A couple of months ago I have been considering getting a Xeon Max ES/QS to avoid paying too much for RAM, but ultimately decided not to go that way. I have done a bit of research that time, and maybe someone finds it useful.
Xeon Max CPUs have the usual TDP of 350W, but, in contrast to Xeon Platinum, the HBM has to be included in the 350W budget. How much does it influence the FP64 peak performance? (This is relevant if you have compute-bound tasks, otherwise you'll very likely benefit from the HBM). I found a
performance report for dual-socket Fujitsu servers with retail CPUs, and here is a brief and very loose summary.
| CPU | cores |
Base freq. [GHz] | Perf. [TFLOPS] |
| 9468 (350W) | 48 | 2.1 | 6.0 |
| 9462 (350W) | 32 | 2.7 | 5.1 |
| 9460 (350W) | 40 | 2.2 | 5.6 |
| 8468 (350W) | 48 | 2.1 | 6.5 |
| 8462Y+ (300W) | 32 | 2.8 | 5.5 |
| 8460Y+ (300W) | 40 | 2.0 | 5.4 |
| 8480+ (350W) | 56 | 2.0 | 7.4 |
A bit offtopic, I am surprised about their high number for 8480+, because the best I could get for it was 3.1 TFLOPS for a single 8480+. Unfortunately, I don't have the complete picture of what they did (LINPACK problem size, RAPL times etc.) and particularly what does their BIOS parameter "CPU Performance Boost = Aggressive" mean. Just wondering if I am leaving any performance on the table with my Gigabyte board...