remove GPU from docs and scripts (#9076)

* remove GPU from docs and scripts

* update net.sh scripts to work without GPU

---------

Co-authored-by: Cavey Cool <caveycool@gmail.com>

docs/src/clusters/benchmark.md

@@ -81,14 +81,6 @@ To run a multinode testnet, after starting a leader node, spin up some additiona
 ./multinode-demo/validator-x.sh
 ```
 
-To run a performance-enhanced validator on Linux, [CUDA 10.0](https://developer.nvidia.com/cuda-downloads) must be installed on your system:
-
-```bash
-./fetch-perf-libs.sh
-SOLANA_CUDA=1 ./multinode-demo/bootstrap-validator.sh
-SOLANA_CUDA=1 ./multinode-demo/validator.sh
-```
-
 
 ### Testnet Client Demo
 

docs/src/clusters/testnet.md

@@ -200,18 +200,3 @@ branch), once the testnet has been created run:
 ```bash
 $ ./net.sh start -t edge
 ```
-
-### Enabling CUDA
-> [!NOTE]
-> CUDA is currently not available on GCE
-First ensure the network instances are created with GPU enabled:
-```bash
-$ ./gce.sh create -g ...
-```
-or
-```bash
-$ ./ec2.sh create -g ...
-```
-
-If deploying a tarball-based network nothing further is required, as GPU presence
-is detected at runtime and the CUDA build is auto selected.

docs/src/implemented-proposals/ed_overview/ed_validation_client_economics/ed_vce_overview.md

@@ -4,6 +4,6 @@ title: Validation-client Economics
 
 **Subject to change.**
 
-Validator-clients are eligible to charge commission on inflationary rewards distributed to staked tokens. This compensation is for providing compute \(CPU+GPU\) resources to validate and vote on a given PoH state. These protocol-based rewards are determined through an algorithmic disinflationary schedule as a function of total token supply. The network is expected to launch with an annual inflation rate around 8%, set to decrease by 15% per year until a long-term stable rate of 1.5% is reached, however these parameters are yet to be finalized by the community. These issuances are to be split and distributed to participating validators, with around 95% of the issued tokens allocated for validator rewards initial (the remaining 5% reserved for Foundation operating expenses). Because the network will be distributing a fixed amount of inflation rewards across the stake-weighted validator set, the yield observed for staked tokens will be primarily a function of the amount of staked tokens in relation to the total token supply.
+Validator-clients are eligible to charge commission on inflationary rewards distributed to staked tokens. This compensation is for providing compute resources to validate and vote on a given PoH state. These protocol-based rewards are determined through an algorithmic disinflationary schedule as a function of total token supply. The network is expected to launch with an annual inflation rate around 8%, set to decrease by 15% per year until a long-term stable rate of 1.5% is reached, however these parameters are yet to be finalized by the community. These issuances are to be split and distributed to participating validators, with around 95% of the issued tokens allocated for validator rewards initial (the remaining 5% reserved for Foundation operating expenses). Because the network will be distributing a fixed amount of inflation rewards across the stake-weighted validator set, the yield observed for staked tokens will be primarily a function of the amount of staked tokens in relation to the total token supply.
 
 Additionally, validator clients may earn revenue through fees via state-validation transactions. For clarity, we separately describe the design and motivation of these revenue distributions for validation-clients below: state-validation protocol-based rewards and state-validation transaction fees and rent.

docs/src/implemented-proposals/ed_overview/ed_validation_client_economics/ed_vce_state_validation_transaction_fees.md

@@ -6,7 +6,7 @@ title: State-validation Transaction Fees
 
 Each transaction sent through the network, to be processed by the current leader validation-client and confirmed as a global state transaction, must contain a transaction fee. Transaction fees offer many benefits in the Solana economic design, for example they:
 
-- provide unit compensation to the validator network for the CPU/GPU resources necessary to process the state transaction,
+- provide unit compensation to the validator network for the compute resources necessary to process the state transaction,
 - reduce network spam by introducing real cost to transactions,
 - open avenues for a transaction market to incentivize validation-client to collect and process submitted transactions in their function as leader,
 - and provide potential long-term economic stability of the network through a protocol-captured minimum fee amount per transaction, as described below.

docs/src/operations/guides/validator-start.md

@@ -29,14 +29,6 @@ solana transaction-count
 View the [metrics dashboard](https://metrics.solana.com:3000/d/monitor-edge/cluster-telemetry) for more
 detail on cluster activity.
 
-## Enabling CUDA
-
-If your machine has a GPU with CUDA installed \(Linux-only currently\), include
-the `--cuda` argument to `agave-validator`.
