How NVIDIA GB10 Performance With the Dell Pro Max GB10 Compares To The GH200

Written by Michael Larabel in Computers on 19 January 2026 at 09:37 AM EST. Page 2 of 3. 4 Comments.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Qwen3-32B-Q8_0, Test: Text Generation 128. GPTshop GH200 was the fastest.

First up was Qwen3 32B with Llama.cpp on these two impressive NVIDIA platforms. For TG128, the GH200 platform was around 10.6x the performance of the GB10. When comparing the exposed GPU power metrics, the GB10 power efficiency is right ahead of the GH200 superchip.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Qwen3-32B-Q8_0, Test: Prompt Processing 512. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Qwen3-32B-Q8_0, Test: Prompt Processing 1024. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Qwen3-32B-Q8_0, Test: Prompt Processing 2048. GPTshop GH200 was the fastest.

For prompt processing with Qwen3 32B the GH200 was around 6x faster.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Llama-3.3-70B-Instruct-Q4_K_M, Test: Text Generation 128. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Llama-3.3-70B-Instruct-Q4_K_M, Test: Prompt Processing 512. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Llama-3.3-70B-Instruct-Q4_K_M, Test: Prompt Processing 2048. GPTshop GH200 was the fastest.

And some reference numbers for Llama 3.3 70B with Llama.cpp on the Dell Pro Max GB10 and GPTshop.ai GH200 platforms. The GB10 GPU performance-per-Watt comes out nicely ahead of the prior-gen GH200.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Text Generation 128. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 2048. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Text Generation 128. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 2048. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: gpt-oss-20b-Q8_0, Test: Prompt Processing 1024. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Text Generation 128. GPTshop GH200 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Prompt Processing 2048. GPTshop GH200 was the fastest.

With smaller models, the GH200 continued to show its still-impressive performance. It will be interesting to compare to GB200/GB300 once having any physical or remote access.

4 Comments - Next Page

Page: 1 2 3 Next Page

URL: https://www.phoronix.com/review/nvidia-gb10-gh200/2

⇱ How NVIDIA GB10 Performance With the Dell Pro Max GB10 Compares To The GH200 Review - Phoronix

How NVIDIA GB10 Performance With the Dell Pro Max GB10 Compares To The GH200