Llama.cpp NVIDIA GeForce RTX 5090

Benchmarks by Michael Larabel for a future article on Phoronix.

Compare your own system(s) to this result file with the Phoronix Test Suite by running the command: phoronix-test-suite benchmark 2501264-PTS-LLAMACPP76
Jump To Table - Results

View

Do Not Show Noisy Results
Do Not Show Results With Incomplete Data
Do Not Show Results With Little Change/Spread
List Notable Results
Show Result Confidence Charts
Allow Limiting Results To Certain Suite(s)

Statistics

Show Overall Harmonic Mean(s)
Show Overall Geometric Mean
Show Wins / Losses Counts (Pie Chart)
Normalize Results
Remove Outliers Before Calculating Averages

Graph Settings

Force Line Graphs Where Applicable
Convert To Scalar Where Applicable
Disable Color Branding
Prefer Vertical Bar Graphs
No Box Plots
On Line Graphs With Missing Data, Connect The Line Gaps

Multi-Way Comparison

Condense Multi-Option Tests Into Single Result Graphs

Table

Show Detailed System Result Table

Sensor Monitoring

Show Accumulated Sensor Monitoring Data For Displayed Results
Generate Power Efficiency / Performance Per Watt Results

Run Management

Highlight
Result
Toggle/Hide
Result
Result
Identifier
Performance Per
Dollar
Date
Run
  Test
  Duration
RTX 3090
January 26
  19 Minutes
RTX 4070
January 25
  27 Minutes
RTX 4070 SUPER
January 25
  26 Minutes
RTX 4070 Ti SUPER
January 25
  22 Minutes
RTX 4080
January 26
  20 Minutes
RTX 4080 SUPER
January 25
  19 Minutes
RTX 4090
January 26
  15 Minutes
RTX 5090
January 25
  12 Minutes
Invert Behavior (Only Show Selected Data)
  20 Minutes

Only show results where is faster than
Only show results matching title/arguments (delimit multiple options with a comma):
Do not show results matching title/arguments (delimit multiple options with a comma):


