Intel's Arc Pro B70 GPU delivers strong performance in DeepSeek R1, beating the RTX 5090D & RTX 4090D in AI LLMs.
Intel Arc Pro B70 32 GB Tested In Quad-GPU Configuration Against The NVIDIA 5090D & RTX 4090D, Offers Superb AI Capabilities In DeepSeek R1 LLM
Graphics card manufacturer Gunnir has posted some interesting tests where they pit the Intel Arc Pro B70 32 GB graphics card against the NVIDIA RTX 5090D 32 GB and the RTX 4090D 24 GB, all in quad-GPU configurations. The purpose of these tests was to evaluate the performance of AI LLMs such as DeepSeek R1 (Distill Qwen 32B FP16). Gunnir wanted to prove that Intel's option isn't just the lowest in cost, but also offers some strong capabilities.
The DeepSeek R1 AI LLM used for the test had a fixed input/output length of 128 tokens, covering the full context concurrency range from 1 to 512. The core metric that was tested was the tokens/sec throughput.
As the tests began, the NVIDIA RTX 5090D immediately dominated the charts with the RTX 4090D in second place. The Intel Arc Pro B70 GPUs were close to the RTX 4090D, but that was only at concurrency levels of under 32. At 32 and 64, the Arc Pro B70 was able to match and exceed the RTX 4090D. As soon as the context currency went past 128, the Arc Pro B70 started beating even the RTX 5090D.
When running at a concurrency level of 128, the Intel Arc Pro B70 GPUs offered a token throughput of 8.6% higher than the RTX 5090D and 34.2% higher than the RTX 4090D. At 256 concurrency levels, the Arc Pro B70 32 GB GPUs were up to 7.5% faster than the 5090D and 48.7% faster than the RTX 4090D. The same performance levels were noticed at concurrency levels of 512.
Intel's Arc Pro B70 GPUs were the strongest of the bunch in terms of token throughput, delivering up to 2320.76 token/s speeds.
This shows some immense capabilities of Intel's Arc Pro B70 GPUs against the competing NVIDIA RTX solutions, which cost more than twice as much. The advantage over the RTX 4090D is simple: the B70 has 32 GB while the 4090D offers 24 GB memory. The other advantage that Intel has is its XMX cores, which specifically boost AI processing capabilities such as FP16 compute.
That is a reason why the Arc Pro B70 is able to beat the 5090D at higher concurrency levels. Now, the RTX 5090D does have NVFP4, which should offer better performance, but for FP16 AI tasks, Intel's Arc Pro GPUs are showcasing very strong performance.
One thing to note is that the standard RTX 5090D with 32 GB of memory is no longer made in China. The focus has now shifted to the newer RTX 5090D V2, which features a cut-down 24 GB memory, so we can expect a similar situation to the RTX 4090D as memory bottlenecks come into play when running larger AI LLMs.
The Arc Pro B70 32 GB is currently listed for $999 US at Newegg, whereas RTX 5090D and RTX 4090D retail for over $4000 and $2000 US, respectively, in Chinese markets. The standard RTX 5090 and RTX 4090 graphics cards also cost similar amounts, which is mainly due to GPU shortages.
About the author: A Software Engineer by training and a PC enthusiast by passion, Hassan Mujtaba serves as Wccftech's Senior Editor for hardware section. With years of experience in the industry, he specializes in deep-dive technical analysis of next-generation CPU and GPU architectures, motherboards, and cooling solutions. His work involves not only breaking news on upcoming technologies but also extensive hands-on reviews and benchmarking.
Follow Wccftech on Google to get more of our news coverage in your feeds.
