Checklist - xilinx-u250-gen3x16-xdma-4.1

Package information

xbtest information

Release Number

3522814

Description (what’s new)

Initial Release

Platform information

Development platform (XSA)

xilinx-u250-gen3x16-xdma-4.1-202210-1-dev-1-3512975

Deployment platform (shell, CMC, SC)

xilinx-cmc-u200-u250-1.2.23-3395909

xilinx-sc-fw-u200-u250-4.6.20-1.28f0c61

xilinx-u250-gen3x16-base-4-3494623

xilinx-u250-gen3x16-xdma-shell-4.1-3494623

xilinx-u250-gen3x16-xdma-validate-4.1-3512975

System information

System information

OS

CentOS / 7.8.2003

Architecture

x86_64

Server type

PowerEdge R740

AUX cable

Yes

Card S/N

21330497P03F

XRT version

2.13.449

Measured download time

8.7 seconds

Requirement - Platform high level features

Platform and card requirements

Platform and card requirements

FPGA part

xcu250-figd2104-2L-e

PCIe speed and width

gen3x16

DMA type

xdma

DDR: speed, size and quantity

4 DDR: 72 bits @ 2400 MT/s; 16GB

HBM: size, quantity and power rail

N/A

Host memory support

Yes

P2P support

Yes

GTs: type, quad, SLR location

GT[0]: GTY, Quad_X1Y11 (SLR2)

GT[1]: GTY, Quad_X1Y10 (SLR2)

Both using refclk_b.

Power rails requirements

Power rail

Card limit (W)

Note

12v_pex

69W

Permanent Connection

12v_aux

  • 78W (AUX 4p)

  • 156W (AUX 6p)

User Selectable

vccint

N/A

Fatal Limit/Card Shutdown @53.5A per phase (6 phases with full AUX power)

Calibration - Power CU - Results

Description

Results

Power, temperature & toggle rate graph

<TBC>

Maximum toggle rate reached (%)

76

Power consumption balance achieved:

  • Yes/no and justification (why are some rails not close to their critical current limit?)

No For the 12V_PEX to draw more power, multiple GTs with QSFP heater modules. Due to FPGA routing and resources, this is not possible.

12v_pex current (A)

Measurement

4.345

Critical limit

5.75

12v_aux current (A)

Measurement

12.828

Critical limit

13.0

vccint current (A)

Measurement

172.5

Critical limit

N/A

Calibration - Memory CU power

Memory

Description

simultaneous_wr_rd

only_rd

only_wr

DDR

Is there enough power for memory?

Yes. The U250 card is capable of supplying full power to the DDR4 memories.

Power graph

<TBC>
<TBC>
<TBC>

BW graph

<TBC>
<TBC>
<TBC>

Latency graphs

<TBC>
<TBC>
<TBC>
<TBC>

Calibration - Memory bandwidth and latency

Note

  • N/A: not applicable

  • 1

    The outstanding write/read is the maximum number of outstanding reads just before the read pipeline is full, giving the best BW and latency results.

Memory

Description

simultaneous_wr_rd

only_rd

only_wr

DDR

QoS / power reduction rate

N/A

<TBC>

N/A

N/A

Best write BW (MBps)

8487

N/A

15182

Best write latency (ns)

1614

N/A

802

Best read BW (MBps)

8487

17178

N/A

Best read latency (ns)

1608

451

N/A

Outstanding write/read 1

Write = 4 / read = 4

3

4

BW/latency graphs (do one for read and one for write)

<TBC>
<TBC>
<TBC>
<TBC>

Checklist questionnaire

Runtime

Question

Answer / justification / problem tracker reference number

Have you updated the default download time?

  • FYI:

    • u250 xclbin (90Mb) downloads in 3.9 sec (PCIe 3x16).

    • HBM requires calibration which may take time.

Yes

"download_time": 20

Sensors

Question

Answer / justification / problem tracker reference number

Do you need to monitor more mechanical sources than the default one (fpga_fan_1)?

No

Do you need to monitor more thermal sources than the default one (fpga0)?

No

Do you need to monitor more electrical sources than the default ones (power_consumption, 12v_pex, 12v_aux, 3v3_pex and vccint)?

No

GT

Question

Answer / justification / problem tracker reference number

Answer / justification / problem tracker reference number

GT

GT[0]

GT[1]

Have you updated GT default settings to maximize signal integrity?

  • Run all GT pre-canned tests.

"0": {
    "name": "top",
    "transceiver_settings": {
        "module": {
            "tx_differential_swing_control": 0,
            "tx_pre_emphasis": 0,
            "tx_post_emphasis": 0,
            "rx_equaliser": "DFE"
        },
        "cable": {
            "tx_differential_swing_control": 0,
            "tx_pre_emphasis": 0,
            "tx_post_emphasis": 0,
            "rx_equaliser": "DFE"
        }
    }
}
"1": {
    "name": "top",
    "transceiver_settings": {
        "module": {
            "tx_differential_swing_control": 0,
            "tx_pre_emphasis": 0,
            "tx_post_emphasis": 0,
            "rx_equaliser": "DFE"
        },
        "cable": {
            "tx_differential_swing_control": 0,
            "tx_pre_emphasis": 0,
            "tx_post_emphasis": 0,
            "rx_equaliser": "DFE"
        }
    }
}

Memory

Question

Answer / justification / problem tracker reference number

Memory

DDR

HOST

Have you defined specific DMA bandwidths (dma_bw)?