-
-When your validator is started look for the following log message to indicate
-that CUDA is enabled: `"[<timestamp> solana::validator] CUDA is enabled"`
-
 ## System Tuning
 
 ### Linux

docs/src/operations/requirements.md

@@ -25,7 +25,6 @@ The hardware recommendations below are provided as a guide.  Operators are encou
 | **RAM**   | Error Correction Code (ECC) memory is suggested<br />Motherboard with 512GB capacity suggested ||
 | | 256GB or more| 512 GB or more for **all [account indexes](https://docs.anza.xyz/operations/setup-an-rpc-node#account-indexing)** |
 | **Disk**  | PCIe Gen3 x4 NVME SSD, or better, on each of: <br />- **Accounts**: 1TB, or larger. High TBW (Total Bytes Written)<br />- **Ledger**: 1TB or larger. High TBW suggested<br />- **Snapshots**: 500GB or larger. High TBW suggested<br />- **OS**: (Optional) 500GB, or larger. SATA OK<br /><br />The OS may be installed on the ledger disk, though testing has shown better performance with the ledger on its own disk<br /><br />Accounts and ledger *can* be stored on the same disk, however due to high IOPS, this is not recommended<br /><br />The Samsung 970 and 980 Pro series SSDs are popular with the validator community | Consider a larger ledger disk if longer transaction history is required<br /><br />Accounts and ledger **should not** be stored on the same disk |
-| **GPUs**  | Not necessary at this time<br />Operators in the validator community do not use GPUs currently | |
 
 A community maintained list of currently optimal hardware can be found here: [solanahcl.org](https://solanahcl.org/). It is updated automatically from the [solanahcl/solanahcl Github repo](https://github.com/solanahcl/solanahcl).
 

docs/src/proposals/ledger-replication-to-implement.md

@@ -18,7 +18,7 @@ The basic idea to Proof of Replication is encrypting a dataset with a public sym
 
 ## Optimization with PoH
 
-Our improvement on this approach is to randomly sample the encrypted segments faster than it takes to encrypt, and record the hash of those samples into the PoH ledger. Thus the segments stay in the exact same order for every PoRep and verification can stream the data and verify all the proofs in a single batch. This way we can verify multiple proofs concurrently, each one on its own CUDA core. The total space required for verification is `1_ledger_segment + 2_cbc_blocks * number_of_identities` with core count equal to `number_of_identities`. We use a 64-byte chacha CBC block size.
+Our improvement on this approach is to randomly sample the encrypted segments faster than it takes to encrypt, and record the hash of those samples into the PoH ledger. Thus the segments stay in the exact same order for every PoRep and verification can stream the data and verify all the proofs in a single batch. This way we can verify multiple proofs concurrently. The total space required for verification is `1_ledger_segment + 2_cbc_blocks * number_of_identities` with core count equal to `number_of_identities`. We use a 64-byte chacha CBC block size.
 
 ## Network
 
@@ -32,11 +32,9 @@ We have the following constraints:
 
 - Verification requires generating the CBC blocks. That requires space of 2
 
-  blocks per identity, and 1 CUDA core per identity for the same dataset. So as
+  blocks per identity. So as many identities at once should be batched with as
 
-  many identities at once should be batched with as many proofs for those
-
-  identities verified concurrently for the same dataset.
+  many proofs for those identities verified concurrently for the same dataset.
 
 - Validators will randomly sample the set of storage proofs to the set that
 

multinode-demo/bootstrap-validator.sh

@@ -8,16 +8,7 @@ here=$(dirname "$0")
 # shellcheck source=multinode-demo/common.sh
 source "$here"/common.sh
 
-if [[ "$SOLANA_GPU_MISSING" -eq 1 ]]; then
-  echo "Testnet requires GPUs, but none were found!  Aborting..."
-  exit 1
-fi
-
-if [[ -n $SOLANA_CUDA ]]; then
-  program=$agave_validator_cuda
-else
-  program=$agave_validator
-fi
+program=$agave_validator
 
 no_restart=0
 maybeRequireTower=true

multinode-demo/common.sh

@@ -16,15 +16,6 @@ if [[ $1 = "--prebuild" ]]; then
   prebuild=true
 fi
 
-if [[ $(uname) != Linux ]]; then
-  # Protect against unsupported configurations to prevent non-obvious errors
-  # later. Arguably these should be fatal errors but for now prefer tolerance.
-  if [[ -n $SOLANA_CUDA ]]; then
-    echo "Warning: CUDA is not supported on $(uname)"
-    SOLANA_CUDA=
-  fi
-fi
-
 if [[ -n $USE_INSTALL || ! -f "$SOLANA_ROOT"/Cargo.toml ]]; then
   solana_program() {
     declare program="$1"
@@ -75,7 +66,6 @@ fi
 solana_bench_tps=$(solana_program bench-tps)
 solana_faucet=$(solana_program faucet)
 agave_validator=$(solana_program validator)
-agave_validator_cuda="$agave_validator --cuda"
 solana_genesis=$(solana_program genesis)
 solana_gossip=$(solana_program gossip)
 solana_keygen=$(solana_program keygen)

multinode-demo/validator.sh

@@ -203,11 +203,6 @@ while [[ -n $1 ]]; do
   fi
 done
 
-if [[ "$SOLANA_GPU_MISSING" -eq 1 ]]; then
-  echo "Testnet requires GPUs, but none were found!  Aborting..."