Llama.cpp NVIDIA GeForce RTX 5090OpenBenchmarking.orgPhoronix Test SuiteIntel Core Ultra 9 285K @ 5.10GHz (24 Cores)ASUS ROG MAXIMUS Z890 HERO (1203 BIOS)Intel Device ae7f2 x 16GB DDR5-6400MT/s Micron CP16G64C38U5B.M8D14001GB Western Digital WD_BLACK SN850X 4000GB + 1000GB Western Digital WDS100T1X0E-00AFY0ASUS NVIDIA GeForce RTX 3090 24GBASUS NVIDIA GeForce RTX 4070 12GBASUS NVIDIA GeForce RTX 4070 SUPER 12GBASUS NVIDIA GeForce RTX 4070 Ti SUPER 16GBASUS NVIDIA GeForce RTX 4080 16GBASUS NVIDIA GeForce RTX 4080 SUPER 16GBASUS NVIDIA GeForce RTX 4090 24GBASUS NVIDIA GeForce RTX 5090 32GBIntel Device 7f50ASUS VP28URealtek Device 8126 + Intel I226-V + Intel Wi-Fi 7Ubuntu 24.106.11.0-13-generic (x86_64)GNOME Shell 47.0X Server 1.21.1.13NVIDIA 570.86.104.6.0OpenCL 3.0 CUDA 12.8.51 + OpenCL 3.0GCC 14.2.0 + CUDA 12.8ext43840x2160ProcessorMotherboardChipsetMemoryDiskGraphicsAudioMonitorNetworkOSKernelDesktopDisplay ServerDisplay DriverOpenGLOpenCLCompilerFile-SystemScreen ResolutionLlama.cpp NVIDIA GeForce RTX 5090 BenchmarksSystem Logs- nouveau.modeset=0 - Transparent Huge Pages: madvise- --build=x86_64-linux-gnu --disable-vtable-verify --disable-werror --enable-bootstrap --enable-cet --enable-checking=release --enable-clocale=gnu --enable-default-pie --enable-gnu-unique-object --enable-languages=c,ada,c++,go,d,fortran,objc,obj-c++,m2,rust --enable-libphobos-checking=release --enable-libstdcxx-backtrace --enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-link-serialization=2 --enable-multiarch --enable-multilib --enable-nls --enable-objc-gc=auto --enable-offload-defaulted --enable-offload-targets=nvptx-none=/build/gcc-14-zdkDXv/gcc-14-14.2.0/debian/tmp-nvptx/usr,amdgcn-amdhsa=/build/gcc-14-zdkDXv/gcc-14-14.2.0/debian/tmp-gcn/usr --enable-plugin --enable-shared --enable-threads=posix --host=x86_64-linux-gnu --program-prefix=x86_64-linux-gnu- --target=x86_64-linux-gnu --with-abi=m64 --with-arch-32=i686 --with-build-config=bootstrap-lto-lean --with-default-libstdcxx-abi=new --with-gcc-major-version-only --with-multilib-list=m32,m64,mx32 --with-target-system-zlib=auto --with-tune=generic --without-cuda-driver -v - Scaling Governor: intel_pstate performance (EPP: default) - CPU Microcode: 0x114 - Thermald 2.5.8- gather_data_sampling: Not affected + itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + reg_file_data_sampling: Not affected + retbleed: Not affected + spec_rstack_overflow: Not affected + spec_store_bypass: Mitigation of SSB disabled via prctl + spectre_v1: Mitigation of usercopy/swapgs barriers and __user pointer sanitization + spectre_v2: Mitigation of Enhanced / Automatic IBRS; IBPB: conditional; RSB filling; PBRSB-eIBRS: Not affected; BHI: Not affected + srbds: Not affected + tsx_async_abort: Not affected

RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 5090Result OverviewPhoronix Test Suite100%149%198%247%296%Llama.cppLlama.cppLlama.cppLlama.cppLlama.cppLlama.cppLlama.cppLlama.cppNVIDIA CUDA - Mistral-7B-Instruct-v0.3-Q8_0 - P.P.1NVIDIA CUDA - Mistral-7B-Instruct-v0.3-Q8_0 - P.P.2NVIDIA CUDA - Llama-3.1-Tulu-3-8B-Q8_0 - P.P.2NVIDIA CUDA - Llama-3.1-Tulu-3-8B-Q8_0 - P.P.1NVIDIA CUDA - Llama-3.1-Tulu-3-8B-Q8_0 - T.G.1NVIDIA CUDA - Mistral-7B-Instruct-v0.3-Q8_0 - T.G.1NVIDIA CUDA - Mistral-7B-Instruct-v0.3-Q8_0 - P.P.5NVIDIA CUDA - Llama-3.1-Tulu-3-8B-Q8_0 - P.P.5

Llama.cpp NVIDIA GeForce RTX 5090llama-cpp: NVIDIA CUDA - Llama-3.1-Tulu-3-8B-Q8_0 - Text Generation 128llama-cpp: NVIDIA CUDA - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 512llama-cpp: NVIDIA CUDA - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 1024llama-cpp: NVIDIA CUDA - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 2048llama-cpp: NVIDIA CUDA - Mistral-7B-Instruct-v0.3-Q8_0 - Text Generation 128llama-cpp: NVIDIA CUDA - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 512llama-cpp: NVIDIA CUDA - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 1024llama-cpp: NVIDIA CUDA - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 2048RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 509090.885030.784784.574437.5795.295046.024783.194437.4954.474502.644198.293798.0657.364533.014212.673803.4654.595419.574997.774436.2957.485475.925024.484452.7570.836328.825889.855301.3574.556391.675924.065313.4074.567606.747034.856228.2878.507685.777067.756246.6977.087782.937201.826380.3481.177862.607236.306410.73100.5111675.0810889.089573.68105.6811826.4110961.069593.96158.7712780.5412385.3411211.19166.1212884.4612462.0311240.15OpenBenchmarking.org

