NVIDIA GH200 Grace Hopper, AI & Machine Learning Benchmarks & Specs

141GB · AI Score 65.6/100 · anchored estimate vs 51 measured cards

65.6 AI Score Includes estimates

We have not run NVIDIA GH200 Grace Hopper on our bench. These figures are anchored estimates, interpolated per workload against the 51 GPUs we did measure (confidence: high (sibling silicon)). On Llama 3.1 8B (Q4_K_M) NVIDIA GH200 Grace Hopper should deliver about 273.9 tokens/sec. Stepping up to Qwen3 32B it should hold roughly 78.2 tok/s. The full Llama 3.3 70B still runs, at about 43.5 tok/s. For image generation, SDXL should run near 18.58 it/s, and FLUX.1-dev at 4.44 it/s. All 12 workloads fit in 141GB. There is no model in our suite this card has to turn down.

AI & Machine Learning benchmark results

Text Generation tok/s 5

Qwen3 4B325.5
Llama 3.1 8B273.9
Qwen2.5-Coder 14B151.5
Qwen3 32B78.2
Llama 3.3 70B43.5
WorkloadResultTelemetryData
Qwen3 4B325.5 tok/sestimatedEst.
Llama 3.1 8B273.9 tok/sestimatedEst.
Qwen2.5-Coder 14B151.5 tok/sestimatedEst.
Qwen3 32B78.2 tok/sestimatedEst.
Llama 3.3 70B43.5 tok/sestimatedEst.

Image Generation images/min 3

Stable Diffusion XL37.16
Z-Image Turbo23.175
FLUX.1 dev9.514
WorkloadResultTelemetryData
Stable Diffusion XL37.16 images/minestimatedEst.
Z-Image Turbo23.18 images/minestimatedEst.
FLUX.1 dev9.51 images/minestimatedEst.

Image Editing images/min 2

WorkloadResultTelemetryData
FLUX.1 Kontext dev4.39 images/minestimatedEst.
Qwen-Image-Edit3.76 images/minestimatedEst.

Video Generation frames/s 2

WorkloadResultTelemetryData
LTX-Video (distilled)17.87 frames/sestimatedEst.
Wan 2.2 5B (720p)1.41 frames/sestimatedEst.
How this estimate is derived. This card hasn’t been through our bench yet, so its numbers are anchored estimates, interpolated from the 51 first-party measured cards (Sibling to measured H200: same GH100-class GPU (141GB HBM3e, 4.9 vs 4.8 TB/s) → ×~1.0). The VRAM “won’t fit” gates are exact, since they’re pure capacity limits. Confidence: high (sibling silicon). Estimates are replaced with measured data as more silicon goes through the bench. Full methodology →

NVIDIA GH200 Grace Hopper specifications

ArchitectureHopper
CUDA cores16,896
VRAM141GB HBM3e
Memory bus6144-bit
Memory bandwidth4900 GB/s
Boost clock1,980 MHz
TDP700 W
ProcessTSMC 4N
InterfaceSXM (Grace superchip)
Release date2024-02-01
Launch MSRP$40,000

Verdict, nothing in our suite slows it down

NVIDIA GH200 Grace Hopper scores 65.6/100, #5 of 102. It ran all 12 workloads. Figures are anchored estimates, not measurements, we flag that on every row.

Relative performance: where the NVIDIA GH200 Grace Hopper lands

100% = this card, AI & Machine Learning headline metric (AI Score). #5 of 21 desktop cards in this vertical.

GPURelative%AI Score
NVIDIA B300
143%93.8
NVIDIA B200
119%78
NVIDIA B100
102%67
NVIDIA H100 NVL
102%67
NVIDIA GH200 Grace Hopper
100%65.6
NVIDIA H200
99%65
NVIDIA H100 80GB HBM3
95%62.6
NVIDIA H800 80GB
95%62.6
NVIDIA RTX PRO 6000 Blackwell Server Edition
76%49.9

← All AI & Machine Learning GPU rankings

What this card can build

Whole-job timings, composed from our measured per-model results on this card.

WorkflowTimeEnergyBasis
24-frame storyboard80 sn/aestimate, 0 of 2 stages measured
60-second AI short film2 minn/aestimate, 0 of 3 stages measured
6-panel comic page2.1 minn/aestimate, 0 of 3 stages measured
Character sheet, 12 poses2.8 minn/aestimate, 0 of 2 stages measured
10 short social clips6.3 minn/aestimate, 0 of 3 stages measured
Full codebase review6.6 minn/aestimate, 0 of 1 stage measured
40-product photo shoot10.2 minn/aestimate, 0 of 2 stages measured
20 long-form articles10.7 minn/aestimate, 0 of 1 stage measured
100-photo restoration batch22.8 minn/aestimate, 0 of 1 stage measured