-  exit 1
-fi
-
 if [[ ${#positional_args[@]} -gt 1 ]]; then
   usage "$@"
 fi
@@ -275,11 +270,8 @@ if [[ $maybeRequireTower = true ]]; then
   default_arg --require-tower
 fi
 
-if [[ -n $SOLANA_CUDA ]]; then
-  program=$agave_validator_cuda
-else
-  program=$agave_validator
-fi
+
+program=$agave_validator
 
 set -e
 PS4="$(basename "$0"): "

net/gce.sh

@@ -14,7 +14,6 @@ gce)
 
   # use n1 instead of n2 so we don't need to spin up >= 4 local SSD's
   cpuBootstrapLeaderMachineType="--custom-cpu 24 --min-cpu-platform Intel%20Skylake --custom-vm-type n1"
-  gpuBootstrapLeaderMachineType="$cpuBootstrapLeaderMachineType --accelerator count=1,type=nvidia-tesla-p100"
   clientMachineType="--custom-cpu 16 --custom-memory 20GB"
   blockstreamerMachineType="--machine-type n1-standard-8"
   selfDestructHours=8
@@ -25,10 +24,6 @@ ec2)
 
   cpuBootstrapLeaderMachineType=m5.4xlarge
 
-  # NOTE: At this time only the p3dn.24xlarge EC2 instance type has GPU and
-  #       AVX-512 support.  The default, p2.xlarge, does not support
-  #       AVX-512
-  gpuBootstrapLeaderMachineType=p2.xlarge
   clientMachineType=c5.2xlarge
   blockstreamerMachineType=m5.4xlarge
   selfDestructHours=0
@@ -38,7 +33,6 @@ azure)
   source "$here"/scripts/azure-provider.sh
 
   cpuBootstrapLeaderMachineType=Standard_D16s_v3
-  gpuBootstrapLeaderMachineType=Standard_NC12
   clientMachineType=Standard_D16s_v3
   blockstreamerMachineType=Standard_D16s_v3
   selfDestructHours=0
@@ -48,7 +42,6 @@ colo)
   source "$here"/scripts/colo-provider.sh
 
   cpuBootstrapLeaderMachineType=0
-  gpuBootstrapLeaderMachineType=1
   clientMachineType=0
   blockstreamerMachineType=0
   selfDestructHours=0
@@ -75,7 +68,6 @@ customMemoryGB="$defaultCustomMemoryGB"
 
 publicNetwork=false
 letsEncryptDomainName=
-enableGpu=false
 customMachineType=
 customAddress=
 zones=()
@@ -123,10 +115,6 @@ Manage testnet instances
    -c [number]      - Number of client nodes (default: $clientNodeCount)
    -u               - Include a Blockstreamer (default: $blockstreamer)
    -P               - Use public network IP addresses (default: $publicNetwork)
-   -g               - Enable GPU and automatically set validator machine types to $gpuBootstrapLeaderMachineType
-                      (default: $enableGpu)
-   -G               - Enable GPU, and set custom GPU machine type to use
-                      (e.g $gpuBootstrapLeaderMachineType)
    -a [address]     - Address to be be assigned to the Blockstreamer if present,
                       otherwise the bootstrap validator.
                       * For GCE, [address] is the "name" of the desired External
@@ -139,15 +127,13 @@ Manage testnet instances
                       DNS name (useful only when the -a and -P options
                       are also provided)
    --custom-machine-type [type]
-                    - Set a custom machine type without assuming whether or not
-                      GPU is enabled.  Set this explicitly with --enable-gpu/-g to call out the presence of GPUs.
+                    - Set a custom machine type.
 $(
   if [[ -n "$defaultCustomMemoryGB" ]]; then
     echo "   --custom-memory-gb"
     echo "                    - Set memory size for custom machine type in GB (default: $defaultCustomMemoryGB)"
   fi
 )
-   --enable-gpu     - Use with --custom-machine-type to specify whether or not GPUs should be used/enabled
    --validator-additional-disk-size-gb [number]
                     - Add an additional [number] GB SSD to all validators to store the config directory.
                       If not set, config will be written to the boot disk by default.
