Cisco Nexus SmartNICs Comparison

Reset

Nexus K3P-Q FPGA SmartNIC

Nexus K3P-S FPGA SmartNIC

Nexus K35-Q FPGA SmartNIC

Nexus K35-S FPGA SmartNIC

Nexus SmartNIC HPT

Nexus SmartNIC GM

Nexus V5P FPGA Application SmartNIC

Nexus V9P FPGA Application SmartNIC

Expand all

General

PCIE adapter dimensions117x68mm (4.65x2.67in)117x68mm (4.65x2.67in)150x68mm (5.9x2.67in)117x68mm (4.65x2.67in)150x68mm (5.9x2.67in)150x68mm (5.9x2.67in)168x69mm (6.6x2.7in)168x69mm (6.6x2.7in)
Ports2x QSFP28 2x SFP28 2x QSFP+ 2x SFP+ 2x SFP+2xSFP+2x QSFP282x QSFP-DD
Supported port speeds8x100M2x100M8x100M2x100M2x100M 2x100M2x100M2x100M
8x1G2x1G8x1G2x1G2x1G2x1G2x1G2x1G
8x10G2x10G8x10G2x10G2x10G2x10G8x10G16x10G
8x25G12x25G12x40G---8x25G116x25G1
2x40G1-------
Memory (on-chip):13Mb UltraRAM13Mb UltraRAMNoneNoneNoneNone133Mb UltraRAM270Mb UltraRAM
Memory (off-chip):9GB DDR4 (optional)4GB DDR4 (optional) NoneNoneNoneNone9GB DDR4 144Mb QDR-IV 9GB DDR4

Latency

2 frame trigger to response (min) @ 64B596 ns568 ns645ns623ns 630ns623ns650ns656ns
2full frame loopback (median) @ 64B714 ns696 ns800 ns780 ns780 ns790 ns810 ns816 ns
2full frame loopback (median) @ 256B927 ns897 ns1020 ns1003ns1003ns1012ns1030 ns1027 ns
3UDP loopback (median) @ 14B828 ns810 ns950 ns880 ns880 ns890 ns960 ns959 ns
3UDP loopback (median) @ 256B1110 ns1100 ns1220 ns1200 ns1200 ns1209 ns1230 ns1226 ns
3TCP loopback (median) @ 14B860 ns850 ns950 ns930 ns930 ns930 ns960 ns958 ns
3TCP loopback (median) @ 256B1130 ns1100 ns1220 ns1200 ns1200 ns1215 ns1230 ns1227 ns

Timing

Timestamp resolution: 4ns4ns6.2ns6.2ns250ps4ns6.2ns6.2ns
Time synchronization: Hardware assisted PTP, PPS in/outHardware assisted PTP, PPS in/outHardware assisted PTP, PPS in/outHardware assisted PTP, PPS in/outHardware assisted PTP, PPS in/outHardware assisted PTP, PPS in/out IEEE1588 PTP Grandmaster (port 0 only)Hardware assisted PTP, PPS in/outHardware assisted PTP, PPS in/out

Extra Features with FDK-PE

4Min packet trigger latency 34ns 34ns 40ns 40ns --34ns 34ns
Ultra-low latency TCP transmission engine------Yes-

Notes

1 40G and 25G Ethernet speeds supported with future firmware update.1 25G Ethernet speeds supported with future firmware update.----1 25G Ethernet speeds supported with future firmware update.1 25G Ethernet speeds supported with future firmware update.
2 Latencies measured from wire-userspace-wire via the libexanic SmartNIC library, on a 3.5Ghz Intel Ivy Bridge processor.2Latencies measured from wire-userspace-wire via the libexanic SmartNIC library, on a 3.5Ghz Intel Ivy Bridge processor.2Latencies measured from wire-userspace-wire via the libexanic SmartNIC library, on a 3.5Ghz Intel Ivy Bridge processor.2Latencies measured from wire-userspace-wire via the libexanic SmartNIC library, on a 3.5Ghz Intel Ivy Bridge processor. 2Latencies measured from wire-userspace-wire via the libexanic SmartNIC library, on a 3.5Ghz Intel Ivy Bridge processor.2Latencies measured from wire-userspace-wire via the libexanic SmartNIC library, on a 3.5Ghz Intel Ivy Bridge processor.2Latencies measured from wire-userspace-wire via the libexanic SmartNIC library, on a 3.5Ghz Intel Ivy Bridge processor.2Latencies measured from wire-userspace-wire via the libexanic SmartNIC library, on a 3.5Ghz Intel Ivy Bridge processor.
3Latencies are median half round trip time latencies for the sockperf benchmark using the exasock socket acceleration library. 3Latencies are median half round trip time latencies for the sockperf benchmark using the exasock socket acceleration library. 3Latencies are median half round trip time latencies for the sockperf benchmark using the exasock socket acceleration library. 3Latencies are median half round trip time latencies for the sockperf benchmark using the exasock socket acceleration library. 3Latencies are median half round trip time latencies for the sockperf benchmark using the exasock socket acceleration library. 3Latencies are median half round trip time latencies for the sockperf benchmark using the exasock socket acceleration library. 3Latencies are median half round trip time latencies for the sockperf benchmark using the exasock socket acceleration library. 3Latencies are median half round trip time latencies for the sockperf benchmark using the exasock socket acceleration library.
4RX + TX time for PMA + PCS + MAC + CDC performing packet trigger based on first data word (i.e. dst MAC) measured with HPT (250ps) precision.4RX + TX time for PMA + PCS + MAC + CDC performing packet trigger based on first data word (i.e. dst MAC) measured with HPT (250ps) precision.4RX + TX time for PMA + PCS + MAC + CDC performing packet trigger based on first data word (i.e. dst MAC) measured with HPT (250ps) precision.4RX + TX time for PMA + PCS + MAC + CDC performing packet trigger based on first data word (i.e. dst MAC) measured with HPT (250ps) precision. 4RX + TX time for PMA + PCS + MAC + CDC performing packet trigger based on first data word (i.e. dst MAC) measured with HPT (250ps) precision.4RX + TX time for PMA + PCS + MAC + CDC performing packet trigger based on first data word (i.e. dst MAC) measured with HPT (250ps) precision.