NVIDIA TITAN Xp, AI & Machine Learning Benchmarks & Specs

12GB · AI Score 2.5/100 · anchored estimate vs 51 measured cards

2.3 AI Score Includes estimates

We have not run NVIDIA TITAN Xp on our bench. These figures are anchored estimates, interpolated per workload against the 51 GPUs we did measure. On Llama 3.1 8B (Q4_K_M) NVIDIA TITAN Xp should deliver about 69.5 tokens/sec. Llama 3.3 70B does not fit. It needs roughly 42GB and this card has 12GB. For image generation, SDXL should run near 0.9 it/s, while FLUX.1-dev won't fit at BF16 (needs ~26GB). 8 of the 12 workloads won't fit on 12GB at the tested precision, Qwen3 32B, Llama 3.3 70B, Z-Image Turbo, FLUX.1-dev and others. We publish those as hard gates rather than quietly dropping to a smaller quant.

AI & Machine Learning benchmark results

Text Generation tok/s 11

Qwen3-4B (power capped)75.21
Llama-3.1-8B (power capped)46.06
Qwen3 8B (power capped)45.11
Qwen2.5-Coder 14B36.2
Gemma 4 12B (power capped)29.41
Qwen3 14B (power capped)24.62
WorkloadResultTelemetryData
Qwen3-4B (power capped)75.21 tok/s
135 W47°CQ4_K_M
✓ Measured
Llama-3.1-8B (power capped)46.06 tok/s
143 W52°CQ4_K_M
✓ Measured
Qwen3 8B (power capped)45.11 tok/s
144 W53°CQ4_K_M
✓ Measured
Gemma 4 12B (power capped)29.41 tok/s
143 W54°CQ4_K_M
✓ Measured
Qwen2.5-Coder 14B36.2 tok/sestimatedEst.
Qwen3 14B (power capped)24.62 tok/s
147 W55°CQ4_K_M
✓ Measured
Gemma 4 26B A4B✕ Won't fit needs ~20 GBVRAM-gated at this precisionEst.
Qwen3 30B A3B✕ Won't fit needs ~20 GBVRAM-gated at this precisionEst.
Gemma 4 31B✕ Won't fit needs ~22 GBVRAM-gated at this precisionEst.
Qwen3 32B✕ Won't fit VRAM-gated at this precisionEst.
Llama 3.3 70B✕ Won't fit VRAM-gated at this precisionEst.

Image Generation images/min 3

WorkloadResultTelemetryData
Stable Diffusion XL1.8 images/minestimatedEst.
FLUX.1 dev✕ Won't fit VRAM-gated at this precisionEst.
Z-Image Turbo✕ Won't fit VRAM-gated at this precisionEst.

Image Editing images/min 2

WorkloadResultTelemetryData
FLUX.1 Kontext dev✕ Won't fit VRAM-gated at this precisionEst.
Qwen-Image-Edit✕ Won't fit VRAM-gated at this precisionEst.

Video Generation frames/s 2

WorkloadResultTelemetryData
LTX-Video (distilled)✕ Won't fit VRAM-gated at this precisionEst.
Wan 2.2 5B (720p)✕ Won't fit VRAM-gated at this precisionEst.
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 (Bandwidth Theil-Sen ladder off NVIDIA anchors × vendor factor (LLM ×0.65, diffusion ×0.3 (GP102 full die, above the 1080 Ti)); measured VRAM floors; recalibrated 2026-09-19 against published llama.cpp (CUDA/ROCm/Vulkan/SYCL) and Stable Diffusion results.). The VRAM “won’t fit” gates are exact, since they’re pure capacity limits. Estimates are replaced with measured data as more silicon goes through the bench. Full methodology →

NVIDIA TITAN Xp specifications

ArchitecturePascal (GP102)
CUDA cores3,840
VRAM12GB GDDR5X
Memory bus384-bit
Memory bandwidth547.7 GB/s
Boost clock1,582 MHz
TDP250 W
Process16nm
InterfacePCIe 3.0 x16
Release date2017-04-05
Launch MSRP$1,200

Verdict, capable, but 12GB sets the ceiling

NVIDIA TITAN Xp scores 2.5/100, #75 of 102. It ran 4 of 12; 8 exceeded its 12GB. Figures are anchored estimates, not measurements, we flag that on every row.

Relative performance: where the NVIDIA TITAN Xp lands

100% = this card, AI & Machine Learning headline metric (AI Score). #36 of 61 desktop cards in this vertical.

GPURelative%AI Score
AMD Radeon RX 6900 XT
100%2.3
AMD Radeon RX 6950 XT
100%2.3
GeForce GTX 1080 Ti
100%2.3
NVIDIA GeForce RTX 3080
100%2.3
NVIDIA TITAN Xp
100%2.3
AMD Radeon RX 7700 XT
96%2.2
NVIDIA GeForce RTX 2080 Ti Founders Edition
96%2.2
NVIDIA GeForce RTX 3070 Ti
96%2.2
NVIDIA GeForce RTX 2080 Super
91%2.1

Same card, other workloads: NVIDIA TITAN Xp Gaming benchmarks

← All AI & Machine Learning GPU rankings

The silicon

Transistors12,000 million
Die size471 mm²
Process node16 nm
Fabricated byTSMC
Transistor density25.5 million per mm²

Denser than 73% of the 746 cards we have silicon data for. Density is the clearest measure of what a process node bought: a card that gained it without growing the die got its speed from the fab rather than the architecture.

Silicon figures from Wikipedia (CC BY-SA 4.0). Benchmarks on this page are our own. Compare every chip.

What this card can build

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

WorkflowTimeEnergyBasis
Full codebase review27.6 minn/aestimate, 0 of 1 stage measured

Can't run: Product photo shoot (needs FLUX.1 Kontext dev), Photo restoration batch (needs FLUX.1 Kontext dev), Restore and enlarge photos (needs FLUX.1 Kontext dev), Character sheet, 12 poses (needs FLUX.1 dev), Short social clips (needs Qwen3 32B), 60-second AI short film (needs Qwen3 32B), 24-frame storyboard (needs Qwen3 32B), 6-panel comic page (needs Qwen3 32B), Long-form article batch (needs Llama 3.3 70B).

Rent or buy?

This card is $1,200 to buy. The cheapest listed rate on Vast.ai is $0.056/hour, but that is the floor: we budget $0.067/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 17,857 GPU-hours. Below it you are paying for idle silicon.

How you would use itGPU-hours a yearRental cost a yearTime to break even
2 hours a day, hobby730$4924.5 years
8 hours a day, working on it2,920$1966.1 years
24/7, always-on agent8,760$5892.0 years

At hobby usage this card is very unlikely to pay for itself before it is superseded. Rent it. Rental figures include a 20% premium over the cheapest listed rate. Ignores electricity, resale and the fact that a rented card can be a newer one tomorrow.

Rental price

$0.056/hr+14.3% since 2026-10-01low $0.049 · high $0.056

Cheapest of the RunPod and Vast on-demand rates we see, sampled daily. Spot and interruptible pricing runs lower.