4 ms·
Don't forget the Sophon SG2380 using 16 SiFive P670 cores, with test chips due in probably September or so. That's equivalent to Arm A78, so a step or two ahead
by brucehoult 3y ago
Don't forget the Sophon SG2380 using 16 SiFive P670 cores, with test chips due in probably September or so. That's equivalent to Arm A78, so a step or two ahead of the RK3588 and Pi 5, as well as have more cores.
No Chinese chip using the SiFive P870 (currently their highest end core, with a SPECInt2k6 of around 18/GHz) has been announced, but that doesn't mean a license agreement doesn't already exist. That's up in Arm Cortex X3 territory.
- camel-cdr 3y agoNot that it would be worth sactioning the P670, since XiangShan already has higher SPECint2006 scores.
- brucehoult 3y agoAiming for, maybe, but I don't think it's achieved, and certainly not in chips someone can buy. The 2nd gen Nanhu, taped out in November so maybe sampling about now, is supposed to be in 2.0 GHz A76 / P550 range. Maybe 3rd gen might be around P670, but that's some years away from being in the market.
- camel-cdr 3y agoFor a v3 tapeout there is certainly still a lot of work to do. But the latest simpoint SPECint2006 rtl simulation [0], reported 44.08@3GHz (14.7/GHz), which is already faster than the P670 (12/GHz). Most of the v3 work seems to now focuses to the rvv implementation, which wasn't used in the SPEC run. The HPCA'24 tutorial reported a similar number, but also a SPECint2006 peak score of 49.96@3GHz. [1] [0] https://www.zhihu.com/people/openxiangshan?utm_id=0 https://www.zhihu.com/people/openxiangshan?utm_id=0 (click on the first post and scroll down till you see the tabel) [1] https://raw.githubusercontent.com/OpenXiangShan/XiangShan-doc/main/tutorial/20240302-HPCA24-1-Introduction-XiangShan.pdf https://raw.githubusercontent.com/OpenXiangShan/XiangShan-do...