@@ -200,7 +186,7 @@ while [[ -n $1 ]]; do
     elif [[ $1 = --validator-additional-disk-size-gb ]]; then
       validatorAdditionalDiskSizeInGb="$2"
       shift 2
-    elif [[ $1 == --machine-type* || $1 == --custom-cpu* ]]; then # Bypass quoted long args for GPUs
+    elif [[ $1 == --machine-type* || $1 == --custom-cpu* ]]; then # Bypass quoted long args
       shortArgs+=("$1")
       shift
     elif [[ $1 == --allow-boot-failures ]]; then
@@ -213,7 +199,7 @@ while [[ -n $1 ]]; do
       evalInfo=true
       shift
     elif [[ $1 == --enable-gpu ]]; then
-      enableGpu=true
+      echo "GPU support has been dropped, --enable-gpu is a noop"
       shift
     elif [[ $1 = --custom-machine-type ]]; then
       customMachineType="$2"
@@ -276,11 +262,10 @@ while getopts "h?p:Pn:c:r:z:gG:a:d:uxf" opt "${shortArgs[@]}"; do
     containsZone "$OPTARG" "${zones[@]}" || zones+=("$OPTARG")
     ;;
   g)
-    enableGpu=true
+    echo "GPU support has been dropped, -g argument is a noop"
     ;;
   G)
-    enableGpu=true
-    customMachineType="$OPTARG"
+    echo "GPU support has been dropped, -G argument is a noop"
     ;;
   a)
     customAddress=$OPTARG
@@ -316,14 +301,12 @@ case $cloudProvider in
 gce)
   if [[ "$tmpfsAccounts" = "true" ]]; then
     cpuBootstrapLeaderMachineType+=" --local-ssd interface=nvme"
-    gpuBootstrapLeaderMachineType+=" --local-ssd interface=nvme"
     if [[ $customMemoryGB -lt 100 ]]; then
       # shellcheck disable=SC2016 # We don't want expression expansion on these backticks
       echo -e '\nWarning: At least 100GB of system RAM is recommending with `--tmpfs-accounts` (see `--custom-memory-gb`)\n'
     fi
   fi
   cpuBootstrapLeaderMachineType+=" --custom-memory ${customMemoryGB}GB"
-  gpuBootstrapLeaderMachineType+=" --custom-memory ${customMemoryGB}GB"
   ;;
 ec2|azure|colo)
   if [[ -n $validatorAdditionalDiskSizeInGb ]] ; then
@@ -363,8 +346,6 @@ fi
 
 if [[ -n "$customMachineType" ]] ; then
   bootstrapLeaderMachineType="$customMachineType"
-elif [[ "$enableGpu" = "true" ]] ; then
-  bootstrapLeaderMachineType="$gpuBootstrapLeaderMachineType"
 else
   bootstrapLeaderMachineType="$cpuBootstrapLeaderMachineType"
 fi
@@ -718,7 +699,7 @@ create)
   printNetworkInfo() {
     cat <<EOF
 ==[ Network composition ]===============================================================
-  Bootstrap validator = $bootstrapLeaderMachineType (GPU=$enableGpu)
+  Bootstrap validator = $bootstrapLeaderMachineType
   Additional validators = $additionalValidatorCount x $validatorMachineType
   Client(s) = $clientNodeCount x $clientMachineType
   Blockstreamer = $blockstreamer
@@ -748,9 +729,6 @@ if [[ -f /solana-scratch/.instance-startup-complete ]]; then
   echo reboot
   $(
     cd "$here"/scripts/
-    if "$enableGpu"; then
-      cat enable-nvidia-persistence-mode.sh
-    fi
 
     if [[ -n $validatorAdditionalDiskSizeInGb ]]; then
       cat mount-additional-disk.sh
@@ -812,10 +790,6 @@ $(
     network-config.sh \
     remove-docker-interface.sh \
 
-  if "$enableGpu"; then
-    cat enable-nvidia-persistence-mode.sh
-  fi
-
   if [[ -n $validatorAdditionalDiskSizeInGb ]]; then
     cat mount-additional-disk.sh
   fi
@@ -900,7 +874,7 @@ EOF
     echo "Bootstrap validator is already configured"
   else
     cloud_CreateInstances "$prefix" "$prefix-bootstrap-validator" 1 \
-      "$enableGpu" "$bootstrapLeaderMachineType" "${zones[0]}" "$validatorBootDiskSizeInGb" \
+      "$bootstrapLeaderMachineType" "${zones[0]}" "$validatorBootDiskSizeInGb" \
       "$startupScript" "$bootstrapLeaderAddress" "$bootDiskType" "$validatorAdditionalDiskSizeInGb" \
       "$maybePreemptible" "$sshPrivateKey"
   fi
@@ -921,7 +895,7 @@ EOF
         numNodesPerZone=$((numNodesPerZone + numLeftOverNodes))
       fi
       cloud_CreateInstances "$prefix" "$prefix-$zone-validator" "$numNodesPerZone" \
-        "$enableGpu" "$validatorMachineType" "$zone" "$validatorBootDiskSizeInGb" \
+        "$validatorMachineType" "$zone" "$validatorBootDiskSizeInGb" \
         "$startupScript" "" "$bootDiskType" "$validatorAdditionalDiskSizeInGb" \
         "$preemptible" "$sshPrivateKey" &
     done
@@ -931,13 +905,13 @@ EOF
 
   if [[ $clientNodeCount -gt 0 ]]; then
     cloud_CreateInstances "$prefix" "$prefix-client" "$clientNodeCount" \
-      "$enableGpu" "$clientMachineType" "${zones[0]}" "$clientBootDiskSizeInGb" \
+      "$clientMachineType" "${zones[0]}" "$clientBootDiskSizeInGb" \
       "$startupScript" "" "$bootDiskType" "" "$maybePreemptible" "$sshPrivateKey"
   fi
 
   if $blockstreamer; then
     cloud_CreateInstances "$prefix" "$prefix-blockstreamer" "1" \
-      "$enableGpu" "$blockstreamerMachineType" "${zones[0]}" "$validatorBootDiskSizeInGb" \
+      "$blockstreamerMachineType" "${zones[0]}" "$validatorBootDiskSizeInGb" \
       "$startupScript" "$blockstreamerAddress" "$bootDiskType" "" "$maybePreemptible" "$sshPrivateKey"
   fi
 

net/net.sh

@@ -60,13 +60,7 @@ Operate a configured testnet
                                       - Number of seconds to wait after validators have finished starting before starting client programs
                                         (default: $clientDelayStart)
    -n NUM_VALIDATORS                  - Number of validators to apply command to.
