I couldn’t find it.
I predict it for 6000 to 8000 CM
I do now own any SpacemiT hardware (yet),
But badly mashing two unrelated benchmarks together I’d guess that CoreMark score might be (+/-200%) the following (scroll down to bottom, for mostly a guesstimate):
JH7110 CoreMark running on StarFive Hardware
CoreMark | 5,545.2 |
---|---|
CoreMark/MHz | 3.6968 |
sbc-bench for JH7110 boards and a SpacemiT K1 board
Device / details | Clockspeed | Kernel | Distro | 7-zip multi | 7-zip single | AES | memcpy | memset |
---|---|---|---|---|---|---|---|---|
VisionFive V2 (JH7110) | 1500 MHzZ | 5.15 | Sid riscv64 | 4180 | 1197 | 25080 | 880 | 770 |
Milk-V Mars CM (JH7110) | 1500 MHz | 5.15 | Bookworm riscv64 | 4110 | 1195 | 25070 | 930 | 830 |
BPi F3 (SpacemiT K1) | 1600 MHz | 6.1 | Bianbu Mantic riscv64 | 6750 | 978 | 27260 | 2620 | 7180 |
Single threaded the JH7110 is about 18% faster then the SpacemiT K1
Multithreaded the JH7110 (quad S76 core) is about 65% slower then the SpacemiT K1 (eight x60 core), but it is clocked about 6% slower.
Mashing these number above together (badly), should possibly only give a very very very very approximate result, instead of actually compiling the coremark source code on real SpacemiT K1 hardware with carefully choosen compiler flags to optimize the performance on a SpacemiT x60 core.
~x60 CoreMark | ~4547.1 |
---|---|
~K1 Coremark (~8 * x60) | ~36376.8 |
~x60 CoreMark/MHz | ~2.8419 |
P.S. My guesstimate is totally ignoring the vector hardware in the K1 which should bump up the result higher, which is why I give it +/-200%.
Thank you.
I too think results may be affected by vector extention.
I think that compiler flags for CoreMark like the following (may need more tweaking, but might be a good starting point) should give optimal results for a SpacemiT based board, but like I said I have no hardware to test it on “-O2 -march=rv64gcv_zba_zbb_zbc_zbs_zvl256b -lm -mabi=lp64d -DPERFORMANCE_RUN=1 -lrt”