous usage under heavy, sustained load, often performing 60% or more faster in tasks like video encoding or code compilation.
Cache and Memory Bandwidth: Server CPUs typically feature much larger L3 caches and support more advanced, higher-bandwidth, ECC memory. This allows them to handle larger data sets and more concurrent, high-demand applications, such as large database systems or complex virtualization, that would cripple a 4th Gen i7.

2) how much larger is the l3 cache on a server compared to an i7-8750H

The L3 cache on a modern server CPU is typically 3x to over 80x larger than the 9 MB to 12 MB L3 cache found on the Intel Core i7-8750H.
While the i7-8750H (a 2018 mobile processor) features 9 MB of L3 cache, enterprise-grade server processors are designed for massive workloads, often featuring hundreds of megabytes of L3 cache to manage data for high core counts.
Comparison Breakdown
i7-8750H L3 Cache: 9 MB (standard), sometimes listed up to 12 MB in similar Coffee Lake models.
Mid-Range Server CPU (e.g., modern Xeon): Often 32 MB to 64 MB (roughly 3x–7x larger).
High-End/AMD EPYC Server CPU: Up to 768 MB or even 1152 MB (using 3D V-Cache technology).
Key Takeaways
Capacity Difference: A modern high-performance server (like AMD EPYC 7773X) has a 768 MB L3 cache, which is over 85 times larger than the 9 MB on the i7-8750H.
Why Server Cache is Larger: Server processors, such as AMD EPYC or Intel Xeon, often utilize multiple chiplets (CCDs) that each have their own large L3 cache, leading to massive aggregated capacities for tasks like database management, virtualization, and complex simulation.
Performance Impact: The massive increase in L3 cache for servers is necessary to reduce the need for the CPU to access significantly slower system RAM, allowing for much faster processing of large datasets.

Ich habe 40/15 gegen 4 CPUs gespielt, wobei Stockfish 17.1 etwas besser war als Level 25 (2900 ELO).

Das oben Genannte entspricht in der Realität mehr als 30 Sekunden pro Zug, da so viele Züge schnell hintereinander ausgeführt werden.

Realistisch betrachtet, könnte man mit NNUE die von SenseRobot angestrebte Reaktionszeit von 1 Sekunde auf einem Server und dessen Ergebnisse recht gut erreichen. Die Unbekannte ist jedoch, wie viele Server man benötigt, wenn mehrere Spieler gleichzeitig spielen.

Soweit ich weiß, wurden weltweit über 100.000 Einheiten verkauft. Wenn also 1 Prozent der Spieler auf dem Server spielt, sind das 1000 Personen pro Tag, und bei einer Spielzeit von 2 Stunden sind das 83 pro Stunde. Könnte man die Nachfrage vielleicht mit 5–10 Servern decken?

Ich fragte auch dieses:

3) at level 25 is your senserobot ai uniquely adapting to your play style

Yes, at level 25 (and across its various difficulty settings), the SenseRobot AI is designed to adapt to your play style, offering a personalized experience that mimics a human opponent rather than just playing at maximum engine strength.
Here is how the SenseRobot AI adapts at higher levels:
Adaptive Difficulty & Style: SenseRobot features 25 to 26 levels of adaptive AI, ranging from beginner to expert (up to 2900-3200 ELO), which adjusts to your skill level and play style.
Human-like Play: Unlike pure chess engines (like Stockfish) designed for maximum strength, the SenseRobot AI is specifically tuned to provide a realistic, "human-like" playing experience.
Personalized Training: The AI system uses data from your games to offer over 1,200 tailored, structured lessons and exercises.
Progressive Learning: The AI adapts as you improve, providing, for instance, endgame practice with 145 classic scenarios and real-time voice coaching.
Unique “Apex Duel” Mode: For advanced players, the "Apex Duel" (3200 ELO) mode allows you to test your skills against a high-level AI designed to challenge even advanced, competitive players.

Angesichts dessen frage ich mich immer wieder, ob ich durch die Ergebnisse dieser hochkarätigen Spiele, die ich gespielt habe, meinen Roboter darauf trainiert habe, sich meinem Spielstil anzupassen. Wenn ja, ist es dann unwahrscheinlich, dass unsere Roboter jemals zu 100 % identisch sein werden, selbst wenn wir alle dasselbe Programm verwenden?

Daher wäre es hilfreich, wenn einige technische Experten bei der Beantwortung dieser Fragen helfen könnten. Ich halte es aber für durchaus denkbar, dass eine Form von NNUE als KI verwendet wird.

Gruss

Nick