-   --gpu-mode GPU_MODE                - Specify GPU mode to launch validators with (default: $gpuMode).
-                                        MODE must be one of
-                                          on - GPU *required*, any vendor *
-                                          off - No GPU, CPU-only
-                                          auto - Use GPU if available, any vendor *
-                                          cuda - GPU *required*, Nvidia CUDA only
-                                          *  Currently, Nvidia CUDA is the only supported GPU vendor
+   --gpu-mode GPU_MODE                - Deprecated, this argument is ignored
    --hashes-per-tick NUM_HASHES|sleep|auto
                                       - Override the default --hashes-per-tick for the cluster
    --no-airdrop
@@ -341,7 +335,6 @@ startBootstrapLeader() {
          ${#clientIpList[@]} \"$benchTpsExtraArgs\" \
          \"$genesisOptions\" \
          \"$maybeNoSnapshot $maybeSkipLedgerVerify $maybeLimitLedgerSize $maybeWaitForSupermajority $maybeAccountsDbSkipShrink $maybeSkipRequireTower\" \
-         \"$gpuMode\" \
          \"$maybeWarpSlot\" \
          \"$maybeFullRpc\" \
          \"$waitForNodeInit\" \
@@ -416,7 +409,6 @@ startNode() {
          ${#clientIpList[@]} \"$benchTpsExtraArgs\" \
          \"$genesisOptions\" \
          \"$maybeNoSnapshot $maybeSkipLedgerVerify $maybeLimitLedgerSize $maybeWaitForSupermajority $maybeAccountsDbSkipShrink $maybeSkipRequireTower\" \
-         \"$gpuMode\" \
          \"$maybeWarpSlot\" \
          \"$maybeFullRpc\" \
          \"$waitForNodeInit\" \
@@ -831,7 +823,6 @@ maybeSkipRequireTower=""
 debugBuild=false
 profileBuild=false
 doBuild=true
-gpuMode=auto
 clientDelayStart=0
 netLogDir=
 maybeWarpSlot=
@@ -925,15 +916,7 @@ while [[ -n $1 ]]; do
       profileBuild=true
       shift 1
     elif [[ $1 = --gpu-mode ]]; then
-      gpuMode=$2
-      case "$gpuMode" in
-        on|off|auto|cuda)
-          ;;
-        *)
-          echo "Unexpected GPU mode: \"$gpuMode\""
-          exit 1
-          ;;
-      esac
+      echo "'--gpu-mode' is deprecated, GPU support was removed from agave"
       shift 2
     elif [[ $1 == --client-delay-start ]]; then
       clientDelayStart=$2

net/remote/remote-node.sh

@@ -22,15 +22,14 @@ numBenchTpsClients="${13}"
 benchTpsExtraArgs="${14}"
 genesisOptions="${15}"
 extraNodeArgs="${16}"
-gpuMode="${17:-auto}"
-maybeWarpSlot="${18}"
-maybeFullRpc="${19}"
-waitForNodeInit="${20}"
-extraPrimordialStakes="${21:=0}"
-tmpfsAccounts="${22:false}"
-disableQuic="${23}"
-enableUdp="${24}"
-maybeWenRestart="${25}"
+maybeWarpSlot="${17}"
+maybeFullRpc="${18}"
+waitForNodeInit="${19}"
+extraPrimordialStakes="${20:=0}"
+tmpfsAccounts="${21:false}"
+disableQuic="${22}"
+enableUdp="${23}"
+maybeWenRestart="${24}"
 
 set +x
 
@@ -75,27 +74,6 @@ ln -sfT validator.log.\$now validator.log
 EOF
 chmod +x ~/solana/on-reboot
 
-GPU_CUDA_OK=false
-GPU_FAIL_IF_NONE=false
-case "$gpuMode" in
-  on) # GPU *required*, any vendor
-    GPU_CUDA_OK=true
-    GPU_FAIL_IF_NONE=true
-    ;;
-  off) # CPU-only
-    ;;
-  auto) # Use GPU if installed, any vendor
-    GPU_CUDA_OK=true
-    ;;
-  cuda) # GPU *required*, CUDA-only
-    GPU_CUDA_OK=true
-    GPU_FAIL_IF_NONE=true
-    ;;
-  *)
-    echo "Unexpected gpuMode: \"$gpuMode\""
-    exit 1
-    ;;
-esac
 
 case $deployMethod in
 local|tar|skip)
@@ -120,14 +98,6 @@ cat >> ~/solana/on-reboot <<EOF
   echo \$! > iftop.pid
   scripts/system-stats.sh  > system-stats.log 2>&1 &
   echo \$! > system-stats.pid
-
-  if ${GPU_CUDA_OK} && [[ -e /dev/nvidia0 ]]; then
-    echo Selecting agave-validator-cuda
-    export SOLANA_CUDA=1
-  elif ${GPU_FAIL_IF_NONE} ; then
-    echo "Expected GPU, found none!"
