Greetings to all,
I’m trying to run various LLM inference engines (vLLM, TensorRT-LLM) on two RTX 5090 GPUs, specifically using tensor parallelism with a size of 2. Inference on a single GPU works fine, but I’m having trouble with two GPUs.
Could someone confirm whether peer-to-peer (P2P) communication is supported by the RTX 5090?
Thanks in advance!
My setup:
PyTorch version: 2.7.0a0+ecf3bae40a.nv25.02
Is debug build: False
CUDA used to build PyTorch: 12.8
ROCM used to build PyTorch: N/A
OS: Ubuntu 24.04.1 LTS (x86_64)
GCC version: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version: 18.1.3 (1ubuntu1)
CMake version: version 3.31.6
Libc version: glibc-2.39
Python version: 3.12.3 (main, Feb 4 2025, 14:48:35) [GCC 13.3.0] (64-bit runtime)
Python platform: Linux-5.14.0-503.29.1.el9_5.x86_64-x86_64-with-glibc2.39
Is CUDA available: True
CUDA runtime version: 12.8.61
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA GeForce RTX 5090
GPU 1: NVIDIA GeForce RTX 5090
Nvidia driver version: 570.124.06
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.7.1
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 52 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 48
On-line CPU(s) list: 0-47
Vendor ID: AuthenticAMD
Model name: AMD Ryzen Threadripper 7960X 24-Cores
CPU family: 25
Model: 24
Thread(s) per core: 2
Core(s) per socket: 24
Socket(s): 1
Stepping: 1
CPU(s) scaling MHz: 17%
CPU max MHz: 5665.0000
CPU min MHz: 545.0000
BogoMIPS: 8387.55
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid overflow_recov succor smca fsrm flush_l1d debug_swap
Virtualization: AMD-V
L1d cache: 768 KiB (24 instances)
L1i cache: 768 KiB (24 instances)
L2 cache: 24 MiB (24 instances)
L3 cache: 128 MiB (4 instances)
NUMA node(s): 1
NUMA node0 CPU(s): 0-47
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Mitigation; Safe RET
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] nvidia-ml-py==12.570.86
[pip3] pytorch-triton==3.1.0+cf34004b8.internal
[pip3] pyzmq==26.2.1
[pip3] torch==2.7.0a0+ecf3bae40a.nv25.2
[pip3] torchvision==0.22.0a0+ecf3bae40a.nv25.2
[pip3] transformers==4.49.0
[pip3] triton2==3.2.0
[pip3] tritonfrontend==2.55.0
[pip3] tritonserver==0.0.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.7.4.dev254+ged6ea065.d20250311
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0 GPU1 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X NODE 0-47 0 N/A
GPU1 NODE X 0-47 0 N/A
Legend:
X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks
NVIDIA_VISIBLE_DEVICES=all
CUBLAS_VERSION=12.8.3.14
NCCL_P2P_DISABLE=1
NVIDIA_REQUIRE_CUDA=cuda>=9.0
CUDA_CACHE_DISABLE=1
NCCL_VERSION=2.25.1
NVIDIA_DRIVER_CAPABILITIES=compute,utility,video
NVIDIA_PRODUCT_NAME=Triton Server
CUDA_VERSION=12.8.0.038
CUDNN_FRONTEND_VERSION=1.10.0
CUDNN_VERSION=9.7.1.26
NVIDIA_TRITON_SERVER_VERSION=25.02
LD_LIBRARY_PATH=/usr/local/lib/python3.12/dist-packages/cv2/../../lib64:/usr/local/lib:/usr/local/lib/python3.12/dist-packages/torch/lib:/usr/local/cuda/compat/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
NVIDIA_BUILD_ID=143749533
CUDA_DRIVER_VERSION=570.86.10
NVIDIA_REQUIRE_JETPACK_HOST_MOUNTS=
VLLM_USE_V1=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_root
NCCL_CUMEM_ENABLE=0
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY
As far as I know, NVIDIA removed p2p capability for consumer RTX 40xx gpus, and it was / is possible to use a modified driver for p2p. I would assume that it is also disabled for RTX 50xx.
