2025-03-04

NVIDIA GeForce RTX 5070 Linux GPU Compute Performance

Ahead of NVIDIA GeForce RTX 5070 graphics cards seeing retail availability tomorrow, today the review embargo expires on the GeForce RTX 5070 Founders Edition graphics cards. I’ve been testing out the GeForce RTX 5070 under Linux and today have a number of GPU compute benchmarks to share.

The GeForce RTX 5070 graphics card is NVIDIA’s new $549 graphics card offering and comes a few weeks after the GeForce RTX 5070 Ti introduction. The GeForce RTX 5070 will be competing head-to-head with the AMD Radeon RX 9070 series that will be available later this week with the Radeon RX 9070 expected to retail starting out at $549 and the Radeon RX 9070 XT at $599. It will be interesting to put the RX 9070 and RTX 5070 head-to-head on Linux but for now it’s just the RTX 5070 review embargo expiry.

Additionally, this initial round of Linux benchmarking for the GeForce RTX 5070 is only looking at the GPU compute performance. Similar to the situation with the recent GeForce RTX 5080 and RTX 5090 launches, no Linux “press” driver was available in advance of today’s embargo expiry / launch. The NVIDIA 570 series does recognize the GeForce RTX 5070 but lacks various optimizations and known 3D graphics/gaming performance handicaps. So only after the next NVIDIA Linux driver release with official RTX 5070 support is it recommended for Linux gaming use. But in any event for those curious about the Linux performance potential, today are some GPU compute numbers.

The GeForce RTX 5070 features 6,144 CUDA cores, down from the 8,960 CUDA cores with the recently launched RTX 5070 Ti. The RTX 5070 has a 2.33GHz base clock with 2.51GHz boost clock, 12GB of GDDR7 video memory with a 192-bit bus, and is rated for 250 Watt total graphics power. The RTX 5070 requires a PCIe Gen 5 power cable or via the included power adapter dual PCIe 8-pin cables.

There are three DisplayPort and one HDMI output with the GeForce RTX 5070 Founders Edition graphics card. Aside from the cut-down GPU engine specs compared to the RTX 5070 Ti, the RTX 5070 has all the same common NVIDIA GPU features for the Blackwell GPU family.

The graphics cards tested for today’s Linux GPU compute comparison included:

- RTX 2070
- RTX 2070 SUPER
- RTX 2080
- RTX 2080 SUPER
- RTX 2080 Ti
- TITAN RTX
- RTX 3070
- RTX 3070 Ti
- RTX 3080
- RTX 3090
- RTX 4070
- RTX 4070 SUPER
- RTX 4070 Ti SUPER
- RTX 4080
- RTX 4080 SUPER
- RTX 4090
- RTX 5070
- RTX 5080
- RTX 5090

NVIDIA GeForce RTX 5070 Linux Compute Benchmarks

I wasn’t seeded with any GeForce RTX 5070 Ti review sample and thus why no RTX 5070 Ti is part of the benchmark comparison. With the Radeon RX 9070 graphics card launch I’ll be working on some benchmarks with a sub-set of the compute tests that can work with the Radeon graphics stack for those wondering how the RTX 5070 performance compares to the competition while this article is a generational look with CUDA and other compute workloads. Again, Linux gaming benchmarks will come upon the proper RTX 5070 Linux driver release.

Blender benchmark with settings of Blend File: Classroom, Compute: NVIDIA OptiX. RTX 5090 was the fastest.

Blender benchmark with settings of Blend File: Pabellon Barcelona, Compute: NVIDIA OptiX. RTX 5090 was the fastest.

With Blender 4.3’s NVIDIA OptiX back-end the GeForce RTX 5070 on Linux was performing between the speed of an RTX 4070 SUPER and RTX 4070 Ti SUPER. Over the RTX 4070 (non-SUPER) were at least some nice performance improvements.

Blender benchmark with settings of Blend File: Barbershop, Compute: NVIDIA OptiX. RTX 5090 was the fastest.

On a performance-per-Watt basis, the RTX 5070 was comparable to the RTX 4070 series just as the other Blackwell cards were relative to the RTX 40 Ada GPU equivalents.

Chaos Group V-RAY benchmark with settings of Mode: NVIDIA RTX GPU. RTX 5090 was the fastest.