-    export SOLANA_GPU_MISSING=1
-  fi
 EOF
 
   case $nodeType in

net/scripts/azure-provider.sh

@@ -174,7 +174,7 @@ cloud_Initialize() {
 
 #
 # cloud_CreateInstances [networkName] [namePrefix] [numNodes]
-#                       [enableGpu] [machineType] [zone]
+#                       [machineType] [zone]
 #                       [bootDiskSize] [startupScript] [address]
 #                       [bootDiskType] [additionalDiskSize] [preemptible]
 #
@@ -183,8 +183,6 @@ cloud_Initialize() {
 # networkName   - unique name of this testnet
 # namePrefix    - unique string to prefix all the instance names with
 # numNodes      - number of instances to create
-# enableGpu     - Optionally enable GPU, use the value "true" to enable
-#                 eg, request 4 K80 GPUs with "count=4,type=nvidia-tesla-k80"
 # machineType   - GCE machine type.  Note that this may also include an
 #                 `--accelerator=` or other |gcloud compute instances create|
 #                 options
@@ -204,13 +202,12 @@ cloud_CreateInstances() {
   declare networkName="$1"
   declare namePrefix="$2"
   declare numNodes="$3"
-  declare enableGpu="$4"
-  declare machineType="$5"
-  declare zone="$6"
-  declare optionalBootDiskSize="$7"
-  declare optionalStartupScript="$8"
-  declare optionalAddress="$9"
-  declare optionalBootDiskType="${10}"
+  declare machineType="$4"
+  declare zone="$5"
+  declare optionalBootDiskSize="$6"
+  declare optionalStartupScript="$7"
+  declare optionalAddress="$8"
+  declare optionalBootDiskType="${9}"
 
   declare -a nodes
   if [[ $numNodes = 1 ]]; then
@@ -274,24 +271,6 @@ cloud_CreateInstances() {
     for nodeName in "${nodes[@]}"; do
       az vm wait --created --name "$nodeName" --resource-group "$networkName" --verbose --timeout 600
     done
-
-    # If GPU is to be enabled, install the appropriate extension
-    if $enableGpu; then
-      for nodeName in "${nodes[@]}"; do
-        az vm extension set \
-        --resource-group "$networkName" \
-        --vm-name "$nodeName" \
-        --name NvidiaGpuDriverLinux \
-        --publisher Microsoft.HpcCompute \
-        --version 1.2 \
-        --no-wait
-      done
-
-      # Wait until all nodes have GPU extension installed
-      for nodeName in "${nodes[@]}"; do
-        az vm wait --updated --name "$nodeName" --resource-group "$networkName" --verbose --timeout 600
-      done
-    fi
   )
 }
 

net/scripts/colo-provider.sh

@@ -129,7 +129,7 @@ cloud_Initialize() {
 
 #
 # cloud_CreateInstances [networkName] [namePrefix] [numNodes]
-#                       [enableGpu] [machineType] [zone]
+#                       [machineType] [zone]
 #                       [bootDiskSize] [startupScript] [address]
 #                       [bootDiskType] [additionalDiskSize] [preemptible]
 #
@@ -138,8 +138,6 @@ cloud_Initialize() {
 # networkName   - unique name of this testnet
 # namePrefix    - unique string to prefix all the instance names with
 # numNodes      - number of instances to create
-# enableGpu     - Optionally enable GPU, use the value "true" to enable
-#                 eg, request 4 K80 GPUs with "count=4,type=nvidia-tesla-k80"
 # machineType   - GCE machine type.  Note that this may also include an
 #                 `--accelerator=` or other |gcloud compute instances create|
 #                 options
@@ -159,16 +157,15 @@ cloud_CreateInstances() {
   #declare networkName="${1}" # unused
   declare namePrefix="${2}"
   declare numNodes="${3}"
-  #declare enableGpu="${4}" # unused
-  declare machineType="${5}"
-  # declare zone="${6}" # unused
-  #declare optionalBootDiskSize="${7}" # unused
-  #declare optionalStartupScript="${8}" # unused
-  #declare optionalAddress="${9}" # unused
-  #declare optionalBootDiskType="${10}" # unused
-  #declare optionalAdditionalDiskSize="${11}" # unused
-  declare optionalPreemptible="${12}"
-  declare sshPrivateKey="${13}"
+  declare machineType="${4}"
+  # declare zone="${5}" # unused
+  #declare optionalBootDiskSize="${6}" # unused
+  #declare optionalStartupScript="${7}" # unused
+  #declare optionalAddress="${8}" # unused
+  #declare optionalBootDiskType="${9}" # unused
