NVIDIA RTX 5000 Ada and NVIDIA L40 Release Date
NVIDIA Announced at SIGGRAPH 2023 the release of 3 new NVIDIA RTX professional GPUs and a data center GPU. The RTX 5000 Ada and NVIDIA L40S general availability starts August 8, 2023, and the RTX 4500 and RTX 4000 Ada will release later this year.
Let's go over the tech and specs of the GPUs announced and what to expect from them.
NVIDIA RTX 5000 and NVIDIA L40S Specs
RTX 5000 Ada replaces the RTX A5500 with more GPU memory, CUDA Cores, RT Cores, and Tensor Cores. The Ada Generation architecture and more memory enable the RTX 5000 Ada to provide ample computing for building highly complex 3D models, render dense 3D environments, and run generative AI smoother to increase workflow. NVIDIA claims a substantial gen-over-gen increase over the RTX 5500:
- 1.5x speedup in AI training
- 2x more throughput in AI inferencing
- 2x performance uplift in graphics and rendering.
The L40S is the boosted version of the earlier released NVIDIA L40. With an increase in TDP the NVIDIA L40S is dubbed the most powerful universal GPU. With an increased demand for AI accelerators, the L40S demonstrates competitive performance to the last generation flagship NVIDIA A100. The L40S benefits by being the GPU accelerator for ease of implementation for any user to get started and boost their computing power to handle any workload. The L40S is perfect for small to medium-scale operations for AI training, LLMs, and environments that employ multiple workloads across their infrastructure. Over the A100 the L40S delivers:
- 1.2x more performance in Generative AI workloads
- 1.7x higher AI inferencing throughput
- 1.2x better performance for LLM models.
The L40S also includes Encode and Decode engines as well as display outputs for 3D media, an aspect absent from NVIDIA A100. Specs of RTX 5000 and L40S as follows:
Model/Specs | RTX 5000 | NVIDIA L40S |
---|---|---|
GPU Memory | 32GB GDDR6 ECC | 48GB GDDR6 |
Bus Interface | 256-bit | 384-bit |
Memory Bandwidth | 576 GB/s | 864 GB/s |
CUDA Cores | 12,800 | 18,176 |
Ray Tracing Cores (3rd Gen) | 100 | 142 |
Tensor Cores (4th Gen) | 400 | 568 |
Single Precision Perf. | 65.3 TFLOPS | 91.6 TFLOPS |
RT Core Performance | 151.0 TFLOPS | 212.0 TFLOPS |
Tensor Core Performance (FP8) | 1044.4 TFLOPS | 1466 TFLOPS |
Encode Decode | 2x NVENC, 2x NVDEC (+AV1 encode & decode) | 3x NVENC, 3x NVDEC (+AV1 E&D) |
System Interconnect | PCIe 4.0 x16 | PCIe 4.0 x16 |
Form Factor | Dual Slot, Full Length | Dual Slot, Full Length |
Power Consumption | 250W | 350W |
NVIDIA RTX 4500 and RTX 4000 Ada Generation
RTX A4500 are very popular high-performance professional GPUs for executing HPC workloads that don’t require a large memory size. The RTX 4500 Ada looks to slot right in to satisfy that exact market workload. The gen-over-gen performance uplift enables designers and creators to utilize generative AI to their advantage. While the RTX 4500 Ada is a dual slot card the RTX 4000 Ada is a single slot card that operates in the same workload range with slightly lower performance. However, the single-slot design enables the flexibility of deployment in small and thinner systems like 1U servers.
RTX 4500 Ada’s performance gains over the RTX A4500 is as follows:
- 1.6x increase in performance in 3D graphics
- 2.7x increase in performance when running NVIDIA Omniverse rendering workloads
- 1.5x performance gain in classic rendering
- 1.5x uplift in throughput for AI training and inferencing
RTX 4000 Ada’s performance gains over the RTX A4000 are as follows:
- 1.5x performance uplift in 3D graphics
- 1.7x increase in classic rendering performance
- 1.2x boost when running generative AI
RTX 4500 and RTX 4000 Ada Generation GPU specifications are as follows:
Model/Specs | RTX 4500 | RTX 4000 |
---|---|---|
GPU Memory | 24GB GDDR6 ECC | 20GB GDDR6 ECC |
Bus Interface | 192-bit | 160-bit |
Memory Bandwidth | 432 GB/s | 360 GB/s |
CUDA Cores | 7,680 | 6,144 |
Ray Tracing Cores (3rd Gen) | 60 | 48 |
Tensor Cores (4th Gen) | 240 | 192 |
Single Precision Perf. | 39.6 TFLOPS | 26.7 TFLOPS |
RT Core Performance | 91.6 TFLOPS | 61.8 TFLOPS |
Tensor Core Performance (FP8) | 634.0 TFLOPS | 427.6 TFLOPS |
Encode Decode | 2x NVENC, 2x NVDEC (+AV1 encode & decode) | 2x NVENC, 2x NVDEC (+AV1 encode & decode) |
System Interconnect | PCIe 4.0 x16 | PCIe 4.0 x16 |
Form Factor | Dual Slot, Full Length | Single Slot, Full Length |
Power Consumption | 210W | 130W |
Buy NVIDIA RTX 5000 Ada and NVIDIA L40S. RTX 4500 and RTX 4000 Coming Soon
At SabrePC we want to deliver the very best solution for the best price to fit your needs optimally. Our customizable solutions are configurable with the NVIDIA RTX 5000 Ada for peak HPC acceleration and our servers can be outfitted with NVIDIA L40S for your AI and 3D workloads today! Contact SabrePC to learn more about how you can get the latest NVIDIA GPUs.
The RTX 4500 and RTX 4000 Ada are not yet released for general availability coming soon later this fall. If you would like to keep in touch to build a system with these not yet released NVIDIA GPUs, contact our team to stay in touch. Looking to boost your computing today? Check out SabrePC’s dedicated server platforms for training Deep Learning AI workloads.