  • This not applicable for host memory.

"dma_bw": {
    "write": {
        "low": 7500
    }
}

It’s possible to increase DMA write performance by using a single DMA channel when concurrent buffers target a single DDR4 channel.

N/A

Have you defined optimum DMA settings (buffer_size and total_size) or are default settings enough?

  • Run dma pre-canned test.

No

N/A

Have you defined specific P2P CARD bandwidths (p2p_card_bw)?

  • This not applicable for host memory.

No

N/A

Have you defined optimum P2P CARD settings (buffer_size and total_size) or are default settings enough?

  • Run p2p_card pre-canned test.

No

N/A

Have you defined specific P2P NVME bandwidths (p2p_nvme_bw)?

  • This not applicable for host memory.

No

N/A

Have you defined optimum P2P NVME settings (buffer_size and total_size) or are default settings enough?

  • Run p2p_nvme pre-canned test.

No

N/A

Have you reduced the CU rate to avoid over-power (cu_rate)?

No

No

Have you defined cu_rate for simultaneous write/read bandwidth QoS?

No

"cu_rate": {
    "only_wr": {
        "write": {
            "nominal": 50
        }
    }
}

Have you defined cu_bw?

  • For each test mode: only_wr, only_rd and simul_wr_rd

For host memory these values are depending on the PCIe speed.

"cu_bw": {
    "only_wr": {
        "write": {
            "average": 15182
        }
    },
    "only_rd": {
        "read": {
            "average": 17178
        }
    },
    "alt_wr_rd": {
        "write": {
            "average": 8500
        },
        "read": {
            "average": 8500
        }
    },
    "simul_wr_rd": {
        "write": {
            "average": 8487
        },
        "read": {
            "average": 8487
        }
    }
}
"cu_bw": {
    "only_wr": {
        "write": {
            "high": 16000,
            "low": 4000
        }
    },
    "only_rd": {
        "read": {
            "high": 16000,
            "low": 4000
        }
    },
    "simul_wr_rd": {
        "write": {
            "high": 16000,
            "low": 4000
        },
        "read": {
            "high": 16000,
            "low": 4000
        }
    }
}

Have you defined cu_latency?

  • For each test mode: only_wr, only_rd and simul_wr_rd

"cu_latency": {
    "only_rd": {
        "read": {
            "high": 586,
            "low": 315
        }
    },
    "simul_wr_rd": {
        "write": {
            "high": 2098,
            "low": 1129
        },
        "read": {
            "high": 2098,
            "low": 1129
        }
    },
    "only_wr": {
        "write": {
            "high": 1042,
            "low": 561
        }
    }
}
"cu_latency": {
    "only_wr": {
        "write": {
            "low": 1,
            "high": 6000
        }
    },
    "only_rd": {
        "read": {
            "low": 1,
            "high": 6000
        }
    },
    "simul_wr_rd": {
        "write": {
            "low": 1,
            "high": 6000
        },
        "read": {
            "low": 1,
            "high": 6000
        }
    }
}

Have you defined cu_outstanding?

  • For each test mode: only_wr, only_rd and simul_wr_rd

"cu_outstanding": {
    "only_wr": {
        "write": {
            "nominal": 4
        }
    },
    "only_rd": {
        "read": {
            "nominal": 3
        }
    },
    "simul_wr_rd": {
        "write": {
            "nominal": 4
        },
        "read": {
            "nominal": 4
        }
    }
}

No

Have you defined cu_burst_size?

  • For each test mode: only_wr, only_rd and simul_wr_rd

No

No

Pre-canned tests

Question

Answer / justification / problem tracker reference number

Have you tested all provided pre-canned tests?

  • If not: list the skipped one and the reason?

Yes

Have you modified the pre-canned test template?

  • If yes:

    • Why?

    • Was it not possible to make a platform definition update (and thus, available for any test as default)?

    • Update next section with your modifications.

No

Have you saved all pre-canned results?

  • Update package information section of your checklist.

Yes

Issues tracking

Question

Answer / justification / problem tracker reference number

Did you modify the platform definition limits and range after running the pre-canned tests?

  • If yes:

    • List the problem tracker reference number

No

Pre-canned tests results

Pre-canned test modifications

Test name

Modification from original template

Justifications

/

No modification

/

dma pre-canned test

Memory

Write BW (MBps)

Read BW (MBps)

DDR

8089

10393

p2p_card pre-canned test

Memory

Write BW (MBps)

Read BW (MBps)

DDR

6258

4762

p2p_card test was ran on R7525 server

p2p_nvme pre-canned test

Memory

Card mode

Write BW (MBps)

Read BW (MBps)

DDR

source

2451

5783

DDR

target

5972

2509

p2p_nvme test was ran on R7525 server

memory pre-canned test

Memory

Test mode

Write BW (MBps)

Read BW (MBps)

Write Latency (ns)

Read Latency (ns)

DDR

alternate_wr_rd

7853

8292

802

452

only_wr

15182

N/A

802

N/A

only_rd

N/A

17178

N/A

451

simultaneous_wr_rd

8487

8487

1614

1608

memory_host pre-canned test

Memory

Test mode

Write BW (MBps)

Read BW (MBps)

Write Latency (ns)

Read Latency (ns)

HOST

alternate_wr_rd

5232

5217

881

3835

only_wr

9155

N/A

881

N/A

only_rd

N/A

12177

N/A

3835

simultaneous_wr_rd

9155

11817

882

3952