+  #declare optionalAdditionalDiskSize="${10}" # unused
+  declare optionalPreemptible="${11}"
+  declare sshPrivateKey="${12}"
 
   declare -a nodes
   if [[ ${numNodes} = 1 ]]; then

net/scripts/ec2-provider.sh

@@ -140,7 +140,7 @@ cloud_Initialize() {
 
 #
 # cloud_CreateInstances [networkName] [namePrefix] [numNodes]
-#                       [enableGpu] [machineType] [zone]
+#                       [machineType] [zone]
 #                       [bootDiskSize] [startupScript] [address]
 #                       [bootDiskType] [additionalDiskSize] [preemptible]
 #
@@ -149,8 +149,6 @@ cloud_Initialize() {
 # networkName   - unique name of this testnet
 # namePrefix    - unique string to prefix all the instance names with
 # numNodes      - number of instances to create
-# enableGpu     - Optionally enable GPU, use the value "true" to enable
-#                 eg, request 4 K80 GPUs with "count=4,type=nvidia-tesla-k80"
 # machineType   - GCE machine type.  Note that this may also include an
 #                 `--accelerator=` or other |gcloud compute instances create|
 #                 options
@@ -170,88 +168,63 @@ cloud_CreateInstances() {
   declare networkName="$1"
   declare namePrefix="$2"
   declare numNodes="$3"
-  declare enableGpu="$4"
-  declare machineType="$5"
-  declare zone="$6"
-  declare optionalBootDiskSize="$7"
-  declare optionalStartupScript="$8"
-  declare optionalAddress="$9"
+  declare machineType="$4"
+  declare zone="$5"
+  declare optionalBootDiskSize="$6"
+  declare optionalStartupScript="$7"
+  declare optionalAddress="$8"
 
   declare region=
   region=$(__cloud_GetRegion "$zone")
 
-  if $enableGpu; then
-    #
-    # Custom Ubuntu 18.04 LTS image with CUDA 9.2 and CUDA 10.0 installed
-    #
-    # Unfortunately these AMIs are not public.  When this becomes an issue, use
-    # the stock Ubuntu 18.04 image and programmatically install CUDA after the
-    # instance boots
-    #
-    case $region in
-    us-east-1)
-      imageName="ami-0a8bd6fb204473f78"
-      ;;
-    us-west-1)
-      imageName="ami-07011f0795513c59d"
-      ;;
-    us-west-2)
-      imageName="ami-0a11ef42b62b82b68"
-      ;;
-    *)
-      usage "Unsupported region: $region"
-      ;;
-    esac
-  else
-    # Select an upstream Ubuntu 18.04 AMI from https://cloud-images.ubuntu.com/locator/ec2/
-    case $region in
-    us-east-1)
-      imageName="ami-0fba9b33b5304d8b4"
-      ;;
-    us-east-2)
-      imageName="ami-0e04554247365d806"
-      ;;
-    us-west-1)
-      imageName="ami-07390b6ff5934a238"
-      ;;
-    us-west-2)
-      imageName="ami-03804ed633fe58109"
-      ;;
-    sa-east-1)
-      imageName="ami-0f1678b6f63a0f923"
-      ;;
-    ap-northeast-2)
-      imageName="ami-0695e34e31339c3ff"
-      ;;
-    ap-northeast-1)
-      imageName="ami-003371bfa26192744"
-      ;;
-    ap-southeast-2)
-      imageName="ami-0401c9e2f645b5557"
-      ;;
-    ap-southeast-1)
-      imageName="ami-08050c889a630f1bd"
-      ;;
-    ap-south-1)
-      imageName="ami-04184c12996409633"
-      ;;
-    eu-central-1)
-      imageName="ami-054e21e355db24124"
-      ;;
-    eu-west-1)
-      imageName="ami-0727f3c2d4b0226d5"
-      ;;
-    eu-west-2)
-      imageName="ami-068f09e337d7da0c4"
-      ;;
-    ca-central-1)
-      imageName="ami-06ed08059bdc08fc9"
-      ;;
-    *)
-      usage "Unsupported region: $region"
-      ;;
-    esac
-  fi
+  # Select an upstream Ubuntu 18.04 AMI from https://cloud-images.ubuntu.com/locator/ec2/
+  case $region in
+  us-east-1)
+    imageName="ami-0fba9b33b5304d8b4"
+    ;;
+  us-east-2)
+    imageName="ami-0e04554247365d806"
+    ;;
+  us-west-1)
+    imageName="ami-07390b6ff5934a238"
+    ;;
+  us-west-2)
+    imageName="ami-03804ed633fe58109"
+    ;;
+  sa-east-1)
+    imageName="ami-0f1678b6f63a0f923"
+    ;;
+  ap-northeast-2)
+    imageName="ami-0695e34e31339c3ff"
+    ;;
+  ap-northeast-1)
+    imageName="ami-003371bfa26192744"
+    ;;
+  ap-southeast-2)
+    imageName="ami-0401c9e2f645b5557"
+    ;;
+  ap-southeast-1)
+    imageName="ami-08050c889a630f1bd"
+    ;;
+  ap-south-1)
+    imageName="ami-04184c12996409633"