Llama.cpp

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4397Backend: NVIDIA CUDA - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Text Generation 128RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 50904080120160200SE +/- 0.06, N = 3SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.02, N = 3SE +/- 0.01, N = 3SE +/- 0.00, N = 3SE +/- 0.02, N = 3SE +/- 0.06, N = 390.8854.4754.5970.8374.5677.08100.51158.771. (CXX) g++ options: -O3

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4397Backend: NVIDIA CUDA - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 512RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 50903K6K9K12K15KSE +/- 2.36, N = 5SE +/- 0.83, N = 5SE +/- 0.71, N = 5SE +/- 0.97, N = 6SE +/- 1.40, N = 6SE +/- 1.43, N = 6SE +/- 4.15, N = 7SE +/- 9.49, N = 85030.784502.645419.576328.827606.747782.9311675.0812780.541. (CXX) g++ options: -O3

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4397Backend: NVIDIA CUDA - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 1024RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 50903K6K9K12K15KSE +/- 4.02, N = 3SE +/- 3.12, N = 3SE +/- 2.88, N = 3SE +/- 3.71, N = 4SE +/- 4.02, N = 4SE +/- 0.54, N = 4SE +/- 2.66, N = 5SE +/- 7.60, N = 64784.574198.294997.775889.857034.857201.8210889.0812385.341. (CXX) g++ options: -O3

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4397Backend: NVIDIA CUDA - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 2048RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 50902K4K6K8K10KSE +/- 6.70, N = 3SE +/- 3.08, N = 3SE +/- 2.42, N = 3SE +/- 4.23, N = 3SE +/- 2.85, N = 3SE +/- 2.45, N = 3SE +/- 6.13, N = 3SE +/- 5.46, N = 44437.573798.064436.295301.356228.286380.349573.6811211.191. (CXX) g++ options: -O3

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4397Backend: NVIDIA CUDA - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Text Generation 128RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 50904080120160200SE +/- 0.02, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.02, N = 3SE +/- 0.03, N = 395.2957.3657.4874.5578.5081.17105.68166.121. (CXX) g++ options: -O3

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4397Backend: NVIDIA CUDA - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 512RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 50903K6K9K12K15KSE +/- 4.47, N = 5SE +/- 1.08, N = 5SE +/- 1.14, N = 5SE +/- 0.47, N = 6SE +/- 1.46, N = 6SE +/- 1.63, N = 6SE +/- 2.51, N = 8SE +/- 4.77, N = 85046.024533.015475.926391.677685.777862.6011826.4112884.461. (CXX) g++ options: -O3

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4397Backend: NVIDIA CUDA - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 1024RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 50903K6K9K12K15KSE +/- 5.32, N = 3SE +/- 3.38, N = 3SE +/- 2.97, N = 3SE +/- 4.77, N = 4SE +/- 3.16, N = 4SE +/- 1.89, N = 4SE +/- 2.42, N = 5SE +/- 7.53, N = 64783.194212.675024.485924.067067.757236.3010961.0612462.031. (CXX) g++ options: -O3

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4397Backend: NVIDIA CUDA - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 2048RTX 3090RTX 4070RTX 4070 SUPERRTX 4070 Ti SUPERRTX 4080RTX 4080 SUPERRTX 4090RTX 50902K4K6K8K10KSE +/- 3.07, N = 3SE +/- 2.69, N = 3SE +/- 2.58, N = 3SE +/- 3.82, N = 3SE +/- 3.30, N = 3SE +/- 1.36, N = 3SE +/- 3.44, N = 3SE +/- 5.63, N = 44437.493803.464452.755313.406246.696410.739593.9611240.151. (CXX) g++ options: -O3