You did see the following, correct?
@njuffa Yeap, that didn’t help. But it did allow the inference to start on a different machine with other GPUs.
Other information
GDS release version: 1.13.1.3
libcufile version: 2.12
Platform: x86_64
============
ENVIRONMENT:
============
=====================
DRIVER CONFIGURATION:
=====================
NVMe P2PDMA : Unsupported
NVMe : Unsupported
NVMeOF : Unsupported
SCSI : Unsupported
ScaleFlux CSD : Unsupported
NVMesh : Unsupported
DDN EXAScaler : Unsupported
IBM Spectrum Scale : Unsupported
NFS : Unsupported
BeeGFS : Unsupported
WekaFS : Unsupported
Userspace RDMA : Unsupported
--Mellanox PeerDirect : Disabled
--rdma library : Not Loaded (libcufile_rdma.so)
--rdma devices : Not configured
--rdma_device_status : Up: 0 Down: 0
=====================
CUFILE CONFIGURATION:
=====================
properties.use_pci_p2pdma : false
properties.use_compat_mode : true
properties.force_compat_mode : false
properties.gds_rdma_write_support : true
properties.use_poll_mode : false
properties.poll_mode_max_size_kb : 4
properties.max_batch_io_size : 128
properties.max_batch_io_timeout_msecs : 5
properties.max_direct_io_size_kb : 16384
properties.max_device_cache_size_kb : 131072
properties.max_device_pinned_mem_size_kb : 33554432
properties.posix_pool_slab_size_kb : 4 1024 16384
properties.posix_pool_slab_count : 128 64 64
properties.rdma_peer_affinity_policy : RoundRobin
properties.rdma_dynamic_routing : 0
fs.generic.posix_unaligned_writes : false
fs.lustre.posix_gds_min_kb: 0
fs.beegfs.posix_gds_min_kb: 0
fs.weka.rdma_write_support: false
fs.gpfs.gds_write_support: false
fs.gpfs.gds_async_support: true
profile.nvtx : false
profile.cufile_stats : 0
miscellaneous.api_check_aggressive : false
execution.max_io_threads : 4
execution.max_io_queue_depth : 128
execution.parallel_io : true
execution.min_io_threshold_size_kb : 8192
execution.max_request_parallelism : 4
properties.force_odirect_mode : false
properties.prefer_iouring : false
=========
GPU INFO:
=========
GPU index 0 NVIDIA GeForce RTX 5090 bar:1 bar size (MiB):32768 supports GDS, IOMMU State: Disabled
GPU index 1 NVIDIA GeForce RTX 5090 bar:1 bar size (MiB):32768 supports GDS, IOMMU State: Disabled
==============
PLATFORM INFO:
==============
IOMMU: disabled
Nvidia Driver Info Status: Supported(Nvidia Open Driver Installed)
Cuda Driver Version Installed: 12080
Platform: TRX50 AI TOP, Arch: x86_64(Linux 5.14.0-503.29.1.el9_5.x86_64)
Platform verification succeeded
You may find this post useful.
Thanks for reaching out to us . RTX 50-series are Geforce GPU series , and p2p is not supported for GPU<->GPU P2P.
For GPU lists that can support p2p , there are mainly 2 series , Quadro RTX and Data Center GPUs .
Quadro RTX - desktop class systems, which would need an actively cooled GPU
Data Center GPUs - server class systems, which would need at passively cooled GPU.
Sample cuda-samples/Samples/0_Introduction/simpleP2P/simpleP2P.cu at master · NVIDIA/cuda-samples · GitHub can check if P2P is successfully enabled on a setup .
Hi @Yuki_Ni, thanks for your reply, but how can we use the shared RAM to make a communcation between RTX 50-series GPUs instead of P2P?
You may wish to keep an eye on this project, based on the open gpu kernel module driver, which enables P2P between RTX 40 series cards. 50 series not currently.
what do you mean by “the shared RAM” ?
Fallback mechanism of NCCL, anyways issue was fixed with the release of the new version of NCCL. See details here: NCCL P2P issue using two RTX 5090 · Issue #1637 · NVIDIA/nccl · GitHub Thank you!