+    ;;
+  eu-central-1)
+    imageName="ami-054e21e355db24124"
+    ;;
+  eu-west-1)
+    imageName="ami-0727f3c2d4b0226d5"
+    ;;
+  eu-west-2)
+    imageName="ami-068f09e337d7da0c4"
+    ;;
+  ca-central-1)
+    imageName="ami-06ed08059bdc08fc9"
+    ;;
+  *)
+    usage "Unsupported region: $region"
+    ;;
+  esac
 
   declare -a args
   args=(

net/scripts/enable-nvidia-persistence-mode.sh

@@ -1,6 +0,0 @@
-#!/usr/bin/env bash
-
-echo "--- Enabling nvidia persistence mode"
-if ! nvidia-smi -pm ENABLED; then
-  echo "^^^ +++"
-fi

net/scripts/gce-provider.sh

@@ -131,7 +131,7 @@ cloud_Initialize() {
 
 #
 # cloud_CreateInstances [networkName] [namePrefix] [numNodes]
-#                       [enableGpu] [machineType] [zone]
+#                       [machineType] [zone]
 #                       [bootDiskSize] [startupScript] [address]
 #                       [bootDiskType] [additionalDiskSize] [preemptible]
 #
@@ -140,8 +140,6 @@ cloud_Initialize() {
 # networkName   - unique name of this testnet
 # namePrefix    - unique string to prefix all the instance names with
 # numNodes      - number of instances to create
-# enableGpu     - Optionally enable GPU, use the value "true" to enable
-#                 eg, request 4 K80 GPUs with "count=4,type=nvidia-tesla-k80"
 # machineType   - GCE machine type.  Note that this may also include an
 #                 `--accelerator=` or other |gcloud compute instances create|
 #                 options
@@ -161,31 +159,16 @@ cloud_CreateInstances() {
   declare networkName="$1"
   declare namePrefix="$2"
   declare numNodes="$3"
-  declare enableGpu="$4"
-  declare machineType="$5"
-  declare zone="$6"
-  declare optionalBootDiskSize="$7"
-  declare optionalStartupScript="$8"
-  declare optionalAddress="$9"
-  declare optionalBootDiskType="${10:-pd-ssd}"
-  declare optionalAdditionalDiskSize="${11}"
-  declare optionalPreemptible="${12}"
-
-  if $enableGpu; then
-    # Custom Ubuntu 20.04 LTS image with CUDA 10.2 installed
-    #
-    # Unfortunately this image is not public.  When this becomes an issue, use
-    # the stock Ubuntu 20.04 image and programmatically install CUDA after the
-    # instance boots
-    #
-    # imageName="ubuntu-2004-focal-v20201211-with-cuda-10-2 --image-project principal-lane-200702"
-
-    # We don't have custom cuda image for ubnutu 22.04
-    echo "Error: Not supported" >&2
-    exit 1
-  else
-    imageName="ubuntu-2404-noble-amd64-v20250709 --image-project ubuntu-os-cloud"
-  fi
+  declare machineType="$4"
+  declare zone="$5"
+  declare optionalBootDiskSize="$6"
+  declare optionalStartupScript="$7"
+  declare optionalAddress="$8"
+  declare optionalBootDiskType="${9:-pd-ssd}"
+  declare optionalAdditionalDiskSize="${10}"
+  declare optionalPreemptible="${11}"
+
+  imageName="ubuntu-2404-noble-amd64-v20250709 --image-project ubuntu-os-cloud"
 
   declare -a nodes
   if [[ $numNodes = 1 ]]; then

scripts/system-stats.sh

@@ -21,25 +21,7 @@ while true; do
   ram_usage=$(awk "BEGIN {print $used / $total * 100}")
   cpu_report="cpu_usage=$cpu_usage,ram_usage=$ram_usage"
 
-  # if nvidia-smi exists, report gpu stats
-  gpu_report=""
-  if [ -x "$(command -v nvidia-smi)" ]; then
-    mapfile -t individual_gpu_usage < <(nvidia-smi --query-gpu=utilization.gpu,memory.used,memory.total --format=csv,nounits,noheader)
-    total_gpu_usage=0
-    total_gpu_mem_usage=0
-    num_gpus=${#individual_gpu_usage[@]}
-    for entry in "${individual_gpu_usage[@]}"
-    do
-      read -r compute mem_used mem_total <<< "${entry//,/}"
-      total_gpu_usage=$(awk "BEGIN {print $total_gpu_usage + $compute }")
-      total_gpu_mem_usage=$(awk "BEGIN {print $total_gpu_mem_usage + $mem_used / $mem_total * 100}")
-    done
-    avg_gpu_usage=$(awk "BEGIN {print $total_gpu_usage / $num_gpus}")
-    avg_gpu_mem_usage=$(awk "BEGIN {print $total_gpu_mem_usage / $num_gpus}")
-    gpu_report=",avg_gpu_usage=$avg_gpu_usage,avg_gpu_mem_usage=$avg_gpu_mem_usage"
-  fi
-
-  report="${cpu_report}${gpu_report}"
+  report="${cpu_report}"
   ./scripts/metrics-write-datapoint.sh "system-stats,hostname=$HOSTNAME $report"
   sleep 1
 done