|
|
Although I don't know much about your specific use-case, personally I wouldn't choose to configure a dual 6444Y system. As tonganlhy mentioned above, the EPYC 9654 processors boost quite reliably, and there is only a small difference in boost frequencies between the 9654 and 6444Y chips. Indeed, CPU frequency is quite often the bottleneck for CPU-intensive scientific computing, but I don't think you'll notice much difference between 3.7 and 4.0 GHz in terms of what you can achieve on this computer system (at most ~10% in single-threaded performance, but probably even less). However, I expect that you'll find the core-count of the 6444Y chip to be more severely limiting: the 96 cores of the EPYC 9654 vs the 16 of the Xeon Gold 6444Y gives you huge headroom for scaling your workflows. It's very often quite easy to find suitable 'scale-up' strategies to parallelise your workflows, but only having 32 cores to play with in the dual 6444Y system would be quite limiting. As code develops, the efficiency of parallelisation is continuously improving, and having lots of cores is a good way to future-proof your systems.
At the center of this, it's worth bearing in mind that there's always a tradeoff between CPU frequency and the number of cores, which can only be somewhat alleviated by spending a lot of money. If you have a big budget, then the AMD Threadripper CPUs bring both high core-counts and high frequencies. However, if CPU frequency is really the most important consideration for you, and the low core-count of the 6444Y CPU is fine for your use case, I would personally choose to build a Ryzen 7000/9000-based system (or two!) for far less money.
Lower core-count server grade CPUs tend to be less suitable for scientific computing applications. Mostly, their advantages over consumer chips (with similar core-counts and even higher frequencies) involve their connectivity (PCI-e lanes, memory lanes, SAS/SATA), and the ability to install large RAM capacities. |
|
发表于 Post on 2025-9-9 14:57:26
收藏 Add to favorites
楼主 Author