Tracked topic
… The CPU and GPU performance is truly a class above standard low power laptop chips, and is even able to compete with larger systems boasting dedicated graphics. CPU performance is especially impressive with a comparable showing to the desktop Zen 5 CPUs. …
… Those differences make Geekbench 6 an interesting suite to evaluate alongside SPEC CPU2017. I looked at SPEC CPU2017 on Chips and Cheese several years ago. …
… GPU shader programs are typically small and simple compared to CPU programs, so I suspect most code fetches will hit in the 32 KB L0 instruction caches. …
… However, it’s still respectable at north of 200 GB/s and likely to be adequate. GB10’s two CPU clusters have asymmetrical external bandwidth, in addition to L3 capacity. …
… It’ll be especially rare in the graphics workloads I’m targeting here because Strix Halo’s second CCD tends to be parked in Windows. CPU-side traffic in general presents a more significant issue. …
… Speaking of HPE’s GX5000 platform, it has 3 initial compute blade configurations: GX250: The GX250 blade has 8 Venice CPUs with up to 40 blades per rack for a total of up to 320 Venice CPUs per GX5000 rack GX350a: The GX350a blade has 1 Venice CPU and 4 MI430X Accelerators with up to 28 blades per … …
… Performance: SPEC CPU2017 Higher memory latency could lead to lower performance, especially in single threaded workloads. But the EPYC 9575F does surprisingly well in SPEC CPU2017. …
… The term “bus lock” dates back to early multiprocessor systems, which placed multiple CPUs on a shared bus. Each CPU had its bus pins connected to the same motherboard traces, which connected them to the chipset. …
… This is a massive increase from the approximately 400mm 2 that the prior EPYC CPUs dedicated for their IO dies. …
… Performance: SPEC CPU2017 Cortex X925 turns in an excellent performance in SPEC CPU2017’s integer suite. …