[DGX Spark] ConnectX-7 firmware download URL is inaccessible

Accelerated Computing Aerial Forum
1

Hi NVIDIA team,

I cannot download the ConnectX-7 firmware specified in the DGX Spark installation guide.

  • PSID: NVD0000000087
  • Current firmware: 28.45.4028
  • Required firmware: 28.47.1088

Download URL:

https://nbu-nfs.gtm.nvidia.com/auto/host_fw_release/fw-4129/fw-4129-rel-28_47_1088-build-001/etc/bin/signed/fw-ConnectX7-rel-28_47_1088-cx7_P4242_HORIZON_PK_Ax-UEFI-14.40.10-FlexBoot-3.8.201.signed.bin

wget fails with:

OpenSSL: error:0A000126:SSL routines::unexpected eof while reading
Unable to establish SSL connection.

Does this URL require NVIDIA internal network access? Could you provide a working public download link for this firmware?

Thank you.

2

Thank you for pointing it out. We are discussing this mismatch internally. I’ll post if a decision is made.

Thank you.

3

Can we just skip this version and use, e.g., 28.45.4028?
Thank you.

4

I think I answered myself, answer is probably not.

[2026-10-09 09:11:11:916399][2573650112][DOCA][INF][CORE][doca_log.cpp:900] DOCA version 3.3.0109
EAL: Probe PCI driver: mlx5_pci (15b3:1021) device: 0000:01:00.0 (socket -1)
09:11:11.916581 CON phy_init 0 [DRV.FH] Dynamic beam id start: 16528, end: 32767, num of beam ids per slots: 4384, number of consecutive slots with unique of beam ids: 3 
PCAP_LOGGER Initialized
09:11:12.429276 CON phy_init 0 [FH.PEER] Doca sem created!: sem_in gpu_addr 0xe1efda436000 cpu_addr 0x38c5d14ebd0
09:11:12.435245 FATAL exit: Thread [phy_drv_init] on core 10 file /opt/nvidia/cuBB/cuPHY-CP/cuphycontroller/examples/cuphycontroller_scf.cpp line 532: additional info: NULL
Stack trace (most recent call last):
#7    Object "/usr/lib/aarch64-linux-gnu/ld-linux-aarch64.so.1", at 0xffffffffffffffff, in 
#6    Object "/opt/nvidia/cuBB/build.aarch64/cuPHY-CP/cuphycontroller/examples/cuphycontroller_scf", at 0xbae5eb2557ef, in _start
#5    Object "/usr/lib/aarch64-linux-gnu/libc.so.6", at 0xe1f2d87c8597, in __libc_start_main
#4    Object "/usr/lib/aarch64-linux-gnu/libc.so.6", at 0xe1f2d87c84c3, in 
#3    Object "/opt/nvidia/cuBB/build.aarch64/cuPHY-CP/cuphycontroller/examples/cuphycontroller_scf", at 0xbae5eb248fc7, in main
#2    Object "/opt/nvidia/cuBB/build.aarch64/cuPHY/nvlog/libnvlog.so", at 0xe1f2d8d03387, in exit_handler::test_trigger_exit(char const*, int, char const*)
#1    Object "/opt/nvidia/cuBB/build.aarch64/cuPHY-CP/cuphycontroller/examples/cuphycontroller_scf", at 0xbae5eb25610b, in nvlog_exit_handler()
#0    Source "/opt/nvidia/cuBB/cuPHY-CP/cuphydriver/src/common/cuphydriver_api.cpp", line 4396, in l1_exit_handler
       4393:         NVLOGC_FMT(TAG,"L1 exit handler: step 3 (CUDA_ENABLE_COREDUMP_ON_EXCEPTION unset)");
       4394:         asm volatile("" : : : "memory"); //Memory barrier inserted here to prevent compiler from reordering
       4395:         MemtraceDisableScope md;
      >4396:         AERIAL_PRINT_BACKTRACE(trace_cnt_max);
       4397:         return;
       4398:     }
       4399: }
09:11:12.435143 ERR phy_init 0 [AERIAL_ORAN_FH_EVENT] [FH.LIB] Exception! Failed to create group 1 RX flow rule with mirroring for Peer e8:c7:4f:25:7f:f3 on NIC 0000:01:00.0: invalid flow action
09:11:12.435147 ERR phy_init 0 [AERIAL_ORAN_FH_EVENT] [DRV.FH] aerial_fh::add_peer error
09:11:12.435208 ERR phy_init 0 [AERIAL_CUPHYDRV_API_EVENT] [DRV.EXCP] /opt/nvidia/cuBB/cuPHY-CP/cuphydriver/src/common/cuphydriver_api.cpp l1_init line 181 exception: Add Peer for cell returned error
09:11:12.435212 ERR phy_init 0 [AERIAL_CUPHYDRV_API_EVENT] [CTL.DRV] Error l1_init
09:11:12.435213 ERR phy_init 0 [AERIAL_CUPHYDRV_API_EVENT] [CTL.SCF] pc_init_phydriver error -1
09:11:12.435236 CON 64 0 [NVLOG.EXIT_HANDLER] [exit_watchdog] thread received exiting notification, wait for 5 seconds to force exit
09:11:12.435256 FAT phy_init 0 [AERIAL_SYSTEM_API_EVENT] [NVLOG.EXIT_HANDLER] FATAL exit: Thread [phy_drv_init] on core 10 file /opt/nvidia/cuBB/cuPHY-CP/cuphycontroller/examples/cuphycontroller_scf.cpp line 532: additional info: NULL
09:11:12.435267 CON phy_init 0 [CTL.SCF] nvlog_exit_handler: nvlog exit handler called in thread [phy_drv_init]
09:11:12.435278 CON phy_init 0 [DRV.API] Triggering L1 exit handler
09:11:12.435315 CON phy_init 0 [DRV.API] L1 exit handler: end of step 1
09:11:12.435322 CON phy_init 0 [DRV.API] L1 exit handler: end of step 2
09:11:12.435323 CON phy_init 0 [DRV.API] L1 exit handler: step 3 (CUDA_ENABLE_COREDUMP_ON_EXCEPTION unset)
09:11:12.514551 CON phy_init 0 [NVLOG.EXIT_HANDLER] test_trigger_exit: exit_cb called
09:11:17.435303 CON 64 0 [NVLOG.EXIT_HANDLER] [exit_watchdog] thread closed FMT log after and force exit
09:11:17.435314 CON 64 0 [NVLOG.CPP] nvlog_fmtlog_close: closing FMT log from [exit_watchdog] thread on core 10, last file path is /var/log/aerial/phy.log
Exiting bg_fmtlog_collector - log queue ever was full: 0
nvlog_fmtlog_close: FMT log closed
09:11:17.437837 EXIT from watchdog after 5 seconds waiting