NVIDIA Quadro GV100

The NVIDIA Quadro GV100 is a professional-grade GPU based on the Volta architecture, manufactured on TSMC's 12 nm process with 21,100 million transistors on an 815 mm² die. It features 5,120 shading units, 320 texture mapping units, 128 raster output units, and 640 tensor cores, delivering 16.66 TFLOPS of FP32 and 33.32 TFLOPS of FP16 (2:1) compute. Memory consists of 32 GB of HBM2 across a 4096-bit interface, providing 868.4 GB/s of bandwidth. With an average benchmark score of 34,677, the card sits in the 78th percentile of all GPUs. Its launch MSRP was 8,999 USD. The card uses a dual-slot design, requires a 600 W power supply, and is powered by a single 8-pin connector. It is now end-of-life, having been released on 2018-03-26.

Benchmark Performance

The GV100's benchmark results reveal a compute-oriented design. In Geekbench OpenCL, it scores 144,393, and in Geekbench Vulkan it reaches 137,547, indicating strong performance in parallel compute and modern graphics APIs. Passmark's G3D score is 19,650, with a GPU compute score of 9,069. However, the DirectX-specific tests are notably lower: DirectX 9 scores 207, DirectX 11 scores 168, DirectX 10 scores 140, and DirectX 12 scores only 84. This pattern suggests the card is not optimized for legacy or gaming-focused DirectX workloads, but rather for compute-heavy tasks that leverage OpenCL, Vulkan, or its tensor cores.

The overall average benchmark score of 34,677 places the GV100 within a tight performance band. It trails the AMD Radeon RX 9070 by 0.3% (34,780 vs. 34,677) and the NVIDIA A2 by 0.5% (34,866 vs. 34,677). Conversely, it leads the AMD Radeon HD 7970 by 0.4% (34,541 vs. 34,677) and the AMD Radeon PRO W6400 by 0.5% (34,511 vs. 34,677). These deltas are all within half a percent, meaning that across the aggregate benchmark suite, the GV100's performance is essentially indistinguishable from these rivals. Yet its underlying architecture and feature set differ substantially, as seen in the low DirectX scores and the presence of tensor cores.

The 78th percentile ranking indicates that the GV100 outperforms the majority of GPUs in the database, but it does not sit at the top. The combination of high compute scores and low DirectX results suggests that the card's value lies in professional applications rather than consumer gaming. The FP32 throughput of 16.66 TFLOPS and FP16 throughput of 33.32 TFLOPS (2:1) further reinforce this compute-first positioning.

How It Compares

AMD Radeon RX 9070 — The RX 9070 posts an average score of 34,780, which is 0.3% higher than the GV100's 34,677. This is a negligible margin, indicating that the two cards deliver comparable aggregate performance in the benchmark suite. However, the RX 9070 is a consumer gaming card, while the GV100 targets professional workloads, so their real-world usage patterns diverge significantly despite similar overall scores.

AMD Radeon HD 7970 — The HD 7970 averages 34,541, which the GV100 surpasses by 0.4%. The HD 7970 is a much older architecture, yet the performance gap is minimal. This highlights how the GV100's compute-focused design does not translate into a large lead over older hardware in general-purpose benchmarks. The GV100's advantage would likely be more pronounced in workloads that utilize its tensor cores or large memory capacity.

AMD Radeon PRO W6400 — The PRO W6400 scores 34,511 on average, placing it 0.5% behind the GV100. Both are professional cards, but the W6400 is a lower-tier product. The small delta suggests that in this benchmark suite, the GV100's extra compute and memory resources do not yield a significant aggregate advantage, possibly because the tests are not fully stress the GV100's capabilities.

NVIDIA A2 — The A2 averages 34,866, which is 0.5% higher than the GV100. The A2 is a compact, low-power accelerator, yet it edges out the GV100 in the overall score. This is surprising given the GV100's larger die and higher power envelope (250 W TDP vs. the A2's unspecified). The close result underscores that aggregate benchmarks can mask architectural differences; the GV100's 32 GB VRAM and tensor cores are not fully exercised by these tests.

