Performance Rating
NVIDIA H200 SXM 141 GB outperforms NVIDIA A100X by 157.27% in the overall GPU ARK performance rating
NVIDIA H200 SXM 141 GB
Expert Comparison
NVIDIA H200 SXM 141 GB значительно выигрывает у NVIDIA A100X по количеству ядер (16896 против 6912), объему памяти (141 ГБ против 80 ГБ) и пропускной способности (4.89 TB/s против 2.04 TB/s). H200 также обладает более высокой производительностью в FP32 (66.91 TFLOPS против 19.91 TFLOPS) и более высокой TDP (700 Вт против 300 Вт). Однако A100X доступен сейчас, а H200 ожидается к выпуску только в ноябре 2024 года.
Contents:
Memory ML Performance Compute Power Architecture & Compatibility ML Software Support Clocks & Performance Power Consumption Rendering Benchmarks AdditionalMemory
Memory Size
+76%
141
ГБ
Memory Type
Memory Bandwidth
+140%
4.89 TB/s
Memory Bus Width
ML Performance
FP16 (Half Precision)
+236%
267.6
TFLOPS
BF16 (Brain Float)
TF32 (TensorFloat)
Compute Power
FP32 (Single Precision)
+236%
66.91
TFLOPS
FP64 (Double Precision)
+236%
33.45
TFLOPS
CUDA Cores
+144%
16,896
RT Cores
Architecture & Compatibility
GPU Architecture
SM (Streaming Multiprocessor)
+22%
132
PCIe Version
ML Software Support
CUDA Version
9.0
CUDA Toolkit (first supported)
CUDA Toolkit status
Clocks & Performance
Base Clock
+89%
1,500
Boost Clock
+38%
1,980
Memory Clock
Power Consumption
Recommended PSU
-36%
700 W
Power Connector
TDP/TGP
-57%
300 W
Rendering
Texture Units (TMU)
+22%
528
ROP
L2 Cache
+60%
80 MB
Benchmarks
MLPerf, llama2-70b-99.9 (UNSET)
MLPerf, llama2-70b-99.9 (fp16)
MLPerf, llama2-70b-99.9 (fp8)
MLPerf, llama3.1-405b (fp16)
MLPerf, llama3.1-405b (fp8)
MLPerf, llama3.1-8b (fp8)
MLPerf, deepseek-r1 (fp8)
MLPerf, mixtral-8x7b (fp8)
Additional
Slots
SXM Module
Release Date
Display Outputs
Rent on Intelion Cloud
Available for rent on Intelion Cloud
