Caffe install on ubuntu 20.04 with RTX3090

[pccillnb4]

"make all" and "make tes"t was completed with no errors:( I'm trying "runtest" shows errors below:

Note: Randomizing tests' orders with a seed of 94164 .
[==========] Running 2301 tests from 299 test cases.
[----------] Global test environment set-up.
[----------] 9 tests from LSTMLayerTest/3, where TypeParam = caffe::GPUDevice<double>
[ RUN      ] LSTMLayerTest/3.TestForward
E0220 16:19:58.250381 155474 common.cpp:113] Cannot create Cublas handle. Cublas won't be available.
E0220 16:19:58.273321 155474 common.cpp:120] Cannot create Curand generator. Curand won't be available.
F0220 16:19:58.294371 155474 syncedmem.hpp:18] Check failed: error == cudaSuccess (46 vs. 0)  all CUDA-capable devices are busy or unavailable
*** Check failure stack trace: ***
    @     0x7f3a89f551c3  google::LogMessage::Fail()
    @     0x7f3a89f5a25b  google::LogMessage::SendToLog()
    @     0x7f3a89f54ebf  google::LogMessage::Flush()
    @     0x7f3a89f556ef  google::LogMessageFatal::~LogMessageFatal()
    @     0x7f3a891ed563  caffe::SyncedMemory::mutable_cpu_data()
    @     0x7f3a8940d805  caffe::Blob<>::Reshape()
    @     0x7f3a8940dc6a  caffe::Blob<>::Reshape()
    @     0x559fa78ae586  caffe::LSTMLayerTest<>::ReshapeBlobs()
    @     0x559fa78aedab  caffe::LSTMLayerTest<>::LSTMLayerTest()
    @     0x559fa78af04b  testing::internal::TestFactoryImpl<>::CreateTest()
    @     0x559fa79bff1a  testing::internal::HandleExceptionsInMethodIfSupported<>()
    @     0x559fa79b90d3  testing::TestInfo::Run()
    @     0x559fa79b9225  testing::TestCase::Run()
    @     0x559fa79b96e0  testing::internal::UnitTestImpl::RunAllTests()
    @     0x559fa79b9827  testing::UnitTest::Run()
    @     0x559fa744f231  main
    @     0x7f3a88c710b3  __libc_start_main
    @     0x559fa7457e8e  _start
make: *** [Makefile:537: runtest] Aborted (core dumped)

also,trying edit "Makefile.config" to "CUDA architecture setting" below: CUDA_ARCH := #-gencode arch=compute_35,code=sm_35 \

-gencode arch=compute_50,code=sm_50 \

         #-gencode arch=compute_52,code=sm_52 \
         #-gencode arch=compute_60,code=sm_60 \
         #-gencode arch=compute_61,code=sm_61 \
         #-gencode arch=compute_70,code=sm_70 \
          gencode arch=compute_86 code=sm_86   <==The Compute Capability of RTX 3090 is 8.6.

System configuration

• Operating system: Ubuntu 20.00
• CUDA version (if applicable): 11.2
• CUDNN version (if applicable): 8.1

[artyom-beilis]

How did you build against cudnn 8.1 some of the API caffe uses was dropped in cudnn 8?

https://github.com/BVLC/caffe/pull/7000

[pccillnb4]

How did you build against cudnn 8.1 some of the API caffe uses was dropped in cudnn 8?

7000

Hello artyom-beilis

I have been successfully install caffe without cudnn kit on Ampere architecture.

Unfortunately, The caffe was not supported in cudnn 8.

[jieunko]

Hi! pccillnb4 I am trying to build Caffe on RTX3080ti and got stuck on same error with you. Did you succeed on building caffe? Can I ask you how???

[pccillnb4]

Sure, if I can. I installed another cudnn of version on my pc, but I forget what version I used.

it's so far away....

if you need to help, add my discord battleship#6542