This product can only be purchased as part of a custom PC build.

Card Specifications:
- ROPs (Render Output Units): 24
- SM Count (streaming multiprocessors): 132
- Shading Units (Cuda Cores): 16896
- Tensor Cores (AI Accelerators): 528
Memory Specifications
- Type: 141GB HBM3e
- Interface: 6144-bit
- Bandwidth: 4.89TB/s
Theoretical Performance
- Pixel Rate: 42.84 GPixel/s
- Texture Rate: 942.5 GTexel/s
- FP16 (half): 120.6 TFLOPS (2:1)
- FP32 (float): 60.32 TFLOPS
- FP64 (double): 30.16 TFLOPS (1:2)
No Display Output
TDP (Thermal Design Power): 600W
PSU Recommendation: 1200W+

PNY NVIDIA H200 NVL 141GB (Passive Cooling)

Card Specifications:
- ROPs (Render Output Units): 24
- SM Count (streaming multiprocessors): 132
- Shading Units (Cuda Cores): 16896
- Tensor Cores (AI Accelerators): 528
Memory Specifications
- Type: 141GB HBM3e
- Interface: 6144-bit
- Bandwidth: 4.89TB/s
Theoretical Performance
- Pixel Rate: 42.84 GPixel/s
- Texture Rate: 942.5 GTexel/s
- FP16 (half): 120.6 TFLOPS (2:1)
- FP32 (float): 60.32 TFLOPS
- FP64 (double): 30.16 TFLOPS (1:2)
No Display Output
TDP (Thermal Design Power): 600W
PSU Recommendation: 1200W+

$55,188.00

SKU: NVH200NVLTCGPU-KIT Category: Dedicated Video Card

$55,188.00

Made in Canada
Tested for performance
Three year warranty on parts and labor

Card Specifications:
- ROPs (Render Output Units): 24
- SM Count (streaming multiprocessors): 132
- Shading Units (Cuda Cores): 16896
- Tensor Cores (AI Accelerators): 528
Memory Specifications
- Type: 141GB HBM3e
- Interface: 6144-bit
- Bandwidth: 4.89TB/s
Theoretical Performance
- Pixel Rate: 42.84 GPixel/s
- Texture Rate: 942.5 GTexel/s
- FP16 (half): 120.6 TFLOPS (2:1)
- FP32 (float): 60.32 TFLOPS
- FP64 (double): 30.16 TFLOPS (1:2)
No Display Output
TDP (Thermal Design Power): 600W
PSU Recommendation: 1200W+

Additional information

The PNY NVIDIA H200 NVL 141GB packs 16,896 CUDA cores and 141 GB of HBM3e memory on a single PCIe Gen 5 card. With 4 PFLOPS of FP8 AI compute, 4.89 TB/s memory bandwidth, and MIG support for up to seven isolated instances, it's built for enterprise inference at scale.