Performance Comparison: Epyc Bergamo vs. Genoa, V-Ray Test, and Intel Xeon Sapphire Rapids

Epyc Bergamo should perform roughly 25% better than Epyc when taking into account lower clocks and smaller L3 cache Genoa. If we ignore clocks and cache size, it could achieve 33% higher performance than Genoa (because it has a third more cores). But in the V-Ray test, it achieved 41-49% higher performance:

graph-1

Such a situation is rather difficult to explain. There are several theoretical possibilities. It does not mind the lower capacity of the L3 cache and, on the contrary, benefits from the mentioned higher throughput of the L3 cache. Also, it looks like it crashes in V-Ray Genoa earlier to the TDP limit than energy optimized Bergamoso you paradoxically Bergamo can allow higher clock rates (result Genoy is the average of eight tests where the processors ran at an average of 2.64 GHz, while the result Pergamon is the average of two tests in which the tested processors ran at an average of 2.83 GHz, which is 7.2% higher). It can be speculated that during the preparation It was 4c some limiting element of the standard was removed as part of the chip design Zen 4which was limiting performance in V-Ray.

Whatever the reason, he’s doing well Bergamo excellent in V-ray as compared to classic Epyc Genoaso compared to Xeon Sapphire Rapids from Intel. One 128-core 360W Bergamo achieves 42% higher performance than two 60-core Xeons Sapphire Rapids with a total TDP of 700 watts. Which would mean 2.76 times higher energy efficiency (for Bergamo).

Preliminarily it looks like that Bergamo even V-ray does well, which is not exactly the type of load this product was designed for. On the other hand, in terms of requirements (a large number of independent parallel calculations), rendering is close to the cloud load.

Competition for Epyc Bergamo (that is, a processor specialized for the cloud) should arrive at Intel in about a year. It will be Xeon Sierra Forest with 144 Atom cores Crestmont.