With the proprietary V-RAY renderer, the RTX 5070 performance matched the RTX 4070 Ti SUPER while enjoying better power efficiency.

IndigoBench benchmark with settings of Acceleration: OpenCL GPU, Scene: Supercar. RTX 5090 was the fastest.

IndigoBench benchmark with settings of Acceleration: OpenCL GPU, Scene: Bedroom. RTX 5090 was the fastest.

But with the IndigoBench renderer the RTX 5070 performance only came out comparable to the prior RTX 4070 GPU for this OpenCL-based renderer.

SHOC Scalable HeterOgeneous Computing benchmark with settings of Target: OpenCL, Benchmark: GEMM SGEMM_N. RTX 5090 was the fastest.

SHOC Scalable HeterOgeneous Computing benchmark with settings of Target: OpenCL, Benchmark: S3D. RTX 5090 was the fastest.

clpeak benchmark with settings of OpenCL Test: Global Memory Bandwidth. RTX 5090 was the fastest.

clpeak benchmark with settings of OpenCL Test: Single-Precision Compute. RTX 5090 was the fastest.

clpeak benchmark with settings of OpenCL Test: Double-Precision Compute. RTX 5090 was the fastest.

clpeak benchmark with settings of OpenCL Test: Integer Compute. RTX 5090 was the fastest.

There was decent generational uplift from the RTX 4070 to RTX 5070 in a number of the OpenCL benchmarks.

cl-mem benchmark with settings of Benchmark: Read. RTX 5090 was the fastest.

cl-mem benchmark with settings of Benchmark: Write. RTX 5090 was the fastest.

FinanceBench benchmark with settings of Benchmark: Black-Scholes OpenCL. RTX 5090 was the fastest.

FinanceBench benchmark with settings of Benchmark: Monte-Carlo OpenCL. RTX 5090 was the fastest.

Most often the RTX 5070 was delivering performance similar to the RTX 4070 SUPER to RTX 4070 Ti SUPER.

FluidX3D benchmark with settings of Test: FP32-FP32. RTX 5090 was the fastest.

FluidX3D benchmark with settings of Test: FP32-FP16C. RTX 5090 was the fastest.

FluidX3D benchmark with settings of Test: FP32-FP16S. RTX 5090 was the fastest.

With the FluidX3D OpenCL CFD software the GeForce RTX 5070 was able to surpass the GeForce RTX 4080 performance.

GpuOwl benchmark with settings of Exponent: 57885161. RTX 5090 was the fastest.

GpuOwl benchmark with settings of Exponent: 77936867. RTX 5090 was the fastest.

GpuOwl benchmark with settings of Exponent: 332220523. RTX 5090 was the fastest.

GpuOwl meanwhile only saw small gains over the GeForce RTX 4070 when using the new RTX 5070 Blackwell graphics card.

Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Text Generation 128. RTX 5090 was the fastest.

Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 512. RTX 5090 was the fastest.

With Llama.cpp there was decent uplift from the RTX 4070 to RTX 5070.

GPU Power Consumption Monitoring Overview benchmark with settings of Accumulated GPU Power Consumption Monitoring.

Across all of the GPU compute benchmarks conducted, the RTX 5070 was consuming similar levels of power to the RTX 4070 and RTX 4070 SUPER graphics cards.

GPU Temp Monitoring Overview benchmark with settings of Accumulated GPU Temp Monitoring.

The GeForce RTX 5070 Founder’s Edition graphics card was also performing well thermal-wise and on-par with the other graphics cards tested.

Geometric Mean Of All Test Results benchmark with settings of Result Composite, NVIDIA GeForce RTX 5070 Linux Compute Benchmarks. RTX 5090 was the fastest.

On the whole with the geometric mean of all the benchmarks carried out for this Linux GPU compute comparison, the RTX 5070 performed similar to the RTX 4070 SUPER but depending upon the particular workload was capable of outperforming the RTX 4080 in some instances. The RTX 4070 SUPER had launched last year at $599 USD so at least the $549 list price for the RTX 5070 is an improvement. The upgrade story is much more compelling if you are currently on NVIDIA RTX 30 class hardware or older. Stay tuned for more Linux (gaming) benchmarks of the GeForce RTX 5070 in the coming days once a proper game ready driver is available. Thanks to NVIDIA for providing this review sample for launch-day Linux testing on Phoronix.