Who Should Consider It

The GV100 is not a card for gamers. Its DirectX scores are among the lowest in its class — DirectX 12 at 84, DirectX 11 at 168 — which would result in poor frame rates in modern titles. Instead, the card is suited for professionals who need massive memory capacity and high compute throughput. The 32 GB HBM2 memory and 868.4 GB/s bandwidth are ideal for large datasets, high-resolution textures, and complex simulations. The 640 tensor cores provide hardware acceleration for AI inference and training, particularly with FP16 workloads that double to 33.32 TFLOPS.

For users running OpenCL or Vulkan applications, the GV100's scores of 144,393 and 137,547 respectively demonstrate strong performance. The card also supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4, making it compatible with modern graphics APIs, though its raw DirectX rasterization performance is limited. The absence of dedicated ray tracing cores (RT core count is null) means ray tracing must be handled via compute shaders, which is less efficient than on cards with dedicated RT hardware.

Given its end-of-life status, the GV100 is a legacy product. It is best suited for users who require its specific combination of high-capacity memory, tensor cores, and professional driver support, and who are working with software that leverages OpenCL, Vulkan, or CUDA-based compute. The card's 250 W TDP and 600 W suggested PSU are moderate for a workstation GPU, and its dual-slot, 267 mm length fits standard chassis.

FAQ

Q: What is the average benchmark score of the NVIDIA Quadro GV100?

A: The average benchmark score is 34,677, which places it in the 78th percentile of all GPUs.

Q: How much memory does the GV100 have and what is its bandwidth?

A: It has 32 GB of HBM2 memory on a 4096-bit bus, providing 868.4 GB/s of bandwidth.

Q: Does the GV100 support DirectX 12?

A: Yes, it supports DirectX 12 (12_1), along with OpenGL 4.6 and Vulkan 1.4.

Q: How many tensor cores does the GV100 have?

A: It has 640 tensor cores, which enable FP16 compute at 33.32 TFLOPS (2:1) and FP32 at 16.66 TFLOPS.

Q: What is the power consumption of the GV100?

A: The TDP is 250 W, with a suggested PSU of 600 W and a single 8-pin power connector.

Q: When was the GV100 released?

A: It was released on 2018-03-26 and is now end-of-life.

Ray Tracing and Feature Set

The GV100 does not include dedicated ray tracing cores; the RT core count is null. This means hardware-accelerated ray tracing is not available. Instead, the card relies on its 640 tensor cores for AI-accelerated tasks, such as denoising or inference. The API support includes DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4, which allow for modern graphics features, but ray tracing would have to be implemented via compute shaders, incurring a performance penalty compared to dedicated RT hardware. The card also supports a 2:1 FP16 ratio, effectively doubling throughput for workloads that use half-precision arithmetic. The pixel rate is 208.3 GPixel/s and the texture rate is 520.6 GTexel/s, indicating strong rasterization capabilities for professional rendering, though the DirectX benchmark scores suggest that driver optimizations are not aligned with gaming workloads.

Memory Subsystem

The GV100 is equipped with 32 GB of HBM2 memory, connected via a 4096-bit bus. The memory clock runs at 848 MHz, with an effective data rate of 1696 Mbps, yielding a total bandwidth of 868.4 GB/s. This is an exceptionally high bandwidth figure, well-suited for memory-intensive applications such as large-scale scientific simulations, high-resolution rendering, and deep learning datasets. The 32 GB capacity allows loading massive models or textures that would exceed the memory of most consumer cards. For high-resolution work, the combination of capacity and bandwidth ensures that data can be fed to the GPU without bottlenecks. However, the memory clock and effective rate are modest compared to newer HBM3 implementations, but for a 2018 product, they were competitive. The card's memory subsystem is a key differentiator, as few rivals offer similar capacity and bandwidth in a single GPU.